UbiCC Journal Volume 3 No 1

OPTICAL COMMUNICATION K.V.S.S.S.S.SAIRAM, Dr. T. JANARDHANA RAO AND Dr. P.V.D SOMASEKHAR RAO ABSTRACT Networks today are the product of reactive evolution to demand. Circuit-switched networks built for providing POTS (Plain Old Telephone Service) evolved into the current multi-layered networks in reaction to pressures for new services and increased capacity. Growth of data traffic in particular has consistently outstripped the most aggressive projections, forcing service providers to continuously build out their networks using this inefficient and costly network model. Responding to these pressures, the telecommunications industry has built networks, which may well be reaching the end of their viability. Multi-layered networks have become so complex that the cost of continuous development to meet demands is becoming prohibitive. This section presents three key issues that have brought traditional networks architectures to a crossroads: complexity, cost and the need to protect investments. Keywords- Optical Communication, Optical Networking and Optical Switching, POTS etc 1. INTRODUCTION Today's telecommunications service providers are faced with what appears to be an insoluble problem. They must maintain and even increase profitability, while responding to an exponential growth in demand with networks that are even more complex and costly to deploy and maintain. Since the requirements for both profitability and the ability to offer new, revenue-rich services are givens, the solution must be found in the networks Service providers must find a way to build networks with the flexibility of IP, the reliability of SONET/SDH, and the scalability of optics at costs that will allow them to offer services at competitive prices. Photonic Service Switching (PSS) offers exactly that: a new network model that permits graceful evolution to simpler, more cost-effective core networks. This article describes key enablers of the Photonic Service Switching intelligent core network solution. It begins with a very brief overview of some of the reasons layered network architecture has reached a crossroads. It then briefly examines network architectures that have been proposed as evolutionary alternatives to traditional layered network architecture. The final section outlines how a Photonic Service Switching network is managed, focusing on specific challenges and benefits, such as traffic engineering, protection and restoration, and possible new services. This section briefly reviews some approaches, including Photonic Service Switching, proposed to move network design out of the current impasse. (i) MESH TOPOLOGIES The ring versus mesh debate is all but closed. SONET rings are reliable, but they lack the flexibility required by today's market, are costly, and are inefficient. They do not make best use of available bandwidth. 1+1 protection, for example, uses only half the available bandwidth, with the other half on reserve as protection against failures. UbiCC Journal, Volume 3, Number 1, February 2008 1 Mesh topologies, on the other hand, can offer great flexibility and more efficient use of resources. They can optimize bandwidth use of Differentiated Services. There is little doubt, therefore, that SONET rings will be replaced by more flexible topologies. Support for pure meshed environments will soon be the norm. Evolution to mesh topologies alone is not a complete solution, however. Traditional layered networks with mesh topologies are still complex, costly and difficult to scale. A more complete solution requires consolidation of network layers. (ii) CONSOLIDATING LAYERS Two important alternatives have been proposed to the traditional layered network model. The first, IP over optics, separates services from the optical transport but must maintain separate network elements for the various services carried. The second, Photonic Service Switching brings together services and transport through one core network element. (iii) LAYERED NETWORKS either necessary tom run TDM over IP, or to maintain parallel TDM and IP networks. Figure 1 : IP Over OXC/DWDM (v) IP over OXC/DWDM Overlay networks require management of their different layers, such as the IP and ATM layers, as though they were separate networks. Intelligence can be implemented for some layers, but is limited to the specific layers where it is implemented. Layers do not communicate with each other, so management and scalability of the network as a whole are severely compromised. (iv) IP over DWDM IP over DWDM uses IP addressing and routing over DWDM networks. Though it appears attractive, IP over DWDM is not a complete architecture and leaves many key issues unresolved. First, since IP over DWDM uses packet line cards for all traffic, network operators deploying this solution are obliged to pay premium price for all traffic. They cannot benefit from the lower cost of using wavelength cards for transit traffic. Second, IP over DWDM requires separate network management systems for routing and wavelength. Third, service types are limited: IP over DWDM cannot provide TDM, private line type connections for example. Since routers do not support TDM, it is UbiCC Journal, Volume 3, Number 1, February 2008 The IP over OXC/DWDM (in fact IP over OXC over DWDM) approach extends the IP over DWDM model to include wavelength services as shown in the Figure 1. It proposes using routers to aggregate traffic and wavelength switches to handle transit traffic. IP over OXC/DWDM requires additional core network devices to aggregate TDM traffic, and hence still relies on overlay network architecture for service integration. As such, it does not scale well. 2. PHOTONIC SERVICE SWITCHING (PSS) Photonic Service Switching introduced in 2001 goes the final step to enabling an intelligent, peerbased core network solution. Using TOPS (TDM + Optical+ Packets) architecture enabled by G-MPLS (Generalized Multi-Label Switching Protocol), PSS supports the G-MPLS view of a common control plane for all layers of services. With Photonic Service Switching, service providers do not have to bet their capital investments against future demand for specific traffic types. PSS does not favor packet, for example, over other traffic types. It is, in fact, traffic-type neutral. Hence, it not only permits complete transition to a more efficient and cost-effective peer-based network model, but offers investment protection as well. 2 (i) PSS NETWORK (PSSN) A Photonic Service Switching network is a peerbased intelligent optical core network. It applies the power of peering, traditionally found in data networks to transport services. All network elements peer with all other elements. Through an InterGateway Protocol (IGP) they have network information (such as link type, link and bandwidth availability, and route reachability) and use this information dynamically and intelligently. The evolution to Photonic Service Switching networks is made possible by three recent innovations as shown in the Figure 2. G-MPLS, which extends the capabilities of MPLS to TDM and wavelength. TOPS architecture, which consolidates timeslot, wavelength and packet traffic onto one network layer. Photonic Burst Switching technology, which has been used to build a new class of intelligent, optical core switch designed to handle packet, TDM and wavelength traffic simultaneously. Together, these innovations permit deployment of an intelligent, optical core network with a common control plane, and management of all network layers with the Photonics Core Manager. Figure 2: PSS NETWORK effectiveness of a Photonic Service Switching network. Core network elements and links can be managed with a single management system and traffic can be directed through paths that make best use of network resources. (iii) PROVISIONING All layers of a Photonic Service Switching core network can be provisioned through a single management system. The architecture and intelligence of a PSS core network make possible restoration through the core. As well G-MPLS- aware edge devices outside the core can become peers of the PSS core devices for dynamic end-to-end provisioning. If the edge devices are not G-MPLS - sware, PSS can nonetheless use G-MPLS to create tunnels in the PSS network. Alternatively, an OSS can be used in concert with the Photonics Core Manager to provide end-to-end provisioning. Consolidation of layers in a Photonic Service Switching network does not mean that traffic can or should be managed in the same way for all layers. PSS manages each layer according to its unique attributes, and uses the inherent differences of the layers to ensure optimal use of network resources. For example to manage path allocation on the SONET/SDH and optical layers, Photonic Service Switching employs a hierarchy of Label Switched Paths (LSPs), developed within the framework of GMPLS. The hierarchy defines simple rules that prevent, for instance, TDM LSPs from being created within packet LSPs. Within this hierarchy, links at all levels are visible to higher levels. Once they have been created, lower level links are made known to higher layer links through IGP advertisements. This visibility permits aggregation across layers: higherlevel LSPs that have the same source and destination nodes can be optimized and their number reduced. (iv) TRAFFIC MANAGEMENT This section outlines how Photonic Service Switching achieves network layer consolidation and some key benefits of this consolidation, describes management of a PSS network with the Photonics Core Manager, and offers some examples of new revenue generating services made possible by PSS. (ii) SERVICES AND TRANSPORT INTEGRATION Layer integration is key to the efficiency and cost UbiCC Journal, Volume 3, Number 1, February 2008 Photonic Service Switching offers network level Quality of Service (QoS). It offers sophisticated capabilities such as Weighted Random Early Discard (WRED) and Weighted Fair Queue (WFQ). Its QoS mappings permit flexibility in the selection of queuing mechanisms, which span the full range of Differentiated Services from those with guarantees to those without guarantees. Switches in the network can be configured to allow establishment of packet LSPs with or without guaranteed bandwidth, as needed. When an LSP requires guaranteed bandwidth, PSS employs the CAC mechanism that checks and reserves resources. Services 3 with guarantees can be maintained at the same time as services without guarantees, permitting great flexibility in design of SLAs. (v) TRAFFIC ENGINEERING Traffic Engineering (TE) is the task of optimizing use of network resources by provisioning LSPs as dedicated paths between two end points. The path an LSP takes in the network can be set up either though the manually specified routes from a management system. or through routes that are computed based on applying constraints to the topology calculation on a switch. Traffic engineering is necessary for delivery of optimal network performance, and it enhances service provides ability to offer Service Level Agreements (SLA). Effective traffic engineering is therefore one of the keys to maximizing return on investment while improving service offerings. Photonic Service Switching extends traffic engineering from traditional packet traffic to include TDM and wavelength services as shown in the Table 1. APSS consolidated service-transport backbone uses G-MPLS TE and optical extensions for routing and signaling protocols. These protocols provide enhanced network information, intelligent path computation and common signaling to packet. TDM and wavelength services. Integration of these protocol extensions in the G-MPLS framework increases flexibility in network planning enables more intelligent decision and better use of resources. Table 1: G-MPLS Optical Extensions (vi) PROTECTION AND RESTORATION One of the challenges facing network operators moving towards more efficient mesh networks is ensuring efficient and reliable fault detection, and adequate restoration times. For many services, `adequate' means restoration speeds comparable to those supported by today's SONET ring networks, where service is typically restored in about 50 milliseconds. The recovery times measured in tens of seconds typical of IP networks using IGP hellos for fault detection are unacceptable for many critical services. The ability to offer SONET level recovery times when they are necessary (but only when they are necessary) and without having in all cases to invest in 1+1 redundant protection will allow service providers to offer their customers tailor-made, comprehensive SLAs. Photonic Service Switching employs a multilevel protection and restoration strategy that leverages `best of breed' protection from all network layers of a given LSP. This means that, depending on the level of protection required; an LSP can be protected at the SONET line layer (opaque wavelength), the SONET path layer or the MPLS layer. The range of protection available at different network layers makes possible protection and restoration management according to QoS requirements. Service providers can adjust the balance of resources against protection required to achieve optimal use of their network. For instance, a SONET LSP is set up with parameters specifying the required protection. If 1+1 protection is required the LSP is set up over 1+1 protected SONET links. In the event of failure, the entire SONET link is switched to the backup link. Restoration time is less than 50 milliseconds. 1+1 protection consumes significant network resources, however. Other options, which consume less resource, are also available. These include 1:1 shared, and 1:N shared protection and other mesh restoration schemes. Similarly, Photonic Service Switching makes possible LSP repair mechanisms for packet LSPs, with local and global repair working in concert to provide rapid network restoration. 3. SCALABILITY Evolution to a consolidated layer network elements capable of scaling to meet demand while UbiCC Journal, Volume 3, Number 1, February 2008 4 continuing to handle the various traffic types required. As well, a consolidated layer network will eventually require a means for managing the large number of links in a G-MPLS, meshed network. (i) OPTICAL SCABLABILITY Photonic Service Switching uses Photonic Burst Switching (PBS) technology, Switches built with this patented technology can scale in three dimensions: space, wavelength and bit-rate as shown in the Fig.3. Figure 3: Photonic Burst Switching uses G-MPLS to bundle multiple links between elements. As well as preserving IP interfaces throughout the network, link bundling reduces control traffic by replacing individual channels for each link with a single IP control channel for the bundle. 4. NETWORK MANAGEMENT One of the most important benefits of the evolution to a single control plane for core networks is a simplification in network management -- an improvement with immediate positive implications for an operator's bottom line. (i) COMMON CONNECTION MANAGEMENT G-MPLS permits the inter-operability of IP routers, legacy SONET/SDH and TDM devices (ADM, BDCS) and optical devices (OXC, OADM, DWDM). G-MPLS provides the flexibility of IP for route advertisement, administration and link discovery while maintaining circuit-like LSPs for traffic across all these devices. Service connection with Photonic Service Switching is greatly simplified. The network operator has one interface for requesting connections at any layer of the network. He need only specify bandwidth, delay and jitter, and reliability required. The network has the intelligence to use lower layer LSPs to provide the requested bandwidth and return notification of successful completion or failure of the request. (ii) A CONSOLIDATED NETWORK MANAGEMENT SYSTEM Though next generation core networks will consolidate layers, service providers may continue to operate along traditional functional partitions. For example, a transport team may manage SONET/SDH, TDM and optical devices, while an IP team manages the core and aggregation routers. The challenge is twofold. Service providers will want to make provisioning of these consolidated networks seamless, while allowing network operators to maintain the expertise and efficiency of traditional teams managing specific network layers. They will also want to consider developing new teams to take advantage of PSS's network layer consolidation, and single system element and network management. CONCLUSION The layered network architecture under 5 The PBS n x n design means fabric can be expanded spatially, with ports added when they are required. Everything between the ingress and egress line cards is optical, and wavelength and bit-rate independent, so switching capacity is virtually unlimited. Thus, PBS switches offer throughput salability to petabit capacity while maintaining the 99.999 percent availability required of core grade, carrier-class network equipment. (ii) LINK BUILDING Photonic Service Switching greatly simplifies network management because G-MPLS extends the use of proven IP protocols for unique identification of elements, signaling, and routing and link to all network layers. However, the combination IP, SONET and wavelength means the number of links in a PSS network is higher than in a typical IP network. The solution to this increase in the number of link bundling. To reduce the number of IP addresses managed in the network, Photonic Service Switching UbiCC Journal, Volume 3, Number 1, February 2008 development since the 1970s to meet demands for new traffic types has come to a crisis. The complexity, cost and inherent limitations of layered networks are such that service providers must begin moving towards simpler, more efficient architectures. Intelligent peer-based networks in which network elements at all layers have full information about all other network elements are the most viable option for future network architecture. Photonic Service Switching brings together the best of many worlds: IP flexibility, SONET/SDB reliability, and optical scalability. Because it enables evolution to a simpler, more efficient network architecture that includes packet, TDM and wavelength traffic, Photonic Service Switching offers a truly evolutionary solution, and true investment protection. REFERENCES 1.Max Ming - Kang Liu, "Principles and Application of Optical Communications". 2.K. Sato "Introduction strategy of Photonic Network Technologies to create bandwidth abundant multimedia networks". Trends in Optics & Photonics; Vol.20, June 1998. 3.G.P. Agarwal, Nonlinear fibre optics, Boston; Academic 1989. 4. H. Masudaetal., "Wideband and low noise optical amplification using distributed Raman amplifiers and EDFA". Proc. Ecoc 98, Paper MO A12, 1998. 5. I. Imoaoka and M. Teshima, "Optical frequency reference in optical path networks based on WD techniques". ICICE tech.rep. OCS96-65, Nov.1996, pp. 37-43. AUTHORS BIOGRAPHY K.V.S.S.S.S.SAIRAM (s5kanduri@rediffmail.com) is working as Senior Associate Professor, ECE Department, Bharat Institute of Engineering & Technology, Mangalpally, Ibrahimpatnam, Hyderabad, Andhra Pradesh State, INDIA. He was previously worked as Lecturer and Assistant Professor in Dr. M.G.R. Deemed University, Chennai. He is pursuing his Ph.D (Optical Communications) under the guidance of Dr. P.V.D Somasekhar Rao and Dr. T. Janardhana Rao, UGCASC Director, J.N.T.University, Kukatpally, Hyderabad – 72 & Professor &HOD of the ECE Department, Sridevi Women’s Engineering College, V.N.Pally, Gandipet, Hyderabad- 75. He got his Bachelors Degree in ECE from Karnataka University, Dharwad in 1996 and Masters Degree from Mysore University, Mysore in 1998.His research interests are Optical Communication, Networking, Switching and Routing and Wireless Communication. He was published 30 PAPERS in IEEE Communication Magazine, IEEE Potentials, International and National Conferences. He is an IEEE REVIEWER and EDITORIAL MEMBER for Optical Society of America, Journal on Photonics and IEEE Journal on Quantum Electronics and IASTED. Dr. P.V.D. Somasekhar Rao B.E. (SVU), M.Tech.(IIT, Kharagpur), Ph.D. (IIT, Kharagpur. Professor and Head of the Department & UGC-ASC Director Specialized in Microwave and Radar Engineering. His research interests include Analysis and design of Microwave circuits, Antennas, Electro Magnetics, Numerical Techniques. He published 20 research papers in National and international Journals and Conferences. He is presently guiding two Ph.D. students. He prepared the source material for School of Continuing and Distance Education, JNTU, in the subjects such as computer programming & Numerical Techniques, Radar Engineering, Antennas and Propagation and Microwave Engineering. He has more than 20 years of teaching and research experience, which include R&D works at Radar Centre, IIT Kharagpur and at Radio Astronomy centre and TIFR. He is a Senior Member of IEEE, Fellow of IETE. He delivered a number of invited lectures. He is a reviewer for the Indian Journal of Radio & Space Physics from 1991. He is the recipient of the IEEE -USA outstanding Branch Counselor/Advisor award for the year 1993-94. He had completed a number of projects aided by AICTE. He has been a visiting faculty at Assumption University, Bangkok, during 1997-99. Dr. T. Janardhana Rao is working as Professor and Head of the Department in Sridevi women’s Engg College, V.N.Pally, Gandipet, Hyderabad, and Andhra Pradesh State, INDIA. His research interests include Optical Networks, Digital Electronics, BioMedical Engg.,&Power Electronics. He published 15 International and National Journal Conferences. Professor Rao was a former a member of faculty of S.V.University with a teaching experience about 45 years. He is a life member of ISI and ISTE. UbiCC Journal, Volume 3, Number 1, February 2008 6 Modeling the Effect of Clipping and Power Amplifier Non-Linearities on OFDM Systems Ashraf A. Eltholth*, Adel R. Mekhail*, A. Elshirbini*, M. I. Dessouki† and A. I. Abdelfattah †† * National Telecommunication Institute, Cairo, Egypt † Faculty of electronic Engineering, Menouf, Egypt †† Faculty of Engineering, Mansoura, Egypt ABSTRACT The high Peak to Average power Ration (PAR) levels of Orthogonal Frequency Domain Multiplexing (OFDM) signals forces to the utilization of linear power amplifiers. However, linear amplifiers have low power efficiency which is problematic considering that battery life is a critical resource in mobile systems; also linear amplifiers have a clipping effect when considered as a limiter. Amplifier nonlinearities affect drastically the performance of OFDM systems. In this paper, the effects of Clipping and amplifier nonlinearities are modeled in an OFDM system. We showed that the distortion due to these effects is highly related to the dynamic range it self rather than the clipping level or the saturation level of the nonlinear amplifier. Computer simulations of the OFDM system using Matlab are completely matched with the deduced model in terms of OFDM signal quality metrics such as BER, ACPR and EVM. 1. INTRODUCTION In OFDM systems, the combination of different signals with different phase and frequency give a large dynamic range that is used to be characterized by a high PAR, which results in severe clipping effects and nonlinear distortion if the composite time signal is amplified by a power amplifier, which have nonlinear transfer function. This degrades the performance of an OFDM system. A measure of the degradation can be very helpful in evaluating the performance of a given system, and in designing a signaling set that avoids degradation. The high PAR sets strict requirements for the linearity of the PA. In order to limit the adjacent channel leakage, it is desirable for the PA to operate in its linear region. High linearity requirement for the PA leads to low power efficiency and therefore to high power consumption. PAs are divided into classes according to the biasing used. A class A amplifier is defined as an amplifier that is biased so that the current drawn from the battery is equal to the maximum output current. The class A amplifier is the most linear of all amplifier types, but the maximum efficiency of the amplifier is limited to 50%. In reality, due to the fact that the amplitude of the input signal is most of the time much less than its maximum value, the efficiency is much less than the theoretical maximum, i.e. only a few percent. This poor efficiency causes high power consumption, which leads to warming in physical devices. This is a problem especially in a base station where the transmitted power is usually high. To achieve a better efficiency, the amplifier can be biased so that current flows only half the time on either the positive or negative half cycle of the input signal. An amplifier biased like this is called a class B amplifier. The cost of the increased efficiency is worse linearity than in a class A amplifier. High demands on linearity make class B unsuitable for a system with high PAR. On the other hand, the large scale of the input signal makes it difficult to bias an amplifier operating in class A. In practice, the amplifier is a compromise between classes A and B, and is called a class AB amplifier. [1] Several options appear in the literature related with OFDM systems and nonlinearities. PAR reduction using clipping or coding or phase optimization techniques or a combination of any two of them [2], are the tools to combat nonlinearities used in the transmitter. Also a good work have been done in modeling the performance of OFDM systems with power amplifiers in [1], but it have assumed the OFDM signal to have Gaussian distribution which is not very accurate description of the composite time OFDM signal. In this paper, the effect of power amplifier nonlinearities is modeled in OFDM systems. Section 2 depicts OFDM signal statistical properties while Section 3 introduces power amplifier models. The derivation of clipping noise and distortion is presented in Section 4, OFDM signal quality metrics are discussed in Section 5. Simulation results are included in Section 5. Finally, the conclusions are drawn. UbiCC Journal, Volume 3, Number 1, February 2008 7 2. OFDM SIGNAL STATISTICAL PROPERTIES It is well known that according to the central limit theory that, the real and imaginary parts of the OFDM signal completely agree with the normal distribution and consequently its absolute agrees with the Rayleigh distribution with Probability density function expressed by: 3. POWER AMPLIFIER MODEL A short description of power amplifier models will be given in this section. Consider an input signal in polar coordinates as [1] x = ρ e jϕ The output of the power amplifier can be written as x − 2 P ( x) = e 2 s , x ∈ [0, ∞] Where s is a s parameter, and Mean µ = s π 2 and variance 4 −π 2 s σ2 = 2 Figure (1) listed below explicitly shows that the measured amplitude histogram of the (a) in-phase component/Quadrature component and (b) amplitude of a 256-subcarrier OFDM signal. x2 g (x) = M (ρ ) e j ( ϕ + P ( ρ )) Where M( ρ ) represents the AM/AM conversion and P( ρ ) the AM/PM conversion characteristics of the power amplifier. Several models have been developed for nonlinear power amplifiers, the most commonly used ones are as follows 3.1. Limiter Transfer characteristics A Limiter (clipping) amplifier is expressed as [4]: 0.012 ⎧ρ M (ρ ) = ⎨ ⎩A ρ a σ SSPA And thus the distortion due to the NLA as a limiter can be represented by an extra Gaussian noise with variance where ∞ ∞ x2 2s2 a 2 ⎛ ⎞ ⎜ − 2a 2 s 2 + 4s 4 + a 4 e 2 s 2 Ei ⎡− a ⎤ − ⎟ ⎢ 2 ⎥ ⎟ ⎜ ⎣ 2s ⎦ ⎜ ⎟ a2 − 2 1 ⎜ ⎡ a ⎤ ⎟ = ⎜ 2 a 2 e 2 s 2π s (a 2 − s 2 )erf ⎢ ⎥ +⎟ 2s ⎜ ⎣ 2s ⎦ ⎟ ⎟ ⎜ a2 ⎜ 2 a 2 e − 2 s 2 2π s (a 2 − s 2 ) ⎟ ⎟ ⎜ ⎝ ⎠ Where Ei(x) is the Exponential integral, is defined as σ lim iter = ∫ ( x − a) 2 P( x)dx = ∫ ( x − a) 2 a a x e s − dx Ei ( x) = et ∫∞ t dt − x It is plotted as in figure (3) UbiCC Journal, Volume 3, Number 1, February 2008 9 Ei x 6 4 2 x -3 -2 -1 -2 -4 -6 1 2 3 depicted in figure (4-b), it is shown that the distortion decays as the value of saturation level increases. And when plotting distortion model versus the saturation level with parameter (s=0.8) figure (4-c) shows a great increase in the distortion despite of the constant value of (A/s). Figure (3) Exponential integral a ⎞ ⎛ − ⎜ 2ae 2 s 2 2π s (a 2 − s 2 )(1 − erf ⎡ a ⎤ ) ⎟ ⎢ ⎥ ⎟ 1 ⎜ ⎣ 2s ⎦ = ⎜ ⎟ 2 a 2s ⎜ ⎟ ⎡ a2 ⎤ 2 4 2 2 4 ⎟ ⎜ + a e 2 s Ei ⎢− 2 ⎥ − 2a s + 4s ⎟ ⎜ ⎣ 2s ⎦ ⎠ ⎝ 2 ∴ σ SSPA σd i s t 0.012 (b) With s= 0.08 0.01 0.008 SSPA And finally 0.006 0.004 0.002 2 2 σ SSPA ⎛ −a2 ⎞ ⎜ ae 2 s 2π (a 2 − s 2 )erfc ⎡ a ⎤ + ⎟ ⎢ ⎥ ⎜ ⎣ 2s ⎦ ⎟ =⎜ ⎟ (5) 2 ⎜ a 4 2as 2 ⎡ a 2 ⎤ ⎟ 2 3 ⎜ e Ei ⎢− 2 ⎥ − a s + 2s ⎟ ⎜ 2s ⎟ ⎣ 2s ⎦ ⎝ ⎠ Clipping A 4 6 8 (c)With s=0. 