ch18

Taylor, L.W.; et. al. “Communications and Information Systems” Mechanical Engineering Handbook Ed. Frank Kreith Boca Raton: CRC Press LLC, 1999 c 1999 by CRC Press LLC Communications and Information Systems Lloyd W. Taylor DIGEX, Inc. 18.1 Introduction ....................................................................18-1 18.2 Network Components and Systems...............................18-2 Electrical and Optical Communications • Wireless Networks • Satellite Communications • Computer Communications Daniel F. DiFonzo Planar Communications Corp. 18.3 Communications and Information Theory...................18-23 Communication Theory • Information Theory A. Brinton Cooper III, PhD Bel Air, Maryland 18.4 Applications..................................................................18-41 Accessing the Internet • Data Acquisition Dhammika Kurumbalapitiya Harvey Mudd College S. Ratnajeevan H. Hoole Harvey Mudd College 18.1 Introduction This chapter provides a broad introduction to and reference resource for the field of communications and information systems. The first section covers the areas of computer networks and their underlying technologies. These technologies include electrical, optical, wireless, and satellite communications channels, as well as the protocols (such as TCP/IP) used to transfer information over these channels. The purpose of this section is to provide a high-level understanding of the infrastructure which underlies all modern electronic communications. The second section introduces Communications and Information Theory. This section provides the mathematical background necessary to better understand the technologies used in electronic communications. Issues such as noise, compression, and error correction are explained. Of necessity, this section is somewhat more theoretical than the others. The third section concludes the chapter with information on two applications of computers and networking. The first, Accessing the Internet, introduces the Internet, provides a summary of the software tools used to access information on the Internet, and discusses methods of finding information using a World Wide Web browser. The second application, Data Acquisition, describes the components of a data acquisition system and shows how they follow the model of a computer network. © 1999 by CRC Press LLC 18-1 18-2 Section 18 18.2 Network Components and Systems There are a variety of network architectures, standards, and transmission methods used to interconnect computers and communications systems. This section provides the information necessary to untangle the internetworking web. It begins by surveying the electrical and optical communications standards and architectures commonly used in modern networks, moves on to discussions of optical fibers and satellite communications, and concludes with a discussion of computer communications standards. Electrical and Optical Communications Lloyd W. Taylor The field of networking is generally broken into two domains: the local area network and the wide area network. The distinction between the two tends to be rather fuzzy. In general, local area refers to networks within the same building, and wide area refers to networks that interconnect buildings, cities, or countries. Local area networks are generally faster than wide area networks. A common rule of thumb used by network engineers is that speed and distance are inversely related; that is, the longer the distance the network must travel, the lower the speed that will be used. This is not an electrical or physical limit, but rather a financial one. Longer links and higher speeds simply cost more. Cabling Architectures All networks are based on underlying cabling architectures. These architectures define how the nodes are interconnected and may also limit the choices of network electrical standards. For example, ethernet requires a bus architecture, but can be implemented in a star architecture with the proper network electrical equipment. Ethernet cannot be implemented over a ring architecture. In a bus architecture (Figure 18.2.1), all members of the network share a common cable for their network signaling. The bus is a shared resource. All traffic can be read by every node. FIGURE 18.2.1 Bus architecture. For a bus-oriented network (such as ethernet) to work, there must be a mechanism for sharing access to the bus. In the ethernet world, the standard for this mechanism is carrier-sense multiple access/collision detection (CSMA/CD). When a computer wishes to transmit on the bus, it first listens to see if the bus is busy. If not, the computer begins to transmit while simultaneously listening to its own message. If the message is heard back as sent, then the transmission has been successful. If there are errors, it is likely that another computer decided to transmit at the same time, resulting in a collision. When a collision occurs, all currently transmitting computers stop transmitting for a random amount of time and then begin the process again by listening to see if the bus is busy. Bus architectures are useful where no one node requires a large portion of the shared bandwidth. If one or two nodes are constantly using the majority of the available bandwidth, performance for all connected nodes can become seriously affected. © 1999 by CRC Press LLC Communications and Information Systems 18-3 Bus architectures also are relatively insecure. It is quite simple to run a program to monitor all traffic on a bus and capture sensitive information such as user IDs and passwords. Because all systems must have access to all transmissions (to determine whether or not the bus is busy), there is no simple way around this limitation. In a ring architecture (Figure 18.2.2), each node has two connections, an inbound connection and an outbound connection. Network traffic passes through each node until it reaches its destination. FIGURE 18.2.2 Ring architecture. For a ring architecture to work, there must be some method of determining which node has the right to transmit at any given time. A common method to do this is by passing a token around the ring. When a node wishes to transmit, it captures the token and sends on its message in the token’s