8 and finally when plotting distortion model versus (A/s) as shown in figure(4-d) that shows that the distortion is reduced as the value of (A/s) increases. 0.01 0.008 Clipping SSPA When plotting the deduced distortion models in eq. (4, 5) versus the distribution parameter (s) with saturation level a= 2 (AIP3 = 10 dB) we notice as shown in figure (4-a) below that the distortion due to the SSPA nonlinearity is much more larger than that of its limiting effect, also it is obvious that the distortion is highley senstive to any variation of the parameter (s) as the slopes of the curves show. 0.006 σ 0.004 0.002 0 1 1.25 1.5 A/s 1.75 2 (d) Distortion Vs. A/s Figure (4) Non-linear Amplifier Distortion From (4-a,b,c,and d) it is clear that the distortion due to these effects is highly related to (s) the distribution parameter, that controls the dynamic range it self, rather than the clipping level or the saturation level of the nonlinear amplifier. (a) With saturation level a= 2 (AIP3 = 10 dB) On the other hand, when plotting distortion model versus the saturation level with parameter (s=0.08) as UbiCC Journal, Volume 3, Number 1, February 2008 10 5. OFDM THE SIGNAL QUALITY METRICS 5.1. Error Vector Magnitude The modulation accuracy of the OFDM signal is measured by Error Vector Magnitude. EVM is a measure for the difference between the theoretical wave and modified version of the measured waveform. The measured waveform is modified by first passing it through a specified receiver measuring filter. The waveform is further modified by selecting the frequency, absolute phase, absolute amplitude and clock timing so as to minimize the error vector. The EVM result is defined as the square root of the ratio of the mean error vector power to the mean reference signal power expressed as a percentage. Mathematically, the error vector e can be written as e = y-x; Where y is the modified measured signal and x the ideal transmitted signal. EVM can be defined as The ACPR can be defined as: f3 ACPR = f2 f2 ∫ S ( f )df ∫ S ( f )df f4 + f1 ∫ S ( f )df ∫ S ( f )df f2 f1 f1 Where f1 and f2 are the frequency limits of the main channel, and f2 and f3 are the frequency limits of the upper adjacent channel, and f1 and f4 are the frequency limits of the lower adjacent channel. 6. SIMULATION RESULTS An OFDM system is implemented using 512 carriers with cyclic prefix length equal to 4. Each carrier is modulated using 16-QAM constellation. AWGN noise is included. BER simulations compared with theoretical results considering the power amplifier distortion models deduced above in equations (4, 5) are shown in Figure 6. From this figure, it is possible to see that it was predicted in the previous analysis. The effect of nonlinear power amplifier is illustrated in figure 6, where a limiter amplifier is included in the simulations with clipping levels of 12 dB. The harmful effect of the nonlinearity can also be clearly seemed in this figure. And finally the figure shows that the computer simulations of BER are completely matched with the deduced models both the limiting and the nonlinearity effect. 0 EVM rms = E[ e ] E[ x ] 2 2 5.2. The Adjacent Channel leakage Power Ratio (ACPR) Another figure of merit, specific to evaluate the out of band behavior of the HPA, is the ACPR; it should stay below the value specified. The ACPR is the ratio of the transmitted power to the power after a receiver filter in the adjacent channel. In order to evaluate the ability of HPA models to reproduce the ACPR, we will use: Comparison of Simulation results and Distortion Models with the same Clip level = 12 dB 10 ∆ACPR = ACPR( S ( f )) − ACPR( S ( f )) ~ ~ Where S ( f ) is the PSD of the output of HPA S ( f ) is the true output. OFDM Signal Spectrum 10 5 0 -5 Magnitude (dB) -10 -15 -20 adjacent channel -25 -30 -35 -40 -0.5 adjacent channel main channel BER 10 -1 10 -2 Without NLA with NLA after clipping clipping distortion model SSPA distortion model 10 -3 f2 f1 2 4 6 8 10 12 Eb/No (dB) 14 16 18 20 Figure (6) BER of OFDM system with HPA f3 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 Normalized Frequency (0.5 = fs/2) 0.3 f3 0.4 0.5 Figure (5) OFDM signal spectrum Table (1) shows The ACPR value for both limiting and non-linear effects with different limiting values relative to the maximum absolute value of the OFDM composite time signal Ymax, it is clear that as the limiting value UbiCC Journal, Volume 3, Number 1, February 2008 11 decreases the ACPR increases. It can also be noted that the effect of nonlinearity on ACPR value is negligible as compared to that of limiting as the clip level varies; this is due to the fact that the spectral leakage that causes the ACPR to increase is mainly due to the clipping that can be viewed as windowing the spectrum by rectangular window. Table (1) ACPR value with different limiting values The EVM is a measure of the total distortion it is highly affected by the nonlinearity rather than the limiting effect. And generally as the limiting value decreases the EVM increases. References 1- F. H. Gregorio and T. I. Laakso, “THE PERFORMANCE OF OFDM-SDMA SYSTEMS WITH POWER AMPLIFIER NON-LINEARITIES”, Proceedings of the 2005 Finnish Signal Processing Symposium - FINSIG'05 August 25, 2005 Kuopio, Finland 2-Curt Schurgers Mani B. Srivastava,”A Systematic Approach to Peak-to-Average Power Ratio in OFDM”, Proc. SPIE Vol. 4474, p. 454-464, Advanced Signal Processing Algorithms, Architectures, and Implementations XI, Franklin T. Luk; Ed, Nov. 2001. 3- Sergey V. Zhidkov, “Performance Analysis of Multicarrier Systems in the Presence of Smooth Nonlinearity”, EURASIP Journal on Wireless Communications and Networking 2004:2, 335–343, ©2004 Hindawi Publishing Corporation. 4- Le LIU, Kiyoshi HAMAGUCHI and Hiromitsu WAKANA, “Analysis of the Combined Effects of Nonlinear Distortion and Phase Noise on OFDM Systems”, IEICE TRANS. COMMUN., VOL.E88–B, NO.1 JANUARY 2005 Saturation Level Ymax 2 Ymax 10 Ymax 0.75 Ymax 0.5 Ymax 0.25 Ymax Limiting (Clipping) -44.7094 dB -44.7094 dB -44.7094 dB -30.4841 dB -23.4736 dB -14.5232 dB Nonlinearity -28.7268 dB -31.0497 dB -32.8791 dB -27.2594 dB -25.0624 dB -22.3549 dB Table (2) shows The EVM value for both limiting and non-linear effects with different limiting values, it is clear that as the limiting value decreases the EVM increases. And as the EVM is a measure of the total distortion it is highly affected by the nonlinearity rather than the limiting effect. Table (2) EVM value with different limiting values Saturation Level Ymax 2 Ymax 10 Ymax 0.75 Ymax 0.5 Ymax 0.25 Ymax Limiting Nonlinearity (Clipping) 1.3110e-016 0.0929 1.3110e-016 0.0273 1.3110e-016 0.0012 0.0126 0.1448 0.0969 0.2474 0.3972 0.4741 7. CONCLUSIONS In this paper, the effects of nonlinearities in the power amplifier over OFDM systems were analyzed and simulated. We can conclude that: The distortion due to high power amplifier, either limiting or nonlinearity effects, is highly related to the distribution parameter (s), that controls the dynamic range it self, rather than the clipping level or the saturation level of the amplifier. Also it is noticed that the effect of nonlinearity on ACPR value is negligible as compared to that of limiting as the clip level varies; this is due to the fact that the spectral leakage that causes the ACPR to increase is mainly due to the clipping that can be viewed as windowing the spectrum by rectangular window. UbiCC Journal, Volume 3, Number 1, February 2008 12 Experimental Study on Time and Space Sharing on the PowerXplorer Sulieman Bani-Ahmad CWRU, Ohio, USA. sulieman@case.edu Ismail Ababneh AABU, Mafraq, Jordan Ismail@aabu.edu.jo ABSTRACT Scheduling algorithms in parallel computers fall into two basic categories: time and space sharing algorithms. Space-sharing based processor allocation algorithms can be contiguous or non-contiguous. Studies show that non-contiguous allocation is superior due to decrease in fragmentation. Other studies have reported that executing jobs on fewer processors (folding) can improve the performance of contiguous and non-contiguous allocation. However, the problem with folding is that it is not always applicable because of parallel programming languages and parallel operating systems limitations. Most of previous studies used simulation. Our study is an experimental one for studying time and space sharing on a real parallel machine (the PowerXplorer), with eight processors arranged as a two-dimensional mesh. A set of five scientific applications with differing communication characteristics were implemented and executed using time and space sharing. The observed execution times were used to study and compare time-sharing and contiguous and non-contiguous space sharing with and without folding. Our study showed that time-sharing gave comparable results to space sharing allocation. Further, non-contiguous allocation gave better results than contiguous allocation when folding is not supported. However, when folding is supported contiguous allocation gave the best mean turnaround times. Keywords: Scheduling algorithms, parallel programmin, parallel computing. 1 INTRODUCTION A job in multiprogrammed parallel systems is characterized along two dimensions; the length, measured by the execution time and the width or size measured by the number of threads such that each thread is executed on a separate processor. Thus, resource sharing in parallel computers takes two levels, (i) time sharing, whereby a thread can be interrupted during execution by other threads of jobs running on the same processor, (ii) space sharing, where a thread has exclusive use of its processor until its execution is complete [11]. Space sharing can be contiguous or non-contiguous depending on whether the processors allocated to a single job are physically adjacent or not [13]. Comparison studies show that non-contiguous allocation is superior to contiguous allocation as the former produces less external fragmentation [13]. However, contiguous allocation is preferred as it provides maximum communication speed between threads of the same job [13]. Folding, that is executing jobs on fewer processors, can improve performance of contiguous and noncontiguous allocation [12]. The problem with folding is that it is not always applicable due to limitations of parallel programming languages and parallel operating systems [12]. Most previous studies used simulation to study and compare different scheduling processor allocation strategies. This study, however, is an experimental one for studying time and space sharing on a real parallel machine (the PowerXplorer) with eight processors arranged in a two dimensional mesh and uses PARIX operating system and parallel programming language. This study passed by the following stages: (i) We selected a set of five scientific applications to implement and later use to comparatively assist different allocation strategies. Two of these applications, namely Matrix multiplication (MM) and LU factorization, are used in well-known parallel benchmarks like NAS [3] and GENESIS [2]. The other applications are (a) 2D Fast Fourier Transform (2D FFT), which is used in image processing, (b) Floyd shortest path algorithm from graph theory, and (c) a simulation of electromagnetic wave propagation in 2D space using Finite-difference in time domain (FDTD) from Electromagnetics. (ii) We implemented the selected applications in PARIX environment. The applications were run on the PowerXplorer using time sharing (TS) and both contiguous and non-contiguous space sharing (CSS, NCSS). (iii) The execution times are used as inputs to a simulator to study and comparatively evaluate contiguous and non-contiguous space sharing (a) with and without folding, and (b) with bounded UbiCC Journal, Volume 3, Number 1, February 2008 13 folding, which limits folding factor. The folding factor for a particular job j that requests nj is defined as the ratio between the number actually allocated to j and nj. The main contribution of this paper is that it comparatively assists time sharing and different processor strategies on a real parallel computer, namely the PowerXplorer. Our major findings are: • TS gave close average execution times to CSS and sometimes performed better with large allocation sizes, i.e. when relatively large allocation sizes. Further, literature shows that TS (i) reduces the job’s wait time (before allocation), (ii) dampens the effect of idle state of processors that occurs when a thread halts waiting for communication or I/O operations. Therefore, TS can be useful to interactive parallel computer systems that require low response times to user commands. • Simulation shows that non-contiguous allocation acheives better results than contiguous allocation when folding is not supported. However, when folding is supported, contiguous allocation gives the best mean turnaround times. The remainder of this paper is organized as follows: Section 2 presents the PowerXplorer and PARIX operating system and programming language. The studied processor allocation strategies are briefly described in section 3. Section 4 presents the implementation details of the five scientific applications used in the experimental part of this study. The simulation details are described in section 5. In section 6, we present our main experimental results and observations. Section 7 provides our conclusions. 2 The PowerXplorer and PARIX The PowerXPlorer is a family of distributed memory systems from Parsytec (http://www.parsytec.com/). This system runs the PARIX operating system and is based on 8 processing units arranged in a 2D mesh, as illustrated in figure 1. Each processing unit has (i) one 80-MHz PowerPC 601 processor, (ii) 8 MB of local memory and (iii) a transputer for establishing and maintaining Figure 1: The 2D mesh of the PowerXplorer with communication links processor IDs. asynchronous communication [7]. 3. Studied Scheduling Strategies Besides time sharing, we study the performance of the following processors allocation strategies of space sharing. (i) Space sharing without folding: We test two strategies under this category, namely, Contiguous Space Sharing (CSS) and Non-Contiguous Space Sharing (NCSS). In general, a CSS strategy starts by allocating the exact requested number of processors. If it fails, the requesting job waits until enough idle processors become available. In NCSS however, if allocating a contiguous set fails, the algorithm looks for a non-contiguous set. If it fails again, the requesting job waits until the needed number of processors becomes available. (ii) Space sharing with folding: Folding is allocating a fewer number of processors than requested in case not enough idle processors are found. We test four strategies under this category: (a) Contiguous space sharing with unbounded folding (CSSUF), which puts no limits on the folding factor. We define the folding factor as the ratio between the requested number of processors and the actual nember allocated after folding. This strategy puts no limits on the folding factor, i.e. it may fold any job requesting ni processors to one processor with folding factor 1/ni. (b) Non-contiguous space sharing with unbounded folding, which differs from the previous one in that it relaxes the contiguity condition and tries to allocate a non-contiguous set of processors if a contiguous set is not available before folding. (c) Contiguous/non-contiguous space sharing with bounded folding (CSSBF, NCSSBF), which are similar to the previous two except in that they allow maximum folding factors of k/ni where ni is the requested number of processors and k is the minimum number of processors that can be allocated to any job. We chose k=2 as we have relatively small mesh (8 processors). For example, we allowed folding factors of 1/4, 1/3 and ½ and 1 on the requests for 8, 6, 4 and 2 processors respectively. PARIX (PARallel UnIX extensions) [7] is the native operating system in the Parsytec PowerXplorer family. It provides UNIX functionality at the front-end with library extensions for the needs of the parallel system. The Parix software package comprises components for the program development environment (compilers, tools, etc.) and runtime environment (libraries). PARIX offers different types of synchronous and 4. Implementation Details Selected Applications of the To facilitate parallelizing the five scientific applications, we chose the size of the input to be dividable by 1, 2, 4, 6 and 8. Further, at run time, one of the instantiated threads of the application, arbitrarily selected, takes divides the data input UbiCC Journal, Volume 3, Number 1, February 2008 14 between all other threads. The same thread is responsible for gathering the final results. We refer to this thread as the main thread. Figure 3 shows the main loops of Matrix Multiplication. Ma and Mb are the input matrices to be multiplied and Mc is the result matrix. The matrices are of size 384*384 each. The algorithm is parallelized by dividing the rows of Ma and the columns of Mb into n slices each, where n is the number of processors assigned to the application (1, 2, 4, 6 or 8). A total of n threads are instantiated at each processor. The main thread distributes the slices such that each thread initially receives a slice of Ma and another of Mb (One-to-All communication). The threads, in turn, start to compute the Mc slice assigned to them. During computation, All-to-All communication occurs between the threads as follows; thread of ID j (i) passes the Mb slice that it has to its neighbor with ID j+1 and (ii) receives its j-1’s Mb slice and (iii) computes the corresponding portion of the Mc slice. The three steps mentioned above, (i), (ii) and (iii) are repeated n-1 times. Finally, the main thread collects back Mc slices (All-to-One communication). As it is quicker than asynchronous communication, we use Synchronous Communication in all applications, i.e. the two communicating threads should be at both sides of the communication channel simultaneously. To prevent deadlock due to communication, the threads with odd IDs start sending first and simultaneously the threads with even IDs receive. Next, the roles are reversed. Figure 2 shows the main loops of Floyd’s shortest path algorithm. |V| is the number of vertices in the input graph G(V,E), where V and E are the vertex and edge lists. We chose |V|=384. M initially is the edge-weight matrix of G. After Floyd’s algorithm is applied on G, M becomes the shortest matrix. At each iteration of loop k (figure 2), the distances between the vertices scanned by loops i and j, M[i][j], are updated as follows: if the distance between vertices i and j passing by k is shorter than the current distance between i and j, then the entry M[i][j] is updated to the new shorter distance. for(k=0;k<|V|;k++) for(i=0;i<|V|;i++) for(j=0;j<|V|;j++) {if(M[i][j] > (M[i][k]+M[k][j])) M[i][j] = M[i][k]+M[k][j];} Figure 2: The main loops of Floyd’s algorithm for(i=0;i Pmax ). Calculate the allowed rate in the system for each request from (4). If there are rates lower than Rmin , reject the request with minimal allowed rate (to achieve fairness) then repeat again till all allowed rates be greater than Rmin . Update all active links in the network to see if they will be disturbed by the new requests or not. If any one be -ve IM, remove the maximum interfering request from the minimal IM to achieve fairness. Then update the IM again and repeat till no -ve IM in the links. Calculate the IM for the residual requests with their calculated power and rate which will be considered as the maximum rate for that request 22 3 and then apply the same procedure as if there are no links available in the network. From pervious procedures, fairness between the contended requests is achieved by WALAC. P0 = min Rallow = IMi Tf σ 2 g0i where 1 ≤ i ≤ N P 0 gi0jo γ ηi + Tf σ 2 N k=1,k=i (3) (4) Pk gkj0 IV. SIMULATION RESULTS AND DISCUSSIONS In this section, we study the behavior of the centralized protocols through simulations. The simulation area is taken as 50m×50m with nodes randomly distributed. The data traffic flows are generated based on a Poisson process with λ call/sec per user. The default parameters used are shown in TABLE I. TABLE I S IMULATION PARAMETERS Parameter Tf σ2 η Pmax λ α γ superframe duration no. of slots in superframe (including Beacon) packet length permission buffer rate minimum rate (Rmin ) maximum rate (Rmax ) Value 10 ns 1.99×10−3 2.56×10−17 7 dBm 30 4 6 dB 20 msec 5 1024 bytes 0.3 10 Mb/s 2 Mb/s 100 Mb/s incoming requests. It is a parameter for the proposed protocol. Fig. 3 and 4 show the system normalized throughput for the data and streaming data traffic respectively. The two curves indicate that the proposed protocol is better than both Slotted Aloha and PRMA systems in both types of traffic. In low traffic or low number of data terminals, Slotted Aloha has better throughput than PRMA. That is because the reserved slots in PRMA decreases the available number of channels for the contented users. While in high traffic or great number of data terminals, PRMA and Slotted Aloha nearly shows the same performance. That is because the collisions for both protocols are assumed to be equal and the reserved channels in PRMA loses their advantage. On the other hand, the proposed protocol shows better throughput than others in both low and high traffic as there are no collisions. For data traffic, Slotted Aloha system can serve 24 data users while PRMA works up to 20 users. The proposed protocol can attend more than 60 data users. For streaming data traffic, number of users is retrograded to only five users for Slotted Aloha and to 10 users for PRMA while they are still above 60 ones for the proposed protocol but with lower throughput. The PRMA enhancement over Slotted Aloha is because the reserved channels in PRMA serve some streaming nature users where this is not found in Slotted Aloha which suffers with a lot of collisions. 1 0.9 0.8 normalized throughput 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 0 10 20 30 40 number of data terminals 50 60 PRMA slotted aloha system throughput proposed protocol Comparison between Slotted Aloha, PRMA, and the proposed protocol in a high rate UWB network is done. Slotted Aloha system is based on the channel organized into uniform slots whose size equals to the packet transmission time. If a station needs to send a packet, it sends in the first next slot. Aloha based channel access was proposed in [14], but there was the contention problem. PRMA system is similar to Slotted Aloha system except that the users can reserve the slots till finishing their transmissions if they success for contention it. To access the channel, it must first have permission over the threshold value [15]. The protocols are compared, for both data traffic and streaming data traffic, according to the following parameters which are considered a measure of the system performance and the protocol effectiveness. • System throughput: the ratio between the successfully transmitted packets and all transmitted packets. • System utilization: the ratio between the successfully transmitted bits averaged over the time. • System loss probability: the ratio between the rejected transmitted packets and all transmitted packets. • System average delay: the average delay per successfully packets. • Admission ratio: admission ratio parameter is defined as the ratio between the admitted requests and the all UbiCC Journal, Volume 3, Number 1, February 2008 Fig. 3. System throughput for data traffic 1 0.9 0.8 normalized throughput 0.7 0.6 system throughput proposed protocol prma saloha proposed protocol 0.5 0.4 0.3 0.2 0.1 0 0 10 20 30 40 number of data terminals 50 60 slotted aloha PRMA Fig. 4. System throughput for streaming data 23 4 The utilization for the three systems is compared for both data and streaming data traffic in Fig. 5 and 6 respectively. The proposed protocol shows better channel utilization than others. From the curves, Slotted Aloha is 16% better than PRMA in low traffic while PRMA is better by 50% in streaming data. For the proposed protocol, it gives better performance than Slotted Aloha more than 60% in low traffic and better than PRMA in streaming data more than 83%. This better performance comes from the use of WALAC which allows the user to share the channel with others without degradation. 10 0 system loss probability slotted aloha 10 loss probability −1 proposed protocol PRMA 10 −2 10 −3 10 10 10 10 10 10 10 10 10 8 −4 0 10 system utilization 20 30 40 number of data terminals 50 60 7 Fig. 7. slotted aloha PRMA proposed protocol System loss probability. 6 system utilization (b/s) 5 4 3 2 1 0 10 20 30 40 number of data terminals 50 60 10 0 system loss probability Fig. 5. The system utilization of the three protocols 10 loss probability −1 proposed protocol PRMA 10 −2 10 10 10 10 10 10 10 10 9 system utilization slotted aloha 8 7 slotted aloha PRMA proposed protocol system utilization (b/s) 10 −3 0 10 6 20 30 40 number of data terminals 50 60 5 Fig. 8. System loss probability under streaming data 4 3 2 0 10 20 30 40 number of data terminals 50 60 Fig. 6. System utilization under streaming data. Fig. 7 shows the loss probability of the compared protocols for data traffic while Fig. 8 shows it for the streaming data traffic. The proposed protocol has the lowest loss probability compared with the other systems. That is because of the use of TH codes for each user and hence no collisions as in the other systems. The loss packets in the proposed system come from only the rejected requests. Furthermore, the loss probability in streaming data is more than data traffic and nearly will be saturated in the proposed protocol at a value lower than the other systems due to the high load of the system. UbiCC Journal, Volume 3, Number 1, February 2008 The average delay for PRMA is larger than Slotted Aloha system and both of them give higher average delay compared with the proposed protocol for data traffic as shown from Fig. 9. That is because the presence of the reserved channels in PRMA which increases collisions for the additional users. For streaming data as shown from 10, The average delay for Slotted Aloha is larger than PRMA system and both of them are still higher than the proposed protocol. That is because the collision increment due to the streaming nature of the data and the absence of the advantage of the reserved channels to serve portion of low traffic streaming data users as in PRMA. Fig. 11 and 12 show the proposed protocol average delay in both types of data to indicate that the average delay increases with the increase of the number of the users and hence saturates. 24 5 10 25 system average delay 6 system average delay 10 20 5 proposed protocol average delay (sec) PRMA 10 10 slotted aloha average delay (sec) 50 60 10 15 4 3 10 5 2 10 0 1 10 −5 0 10 20 30 40 number of data terminals 0 0 10 20 30 40 number of data terminals 50 60 Fig. 9. System average delay Fig. 12. The proposed protocol average delay for streaming data 7 6 5 average delay (sec) 4 3 2 1 0 x 10 21 system average delay The admission ratio of the proposed system is decreased with the increase of the number of data terminals. For more users, the active links will increase and the probability of the admission of new user will decrease. Whenever more links available, interference will increase. The interference margins will decrease and hence low probability of admission of new requests. Fig. 13 depicts the admission ratio for both types of data versus the number of link terminals. For streaming data traffic, low admission ratio can be noticed. 1 admission ratio slotted aloha PRMA 0.9 0.8 proposed protocol admission ratio 0.7 0.6 0.5 0.4 0.3 0.2 data streams data traffic 0 10 20 30 40 number of data terminals 50 60 Fig. 10. System average delay for streaming data 0.1 0 0 10 20 30 40 number of data terminals 50 60 Fig. 13. Admission ratio V. 10 −1 CONCLUSION system average delay average delay (sec) 10 −2 In this paper, a new protocol was proposed. By using this proposed protocol, the network performance will be enhanced due to the increment of capacity. Comparisons between the proposed protocol, Slotted Aloha and PRMA were made. From these comparisons, we conclude that the proposed protocol gives better performance in both data traffic and streaming data of system throughput, channel utilization, average delay, and loss probability than others. 10 −3 0 10 20 30 40 number of data terminals 50 60 Fig. 11. UbiCC Journal, Volume 3, Number 1, February 2008 The proposed protocol average delay 25 6 R EFERENCES [1] First report and order in the matter of revision of part 15 of the commission’s rules regarding ultra-wideband transmission systems. Fedral Communications Commission (FCC 02-48), Apr. 2002. [2] X. S. Shen, H. Jiang, and J. Cai, “Medium access control in ultrawideband wireless networks,” IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, vol. 54, pp. 1663–1677, 2005. [3] DS-UWB Physical Layer Submission to 802.15 Task Group 3a. IEEE P802.15-04/0137r5, September 2005. [4] MultiBand OFDM Physical Layer Proposal for IEEE 802.15 Task Group 3a. MultiBand OFDM Alliance SIG, 2004. [5] Part 15.3: Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications for High Rate Wireless Personal Area Networks (WPANs). IEEE Computer Society, 29 September 2003. [6] Part 15.3: Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications for High Rate Wireless Personal Area Networks (WPANs) Amendment 1: MAC Sublayer. IEEE Computer Society, 5 May 2006. [7] M. win and R. Scholdtz, “Ultra-wide bandwidth time-hopping spreadspectrum impulse radio for wireless multiple-access communications,” IEEE Transactions on communication, vol. 48, pp. 649–691, April 2000. [8] Y. Chu and A. Ganz, “Mac protocols for multimedia support in uwb-based wireless networks.” Available at: www.broadnets.org/2004/workshop-papers/Broadwim/broadwim2004Paper09-yuechun.pdf. [9] F. Cuomo, C. Martello, and A. Baiocchi, “Radio resource sharing for ad hoc networking with uwb,” IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, vol. 20, NO. 9, pp. 1722–1732, Dec. 2002. [10] H. Yomo, P. Popovski, C. Wijting, I. Z. Kovacs, N. Deblauwe, A. F. Baena, and R. Prasad, “Medium access techniques in ultra wideband ad hoc networks.” This paper describes work undertaken partly in the context of the IST-2001-34157 Power aware Communications for Wireless OptiMised personal Area Network (PACWOMAN). The IST program is partially funded by the EC. Available at: http://cpk.auc.dk/FACE/documents/Publication/hiro/etai.pdf. [11] M. Mezzour, “Direct spread spectrum (ds) / time hopping (th) uwb performances comparison in a multi-user ad hoc environment.” Available at: www.mics.ch/SumIntU04/MohamedMezzour.pdf. [12] H. Jaing, W. Zhuang, and X. S. Shen, “Effective interference control in ultra-wideband wireless networks,” IEEE VEHICULAR TECHNOLOGY MAGAZINE, pp. 39–46, SEPTEMBER 2006. [13] B. Radunovic and Jean-Yves, “When power adaptation is useless or harmful,” tech. rep., IC, 2004/60. [14] F. Legrand, I. Bucaille, S. Héthuin, L. D. Nardis, G. Giancola, M. D. Benedetto, L. Blazevic, and P. Rouzet, “Ultra wide band system: Mac and routing protocols,” International Workshop on Ultra Wideband Syatems, IWUWBS, Oulu, Finland, April 25, 2003. [15] D. J. Goodman and S. X. Wei, “Efficiency of packet reservation multiple access,” IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, vol. 40, no. 1, pp. 170–176, Feb. 1991. UbiCC Journal, Volume 3, Number 1, February 2008 26 Study on RNN Query in Broadcasting Environment Lien-Fa Lin*, Chao-Chun Chen Department of Computer Science and Information Engineering National Cheng-Kung University, Tainan, Taiwan, R.O.C. Department of Information Communication Southern Taiwan University of Technology, Tainan, Taiwan, R.O.C. lienfa@cc.kyu.edu.tw,chencc@mail.stut.edu.tw ABSTRACT Location-based services (LBSs) provide information based on location information specified in a query. Queries that support for LBS are called Location-Dependent Queries (LDQ). One such query is the Reverse Nearest Neighbor (RNN) query that returns the objects that have a query object as their closest object. Just like the Nearest Neighbor (NN) queries, the RNN queries appear in many practical applications such as decision support system, continuous referral systems, profilebased marketing, maintaining document repositories, bioinformatics, etc. Thus efficient methods for the RNN queries in database are required. While the RNN is well studied in the traditional wired, disk-based client-server environment, it has not been tackled in a wireless broadcasting environment. The liner property of wireless broadcast media and power conserving requirement of mobile devices make the problem particularly interesting and challenging. In this paper, the issues involved with organizing location dependent data and answering RNN queries on air are investigated. An efficient data organization, called Jump Rdnntree, and the corresponding search algorithms are proposed. Performance of the proposed Jump Rdnn-tree and other traditional indexes (enhanced for wireless broadcast) is evaluated using both uniform and skew data. The results show that Jump Rdnn-tree substantially outperforms the traditional indexes. Keywords: location-dependent services, data broadcast, energy-conserving, mobile computing 1 INTRODUCTION Owing to the popularity of personal digital devices and advances in wireless communication technologies, location-based services (LBSs) have received a lot of attention from both of the industrial and academic communities [6,11,12,16,17,18]. With the maturation of necessary technologies and the anticipated world wide deployment of 3G wireless communication infrastructure, LBSs are essential applications in wireless networking environment. The query concerns LBSs, we called it LDQ (locationdependent query). The LDQ’s applications contains range query, nearest neighbor (NN), k-nearest neighbor (KNN) query and reverse nearest neighbor (RNN) query etc. In the past study on LDQ includes NN [5,15] query, KNN [3,4,8,12] query, CNN [21,23] query and CKNN [22,23] query are abundant and successful. And in recent years, the researchers have considerable attention on RNN query questions too. The query concerns LBSs, we called it LDQ (location-dependent query). 1 The RNN problem has been introduced in database setting by Korn and Muthukrishman [9] along with several applications. For example, the bank plans to establish a new branch. If customers always prefer the nearest branch, then the new branch should be established on the location where the distance to such location for the majority of customers is shorter than that to other banks. Another common example is how a taxi driver chooses customers. By using wireless devices, a taxi driver may know the location of a customer who is looking for a taxi. From the view of competition, RNN is * Lien-Fa Lin is also a lecturer at the Department of Information Communication, Kao Yuan University, No.1821, JhongShan Rd, Lujhu Township, Kaohsiung Country 821, Taiwan, R.O.C. UbiCC Journal, Volume 3, Number 1, February 2008 27 more meaningful than NN. As shown in Figure 1, the nearest neighbor to Taxi A is Customer C, but that does not necessarily mean Taxi A is the most likely to get to Customer C because Taxi B is even closer to Customer C. On the contrary, Taxi A should head for Customer D because Taxi A is the nearest neighbor in relation to Customer D. That is, the RNN for Taxi A is Customer D, and Taxi A may get to Customer D faster than all other taxis. the experiment environment. Performance results are shown in Section 6. Finally, we summarize the paper and describe our future work in Section 7. 2 RELATED WORK In this section, we shall introduce RNN query, and research topics that relate to on air index and RNN query in broadcasting environment in the following subsections 2.1 Reverse Nearest Neighbor Query Figure 1: Example of RNN query application As mobile device users increase, it has become a great challenge to availability of LBSs regardless of the increasing number of users. Wireless broadcasting technology is a solution to this problem [1,4,7]. Data delivery via broadcasting channels allows any number of mobile users (MU) to receive data at the same time. In addition, to effectively conserve power of mobile devices, the common practice is to broadcast data and on air index through broadcasting channels in an interlaced fashion. With on air index, MU knows when the required data will be broadcasted; the doze mode of mobile device, therefore, can be selected first, and then the active mode can be switched on until the arrival of the required data without wasting power by maintaining active mode to wait for the arrival of required data. The studies of using on air index technology, making MU use selective tuning to conserve power are plenty and popular [3,4,8,11,12,15,20]. Effective power conservation for mobile devices in wireless environment is a critical issue. Therefore, there is much literature dedicated to general query processing on mobile devices with effective power management [13,14,15,18,20]. From these studies we have deduced some principles for designing a good on air index. We use these principles to design an on air index method that can process RNN query efficiently. In addition, simulation experiments proved that our method may significantly improve efficiency when compared to Rdnn-tree modified for broadcasting environment. The rest of paper is organized as follows. Section 2 is an overview of related work. In Section 3, we describe the effectiveness on air indexing design rules. The details of Jump Rdnn-tree index structure are introduced in Section 4. In Section 5, we describe The so-called RNN query that means offers a certain objects set S and a query object q to find out the objects which q is their nearest neighbor (NN) object. The RNN query application is quite widespread, including decision-making support system, biological information and so on. In [5,9,10] mention many about the RNN query application example. A straightforward solution to computing reverse nearest neighbor (RNN) queries is to check for each point whether it has a given query point as its nearest neighbor. However, this method is not practical when processing large amount of data, because the time complexity involved is O (N3). Therefore, the general method is use a specially R-tree (is called RNN tree [9]) to process query. Conjun Yang [5] proposed Rdnn-tree index structure to improve the method in [9]. This Rdnn-tree index structure can be applied to solve NN and RNN query problem simultaneously. The difference between Rdnn-tree and R-tree is that Rdnn-tree has recorded each object’s NN information which can be used to process RNN query effectively. 2.2 Wireless Data Broadcas MU may access LBSs information in wireless broadcasting environment with two methods: On-demand Access: MU submits query to server, and server may use disk-based spatial index to accelerate query processing and increase data access efficiency. Server side is responsible to filter out data requested by MU and return the result to MU. Broadcast and Filter: Data is broadcasted on public wireless channels periodically. MU simply tunes into the broadcast channel to access required data instead of constantly submitting query to server. On-demand access uses basic client-server model, where server is responsible for query processing and returning results to users via point-topoint dedicate channel. However, on-demand access is more adequate on the system with less contention for wireless bandwidth, server processing, and workload. When the number of users increases, UbiCC Journal, Volume 3, Number 1, February 2008 28 system efficiency will reduce rapidly. As for wireless broadcasting applied to the radio and TV industry, the workload of a server is same and isn’t affected by the number of users; the server still delivers one set of data only. It is a natural solution for user scalability and bandwidth problems. On top of that, because mobile devices have a very limited supply of power, efficient power conservation is a major issue for mobile devices in wireless environment. In order to conserve power, it is common that mobile device design includes operation modes of active mode and doze mode [20]. A typical wireless PC card consumes 60mW in doze mode and 805–1400mW in active mode [16]. Power conservation in wireless broadcasting environment is achieved by adding index data to broadcasted data. By querying index data, users may know the time when required data will be broadcasted and select doze mode to save power and turn to active mode to access required data when schedule broadcast time is on. not a very effective method in term of access latency. An alternative, as shown in Figure 3 (b), is direct access to MBRs sequentially. However, this method will cause unnecessary traversal of MBRs, and index search performance will not be optimized. For example, the search of NN for q1, the real NN is O4 of MBR R2, and accessing to R1 is obviously a waste of resource. Therefore, a new index method must be designed for wireless broadcasting environment to effectively adopt the feature of linear access in broadcasting environment and satisfy the need of power conservation for mobile devices. In this paper, we have proposed a set of better broadcast index design principles and a modified Rdnn-Tree structure by adding the so-called jump pointer to make index tree accommodate linear access and to eliminate several unnecessary indexes to shrink the size of index data. We call this new index structure Jump-Rdnn Tree. 2.3 On Air Index In traditional disk-based access environment, back-tracking is often used in query algorithm to enhance query efficiency. However, it makes problems if used in broadcast channels where only linear access is available. In a wireless broadcasting environment, users may access data only when index data is being broadcasted. Therefore, when the sequence that the algorithm obtains index data is opposite to that of broadcasting, users must wait until the next broadcasting of such index data. For traditional database, on the contrary, index data that is stored on resident storage media, such as disk or memory chip, can be accessed in any time. Because linear access is not considered in the design of traditional index structure, the algorithm that is currently adopted in disk-based spatial index can not satisfy the need of effective power conservation. Shown in Figure 2 is R-tree index; its broadcasting sequence is root, R1, and R2. The visit sequence for searching for NN with a given query point of q2 is shown in Figure 3 (a). Root is first visited because the distance between q2 and R2 is shorter than that to R1. Therefore, R1 is skipped and R2 is visited first. However, the shortest object to q2 is o3 of R1 in MBR, and therefore R1 must be first visited. However, at this time R1 has just been broadcasted and it can only be accessed in the next broadcasting cycle. With the feature of linear access in broadcasting environment, if the broadcasting sequence differs from the sequence of query, then long access latency will occur. Therefore, branch-andbound query method in broadcasting environment is (a) MBR structure. (b) R-tree index Figure 2: A running example of R-tree Indexing. Figure 3: Linear access in wireless broadcasting environment 3 EFFECTIVE BROADCAST INDEX DESIGN Access latency of accessing to data and tuning time that a mobile device requires in active mode are the two benchmarks for broadcast index efficiency measurements. Access latency is the time required for accessing to data from the moment a user gives the query command to the data that satisfies the query is accessed. Tuning time is the time required for users to receive requested data in active mode. Broadcast index is mixed with broadcast data and sent out together, and MU receives data in the following three steps [12]: (1) Initial probe: during any point in time of broadcasting, a user tunes into a broadcast channel and wait for the index data to be broadcasted. This UbiCC Journal, Volume 3, Number 1, February 2008 29 period of time is called initial probe waiting. (2) Index search: When index data arrives, a user receives the index data, selectively accesses some index data according to his/her needs, and finds the location of the requested data. (3) Data retrieval: When the requested data arrives, a user downloads and accesses to the data. The time required for these three steps shall influence broadcast index efficiency. Therefore, a design of effective broadcast index must reduce the time required for these three steps. Reduction of initial probe time: Initial probe waiting is the time that a user waits for index data. By duplicating multiple indexes in the entire broadcast cycle, the possibility of index appearing may increase, and the initial probe waiting time can be reduced. Imielinski et al. [7] used interleaving method, as shown in Figure 4 (1: m), to duplicate m copies of index data in order to reduce initial probe waiting time. Reduction of index data size: Index searching time is related to the size of index data; the smaller the index data size is, the shorter the search time will be. Consequently, the entire broadcast cycle will be shorter, and the average access latency will be smaller. For example, Imielinski et al. [20] only duplicated k layers of index tree to reduce the size of index data. Hu et al. [18] used the signature capture technique to reduce index data size. A RNN query that searches for q returns a collection of objects of nearest neighbors in relation to q. If we may know the distance between every object and its NN in advance, then all we have to do is to find out the distance between q and the objects which are closer than that between the objects and its NN, and then the objects are the results for the RNN query that searches for q. The difference between Rdnn-tree and R-tree is that Rdnn-tree stores the information of every object (such as distance of neighbor, or DNN), and it may directly determine whether a leaf node is the result of the query, while R-tree cannot directly determine whether a leaf node is the result of the query and must use branch-and-bound technique, which may cause back-tracking problem. Therefore, we further improve Rdnn-tree with the principles for a better broadcast index that we have proposed to make it an index structure that can effectively support the RNN search in wireless broadcasting environment. 4.1 Rdnn-Tree Figure 4: Data and Index Organization using the (1: m) Interleaving Technology Efficient data placement: Chen et al. [15] has proved that different broadcasting sequence of different data would affect average access latency of data retrieval, and proposed ORD algorithm to reduce average access latency of data retrieval. Jianting and Le Gruebwald [14] proposed to reduce access latency by arrangement of the sequence of broadcast data according to retrieval frequency. Currently broadcast index studies focus on one single step to enhance efficiency without considering improving the efficiencies of the three steps. This paper has designed a new broadcast index to handle RNN query in broadcasting with considerations for the three steps. R-tree [2] in the early stage was an index structure developed for spatial database, and was modified to Rdnn-Tree by Yang and Lin [5] to accelerate NN and RNN queries. Rdnn-Tree structure, as shown in Figure 5, groups objects that share similar coordinates and places them on leaf node. That is, objects with similar coordinates are grouped. Then, a group of objects is contained in a smallest rectangle, which is called minimum bounding rectangle, “MBR” for short. Next, similar MBRs are further grouped; a group of MBR is contained in one even larger MBR, and the process continues until all objects are contained in the same MBR. What is stored on the internal node within an Rdnn-Tree is MBR; all nodes under it will be contained by it, and all objects will be contained by the root of Rdnn-Tree Every MBR will record the coordinate at its bottom left (Ml，Md) and upper right (Mr，Mu), and the size and scope of a MBR can be obtained. Ptid and dnn are stored at the leaf node. Ptid is the reference number of data collection point, while dnn is the distance between the object and its NN. Ptr, rect, and MaxDnn are stored on non-leaf node. Ptr points to the address of child node, Rect is the MBR contained in the node and the child nodes underneath, and MaxDnn is the maximum value of the dnn of all objects in this sub-tree; the greatest distance between all objects and their NN under the sub-tree will not be greater than MaxDnn. 4 A NEW INDEX FOR RNN QUERY UbiCC Journal, Volume 3, Number 1, February 2008 30 Figure 5: Data structure of Rdnn-Tree Figure 6: Data structure of Rdnn-tree 4.2 Jump Rdnn-tree The design of a good broadcast index as mentioned in Section 3 includes three steps: reducing initial probe time, reducing size of index data, and effective placement of broadcast data. We shall explain how we improve Rdnn-tree with these three steps. Reduction of initial probe time: The traditional approach is to increase the possibility of the appearance of index by duplicating index. However, this approach will cause longer broadcast cycle and longer average data access latency. Our approach is to build a Jump Rdnn-tree with our index structure for broadcast data. Data and index will be mixed together and broadcasted based on every sub-tree. After the index of such sub-tree has been broadcasted, the data under the sub-tree will be broadcasted in order to reduce the distance between data and index instead of broadcasting all data after the index broadcasting is completed. Taking Figure 6 as example, the broadcasting sequence is B, B1, b1, a, b, c, d, b2, e, f, g, B2, b3, h, i, b4, j, k, and l. The cyclic index structure is also adopted. The location of each index node to be visited next is calculated with Depth First Search (DFS) traversal according to depth priority. This approach is also called jump pointer, as shown in Figure 7. Because every index node has a jump pointer, index trees are interlined through jump pointers. Therefore, index tree search may begin from any index node instead of root node. In the same time, index tree search follows pointer sequence, and there will be no back-tracking problem. Figure 7: Data Structure of Jump Rdnn-tree Reduction of index data size: As mentioned earlier, we adopt cyclic index structure, replication of index data for the entire broadcast cycle is not required and only one copy of index data is needed. Therefore, the size of index data is very small. Also, if the largest sub-tree at every layer of the structure of a traditional index tree has f fan-out that represents the index tree needs f pointers are needed to record the address of every sub-tree. Because of linear access feature of wireless broadcasting environment, MU can only access data in active mode, or skip the current data in doze mode and wait for the next data access in active mode. That is, only two behaviors are available: Next, which proceeds to next action, and Jump, which skips data retrieving. If the criteria for Jump are not met, then Next takes place and Next behavior does not have to be recorded. Therefore, any search for nonleaf node of a tree requires only keeping one jump pointer, and other f-1 pointers for sub-trees can be eliminated, leading to reduction of index data size. Effective data placement: Query point is produced based on the entire search space. The possibility of the occurrence in every area shall affect searching efficiency. If the corresponding index data for the UbiCC Journal, Volume 3, Number 1, February 2008 31 area where the possibility of the occurrence of query point is higher are broadcasted earlier, then the index search efficiency will be better and if the location of the query point is in uniform distribution. Therefore, the broadcast sequence of a sub-tree is determined according to the area of MBR where the sub-tree is, because the larger the area of MBR is, the higher the possibility of the query occurrence will be. After Rdnn-tree is improved by the abovementioned approach, the problem of back-tracking is eliminated, and Rdnn index tree with cyclic structure fits with linear access feature of broadcast better. The detailed algorithm of RNN queries of cyclic broadcast is illustrated in Algorithm 1: RNN-SearchOn-Air. D (q, ptid) represents the distance between query point q and objects ptid, wile D (q, rect) represents the distance between query point q and 5 EXPERIMENT ENVIRONMENTS 5.1 Two Different Experiment Data Sets rectangle rect. RNN-Search-On-Air (Node n, Point q) Case 1: If n is leaf node then For all (data-item, dnn) in n If D (q, ptid) < dnn, then data-item is the RNN for q. Case 2: If n is non-leaf node, then For all branch B = (ptr, MBR, Maxdnn) in n If D (q, rect) < MaxDnn, then call RNNSearch-On-Air (B.ptr, q). Algorithm 1: RNN Search on Air 5.3 Performance Metrics We use two different data sets for the experiment as shown in Figure 8. For the first dataset, UNIFORM, we produce 1,000 points in square Euclidean space uniformly. For the second data set, SKEW, we produce 1,000 points with Zipf distribution, and the skewness parameter of Zipf distribution is 1.2. Figure 8: Uniform and Skew data sets The metrics used for measuring the effectiveness of On Air Index are access time and tuning time, and the unit of measurement is packet (the unit of broadcast). Although tuning time in general may reflect the power consumption of mobile devices, but it only records the power consumed on active mode; therefore, it may not reveal the actual power consumption. Although less power is consumed in doze mode, but as the waiting time is prolonged, the power consumption is also very huge. We believe power conservation should be evaluated with total power consumption; therefore, total power energy metric is added to the performance metrics used in our experiment. Total power energy is P=1200*Timeactive mode + 60* Timedoze mode. In order to simplify the complexity of this experiment, we ignore the power consumed in query processing under the premise that the result of the experiment is not affected. Assume 1200mW includes the power required for accessing one unit of broadcast packet, 60mW is the power required for waiting for one unit of broadcast packet. 5.2 Compared Algorithm 5.4 Parameters Setting Due to the feature of linear access of wireless broadcasting, and therefore Rdnn-Tree is modified to fit in the air indexing model. In a node of Rdnn-Tree, it will access to data by DFS sequentially according to depth searching. The sub-tree branches that do not match with the condition of distance heuristics search in the process of searching will be pruned. In order to reduce access latency, the ratio of Rdnn-tree and sorted list of the broadcast index here is set to 1:m. Interlace techniques are called Rdnn-Tree (1: m). Parameters setting are shown in Table 1. The packet id of each packet is 2 bytes; one coordinate is 4 bytes, and index takes up 2 bytes. Packet size varies from 64 bytes to 1024 bytes. Fan out of Jump Rdnn-Tree is set to 6. The number (object #) of the parameter object varies from 1000 to 5000. Query is randomly produced from the entire search space. User’s initial probe time is randomly produced from 1 to 5000 broadcast unit. The final statistic result is an average value of 30 queries [19]. The program used for the experiment is modified with the R-tree codes of R-Tree Portal (http://www.rtreeportal.org/). UbiCC Journal, Volume 3, Number 1, February 2008 32 access latency. Table 1: Experiment Metrics Setting Parameter Object# Description number of data object size of a broadcast packet number of the sub-trees of Rdnn-Tree Setting default:1000 vary from 1000 to 5000 default:256 bytes vary from 64 to 1024 bytes default t:6 For tuning time, as broadcast packet capacity increases, the average packet access time decreases. However, when object data is in skew distribution, if broadcast packet capacity is larger than 1024 bytes, then tuning time will increase slightly. We adopt the approach in which data is mixed with index and broadcasted, and the entire index data in DFS sequence is scattered in the entire broadcast program; when the broadcast packet capacity is large enough, index sections may not fill a broadcast packet to the fullest. Because a broadcast program with index data must separate index and data packet, a broadcast packet not fully loaded still occupies one broadcast packet. This situation tends to be more severe when the object data is in skew distribution. Therefore, when broadcast packet capacity becomes larger, the entire average broadcast packet access time decreases, and when average packet access time is shorter, its efficiency is more significant. For total power consumption, regardless of data object distribution, our approach is significantly better than Rdnn-Tree (1: m). Even though the tuning time in our approach is slightly larger than Rdnn-Tree (1: m) when object data is in skew distribution; the broadcast packet capacity is larger than 1024 bytes, the total power consumption in our approach is still better. This matches with our idea. When considering broadcast index efficiency, total power consumption must be also considered because it reveals the real power consumption of mobile devices. Packet Size Fan out 6 PERFORMANCES RESULTS 6.1 Influence of Packet Size on Performance This experiment is to measure the performance efficiency for different packet size under two different data sets. Experiment results of different data are shown in Figure 9 and Figure 10. For access time, no matter data object is in uniform distribution or skew distribution, access time becomes smaller as the size of broadcast packet changes, because larger packet capacity of broadcast packet allows more data. Therefore, the broadcast cycle of entire broadcast program will be shorter. Besides, our approach is obviously better than Rdnntree (1:m). In our approach, index data is broadcasted only once. Compared to Rdnn-tree that adopts (1: m) and broadcasts index data m times, our approach has relatively shorter broadcast program cycle and less Figure 9: Influence of packet size on performance when data objects are in normal distribution UbiCC Journal, Volume 3, Number 1, February 2008 33 Awerage tuning time(#Bucket) Access Latency(Bucket#) 180 160 140 120 100 80 60 40 20 0 64 128 256 512 1024 2048 Jump-Rdnn Tree Rdnn Tree(1:m) 110 100 90 80 70 60 50 40 30 20 10 0 64 Jump-Rdnn Tree Rdnn Tree(1:m) 105 104 Jump-Rdnn Tree Rdnn Tree(1:m) W 103 102 101 128 256 512 1024 2048 64 128 256 512 1024 2048 Packet Size Packet Size Packet Size (a) access time (b) tuning time (c) power energy Figure 10: Influence of packet size on performance when data objects are in skew distribution 6.2 Effect of Data Object on Experiment Effectiveness This experiment is to measure the performance efficiency for different number of data objects (object#) under two different data sets. Experiment results of different data are shown in Figure 11 and Figure 12. As the number of data objects increases, broadcast program cycle lasts longer, making access latency, tuning time, and total power consumption increase. 14000 For access latency, because the entire broadcast program cycle is longer, average waiting time for packet broadcast is longer, and the size of index data is larger, the performance ratio of Rdnn-Tree (1: m) is m times. For tuning time, as mentioned earlier, because a user turns into active mode for data retrieval or doze mode for skipping the retrieval according to broadcast data, the capture of selective tuning data must separate index packet and data packet. For total power consumption, our approach is significantly better than Rdnn-Tree (1: m). 104 Jump-Rdnn Tree Rdnn Tree(1:m) Awerage tuning time(#Bucket) Access Latency(Bucket#) 45 40 35 30 25 20 15 10 5 0 1000 2000 3000 4000 5000 Jump-Rdnn Tree Rdnn Tree(1:m) 12000 10000 8000 6000 4000 2000 0 1000 Jump-Rdnn Tree Rdnn Tree(1:m) 103 W 102 101 2000 3000 4000 5000 1000 2000 3000 4000 5000 number of data objects number of data objects number of data objects (a) access time (b) tuning time (c) power energy Figure 11: Effect of number of data object on performance when data objects are in normal distribution 1000 800 600 Jump-Rdnn Tree Rdnn Tree(1:m) 14000 12000 10000 8000 6000 4000 2000 Jump-Rdnn Tree Rdnn Tree(1:m) Awerage tuning time(#Bucket) 40 35 30 25 20 15 10 5 0 1000 2000 3000 4000 5000 Jump-Rdnn Tree Rdnn Tree(1:m) Access Latency(Bucket#) W 400 200 1000 0 1000 2000 3000 4000 5000 2000 3000 4000 5000 number of data objects number of data objects number of data objects (a) access time (b) tuning time (c) power energy Figure 12: Effect of number of data object on performance when data objects are in skewed distribution UbiCC Journal, Volume 3, Number 1, February 2008 34 7 CONCLUSIONS AND FUTURE WORK In this work we have discussed how to effectively organize and deploy data in wireless broadcasting environment and answered questions about RNN query. Based on previous studies, we have summarized the principles for designing broadcast index. Based on these principles, a new index structure, Jump-Rdnn tree that is idea for RNN query in broadcast environment, is designed. Because this work focuses on location-dependent data access in broadcasting environment and is different from traditional on-demand access model, we only discuss issues concerning static RNN query. We shall further extend to more advanced and more dynamic location-dependent queries. 8 REFERENCES based on reverse nearest neighbor queries “， Proceedings of the ACM SIGMOD International Conference on Management of Data，,May 16-18， 2000 ，p.p. 201-212. [10] I. Stanoi ，D. Agrawal and A.E. Abbadi，”Reverse Nearest Neighbor Queries for Dynamic Databases”，ACM SIGMOD Workshop on Research Issues in Data Mining and Knowledge Discovery，2000，p.p.44-53. [11] J. Xu , B. Zheng , W.C.Lee and D.L. Lee", "Energy efficient index for energy query locationdependent data in mobile environments", In Proceedings of the 19th IEEE International Conference on Data Engineering, Bangalore, India March 2003, p.p. 239-250. [12] J. Zhang, M. Zhu ,D. Papadias ,Y. Tao and D.L. Lee, "Location-based Spatial Queries", Proceedings of the 2003 ACM SIGMOD international conference on Management of data, San Diego, California, USA , June 9-12 2003,p.p. 443-454. [13] J. Xu and W.C. Lee and X. Tang ,"Exponential Index: A Parameterized Distributed Indexing Scheme for Data on Air”., "Proceedings of the 2nd ACM/USENIX International Conference on Mobile Systems, Applications, and Services , Boston, MA, June 2004,p.p. 153-164. [14] J. Zheng and L. Gruenwald，”Prioritized Sequencing for Efficient Query on Broadcast Geographical Information in Mobile Computing”, ACM GIS，2002,p.p.88-93. [15] M.S. Chen ,P.S. Yu and K.L. Wu，”Optimizing index allocation for sequential data broadcasting in wireless mobile computing”， IEEE Transactions on Knowledge and Data Engineering，Vol. 15，No. 1,January-February 2003,，p.p. 161-173. [16] M. A. Viredaz，L. S. Brakmo and W. R. Hamburgen，”Energy management on handheld devices”, ACM Queue，Vol. 1，No. 7，October 2003，p.p. 44-52. [17] N. Roussopoulos，S. Kelley and Fr'ed'eric Vincent，Nearest neighbor queries，Proceedings of ACM SIGMOD International Conference on Management of Data，June 1995，p.p. 71-79. [18] Q. Hu , W.C. Lee and D.L. Lee，"Index Techniques for Power Management in MultiAttribute Data Broadcast"，Mobile Networks and Applications，Vo. 6,No. 2,2001,p.p.185-197. [19] Sheldon Ross，Introduction to Probability and [1] A. Datta, D.E. Vandermeer, A. Celik, and V. Kumar,”Broadcast Protocols to Support Efficient Retrieval from Database by Mobile Users”, ACM Transactions on Database Survey [2] A. Guttman ， ”R-Trees: A Dynamic Index Structure for Spatial Searching”，Proceedings of the 1984 ACM SIGMOD international conference on Management of data，1984， p.p.47-57. [3] B. Zheng , W.C. Lee and D.L. Lee, "Search K Nearest Neighbors on Air", Proceedings of the 4th International Conference on Mobile Data Management, Melbourne, Australia January 2003,p.p. 181-195. [4] B. Zheng and D.L. Lee,”Information Dissemination via Wireless Broadcast”, Communication of ACM, 48(5), May 2005, p.p. 105-110. [5] C. Yang and K.L. Lin, "An index structure for efficient reverse nearest neighbor queries", Proceedings of the 17th International Conference on Data Engineering,2001,p.p. 485-492. [6] Computer Science and Telecommunication Board. IT Roadmap to a Geospatial Future, the National Academies Press, 2003. [7] D. Barbara，”Mobile Computing And Databases - A Survey”，IEEE Transactions on Knowledge and Data Engineering，Vol. 11，No. 1， February 1999，p.p. 108-117. [8] D.L. Lee，W.C. Lee，J. Xu and B. Zheng，”Data Management in location dependent services”，IEEE Pervasive Computing，Vol. 1， No. 3，September 2002，p.p. 65-72. [9] F. Korn and S. Muthukrishnan，”Influence sets UbiCC Journal, Volume 3, Number 1, February 2008 35 Statistics for Engineers and Scientists，2000. [20] T. Imielinski ,S. Viswanathan and B. R. Badrinath，Data on Air:organization and access， IEEE Transactions on Knowledge and Data Engineering，Vol. 9，No. 3，May 1997，p.p. 353-372. [21]R. Benetis，C.S. Jensen，G. Karciauskas and S. Saltenis， Nearest Neighbor and Reverse Nearest Neighbor Queries for Moving Objects， International Database Engineering and Applications Symposium，Canada，July 17-19， 2002，p.p. 44-53. [22] G. Iwerks，H. Samet and K. Smith，”Continuous K-Nearest Neighbor Queries for Continuously Moving Points with Updates”， International Conference on Very Large Data Bases， 2003，p.p. 512-523. [23] Z. Song and Nick Roussopoulos, K-Nearest Neighbor Search for Moving Query Point, Proceedings of 7th International Symposium on Advances in Spatial and Temporal Databases, LNCS2121, Redondo Beach, CA, USA, July 1215,2001,p.p. 79-96. UbiCC Journal, Volume 3, Number 1, February 2008 36 Performance Evaluation for Mobility Management Protocols in Cellular IP and Hawaii Mobile Networks M.Mansour, A.Ghneimat ,J. E. Mellor Department of Computing University of Bradford Bradford BD7 1DP, UK. [Mmansour, Ghneimat, j.mellor]@bradford.ac.uk ABSTRACT Handover management is one of the most critical issues that mobility management protocols are concerned with. This process becomes even more critical in the case of Micro-mobility where the Mobile host is expected to encounter frequent handovers. Several IP micro-mobility protocols have been proposed to enhance the performance of Mobile IP in an environment with frequent handoffs. In this paper we make a details study of how the performance of the handoff schemes of two candidates for micromobility protocols, namely HAWAII and Cellular IP. The aim of the paper is to investigate the impact of handoffs on TCP by means of simulation traces that show the evolution of segments and acknowledgments during handoffs. Application of these models allows a comparison of two important handoff schemes: the Multiple Stream Forwarding scheme of HAWAII and the Semi-soft Handoff scheme of Cellular IP. Keywords: IP micro-mobility, CIP, HAWAII, handoff 1 INTRODUCTION introduce considerable delays in the handoff process due to the round-trip time between the Foreign Agent (FA) and the Home Agent (HA) during the registration process. Applied in an environment with frequent handoffs, this may lead to unacceptable disturbance to ongoing sessions in terms of handoff latency and packet loss. Over the past several years a number of IP Micromobility protocols have been discusses in the IETF mobile IP working group. Most of these protocols adopt a hierarchal approach by dividing the network into domains. Mobile IP is used to support mobility between two domains called macromobility. While Local movements of a mobile within an administrative domain called micromobility. IP micro-mobility protocols are designed for environment where mobile hosts change their point of attachment to the network so frequently. Micro-mobility protocols reduce the number of registrations that an MH has to do by allowing the MH to avoid having to register with the home agent (HA) every time it moves within the same domain [5]. This has the benefit of reducing delay and packet loss during handoff and eliminating registration between mobile hosts and possibly distant HAs when mobile hosts remain inside their local coverage areas. The Cellular IP [6] and Hawaii IP [4] protocols achieve faster and more seamless local mobility support in limited geographical areas. However there are still shortcomings and Today the most world’s powerful technology trends, the Internet and mobile communications. With the majority of information and new services being deployed over IP, the increasing variety of wireless devices offering IP connectivity, such as PDAs, handhelds, and digital cellular phone, is beginning to change our perceptions of the internet. Mobile data communication will likely emerge as the technology supporting most communication including voice and video. The third generation mobile radio communication systems (UMTS, CDMA2000) use a packet data transfer and switching technology as preferred solution. At the same time, the research community is investing a major effort to provide an “all-IP” end-to-end solution, commonly referred to as fourth-generation (4G) systems. 4G systems will support IP services transparently in a highly heterogeneous infrastructure independent of the underlying physical layer [1]. Mobile IP [MIP] [2] is the current standard solution designed for mobility management in IP network. It allows a Mobile Host (MH) to change its point of attachment from one access router to another. However Mobile IP is not designed to support fast handoff and seamless mobility because after each migration a local address must be obtained and communicated to the home agent (HA). This will UbiCC Journal, Volume 3, Number 1, February 2008 37 inefficiencies with these protocols. The Cellular 1P protocol does not support seamless handoff when a mobile host attaches to a new access point and loses contact with the previous one. The Hawaii IP protocol can handle this situation but does not handoff as quickly as the Cellular IP protocol does. Both the Cellular IP and Hawaii IP protocols include their own handoff schemes for local mobility. Cellular IP support two handoff schemes while Hawaii IP protocols use four handoff schemes. In this paper we present a performance comparison of the handoff schemes of two micro-mobility protocols Hawaii and Cellular IP (CIP). We show that the difference in handoff quality depending on the design of these protocols. The rest of the paper is organized as follows. The second section of this paper we describe the Cellular IP and their handoff schemes. The third section of this paper we describe the Hawaii protocols with two handoff schemes (MSF and UNF). The fourth section we then present the simulation using the ns simulator and compare some performance aspect of the MSF and UNF schemes of HAWAII and the hard handoff and Semis-oft handoff schemes of CIP. Finally Conclusion remark and future work. reduced through the introduction of paging. Mobile hosts are expected to typically operate on batteries with limited lifetime. This makes it important to save idle mobile hosts from having to transmit frequent location update messages. This requires explicit support from networking protocols, such as the ability to track location approximately and the ability to page idle mobile hosts. Idle mobile hosts do not have to register if they move within the same paging area. Rather, they only register if they change paging area. Handoff management is performed in three steps. The first stage is the initiation, in which the MN changes its point of attachment with a base station, it sends request to the current base station for handoff to the target BS. Then the connection generation follows and the network must find new resources for the handoff connection and perform additional routing operations. Finally the third step is the data flow control, where the delivery of the data from the old connection path to the new one is maintained according to the service agreement. In cellular environment there two kinds of handoff, the first one (Intracell) when the MN moves to a new AP belonging to the same subnet, it performs a Layer2 handoff. The second (Intercell) when the MN moves to anew AP belonging to another subnet, it performs a Layer3 handoff. . 3 CELLULAR IP 2 MOBILITY MANAGEMENT Mobility management consists of two components namely location management and handoff management [11, 12, and 13]. Location management, this is refers to the registration or location update (LU) and paging. Handoff management is the most important issue in mobility management. It is refers to the ability of the network to allow a call in progress to continue as the MN continues to move and change its point of attachment. Location management concerns the discovery of the current point of attachment of a mobile user for the delivery of incoming calls. When the MN move to other location it must informs the network of its location at times triggered by movement, timer expiration, this procedure perform by registration or location update. In the case of mobile hosts maintaining location information in support of being continuously reachable would require frequent location updates which would consume precious bandwidth and battery power. This signaling overhead can be The Cellular IP [5, 6 and 7] is a proposal to the IETF made by researchers from Colombia University and Ericsson. A Cellular IP network consists of interconnected Cellular IP nodes. A Cellular IP node that has a wireless interface is also called a Base Station. Mobility between gateways is managed by mobile IP while mobility within access networks is handled by Cellular IP. Cellular IP access networks are connected to the Internet via gateway router. Mobile hosts attached to an access network use the IP address of the gateway as their Mobile IP care-of address. Figure 1 illustrates the path taken by packets addressed to a mobile host. 3.1 Routing and Paging The Cellular IP base stations emits beacons on regular basis, this allows the mobile host to locate their nearest base station. When a Mobile host finds its nearest base station it has to register to this cellular IP network and sends a Route Update message to its connected base station, this route update message is routed internally in the cellular IP network from the base stations to Cellular IP Gateway by using a hop-byhop shortest path routing mechanism. As the route UbiCC Journal, Volume 3, Number 1, February 2008 38 update message travels to the gateway via base stations, route cache mappings are created in the base station, the path taken by these packets is cached by all intermediate base station. These cached routing maps are also used for packets destined for the Mobile host. The routing entries in the Cellular IP nodes are soft state. In order not to lose its routing path, a mobile host has to refresh its routing entries periodically. In Cellular IP this is done by a regular data packet or by sending a route update message if the mobile host has no data to transmit [6]. are also updated and refreshed by any packet sent by the mobile host. When an incoming call is detected at the gateway, a paging message is transmitted to the mobile host’s current paging area to establish the call. The mobile host receives the paging packet; it moves to active state and creates its Route Cache mappings as shown in Figure 2 Figure 2 Paging Update/Routing Update Figure 1Cellular IP access network If the mobile host wishes to maintain these mapping but not transmit data it can send a special Internet Control Message (ICMP). A typical example of this is when a mobile host receives UDP stream of packets on the downlink but has no data to transmit on the uplink. When a mobile host changes to a new base station it must send a route update packet, which contains authentication information, to its new Base Station, this will modify the Route cache mappings. Since data packets do not contain authentication information and hence they cannot modify the route cache mapping but only refresh the mappings. Packet send by Corresponding host to mobile host are first routed to the host HA and then tunneled to the gateway. The gateway detunnels packets and forwards them toward a base station. Inside a Cellular IP network, mobile hosts are identified by their home address, and data packets are routed without tunneling or address conversion. Mobile hosts that are not actively transmitting or receiving data but want to stay reachable for incoming packets, have the opportunity to let their route cache entries time out and to maintain paging cache entries. The difference between the route cache and the paging cache is that paging caches are not necessarily maintained on each Cellular IP node and have longer timeout values. Paging caches If a Cellular IP node, that does not maintain a paging cache, receives a downlink packet for a mobile host for which it has no routing entry in its route cache, it broadcasts the packet to all its downlink neighbors. A node that has paging cache but has no mapping in it for the mobile host discards the packet. Idle mobile hosts do not have to register if they move within the same paging area. Rater they only register if they change paging area. In order to remain reachable mobile hosts transmit paging-update packet at regular intervals defined by paging-update time. 3.2 Handoff Scheme in Cellular IP The Cellular IP support three types of handoff scheme, hard handoff, indirect handoff and semisoft hand off. In Cellular IP a handoff is always initiated by the mobile host. Mobile host listen to beacons transmitted by base station and initiate handoff based on signal measurements by sending a route update packet to the new base station. 3.2.1 Cellular IP hard handoff Cellular IP hard handoff is based on simple approach that trades off some packet loss in exchange for minimizing handoff signaling rather than trying to guarantee zero packet loss. When Mobile hosts switches to a new base station, it send a route-update packet to the new base station, hence UbiCC Journal, Volume 3, Number 1, February 2008 39 it has stopped listening to the old base station. Packets that are traveling on the old path to the old base station will be lost. No packets are transmitted along the old path once the route-update massage has created a new mapping at the cross-over base station that point toward the new base station. 3.2.2 Cellular IP Indirect handoff The [8] indirect handoff based on some wireless technology can not listen to the current base station while sending a route-update packet to the new base station. It assumed the network can obtain the IP address of the new base station. When the mobile host decides to make a handoff instead of sending a route-update packet to the new base station directly, it sends the packet to the current base station. This packet will have as its destination IP address, the IP address of the new base station. The old base station forwards this packet with a flag indicating indirect handoff to the gateway. The gateway delivers the packet to the new base station using normal IP routing. 3.2.3 Cellular IP Semi-soft handoff On the semi-soft handoff the mobile host switches to the new base station, transmits a route update message with a flag indicating the semi-soft handover while listening to the old base station. The route update message reconfigures the route caches on the way to the gateway router and adds an entry to the cache of crossover node. Downlink packets for the specific mobile host are duplicated and sent along paths, the new one and the old one. After a fixed amount of time, the mobile host finally migrates to the new base station and then sends another route update message to complete the semi-soft handover. This second route update message sets up a proper path to the new base station and stops the crossover node duplicating packets. If the path to the new base station is shorter than to the old one, some packets may not reach the mobile host. To overcome this problem, packets sent along the new path need to be delayed. A mobile host may get some duplicate packets or detect a loss of few packets. To prevent this delay device can be used to keeps the packets looping a specified time before sending them towards the MH through the new route. This delay ensures data stream on the new route does not get ahead of the stream traveling the old route. A delay device mechanism, located at the crossover node, should provide sufficient delay to compensate for the time difference between the packets travelling on the old and new paths. 4 HAWAII The Hawaii [4] Handoff-Aware Wireless Access Internet Infrastructure was proposed to the IETF by researchers from Lucent Bell Labs. Like in Cellular IP, HAWAII responsible for the micro-mobility within a domain while the macro-mobility is handled by Mobile IP. 4.1 Routing and Paging In HAWAII a hierarchy based on domains is used. The gateway into each domain is called domain root router (DRR). A HAWAII domain comprises several routers and base stations running the HAWAII protocol, as well as mobile hosts. When the Mobile host enters into a foreign domain the mobile host is assigned a co-located care-of-address. The mobile hosts keep its network address unchanged while moving within a domain. The Corresponding Node (CN) and the Home Agene (HA) do not need to be a ware of the host’s mobility within this domain. A mobile host in a HAWAII environment runs a standard Mobile IP protocol. Therefore mobile host in the Hawaii domain exchanges only MIP control messages with the network, while the routing within a Hawaii domain is realized with UDP-based Hawaii messages. Hawaii messages are never forwarded outside the domain, not even to the MH. The MH must support a bit modified way to handle some MIP messages. There are three types of HAWAII path setup messages: power-up, update and refresh. On power up a mobile host sends a Mobile IP registration request message to the corresponding base station. The base station then sends a HAWAII path setup power-up message to the domain root router which is processed in a hop-by-hop manner. On all routers on its way to the domain root router this power-up message adds a routing entry for the concerned mobile host. The domain root router finally acknowledges this path setup power-up message to the base station which finally notifies the mobile host with a Mobile IP registration reply. Thus, the connectivity from that domain root router to the mobile hosts connected. The routing entries in the routers are soft-state, i.e. they have to be refreshed periodically by path setup refresh messages, which are sent independently by each network node and which can be aggregated. Routers, not passed by a path setup message related to a mobile host; do not have any knowledge about its whereabouts. Whenever a router receives a packet for such an unknown mobile host, e.g. from another mobile host within the domain, it uses a preconfigured default interface pointing towards the UbiCC Journal, Volume 3, Number 1, February 2008 40 domain root router. This packet will be forwarded in this direction until it will arrive at a router knowing a route to the addressed host. In worst case this will be the domain root router. Mobile hosts in standby state only have to notify the network on a change of paging area and not on each base station handover. When a packet arrives for a mobile host in standby state, the network has to page it before it delivers the packet. This paging induces the mobile host to switch to active state immediately. For using Hawaii’s paging support, it is necessary to have link-layer paging functionality on the wireless link which means that the mobile host is able to identify its paging area and to detect paging requests. A typical solution for identifying the paging area is, that base stations periodically send beacon signals including the paging area identities on a broadcast channel, so a mobile host listening to this channel can easily detect a change. The paging requests of the base stations can be sent on separate paging channels to which the mobile hosts are listening. The network has to maintain paging information for each mobile host and has to deliver paging requests for these hosts up to the base stations from where on link-layer paging mechanisms are responsible. One way to achieve this HAWAII relies on the IP multicast routing protocol. Each paging area is assigned a multicast group address and all base stations within that paging area join this multicast group. 4.2 Handover Mechanisms Four alternative setup schemes control handoff between access points. The appropriate path setup scheme is selected depending on the service level agreement (or operator’s priorities among QoS parameters, e.g., eliminating packet loss, minimizing handoff latency, and maintaining packet ordering). The four path setup schemes can be classified into two classes. The first class includes two forwarding path setup schemes: MSF and an alternative the SSF schemes. The second class includes two non forwarding schemes: the UNF scheme and MNF scheme [3]. 4.2.1 Forwarding path setup schemes In these path setup schemes, packets are first forwards from the old base station to the new base station before they are diverted at the crossover router. We define the crossover router as the router closest to the mobile host that is at the intersection of two paths, one between the domain root router and the old base station, and the second between the old base station and the new base station. As mentioned above two variants of forwarding schemes in HAWAII are proposed, one that works with standard IP routing tables to update the hostbased entries and another scheme where the IP routing table is extended to accommodate interfacebased information. Figure 3 Hawaii path setup scheme MSF These schemes are known as Multiple Stream Forwarding (MSF) and Single Stream Forwarding (SSF). In the following the MSF scheme analyzed in this paper is describe. The MSF scheme is illustrate in Figure 3. When handoff occurs the mobile host loses contact with the old base station and at the same time MH sends a Mobile IP registration message (Message 1) to new base station. The new base station then sends a HAWAII path setup update message (Message 2) directly to the old base station. This message contains the new base stations address. The old base station performs a table look-up for a route to the new base station and determines the interface, interface A, and next hop router, Router 1. The old base station adds forwarding entry for the mobile host IP address with outgoing interface set to interface A. From now on the old base station forwards all data packets, including packets arrived after the handoff and stored in a forwarding buffer at the old base station to the new base station according to the new forwarding entry. The old base station then forwards Message 3 to Router 1. Router 1 performs similar actions and forwards the message 4 to Router 0. Router 0, the cross-over router in this case, changes the forwarding entry that results in new packets being diverted to the MH at the new base station. It then forwards the message towards the new base station. Eventually Message 6 reaches the new base station that changes it’s forwarding entry and sends an acknowledgment of the path setup message to the MH, shown as Message 7 Note that this order of updating the routers can lead to the creation of multiple streams of disordered packets arriving at the MH. For example, during UbiCC Journal, Volume 3, Number 1, February 2008 41 transient periods newer packets forwarded by Router 0 may arrive at the MH before older packets forwarded by Router 1 which might in turn arrive before even more older packets forwarded by the old base station. This scheme can also result in the creation of transient routing loops (for example, after old base station has changed its entry to forward packets but before the Router 1 processes Message3). However, note that the disordered streams and routing loops exist for short periods of time. The main benefit of this scheme is that it is simple and results in no loss. The BSs use a forwarding buffer for each MH in order to store the packets to be forwarded in the handoff procedure. All packets addressed to a MH are stored in the buffer (even after being transmitted to the MH). This allows that packets sent to the MH but lost because the MH moved out of coverage, will have the opportunity to reach the MH when forwarded to the new BS. Furthermore, the forwarding buffer is provided with a time out mechanism such that the buffer holds a packet only for a limited time period. When the path setup update message arrives at the old BS, all packets outstanding in the buffer for which the time out is not expired are forwarded to the new BS. 4.2.2 Non-forwarding path setup schemes In these path setup schemes, as the path setup message travels from the new BS to the old BS, data packets are diverted at the cross-over router to the new BS, resulting in no forwarding of packets from the old BS. There are two variants of the Non-Forwarding scheme, motivated by two types of wireless networks. The Unicast Non-Forwarding (UNF) scheme is optimized for networks where the MH is able to listen/transmit to two or more BSs simultaneously for a short duration, as in the case of a WaveLAN or Code Division Multiple Access (CDMA) network. The Multicast Non-Forwarding (MNF) scheme is optimized for networks where the MH is able to listen/transmit to only one BS as in the case of a Time Division Multiple Access (TDMA) network. In the following the UNF scheme analyzed in this paper is described. The UNF scheme is illustrated in Figure 4. In this case, when the new BS receives the path setup message, it adds a forwarding entry for the MH’s IP address with the outgoing interface set to the interface on which it received this message. It then performs a routing table lookup for the old BS then forwards Message 2 to Router 2. This router performs similar actions and forwards Message 3 to Router 0. At Router 0, the cross-over router in this case, forwarding entries is added such that new packets are diverted directly to the MH at the new BS. Eventually Message 5 reaches the old BS that then changes its forwarding entry and sends an acknowledgment, Message 6, back to the MH. Figure4 Hawaii path setup scheme UNF 5. Simulation Model To compare different handoff schemes performance, we implement both hard and semisoft handoff of Cellular IP and multiple stream forwarding (MSF) and unicast nonforwarding (UNF) of Hawaii. We used the Columbia IP Micro-Mobility Software (CIMS) [11] based on version 2.1b6 of network simulation [10]. The platform used for the simulation is shown in Figure 5. Figure 5 The Simulated network topology In this network the base station broadcast their beacons at interval of 1 s. The connection between wired nodes uses a 10 Mb/s duplex link with 2 ms delay. Mobile host connect to access point using the ns-2 carrier sense multiple access with collision avoidance (CSMA/CA). The results were obtained using a single mobile host moving between access points (AP1-AP4) at a speed of 20meter/s, the overlap area is 30 m. We compare these schemes for UDP and TCP applications with regarding to UbiCC Journal, Volume 3, Number 1, February 2008 42 delay, packet lost and throughput. During the simulation mobile host receives UDP packets transmitted by CH consist of 210 byte at 10 ms interval. The Reno congestion control was used for TCP with packet size 1460 bytes. The full time for the simulation is 70 second. 6 RESULTS buffers for storing and re-directing the packets from the old base station to the new base station. 6.1 UDP Performance While mobile host making handoff when moving between access points (AP1-AP4) and vice versa and during the simulation time the mobile host performs 6 handoffs in total, the first handoff occurs at 6.14 second, while the last one at 53.08 second. We use UDP probing traffic between the CH and MH and we observe for higher data rates. In each case, the same scenario applied: CBR (constant bit rate) traffic packet size was 210 bytes and the mobile host moved at a speed of 20 meters/second and the full time for the simulation is 70 Second. The number of packet losses for Hard and Semi-soft handoff schemes respectively. Under the Hard handoff scheme, 5 packets were lost during the handoff as shown in Figure 6, while in Semi-soft handoff under the same conditions, by contrast, no packets at all were lost because the mobile host able to listen to two base stations simultaneously for a short duration. In this way those packets which lost in hard handoff scheme during the establishment of new link are saved and arrive at the mobile host. Therefore as the mobile host moves faster it spends less time in the overlap region so the packet loss increases. The packets lost under MSF Hawaii were 5 packets during the handoff this is depend on the buffering as increasing the buffering time results in increasing the number of packets being buffered and forwarded until loss eliminated as in our simulation the buffering time is 5 ms, while under UNF results in 20 packets were lost. This lost depend of the packet size as the higher data rate the greater the packet loss. For UNF and Hard handoff the delay related to the packet delay between the APs and the cross-over node. 6.2 TCP Performance Handoff latency is proportional to the round-trip time between mobile host and Crossover router. When the mobile host handoff is completed over shorter handoff distances, the time in which packets can be lost is shorter, resulting in fewer packets lost in total compared to longer handoff distances. The delay for the four micro-mobility shown in Figure 7a and 7b, and from this figure we can observe that the MSF have the maximum delay. This increase in the delay time is due to the use of Figure 6 UDP Packet lost CIP-Hard handoff Figure 7a. Packet Delay Figure 7b. Packet Delay In the next experiment we study the impact of handoff performance on TCP throughput. The mobile host performs handoffs between AP1 and AP4 and via versa. The average TCP throughput at the mobile host for all handoff schemes is shown in UbiCC Journal, Volume 3, Number 1, February 2008 43 Figure 8 and Figure 8. We observe that the performance of TCP degrades as the handoff frequency increases, due to packet loss. As the handoff rate increases TCP has less time to recover from loss. This forces TCP to operate below its optimal operational point From Figure 8 we can observe that semi-soft handoff reduces packet loss and significantly improves the transport throughput in relation to the hard handoff scheme, this performance results associated with using the large buffer. In contrast to the semisoft handoff for UDP traffic, packet loss is not entirely eliminated with TCP. Even using semisoft handoff losses packet can occur, remember that the time between beacons are generated is 1 second. Therefore it can occur that MH receives the beacon from the new base station triggering the handoff after crossing the overlapping area. The advantage of MSF over UNF is that in Hawaii MSF the protocols tries to a voide losing the packets outstanding at the old base station when the connection is switched to the new base station. This is accomplished by forwarding these outstanding packets to the new base station. When using Hawaii with the UNF path setup scheme the MH is able to listen to both the new and the old base station. This scheme and CIP with semisoft handoffs have only small differences in the handoff procedure that is not relevant for their performance evaluation. 7 CONCLUSION Figure 8 TCP Throughput From Figure 7 we can observe that semi-soft handoff reduces packet loss and significantly improves the transport throughput. Figure 9 TCP sequence number From Figure 9 we can observe that semisoft handoff reduce packet loss, the Figure shows that the packet loss in the CIP hard handoff caused by the handoff results in a TCP timeout. The degradiation caused by packet loss increase with increasing handoff. As each handoff packets get lost when the MH switches between AP1 to AP4. In this paper, we discuss the handoff issue in Mobile IP thoroughly and analyze the key factors that affect MH’s handoff performance. The great the handoff distance, the more packet loss results in all handoff schemes in Cellular IP. If the handoff delay is less than the time that the mobile host spent in the overlap region then the Semi-soft handoff scheme has the best performance for Cellular IP protocol. In a Cellular IP network the number of hops from source to destination is constant since all routing-update packets must reach the gateway. However in Hawaii, nodes upper than Crossover router are not involved when handoff is happening. Therefore Hawaii is more reliable and, compared with Cellular IP, has less control signalling in the nodes upper than the Crossover router compared to Cellular IP. The longer handoff delay in the Hawaii handoff scheme causes greater packet loss than in Cellular IP handoffs. Comparing packet loss numbers for UDP applications in Cellular IP and Hawaii; at higher handoff rates the performance of both the forwarding and the nonforwarding schemes in Hawaii is worse than that of all handoff schemes in Cellular IP. The MH in semisoft must able to send route update packet to new station while listening to the old base station, whereas in MSF scheme the MH switches instantaneously from the old base station to the new base station. It is shown Cellular IP needs less buffers than Hawaii, and the Cellular IP semi-soft can eliminate packet loss but it may cause packet duplicated. Cellular IP employs special signaling even for mobile host while Hawaii keeps Hawaii signaling apart from mobile hosts. UbiCC Journal, Volume 3, Number 1, February 2008 44 References [1] P. Reinbold, O. Bonaventure, "IP micro-mobility protocols" IEEE Communications Surveys & Tutorials (2003) 40-56 [2] C.Perkins "IP Mobility Support for IPv4. RFC 3344, IETF Network Working Group, August 2002. Oct. http://www.ietf.org/rfc/rfc3344.txt [3] C. Blondia, O. Casals, P. De Cleyn and G.Willems, “Performance analysis of IP Micro-Mobility Handoff Protocols”, Proceedings of Protocols for High Speed Networks 2002 (PfHSN 2002), Berlin 2002, pp 211-226. [4] Ramjee, R.; Varadhan, K.; Salgarelli, L.; Thuel, S.R.; Shie-Yuan Wang; La Porta, T., “HAWAII: A domain-based approach for supporting mobility in wide-area wireless networks,” IEEE/ACM Transactions on Network,Volume:10, issue 3, June 2002 pp.396-410. [5] A. T. Campbell et al., "Comparison of IP Micromobility Protocols," IEEE Wireless Commun, vol. 9, no. 1, Feb. 2002. [6] A. Campbell, J. Gomez, C-Y. Wan, S. Kim, Z. Tur´anyi, A. Valk´o, "Cellular IP", Internet Draftietf-mobileip-cellular-00-txt, December 1999. [7] A. T. Campbell et al., "Design, Implementation, and Evaluation of Cellular IP," IEEE Pers. Commun., vol. 7, no. 4, Aug. 2000, pp. 42–49. [8] Mona Ghassemian; A. Hamid Aghvami “ Comparison different Cellular IP with Hawaii Handoff schemes”, 3G Mobile Communication Technologis,2002.Third International Conference on (Conf. Publ. No.389),8-10 May 2002 Page(s):52-57 [9] The Network Simulator – ns-2 home page, http://www.isi.edu/nsnam/ns [10] Columbia IP Micro-mobility Software: http://www.comet.columbia.edu/micromobility/ [11] J. S. M. Ho et al., “Mobility management in NextGeneration Wireless system”, Proc. IEEE, vol. 87, Aug. 1999, pp 1347-84. [12] J. S. M. Ho et al., “Mobility Management in Current and Future Communication Networks”, IEEE Network, vol. 12, Aug. 1998, pp.39-49. [13] D. Saha, et al, "Mobility Support in IP: A Survey of Related Protocols",IEEE Network, vol. 18, no. 6, Nov/Dec 2004. UbiCC Journal, Volume 3, Number 1, February 2008 45 OVERVIEW OF REFERENCE REGION GROUP MOBILITY MODEL FOR AD HOC NETWORKS Author: Diouba Sacko, Prof.Huang Benxiong and Prof. Wang Furong Postal Address: Department of Electronic and Information Systems, Communication Software and Switch Technology Center, Huazhong University of Science and Technology, Friendship apartment, room 411, Wuhan430074, Hubei, P.R. China. Phone: E-mail: 00-86-27-13871244974 sacko_dioba@hotmail.com ABSTRACT In this paper, we present and visit the limitation of reference point group mobility model. It assumes that nodes in the same group always stay together throughout the simulation process. However, in many real life applications, the nodes’s movement within a group is not always common. In particular, in a military operation, initially there is only one group. With multiple missions assigned to it, the group may be divided into a number of subgroups with each subgroup moving to a different location for accomplishing its task. A subgroup may be further divided into smaller groups or merge with other subgroups after completing its task. Therefore, in many scenarios it is necessary for a group to partition itself into smaller groups or a number of smaller groups to merge. Some recent researches present mobility models, which model possible group partitioning and group merging. We call this kind of mobility model a reference region group mobility model for ad hoc networks. Keywords: Group, partition, Merging, MANET, review. 1 INTRODUCTION random walk. One such model is the Random WayPoint mobility (RWP) model, which is the most Node mobility is one of the inherent characteristics of mobile ad hoc networks (MANET). It is also one of the parameters that most critically affect the performance of network protocols (e.g., routing). Today, in most simulation experiments, node movement is modeled as an independent popular mobility model used in the literature [2]. However, in real military scenarios, node mobility is not always independent. Mobility correlation among nodes is quite common. One typical example is group mobility. In battlefield, nodes with the same mission usually move in group such as tank Ubiquitous 1, February 2008 UbiCC Journal, Volume 3, NumberComputing and Communication Journal 1 46 battalions. For the modeling of military assets, group mobility models have drawn a lot of interest recently. The mobility models proposed so far in the literature assume some kind of permanent group affiliation. Also they require that each node belongs to a single group. In reality in a typical military scenario, a much more complex mobility behavior is observed. Some nodes move in groups; while others move individually and independently. Moreover, the group affiliation is not permanent. The mobile groups can dynamically re-configure themselves triggering respectively. Section 10 analyzes the impact of mobility model on the performance evaluation of various routing protocols. Conclusions and References appear in sections 11 and 12, respectively. 2 REFERENCE POINT GROUP MOBILITY MODEL: RPGM Group movements are based upon the path traveled by a logical center for the group. The logical center for the group is used to calculate group motion via a group motion vector, GM. The motion of the group center completely characterizes the movement of its corresponding group of Mobile Nodes (MNs), including their direction and speed. Individual MNs randomly move about their own predefined reference points, whose movements depend on the group movement. As the individual reference points move from time t to t+1, their locations are updated according to the group’s logical center. Once the updated reference points, RP (t+1), are calculated, they are combined with a random motion vector, RM, to represent the random motion of each MN about its individual reference point [6]. Figure 1 gives an illustration of three MNs moving with the RPGM model. The figure illustrates that, at time t, three black dots exist to represent the reference points, RP (t), for the three MNs. The RPGM model uses a group motion vector GM to calculate each MN’s new reference point, RP (t+1), group partition and mergence. All these different mobility behaviors coexist in military scenarios. A good realistic mobility model must capture all these mobility dynamics in order to yield realistic performance evaluation results, which, unfortunately, is not satisfactorily captured in any of the existing models [1]. In this paper, we present group mobility model, which includes all these “heterogeneous” mobility behaviors. We discuss in section 2, group mobility model; called Reference Point Group Mobility model RPGM. It assumes that a group of nodes always move together [10]. Section 4 presents the Reference Region Group Mobility (RRGM) model, which models possible group partitioning and group merging [3, 4].. The remainder of the paper is organized as follows. Section 5 presents a group partition and merging processes. Sections 6, 7, and 8 provide firefighters operating in a building, room searching or exhibition hall visiting, and battlefield, Ubiquitous 1, February 2008 UbiCC Journal, Volume 3, NumberComputing and Communication Journal 2 47 at time t+1; as stated, GM may be randomly chosen or predefined. The new position for each MN is then calculated by summing a random motion vector, RM, with the new reference point. Figure 2 is an illustration of three MNs moving together as one group. The movement of the logical center and the random motion of each individual MN within the group are implemented via the RWP mobility model. One difference, however, is that individual MNs do not use pause times while the group is moving. Pause times are only used when the group reference point reaches a destination and all group nodes pause for the same period of time [9]. group reference point reaches a destination and all group nodes pause for the same period of time [9]. The RPGM model was designed to depict scenarios such as an avalanche rescue. During an avalanche rescue, the human guides tend to set a general path for the dogs to follow, since they usually know the approximate location of victims. The dogs each create their own “random” paths around the general area chosen by their human counterparts [9]. The RPGM model can generate topologies of ad hoc networks Figure 1: Movements of three MNs using the with group-based node mobility for 3 DISCUSSION Figure 2: Traveling pattern of one group (three MNs) using the RPGM model simulation purposes, but for mobility or partition prediction purposes, it has two disadvantages. First, this model is used in the scope of an omniscient observer or a God, where the complete information about the mobility groups including their member nodes and movements are known. Given the distributed RPGM model Ubiquitous 1, February 2008 UbiCC Journal, Volume 3, NumberComputing and Communication Journal 3 48 nature of the ad hoc network, such global information about the mobility groups are not conveniently available to any mobile nodes at run-time. For example, a mobile user traveling to a destination does not know all the other users that are heading in the same direction. Therefore, the lack of prior knowledge about the mobility groups make the RPGM model waiting for the arrival of others. After a reference region has been stationary for some time at an intermediate location, a new location for the reference region will be generated. As such, the reference region moves gradually towards the destination with its path defines the trajectory of the movement of the group. The size of the region is defined based on the node density as given by the user according to the specific scenario. In RRGM, new destinations may be created at times so that if multiple destinations are assigned to a group, this group will be partitioned into a number of smaller subgroups, each with a new reference region associated to a different destination. When a group has reached its destination, the group could merge with another group. RRGM also defines two group types: active groups and standby groups. Active groups are those that have destinations assigned to them and nodes are actively either moving toward their reference region or moving within the regions. Whereas standby groups have no destination assigned yet and nodes only move within the stationary reference regions. The standby groups model situations where some groups inapplicable for run-time partition prediction. Second, the RPGM model represents the mobile nodes by their physical coordinates. Given only the instantaneous physical locations of the nodes, it is difficult to discern the nodes’ group movement patterns and the trend in the network topology changes [6]. Moreover, because the RPGM model is based on RWP model, it still cannot overcome the shortcomings caused by the characteristics of the RWP model, such as non-uniform network density, and it is not adequate to simulate the group movement in reality, such as group partition and mergence. Thus, several other mobility models such as RRGM model were proposed. We shall discuss this model in this paper. 4 REFERENCE REGION GROUP MOBILITY MODEL are waiting for their task assignments or where nodes have reached the destination and are waiting for a new task [3]. Two group-partitioning modes have been In this section, we present Reference Region Group Mobility (RRGM) model. In this model, every group is associated with a reference region which is an area that nodes will move towards to a once they arrive, the nodes will move around within the region designed: 4.1 Group partition when a new destination is generated (First mode) In some applications it is necessary for a group to partition itself into a number of smaller Ubiquitous 1, February 2008 UbiCC Journal, Volume 3, NumberComputing and Communication Journal 4 49 groups to accomplish different tasks at different locations. For instance, when an army unit is moving towards an enemy’s citadel, a command is received that a team of soldiers has to be separated from the main force to accomplish another task. A new team would then be formed and partitioned from the current team. To support group partitioning in RRGM, new destinations will be generated and placed at some time interval as specified by the users. Once a new destination is generated, the distance from the destination to every standby group is calculated. Again, the closest standby group is selected and becomes active and will move towards the destination. If no standby group exists, the active group that is closest to the new destination is chosen, and a number of nodes are randomly selected to form the new group. Thereafter, a new reference region is generated between the original group and a newly created destination. Members of the newly formed groups will than change their directions and move towards the new reference regions. To ensure each group has a minimum number of nodes, a threshold nmin, this group cannot be chosen for partition. In RRGM, if a group has reached its destination for some time, the group will become a standby group and will merge with another group. Two conditions need to be satisfied before a group could merge into other groups. Firstly, the number of nodes in the standby group is less than nmin. This is to ensure that we have either two small groups merge with each other or a small group merges into a large group. Secondly, the group has paused at the destination for a period of time τ as specified by users. This is to ensure that the nodes have spent some time at the destination to complete their assigned tasks before the group becomes a standby group. Once the two conditions are met, the group will select the nearest reference region as its new reference region, and its nodes become members of the target group [4]. 4.2 Group partition when a group passes by a destination (Second mode) The second mode of group partitioning is useful in scenario such as building search where locations of the destinations (e.g. rooms) are in general predefined by the user. Under this mode of operations, generating a reference region for each destination will not initialize the model. Instead, only one reference region for the whole group will be created initially. A set of coordinates pairs {(dx1, dy1), (dx2, dy2)… (dxk, dyk)} will be used to define the intermediate checkpoints for the path of the reference region. Such checkpoints represent turnings in a building where the group may turn left or right to move into another corridor. The initial reference region will be placed along the path between the initial group position and the first checkpoint [4]. 5 GROUP PARTITION AND MERGING Ubiquitous 1, February 2008 UbiCC Journal, Volume 3, NumberComputing and Communication Journal 5 50 Figure 3 shows us a general group mobility scenario where a group may partition and merge. newly formed subgroup moves towards D4 as shown in figure 3(c). At time 20, the biggest group on the right side in figure3(c) has arrived its destination and became a standby group, while other subgroups are still moving towards their destinations. Figure3 (d) to (f) illustrate the process of mergence. Figure 3(d) shows that the two smaller groups are standby groups while the third one is an active group moving toward the destination D. In figure3(e), one of the smaller standby groups starts to merge into its nearest reference region, and the merging is completed at time 85 as shown in figure3(f). The scenario given above can be used to model application scenario such as search and rescue. Destinations represent the areas where rescue teams move towards the destinations, some members may be called upon to provide help in other areas. Another application Figure 3: General Group Mobility Pattern with Group partition and merging. As shown in figure3 (a), initially at time 0, for the three destinations, D1, D2 and D3, three reference regions are generated. The initial group is partitioned into three subgroups and they is to model battlefield scenario where a number of enemies’ defenses are deployed around. After the units get to their destinations and finish their tasks, they may reassemble again and be deployed to other areas [4]. gradually move into their corresponding reference regions. Figure 3(b) shows that at time 15, while the groups are moving towards their destinations, a new destination D4 has been generated. The closest subgroup, which is moving towards D2, is now partitioned into two subgroups with the 6 FIREFIGHTERS OPERATING IN THE BUILDING As firefighting agencies become more advanced, they are using sophisticated location determining, tracking and communications systems Ubiquitous 1, February 2008 UbiCC Journal, Volume 3, NumberComputing and Communication Journal 6 51 that are often based on packet radio networks. Firefighting teams themselves are typically small elements of not more than five firefighters, operating in concert with other small teams as they enter buildings and attack the fire. Group structure and control is critical. Individual nodes stay fairly close together in this scenario, but barriers and node failure can easily lead to link breakages that will stress the routing protocol. It is also common for two members to break off from the group to clear a room or search an obscured area, for example. Figure 4 depicts a typical tactic employed by firefighting teams, wherein a command element of a team stations itself at the entrance to a room and a smaller clearing team moves through the room to search for fire and victims [7]. The destinations shown on the two sides of the figure 5 represent rooms or exhibition counters. During a building search, the police officers will move along the corridor, and a small team will be formed to search the rooms as they pass by. After searching a room, the team will join back the main force to move toward. Similarly, in an exhibition hall, delegates from a company may gather together when they enter an exhibition hall. When the group passes by a counter that some may be interested in, the small group may visit the counter while others may continue to walk forward. After visiting a counter for a while, the members will rejoin the main group again. The circles with arrows indicate the movement direction of each subgroup [4]. Figure 4: Firefighting team in a building: clearing a room Figure 5: Building search 7 ROOM SEARCHING OR EXHIBITION HALL VISITING 8 BATTLEFIELD Ubiquitous 1, February 2008 UbiCC Journal, Volume 3, NumberComputing and Communication Journal 7 52 During battlefield planning, topographical teams and support staff are responsible for conducting thorough terrain analyses to support commanders in battlefield planning. This analysis can range from elevation calculations and wireless ad hoc networks. As ad hoc network is most likely to be deployed to support group communication, such as in search and rescue, battlefield operations, etc., it is very unlikely that the mobile nodes will move around independently. Furthermore, in-group operations, groups may frequently sub-divide or merge whenever necessary. As most mobility models fail to describe such mobility patterns, our mobility model attempts to provide a better reflection of the group movement pattern with group partition and mergence. Examples have been provided to illustrate the applications of the model for different scenarios. With this mobility model, the effectiveness and the efficiency of group communication routing protocols could be evaluated under a more realistic environment. There are a number of ways to extend this initial work. The first of these relates to the size of coverage region. By using the density-based approach, our model can control the size of the region to be covered by a group. Density-based routing is of particular interest in mobile and unstable networks. In mobile networks, the closest node might leave or move to another location. In such scenarios, density-based routing increases the probability of successful packet specifications of restricted and unrestricted terrain, to soil and vegetation data depending upon the specific needs of the commander and the battle situation. The commander’s task of terrain analysis for the purpose of battlefield planning is usually two fold: 1) the analysis of the military aspects of the terrain, and 2) evaluation of the terrain’s effects on military operations. On the battlefield, RRGM model is very useful. Each vehicle or in some cases each soldier represents a node in a larger tactical internet. Military units are fundamentally hierarchical, and they deploy, move and operate in groups that display tight adherence to a group structure that is known a priori [8]. Many other application scenarios, such as a fleet of warships or fighter planes in a combat maneuver, can also be modeled using RRGM. As such, all nodes will move within the area based on the random waypoint mobility model. 9 DISCUSSION delivery. This work can also be improved through further investigation Network on network disconnect causes the In this section, we have discussed a Reference Region Group Mobility model that is used in the description of group movement in mobile prediction. disconnection network to separate into completely disconnected portions. It is a widescale topology change that can Ubiquitous 1, February 2008 UbiCC Journal, Volume 3, NumberComputing and Communication Journal 8 53 cause sudden and severe disruptions to on-going network routing and upper layer applications. Using this model, we can predict the future network partitioning, and thus minimize the amount of disruptions. Finally, according to the fact that multicasting, in general, works well if the density of group members is sparse and in low mobility, this work can be improved through multicast routing based on cluster formation information in-group communications. when the node mobility is high. As a result, such invalid route information will cause the generation of route errors and initiate new route requests resulting in the relatively higher overhead than AODV as show in figure 7. It is worth noting that the amont of control packets generated by DSR under RRGM is much less than that under RWP, as paths generated for intra-group and inter-group communications for RRGM will mostly likely remain valid as long as the groups are not partitioned. Figure 8 shows that the end-to-end delay of DSR under RRGM is lower than 10 THE IMPACT OF MOBILITY MODEL that under RWP. Again, the lower delay is achieved with the possible intra-group communications and less control packets being generated under RRGM. It has been shown that mobility Similarly, figure 9 shows that DSR has a smaller patterns can affect the performance of ad hoc jitter under RRGM. On the other hand, the end-tonetwork routing protocol significantly. In this section, end delays and jitters of AODV under the two we will evaluate the performance of two routing models do not differ significantly. This illustrates protocols, AODV and DSR, under the Random that AODV performs rather stable under different WayPoint mobility model and the Reference Region environment and is not very sensitive to group Group Mobility model. The performance metrics physical changes. Note that as velocity increases, the collected include packet delivery ratio, average jitter of DSR is much greater than that of AODV. control packets per data packet delivered, end-to-end Figure 10 shows that when the group density is low, delay and average jitter. As shown in figure 6, as nodes are moving randomly around in a larger region speed increases, the packet delivery ratio for RRGM and DSR performs badly. The performance of DSR degrades rapidly for both AODV and DSR as group improves partitioning occurs more frequently. For RWP, information in the route cache will remain valid for a DSR’s performance deteriorates rapidly as speed longer period of time with the area covered a group increases as DSR relies on the information stored in reduces. However, with further reduce in the group the route cache that may become invalid very soon coverage area; the overlapping area among groups is as the density increases because Ubiquitous 1, February 2008 UbiCC Journal, Volume 3, NumberComputing and Communication Journal 9 54 reduced resulting in group partitioning. Hence, the packet delivery ratio reduces as group density further increases. As AODV does not rely on the cache information, it manages to achieve a higher delivery ratio. Similarly, figure 11 shows that the end-to-end delay of DSR decreases as density increases initially. This is because at low density, the overlapping area among groups is so large that even intra-group communication may employ members from other groups’ as relays and the lifetime of routes constructed with nodes from different groups would not last long. As a result, the end-to-end delay at low group density is high. As the group density increases, the overlapping area becomes smaller and shorter routes for intra-group communication are more readily available resulting in the decrease in delay. However, with further increase in density, transient network partition occurs frequently resulting in a graduate increase in delay. On the contrary, AODV is not affected much by the change in density and the end-to-end delay is stabilized at a low value. Although AODV out performs DSR in the studies showed here, we can see that under RRGM, the difference in performance between DSR and AODV is not as drastic as in the case of RWP. With nodes moving in a smaller region covered by a group, the cached information kept by DSR remains valid for a longer while. Furthermore, if the group density is high, using DSR for intra-group communication will even outperform AODV [5]. Figure 8: End-to-end delay vs. speeds Figure 7: Average control packet overhead vs. speeds Figure 6: Packet delivery ratio vs. speeds Ubiquitous 1, February 2008 UbiCC Journal, Volume 3, NumberComputing and Communication Journal 10 55 protocol can vary significantly due to the selected mobility model. It should be evaluated with the mobility model that most closely matches the expected real life system. Over the years, a number of group mobility models have been proposed for ad hoc networks. Most of them such as Reference Point Group Mobility model, model the movement of preFigure 9: Average jitter vs. speeds defined groups, where nodes in the same group always stay together throughout the simulation process. Such models fail in modeling scenarios where groups may be partitioned and merged those are most likely to be found in ad hoc networks. These kinds of application scenarios can be found in search and rescue operations, conference seminar sessions, and conventional events. In this paper, in section 4 we presented RRGM model, which Figure10: Packet delivery ratio vs. node density provides a better reflection of group movement behavior with possible group partition and mergence. Section 5 shows a group partition and merging processes. Some practical applications of RRGM model such as firefighters operating in a building, room searching or exhibition hall visiting, and battlefield are provided in sections 6, 7, and 8 respectively. In section 10 we have shown how two Figure 11: End-to-end delay vs. node density typical ad hoc routing protocols, AODV and DSR, perform in a group environment. From the simulation results, we see that AODV performs 11 CONCLUSIONS better than DSR in general, and for AODV, less data packets are delivered and more control packets are The performance of an ad hoc network required under frequent network partitioning. Ubiquitous 1, February 2008 UbiCC Journal, Volume 3, NumberComputing and Communication Journal 11 56 ACKNOWLEDGEMENT This work was supported by national natural science foundation of China under grant No.60572047. [5] Jim M.Ng and Yan Zhang. Impact of Group Mobility on Ad hoc Networks Routing Protocols. ICACT 2006, ISBN 89-5519-129-4. [6] Karen H. Wang and Baochun Li. Group Mobility and Partition Prediction in Wireless Ad-Hoc 12 REFERENCES Networks. Department of Electrical and Computer Engineering, University of Toronto. 0_7803-7400- [1] Biao Zhou, Kaixin Xu, and Mario Gerla. Group and Swarm Mobility Models for Ad Hoc Network Scenarios Using Virtual Tracks. MILCOM 2004 – 2004 IEEE Military Communications Conference, 0_7803_8847_X/04/$20.00 2004 IEEE. 2/02/$17.00 2002 IEEE. [7] Ken Blakely and Bruce Lowekamp. A Structured Group Mobility Model for the Simulation of Mobile Ad Hoc Networks. MobiWac’o4 Philadelphia, Pennsylvania, USA. ACM 1-58113-920- [2] J. Broch, D.A.Maltz, D.Johnson, Y. –C. Hu, and J. Jetcheva. A Performance Comparison of MultiHop Wireless Ad Hoc Network Routing Protocols. In Proceedings of MobiCom’98, Dallas, Texas (1998). [3] Jim M.Ng and Yan Zhang. Reference Region Group Mobility model for Ad hoc Networks. School of Electrical and Electronic Engineering, Nanyang 9/04/0010…$5.00 (2004). [8] Rachel Banks and Christopher D. Wickens. Commanders’ Display of Terrain Information: Manipulations of Display Dimensionality and Frame of Reference to Support Battlefield Visualization. Technical Report ARL-97-12/ARMY-FED-LAB-972 (1997). [9] Tracy Camp, J.Boleng, and V. Davies. A Survey Technological University. Nanyang Ave., Singapore of Mobility Models for Ad Hoc Network Research. 639798, 0_7803_9019_9/05/$20.00 2005 IEEE. Wireless Communication & Mobile Computing (WCMC): Special issue on Mobile Ad Hoc Networking: Research, Trends and Applications, Vol.2, no.5, pp.483-502 (2002). [10] X. Hong, M. Gerla, G. Pei, and C. Chiang,. A Applications 0_7695_2316_1/05$20.00 (ICITA’05), group mobility model for ad hoc Wireless networks, 2005 IEEE (2005). Proceedings of the ACM International Workshop on [4] Jim M.Ng and Yan Zhang. A Mobility Model with Group Partitioning for Wireless Ad hoc Networks. Proceedings of the Third International Conference on Information Technology and Ubiquitous 1, February 2008 UbiCC Journal, Volume 3, NumberComputing and Communication Journal 12 57 Modeling and Simulation of Wireless and Mobile Systems (MSWiM) (1999). Ubiquitous 1, February 2008 UbiCC Journal, Volume 3, NumberComputing and Communication Journal 13 58 Ubiquitous Computing and UbiCC Journal, Volume 3, Number 1, February 2008 Communication Journal 2 59 Medical Care System Using VORD Methodology Mr. Fiaz Ahmad Dr. Mohamed Osama Khozium Assistant Lecturer Assistant professor Faculty of Information Technology MISR University for Science & Technology, 6th of October City, EGYPT Osama@Khozium.com fiaz.ahmad@yahoo.com ABSTRACT Is technology serving the humanity? Current IT developments are affecting the every walk of life, of the human-being of this planet. As the computer technology is impacting very valuable effects on today’s life, it’s also playing a very positive and beneficial role in the field of medical. This paper brings to light, how the information technology can play a vital role in saving the precious human-lives, as well as how IT can help out the people in any emergency situation in their homes, before reaching to the hospitals. The paper describes the design and implementation phases of the” Medical Care System” (MCS). It reveals the momentous aspects of computer technology and its development effects on today’s society. The proposed system also brings to light that how the computer technology is mounting the luxuries of today’s life by introducing new amazing aspects in every field of life. The system is developed to help out the people in a rapid, easy and a cheap way, when they faced any emergency situation in their homes, like “Asphyxia and Obstruction of Air Passages”, Bites and Stings, Electricity Shock etc. The system also has a unique interface, that can help out to any individual, that how to give “First Aid” on the road in an accident situation, by using a cellular phone. The successful implementation of this system can play a vital role in saving human-lives as well as to lend a hand to the people in a very rapid way in any emergency situation, on their doorsteps. Keywords: Portability, Litheness, Fault Tolerant, Asphyxia and Obstruction of Air Passages, Internet Security, Software Standards. 1. INTRODUCTION Information technology continuous advancements have opened the number of constructive possibilities in our today’s life that were a trance in the past. Now a day’s computerized system are playing a very valuable role in every walk of life. Whether it’s an educational field, business or entertainment side, no one can deny from their priceless contribution. Computer technology is also playing a vital role in the medical field and is caused in saving, millions of precious human-lives. “Medical Care System” (MCS) is also a contribution from the computer technology in saving the humanity. “First-Aid”, immediate and temporary treatment of a victim of sudden illness or injury while awaiting the arrival of medical aid. Proper early measures may be instrumental in saving life and ensuring a better and more rapid recovery. The avoidance of unnecessary movement and over-excitation of the victim often prevents further injury. Conditions that require immediate attention to avert death include cessation of breathing (asphyxia), severe bleeding, poisoning, strokes, and heart attack. The essentials of first aid treatment also include the correct bandaging of a wound; the application of splints for fractures and dislocations; the effective methods of cardiopulmonary resuscitation (CPR) and artificial respiration; and treatment of shock, frostbite, fainting, bites and stings, burns, and heat exhaustion [1]. An emergency medical technician-paramedic is a licensed and/or certified out-of-hospital health-care provider. EMTs represent the uppermost level of prehospital health care providers and serve as managers of pre-hospital treatment teams. They work under the direction of a physician—often by two-way radio—to evaluate and manage acutely ill or injured patients in ambulance services or other life-support units [2]. UbiCC Journal, Volume 3, Number 1, February 2008 60 First- Aid awareness is much more essential in today’s life. People have to face so many emergency situations in their daily lives, like Burns, Drowning and NearDrowning, Fainting, Foreign Body in the Eye, Fractures and Joint Injuries, Frostbite, Heat Exhaustion and Heatstroke etc. All these symptoms required immediate, proper and correct attention, because proper early measures may be instrumental in saving life and ensuring a better and more rapid recovery. By getting data facts, visiting different hospitals, it comes to know, that most of the cases are in a very bad situation due to the absence or inaccurate First-Aid procedure. So many death cases occur due to the unavailability of the medical treatment on time. “Medical Care System” (MCS) is an attempt to provide the awareness of the “First-Aid” to the community in an easy, cheap and rapid way, at their door steps. People can interact with the system, simply by connecting with the internet and the system will show them “First-Aid” procedures for different situations that required medical treatment, in various forms like written instructions as well as visual representation, “how to provide “first-aid” to the victim by visualization”. The people can access the experiences and guidelines of the specialist and experienced doctors 24 / 7, for any emergency situation. The system will also provide a unique interface for the cellular phone, so that any individual can access the system, to provide “first aid” on the road during a road accident. The system will also facilitate the people by giving the facility for connecting with the specialist doctor for the advices in any worst situation while they are unable to get satisfactory information from the system for the specific patient. A successful implementation of the System can improve the image of the hospital, doctors as well as catch the attention of more patients. 2. MCS ANALYSIS AND DESIGN This section is designed to give an idea about the analysis and design of the proposed system. It starts with introduction to the requirements elicitation and analysis for MCS. Viewpoint-Oriented Requirements Definition (VORD) method is explained and used to analyze MCS. 2.1. MCS Analysis Requirements elicitation and analysis is the next stage after the initial feasibility studies. In this activity, we work with the proposed system end-users to find out the application domain, what services the system should provide, the required performance of the system, hardware constraints, and so on. Requirements elicitation and analysis may involve a variety of different kinds of people (Stake-holders) in the application. Stake-holders include end-users who will interact with the system and everyone else in an organization which will be affected by it. Elicitation and analysis is a difficult process for a number of reasons: Stakeholders often don’t really know what they want from the computer system except in the most general terms; they may find it difficult to articulate what they want from the system. They may take unrealistic demands because they are unaware of the cost of their requests. Stakeholders in a system naturally express requirements in their own terms and with implicit knowledge of their own work. Requirements engineers, without experience in the customer’s domain, must understand these requirements. Different stakeholders have different requirements and they may express these in different ways. Requirements engineers have to discover all potential sources of requirements and discover commonalties and conflicts. The economic and business environment in which the analysis takes place is dynamic. It inevitably changes during the analysis process. Hence the importance of particular requirements may change. New requirements may emerge from new stakeholders who were not originally consulted. The VORD (Viewpoint-Oriented Requirements Definition) method has been chosen as an activityoriented framework for MCS elicitation and analysis [3], [4]. 2.2 VORD (Viewpoint–Oriented Requirements Definition) Method For any medium sized or large systems, there are usually different types of end–user. Many stakeholders have some kind of interest in the system requirements. Different viewpoints on a problem see the problem in different ways. However, their perspectives are not completely independent but usually overlap so that they have common requirements. A key strength of view point – oriented analysis is that it recognizes the existence of multiple perspectives and provides a UbiCC Journal, Volume 3, Number 1, February 2008 61 framework for discovering conflicts in the requirements proposed by different stakeholders VORD method considered viewpoints as a receiver of services. In this case, viewpoints are external to the system and receive services from the system [3]. Viewpoints may provide data for these services. The analysis involves examining the services received by different viewpoints, collecting them and resolving their conflicts. Interactive systems deliver services to end-users. Consequently, the most effective viewpoint – oriented approach for interactive systems analysis uses external viewpoints. These viewpoints interact with the system by receiving services from it and providing data to the system. 2. 2.1 The Advantages of VORD Method Viewpoints are external to the system so they are a natural way to structure the requirements elicitation process. It is relatively easy to decide if something is a valid viewpoint. Viewpoints must interact with the system in some way. Viewpoints and services are useful ways of structuring non-functional requirements. Each service may have associated non-functional requirements. Multiple viewpoints allow the same services to have different non-functional requirements in different viewpoints. The VORD (viewpoint – oriented requirements definition) method [4] has been designed as a service-oriented framework for requirements elicitation and analysis. 2. 2.2 The Principle Stages of the VORD Method As shown in figure 1 the principle stages of the VORD method are viewpoint identification, structuring, documentation and mapping. Viewpoint identification, which involves discovering viewpoints that receive system services and identifying the specific services provided to each viewpoint. Viewpoint structuring, which involves grouping related viewpoints into a hierarchy. Common activities are provided at levels in hierarchy and are inherited by lower-level viewpoints. Viewpoint documentation, which involves refining the description of the identified viewpoints and services. Viewpoint-system mapping, which involves identifying, objects in an object-oriented design using services information, which is encapsulated in viewpoints. V iewpoints identified V iewpoints structuring Viewpoint system documentation Viewpoint mapping Figure 1 The Principle Stages of The VORD Method UbiCC Journal, Volume 3, Number 1, February 2008 62 2.3 Viewpoints in MCS 2.3.1The Victim This is the person who is endured in sudden illness or injury. 2.3.2 Paramedic He is usually the key person who provides “first aid” in any emergency situation. In any sudden case, this person will interact with the system and try to find out the proper and accurate first aid for any injury or illness. 2.3.3 Hospital Staff (Nurse/Receptionist) It will be the hospital staff who is interacting with the system and facilitating the people who want to get “first aid” information, about any sudden case. He or she is also responsible to establish conversation between a specialist doctor and a first aid provider, if he/she failed to get satisfactory material about any special case from the system or he tried the present first aid methods but the victim did not get any pleasing results. 2.3.4 Specialist doctor He is a specialist doctor who is responsible for the medical treatment of any sudden case, or to provide online help to any “first aid” provider. 2.3.5 System Administrator System Administrator is responsible for the maintenance and the technically issues related to the system, and he is responsible to make sure the availability of the system to the users, 24/7. 2.4 Viewpoints Structuring The above mentioned viewpoints can be grouped and structured in a hierarchal form which can represent the activities for each viewpoint. The structure is given below in figure 2. All VPs Hospital staff Activities *Diagnoses & Advices Activities Specialist Dr System Admin Activities *Maintenan ce & technical support * interact with the system, arrange conversation with doctor, F-aid person Nurse/Recep tionist Paramedic Activities * Interact with system and find first-aid methods.. Figure 2 Viewpoints Hierarchy UbiCC Journal, Volume 3, Number 1, February 2008 63 2.5 MCS Events Sequence To show how the proposed system could interact with its stakeholders we present the events sequence activity diagram indicated in figure 3 by using Unified Modeling Language (UML) [5]. Interaction with MCS Get satisfied information Need more information Satisfy Need medical treatment To hospital Figure 3 Events Sequence diagram. UbiCC Journal, Volume 3, Number 1, February 2008 64 3. DESIGN AND IMPLEMENTATION 3.1 Development Environment Java is a programming language that is well suited for designing such type of software that work in conjunction with the internet [6]. Additionally it’s a cross platform language, which means its program can be designed to run the same way on Microsoft Windows, Apple Macintosh and most versions of UNIX, including Solaris. Java extends beyond desktops to run on devices such as televisions, wristwatches, and cellular phones as it is small, secure, and portable [7]. Java’s strength include platformindependence, object oriented nature, as well as easy to learn. [8]. Due to the above mentioned powerful features of the java programming language, it is desired language for the development of the proposed system. Furthermore, java has JSP (Java Server Pages), Struts, EJBeans (Enterprise Java Beans), like dominant technologies that create attraction for the development of distributed web applications. 3.2 Structure of the System The proposed system is a distributed web application, containing three modules. 1. Web Application 2. Cellular Phone Application 3. Desktop Application ( Server Side Application ) Struts are used as architecture for the said system, which is famous model view controller pattern. Apache Struts is an open-source web application framework for developing Java EE web applications. It uses and extends the Java Servlet API to encourage developers to adopt a modelview-controller (MVC) architecture. It was originally created by Craig McClanahan and donated to the Apache Foundation in May, 2000. [9]. Through the web application of the system the Paramedic can interact with the system and can find the first-aid methods according to his/her needs. In a standard Java EE web application, the client will typically submit information to the server via a web form. The information is then either handed over to a Java Servlet which processes it, interacts with a database and produces an HTML-formatted response, or it is given to a Java Server Pages (JSP) document which intermingle HTML and Java code to achieve the same result. Both approaches are often considered inadequate for large projects because they mix application logic with presentation and make maintenance difficult. The goal of Struts is to cleanly separate the model (application logic that interacts with a database) from the view (HTML pages presented to the client) and the controller (instance that passes information between view and model). Struts provide the controller (a servlet known as ActionServlet) and facilitate the writing of templates for the view or presentation layer (typically in JSP, but XML/XSLT and Velocity are also supported). The web application programmer is responsible for writing the model code, and for creating a central configuration file struts-config.xml which binds together model, view and controller. [9]. EJBeans (Entity Java Beans) are used an application layer between browser and data base. Enterprise JavaBeans (EJB) technology is the server-side component architecture for Java Platform, Enterprise Edition (Java EE). EJB technology enables rapid and simplified development of distributed, transactional, secure and portable applications based on Java technology. [10]. The Enterprise Java Beans (EJB) 3.0 specification vastly improves the simplicity of programming enterprise beans. This promises to increase your productivity as a developer. [11]. The cellular phone application is developed using J2ME (Java to Micro Edition) to facilitate any individual who wants to interact with the system using cellular mobile phone in any accident situation on the road. That is basically a Midlet and data moved from Midlet to JSP and from JSP to EJBeans (inside application server which is Bea Web Logic) and then to the database. The basic functionality is to display a unique interface on a constrained memory and user interface cellular device. The desktop application (server side application) that is communicating with the database through Bea Web Logic, which is an application server for sending and retrieving data from the data base. 4. CONCLUSION The design and development phases of the proposed system (MCS) are described in this paper. The paper brings to light the salient features of the system. MCS is a very strong idea in the medical field. Currently available online “first-aid” information is not according to the specific need and not to the point according to the specific situation e.g. the paramedic has to UbiCC Journal, Volume 3, Number 1, February 2008 65 spend too much time to find any particular info about any sudden case. The proposed system will be developed according to the needs of the community. The system will provide the exact first-aid information about any sudden situation, so that any individual could get completely to the point information. There will some visual presentations, “that how to provide first-aid in particular situation” so that any individual can see and can act according to that. One of the salient features of the proposed system is that, it will provide a facility to the people to communicate with the specialist doctors, if they are unable to get satisfactory results after acting upon the advices that were provided by the system. The successful implementation of the system can play a vital role in saving the human-lives as well as improving the image of any hospital and the medical professionals. REFERENCES: [1] http://www.answers.com/first%20aid , Encyclopedia. [2] http://www.healthline.com/galecontent/ emergency-medical-technicians, mergency Medical Technicians Health Article. [3] Kotonya, G. and Sommerville I., Requirements Engineering with Viewpoints. BCS/IEE software Enineering, J., Ch6, 1996. [4] Kotonya G. and Sommerville I., Requirements Engineering: Process and Techniques, Chichester, UK, Chs 5,6, 1998. [5] Dittman W. Bentley, System Analysis and Design Methods 6th Edition, Irwin/ McGraw-Hill, 2004. [6] Java web site, Sun Microsystems, java.sun. com, 2007. [7] Newman A., A Special Edition Using Java, Indianapolis, IN, Que Corporation, 1996. [8] Horstmann Charlie., Core Java 1.2, Sun Microsystems’s Press, California, 1999. [9] ApacheStruts.Wikipedia, http://en.wikipedia. org/ wiki /Apache_Struts, 2007. [10] Enterprise JavaBeans Technology. http://java. sun.com/products/ejb/index.jsp, 2007. [11] Article, Writing Per formant EJB Beans in the Java EE 5 Platform,http://java.sun.com/ developer / technical Articles / ebeans / ejb_30 / index.html, 2007. UbiCC Journal, Volume 3, Number 1, February 2008 66 1 Quality of Service Provisioning for Multimedia Transmission over UWB Networks N. El-Fishawy, M. Shokair, and W. Saad El-Menoufia University, Faculty of Electronic Engineering, Communication Department Email: nelfishawy@hotmail.com, {i_shokair, waleedsaad100}@yahoo.com Abstract— In this paper, the Quality of Service (QoS) for multimedia traffic of the Medium Access Control (MAC) protocol for Ultra Wide-Band (UWB) networks is investigated. A protocol is proposed to enhance the network performance and increase its capacity. This enhancement comes from using Wise Algorithm for Link Admission Control (WALAC). The QoS of multimedia transmission is determined in terms of average delay, loss probability, utilization, and the network capacity. In addition, a new parameter is aroused for the network performance. Index Terms— Ultra Wide Band, Medium Access Control, resource allocation, and Quality of Service. traffic using the proposed protocol are made in Section IV. Finally, conclusion will be shown in Section V. II. RESOURCE ALLOCATION I. INTRODUCTION UWB is a technology for transmitting information spread over a large bandwidth (>500 MHz) under the right circumstances. A February 2002 Report and Order by the Fedral Communication Commission (FCC) [1] authorizes the unlicensed use of UWB in 3.1–10.6 GHz. This is intended to efficiently usage of exceptional radio bandwidth while enabling both low and high data rates. The FCC defines UWB signal as the emitted signal bandwidth exceeds the lesser of 500 MHz or 20% of the center frequency. Over there, pulse-based systems can access the UWB spectrum under these rules. Each pulse in a pulse-based UWB system occupies the entire UWB bandwidth, thus reaping the benefits of relative immunity to multipath fading (but not to intersymbol interference) [2]. Multiple Band Orthogonal Frequency Division Multiplexing (MB-OFDM) and Direct Sequence- UWB (DS-UWB) were proposed for the physical layer in IEEE 802.15.3a Task Group [3], [4]. The 802.15.3 MAC mainly works within a piconet which is a small network [5], [6]. It consists of data devices and one of them is taken as the piconet coordinator (PNC). The PNC is responsible for devices association/disassociation and the basic timing of the network by sending the beacon to all devices [5]. One major challenge in UWB MAC design is the QoS investigation with efficient resource scheme. Very limited work takes into account the characteristics of UWB for real time traffic. In this paper, the proposed protocol in [7] has been modified to achieve QoS requirements for multimedia traffic. Furthermore, additional proposed algorithm had been realized for real time traffic. The paper is organized as follows; Section II gives an overview of UWB physical model and resource allocation. Section III introduces the detail description of the proposed protocol for QoS provisioning. Simulation results and comparison discussions between data, voice, video, and multimedia UbiCC Journal, Volume 3, Number 1, February 2008 For UWB networks, to utilize the bandwidth and achieve desired QoS, an effective resource allocation scheme is needed to specify power level and transmission rate of each node to access the wireless medium. In [8], the general approach used for resource allocation is based on a joint management of rates and powers of the nodes. Specifically, the channel capacity for UWB network is bounded by the Signal to Interference plus Noise Ratio (SINR) threshold which is given by: P i gij Ri ηi + Tf σ 2 N k=1,k=i SIN R = ≥ γi Pk gkj (1) where P i is the average transmitted power for the link i , g ij is the path gain from the transmitter i to the receiver j which can be calculated as d−α where α is the path gain constant usually ij between 2–4 and dij is the distance between the transmitter i and the receiver j , ηi is the background noise energy, T f is the pulse repetition frequency, σ 2 is an operation parameter depending on the shape of the pulse, Ri is the rate of the link i , N is the number of active links in the network, and γi is the threshold value of the SINR [9]. Then powers and rates are chosen in order to match the the maximum allowed power (0 ≤ P i ≤ Pmax ) and the threshold value of SINR [8], [2], [10]. In [10], [11], Interference Margin (IM) approach has been assumed to avoid the frequent power reconfigure for each new admitted link. Each active link has an IM given by (2), which donated the additional interference by the new links. P i gij − ηi − Tf σ 2 Ri γi N IM i = Pk gkj k=1,k=i (2) One major challenge in UWB MAC design is the QoS provisioning with an efficient resource allocation scheme [11], [12], [13]. Although there have been large researches on real time traffic (voice and video) [13], [14], very restricted work takes into account the unique characteristics of UWB. The proposed protocol in [7] is modified to achieve the QoS requirements for multimedia traffic. 67 2 Fig. 1. Flowchart of WALAC1 and WALAC2 case the PNC calculates IM for all incoming requests using the maximum rate and power from (2). It checks the negative IM and applies the iteration procedure to the maximum negative IM link, if there are negative IM found. It updates IM for that link using the median then the minimum value of the rate. If it still negative, the PNC rejects that link and update the other IM and repeats this procedure till there are no negative IM links. All the residual positive IM links will be admitted. The other case is there are available links in the system. In this case, the PNC calculates the allowed power for each request from the minimal IM of active links from (3). Then remove the links with zero power value and let P0 = Pmax (if P0 > Pmax ). Calculate the allowed rate in the system for each request from (4). If there are rates lower than the minimum allowed rate in the system (Rmin ), reject the request with minimal allowed rate (to achieve fairness) then repeat again till all allowed rates be greater than Rmin . Update all active links in the network. If any one be negative IM, remove the maximum interfering request from the minimal IM to achieve fairness. Then update the IM again and repeat till no negative IM in the links. Calculate the IM for the residual requests with their calculated power and rate which will be considered as the maximum rate for that request and then apply the same procedure as if there are no links available in the network. As shown from Fig. 1, there are no great differences between them except that in WALAC2, there are no iterations as in WALAC1. In addition, IM is calculated using requested rate not the maximum or allowed rates as in WALAC1. Furthermore, the PNC has three queues for the incoming requests. The highest priority is placed for the voice queue, then the video one followed by the data one will be served respectively. That is to achieve QoS requirements. P0 = min IMi Tf σ 2 g0i where 1 ≤ i ≤ N P 0 gi0jo γ ηi + Tf σ 2 N k=1,k=i (3) III. THE PROPOSED PROTOCOL OVERVIEW The proposed protocol in [7] shows superior performance for UWB network when compared with Slotted Aloha and Packet Reservation Multiple Access systems in data traffic. It is based on two channels (data and control channels). The data channel is used for traffic transmission while the control channel is existed for requesting a link. The piconet coordinator (PNC) applies Wise Algorithm for Link Admission Control (WALAC) for the link requesting. The transmission is based on the superframe which consists of a number of slots in addition to the beacon in its header for both synchronization and broadcasting the piconet information. Furthermore, the control channel is divided into the same number of slots and each one is subdivided into uplink (for requesting) and downlink (for acknowledgment) subslots. This proposed protocol is modified to cope with multimedia challenges. Actually, the modified protocol has two proposed algorithms. One is for non real time traffic (data) which is the same as used in [7], and it is named WALAC1. The other one is made for real time traffic (voice and video) and it is named WALAC2. In WALAC1, if a data request is valid, there are two cases. Firstly, there are no available links in the system, and in this UbiCC Journal, Volume 3, Number 1, February 2008 Rallow = (4) Pk gkj0 The modified proposed protocol can be summarized as follows; 1) Terminal with traffic desired to be sent, requests a link from PNC using the uplink subslot in the control channel. This request includes the transmitter and receiver identifications as well as the traffic type. Each terminal transmits with a certain code. Therefore there are no collisions. 2) The PNC collects all requests and places them in the correct queue. Subsequently, it applies WALAC2 for voice and video requests respectively then WALAC1 for non real time one. Over there, The PNC informs the requesting terminals about its state, i.e., admitted or rejected. 3) The admitted links transmit in the next slot in the data channel while the rejected ones request again in the next slot in the control channel. 4) For the link termination, the PNC is informed through the control channel. 68 3 IV. SIMULATION RESULTS AND DISCUSSIONS In this section, we study the behavior of the proposed centralized protocol through simulations. The simulation area is taken as 50m×50m with nodes randomly distributed. Three types of traffic are considered. First of all, the constant bit rate source model (voice traffic) which has the highest priority according to its real time characteristics. It generates a signal of talkspurts separated by silentspurts with a rate of 32 Kb/s. A speech activity detector can be used to detect this pattern. Durations of talkspurts and silentspurts are exponential distributions with mean values of 1 and 1.35 seconds respectively [15], [16]. The second priority traffic is the variable bit rate source model, i.e., video traffic. It generates stream traffic with a variable time rate. The source rates as generated based on truncated Gaussian distribution between 128-384 Kb/s with mean rate of 256 Kb/s. The slice time is 33 msec. The last priority traffic is held for the data traffic which is generated based on Poisson process with λ call/sec per user. Furthermore, the buffering rate is 9600 b/s [17]. The rest of the default parameters used are shown in TABLE I. TABLE I SIMULATION PARAMETERS . users (because there is no streaming traffic here). While from Fig. 3, the average delay for voice traffic is directly proportional to the number of users then will be saturated. That is because the channel interference is increased with the number of active users and hence less admission ratio which leads to more delay. While from Fig. 4, the average delay is too low compared with the above due to the lowest transmission time. Although the streaming nature of the traffic, the high channel coding rate prohibits users to dominate the channel (and hence, there is fairness among users). More delay can be noticed for large number of users because of the channel congestion. The average delay for multimedia traffic can be shown from Fig. 5. A slightly decrease for the average delay of both data and voice traffic can be noticed (on the contrary of video traffic). That is because the highest priority of the voice traffic. While for data traffic, although it has the lowest priority, it has non QoS nature. Despite of the users’ possession of the channel, the average delay for the data users is not greatly affected like voice users because data traffic can be transmitted with the available rate. On the contrary, video traffic must achieve the QoS requirements. Larger number of active users, larger interference in the channel will be deduced and hence, less probability of admission and more delay can be noticed. 10 2 Parameter Tf σ2 η Pmax λ α γ superframe duration slot duration packet length voice life time video life time data life time voice channel coding rate video channel coding rate minimum rate (Rmin ) maximum rate (Rmax ) Value 10 ns 1.99×10−3 2.56×10−17 7 dBm 30 4 6 dB 10 msec 128 µsec 32 bytes 20 msec 50 msec 6 sec 8.33 Mb/s 33.3-100 Mb/s 2 Mb/s 100 Mb/s system average delay average delay (sec.) 10 1 10 0 10 20 30 40 number of data terminals 50 60 Fig. 2. System average delay for data traffic. The performance of the proposed protocol is measured according to QoS parameters such as the average delay and the loss probability. In addition, the system utilization (the ratio between the successfully transmitted bits averaged over the time) and the network capacity are considered. Furthermore, the admission ratio (the ratio between admitted requests and all incoming requests) as a new parameter for the network performance is perused. Figs. 2 to 4 show the average system delay (the average delay per successfully packets) for data, voice, and video traffic respectively. Due to the low buffering rate for the data traffic, its transmission time is high (26.7 msec) compared with voice and video traffic (7.8 msec and 2 msec maximum respectively) and hence, its average delay is somewhat large compared with voice and video traffic. Furthermore, from Fig. 2, the average delay for data traffic is nearly saturated. That is because the channel can not be dominated by certain UbiCC Journal, Volume 3, Number 1, February 2008 10 1 system average delay 10 average delay (sec.) 0 10 −1 10 −2 10 −3 0 10 20 30 number of voice terminals 40 50 Fig. 3. System average delay for voice traffic. 69 4 10 0 system average delay 10 0 system admission ratio 10 −1 average delay (sec.) 10 −1 10 admission ratio −2 10 −3 10 −2 10 −4 10 −5 10 −3 10 20 30 40 50 60 number of video terminals 70 80 90 −6 10 10 20 30 40 number of data terminals 50 60 Fig. 4. System average delay for video traffic. system average delay Video Traffic Voice Traffic Data Traffic Fig. 6. Admission ratio for data traffic. 10 2 10 system average delay (sec.) 1 10 0 10 10 −1 0 system admission ratio 10 10 −2 −1 admission ratio 10 −2 10 −3 10 20 30 40 number of active terminals 50 60 10 −3 Fig. 5. System average delay for multimedia traffic. 10 −4 Figs. 6 to 9 depict the system admission ratio versus the number of active terminals for data, voice, video, and multimedia traffic respectively. For data, voice, and video traffic, the admission ratio is inversely proportional to the traffic then it will be saturated between 10−5 to 10−6 for both data and video traffic and nearby 10−4 for voice traffic. That is because the highest priority of the voice traffic. While the low admission ratio for the data traffic is due to its very low buffering rate. Furthermore, from Fig. 8, the degradation in the admission ratio for the video traffic can be noticed for larger number of users due to the channel congestion and hence the delay will be increases as shown in Fig. 4. For multimedia traffic as shown from Fig. 9, a slightly decrease in the admission ratio for voice traffic can be noticed; because the presence of the other traffic admitted to the channel (video and data) lowers the probability of the admission. While there are no effective changes in the admission ratio for the video traffic. That is because its streaming nature besides its high transmission rate. While for the data traffic, the admission ratio is increased and hence slightly less delay can be noticed from Fig. 5. That is because the presence of other traffic in the system which prohibits the data users to take possession of the channel as happened when it stood alone. However, the admission ratio for the video traffic still the minimum one then for the data and voice traffic respectively. UbiCC Journal, Volume 3, Number 1, February 2008 10 −5 0 10 20 30 number of voice terminals 40 50 Fig. 7. Admission ratio for voice traffic. 10 −3 system admission ratio 10 admission ratio −4 10 −5 10 −6 10 −7 10 20 30 40 50 60 number of video terminals 70 80 90 Fig. 8. Admission ratio for video traffic. 70 5 10 −2 system admission ratio Video Traffic Voice Traffic Data Traffic 1 0.8 0.6 0.4 system loss probability 10 admission ratio −3 loss probability 20 30 40 number of active terminals 50 60 0.2 0 −0.2 −0.4 −0.6 −0.8 10 −4 10 −5 10 −6 10 −1 10 20 30 40 number of data terminals 50 60 Fig. 9. Admission ratio for multimedia traffic. Fig. 10. System loss probability for data traffic. The system loss probability (the ratio between the rejected transmitted packets and all transmitted packets) for data, voice, video, and multimedia traffic can be shown from Figs. 10 to 13 respectively. For data traffic, a very large number of data terminals can be supported. Because of the large threshold value of the maximum delay for data traffic, in addition to its non QoS nature. Therefore there are nearly no lost packets. While for voice traffic, the lowest threshold value of the maximum delay (to achieve real time requirements) plays a great role in the probability of loss increase. Fig. 11 shows that the system can support up to 37 voice users taking 10−2 as the threshold value of the loss probability. For video traffic, because of the lowest transmission time delay in the buffer, in addition to the large maximum delay threshold value, the system can support more than 90 users taking 10−4 as the threshold value of the loss probability as shown in Fig. 12. For multimedia traffic shown from Fig. 13, the system can support nearly 43 voice users due to its highest priority. This increase in the number of users due to the slightly decrease in the admission ratio beside the delay decrease which prohibits users to dominate the channel. While a large degradation in the video traffic is noticed as nearly 45 video users can be supported. That is because of the long time channel usage for the voice users (for the low channel coding). Therefore more delay can be noticed which leads to more losses. Despite of its non changeable admission ratio, the delay is increased, then the admitted links will be terminated due to the threshold value of the delay for video traffic. While for data users, there is no degradation noticed due to its non real time nature. Furthermore, users enhancement is predicted due to its admission ratio increase besides its average delay decrease. The system utilization for data, voice, and video traffic can be shown from Figs. 14 to 16 respectively, while for multimedia traffic is shown from Fig. 17. The system utilization for data traffic is nearly saturated around 104 b/s. The low utilization because of the low traffic rate for data users. While for voice users, the system utilization is directly proportional to the number of active users and saturated around 106 b/s. That is because the saturation of the admission ratio and hence more users admitted for more traffic, therefore more successful transmission packets over the time. For video traffic, the utilization will be saturated around 106 b/s. The UbiCC Journal, Volume 3, Number 1, February 2008 10 0 system loss probability 10 −1 loss probability 10 −2 Voice Threshold 10 −3 10 −4 0 10 20 30 number of voice terminals 40 50 Fig. 11. System loss probability for voice traffic. more utilization for lower users is due to the streaming nature for video traffic. For multimedia traffic as shown from Fig. 17, the saturated utilization and the better utilization for video traffic over voice and data can be noticed due to the streaming nature of the video traffic. The data traffic has the lowest utilization because of its low buffering rate. From these discussions, the system performance is controlled by both the admission ratio and the system average delay. It can be ordered from the best to the worth as follows; High admission ratio with low average delay. It is like the case of the data traffic. Low admission ratio with low average delay. It is like the case of the voice traffic when it is alone and with the multimedia traffic. The voice enhancement can be noticed. High admission ratio with high average delay. Low admission ratio with high average delay. It is like the case of the video traffic when it is alone and with the multimedia traffic. The video degradation can be noticed. 71 • • • • 6 1 0.8 0.6 system loss probability 10 6 system utilization 0.2 0 −0.2 −0.4 −0.6 −0.8 −1 10 20 30 40 50 60 number of video terminals 70 80 90 system utilization (b/s) 0.4 loss probability 10 5 10 4 0 10 20 30 number of voice terminals 40 50 Fig. 12. System loss probability for video traffic. Fig. 15. System utilization for voice traffic. 10 9 system utilization 10 system loss probability Video Traffic Voice Traffic Data Traffic Voice Threshold system utilization (b/s) 8 10 0 10 7 10 −1 system loss probability 10 −2 10 6 10 −3 5 10 −4 Video Threshold 10 10 20 30 40 50 60 number of video terminals 70 80 90 10 −5 Data traffic Fig. 16. 50 60 System utilization for video traffic. 10 −6 10 20 30 40 number of active terminals 10 8 system utilization Video Traffic Voice Traffic Data Traffic Fig. 13. System loss probability for multimedia traffic. 10 system utilization (b/s) 10 6 7 10 6 10 system utilization 5 10 10 system utilization (b/s) 5 4 10 20 30 40 number of active terminals 50 60 Fig. 17. 10 4 System utilization for multimedia traffic. V. 10 3 CONCLUSION 10 2 10 20 30 40 number of data terminals 50 60 Fig. 14. UbiCC Journal, Volume 3, Number 1, February 2008 System utilization for data traffic. Extensive simulation programs were performed to investigate the possibility of transmitting multimedia over UWB networks. A proposed protocol was explained to achieve QoS for multimedia transmission over UWB networks. The extended results showed evaluation of sensitive parameters affecting real time traffic transmission such as the delay guarantee and the loss probability, as packets with a large delay should be 72 7 discarded. The number of stations the network can support was determined. In addition, the admission ratio parameter and the system utilization were aroused for the system performance. Furthermore, the system performance can be managed by both the admission ratio and the average delay. The best performance is for the highest admission ratio with the lowest average delay, while the worth performance is for the lowest admission ratio with the highest average delay. UbiCC Journal, Volume 3, Number 1, February 2008 73 8 R EFERENCES [1] First report and order in the matter of revision of part 15 of the commission’s rules regarding ultra-wideband transmission systems. Fedral Communications Commission (FCC 02-48), Apr. 2002. [2] X. S. Shen, H. Jiang, and J. Cai, “Medium access control in ultrawideband wireless networks,” IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, vol. 54, pp. 1663–1677, 2005. [3] DS-UWB Physical Layer Submission to 802.15 Task Group 3a. IEEE P802.15-04/0137r5, September 2005. [4] MultiBand OFDM Physical Layer Proposal for IEEE 802.15 Task Group 3a. MultiBand OFDM Alliance SIG, 2004. [5] Part 15.3: Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications for High Rate Wireless Personal Area Networks (WPANs). IEEE Computer Society, 29 September 2003. [6] Part 15.3: Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications for High Rate Wireless Personal Area Networks (WPANs) Amendment 1: MAC Sublayer. IEEE Computer Society, 5 May 2006. [7] N. El-Fishawy, M. Shokair, and W. Saad, “A novel mac protocol for high rate uwb network,” in IEEE Conference on Wireless Rural and Emergency Communications Conference (WRECOM 2007), Rome, Italy, October 1-2, 2007, to be published. [8] C. Martello, “Uwb radio resource control: Mac funcational model and resource sharing approach.” Available at: http://net.infocom.uniroma1.it/papers/uwbnwu01.pdf. [9] M. win and R. Scholdtz, “Ultra-wide bandwidth time-hopping spreadspectrum impulse radio for wireless multiple-access communications,” IEEE Transactions on communication, vol. 48, pp. 649–691, April 2000. [10] F. Cuomo, C. Martello, and A. Baiocchi, “Radio resource sharing for ad hoc networking with uwb,” IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, vol. 20, NO. 9, pp. 1722–1732, Dec. 2002. [11] Y. Chu and A. Ganz, “Mac protocols for multimedia support in uwb-based wireless networks.” Available at: www.broadnets.org/2004/workshop-papers/Broadwim/broadwim2004Paper09-yuechun.pdf. [12] I. F. Akyildiz, D. A. Levine, and I. Joe, “A slotted cdma protocol with ber scheduling for wireless multimedia networks,” IEEE/ACM Trans. Net., vol. 7, no. 2, pp. 146–158, Apr. 1999. [13] V. Huang and W. Zhuang, “Qos-oriented packet scheduling for wireless multimedia cdma communications,” IEEE Trans. Mobile Comput., vol. 3, no. 8, pp. 73–85, Mar. 2004. [14] S. Jiang, J. Rao, D. He, X. Ling, and C. C. Ko, “A simple distributed prma for manets,” IEEE Trans. Veh. Technol, vol. 51, no. 2, pp. 293–305, Mar. 2002. [15] D. J. Goodman, R. A. Valenzuela, K. T. Gayliard, and B. Ramamurthi, “Packet reservation multiple access for local wireless communications,” IEEE Trans. Commun., vol. 37, pp. 885–603, Aug. 1989. [16] D. J. Goodman and S. X. Wei, “Efficiency of packet reservation multiple access,” IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, vol. 40, no. 1, pp. 170–176, Feb. 1991. [17] Ellis, Siwiak, and Roberts, “Tg3a technical requirements.” IEEE P802.15-03/030r0. December 27, 2002. UbiCC Journal, Volume 3, Number 1, February 2008 74