place. When the message reaches the intended destination, the recipient node marks it “read” and sends it on around the ring. When the originating node receives back its own message, it removes it from the ring and replaces the token on the ring. Note that the node that has just transmitted will not have another chance to transmit until all other nodes on the ring have had a chance to transmit. In other words, until the token makes it all the way around the ring, the first node may not transmit again. This has the effect of making a ring architecture fairer than a bus architecture, as no one node can co-opt the entire available bandwidth. Ring architectures suffer from the same security problems as bus networks: all traffic flows through all nodes. In addition, ring architectures are more difficult to install and maintain. If any one node becomes disconnected from the ring, the entire ring can fail, as there will be no path for the data. Some rings are implemented redundantly, with two separate traffic paths rotating in opposite rotations. Should any one node fail, the nodes on either side of the failure will wrap the ring, sending data back around the second ring to reach the computer on the other side of the failed node. In this way, no single-point failure can disrupt the ring. In a star architecture (Figure 18.2.3), each node has a separate connection to a central point. Each of these connections is to an active hub, which is a networking device designed to manage and switch network traffic. FIGURE 18.2.3 Star architecture. © 1999 by CRC Press LLC 18-4 Section 18 FIGURE 18.2.4 Star implementation of other architectures. A star architecture can be used to implement both bus and ring architectures. For a ring architecture (Figure 18.2.4a), the central hub simply sends data out one wire to the computer and receives it back on another wire. For a bus architecture (Figure 18.2.4b), the hub electrically or logically interconnects all wires together. For implementations of both architectures, the hub can monitor the bandwidth being used by each attached node, and in some implementations can actually limit the total bandwidth used by each. Many hubs can also “censor” the data going to each node, sending unchanged data only to the destination system. All other nodes get scrambled data, making it impossible for them to capture sensitive information. Most larger networks are now implemented as stars, because of the increased flexibility and manageability of a star architecture. The same cabling system can be used to implement bus and ring architectures, as well as point-to-point connections such as telephone lines. In addition, the costs of operation are lower, as all moves and changes can be made in a central location, rather than requiring that new cable be pulled for each new computer or terminal. Complex systems are larger networks generally using a combination of architectures. For campuswide networks, a common architecture is the “ring of stars” (Figure 18.2.5a). In this architecture, a star architecture is implemented in each building, and the buildings are interconnected in a ring, usually using fiber optic cables. The key advantage of this configuration is that no single-point failure will disrupt the network. If the ring architecture has been implemented as a redundant ring, it will self-heal and route around the damaged section. FIGURE 18.2.5 Complex architectures. Another common architecture is the “star of stars” (Figure 18.2.5b). This architecture is implemented identically in each building, but each building is interconnected to a single central location in a star configuration. The key advantage of this architecture is that it can support very high bandwidth networks. The key disadvantage is that the loss of the central hub will disrupt the entire network. Local Area Networks Copper Standards. Local networks are still most commonly implemented using copper wire, rather than fiber optics, because of the significantly higher cost of fiber. As the demand for higher bandwidth © 1999 by CRC Press LLC Communications and Information Systems 18-5 connections to the desktop increases and the cost of fiber decreases, we can expect to see more fiber to the desktop. There are three common categories of copper cable used in networking: unshielded twisted pair (UTP), shielded twisted pair (STP), and coaxial (Coax). Each of these has its unique advantages and disadvantages. All copper cabling is available in two general grades of insulation. Normal (nonplenum) insulation is usually made from PVC (polyvinyl chloride) and may be used anywhere other than in air handling spaces. The more expensive plenum cable (Teflon®-based insulation) is required in air handling spaces because of its fire-resistant characteristics. Unshielded twisted pair cable is the most commonly used network cable today. It comes in a variety of qualities, called categories. These categories (Table 18.2.1) define the electrical characteristics of the cable and specify the maximum data rate that they support. TABLE 18.2.1 UTP Cable Categories Category 1 2 3 4 5 # Pairs Any 2–25 1–6 1–6 1–6 Max. Bandwidth 1 4 10 20 100 Mbps Mbps Mbps Mbps Mbps Uses Telephone Token Ring — 4 Mbps Ethernet Token Ring — 16 Mbps 100BaseT, CDDI UTP cable is always installed in a star architecture. It requires active electronics in the wiring closets to operate the connections to each computer. As the category number gets higher, the manufacturing standards for such things as the number of twists per meter, variations in impedance, insulation consistency, etc. become stricter. This is because the cables must be of uniformly high quality to handle the higher frequencies required for higher data rates. For higher category cabling, installers must be specially trained. The rules for installation of highbandwidth cable limit the number of twists that can be unwrapped for termination, the minimum bend radius, and the maximum force that can be applied while pulling the cable through walls or conduits. Once the cable is installed and terminated, it must be tested and certified to the appropriate level. Improperly handled or installed Category 5 cabling may perform at only a Category 3 level, through no fault of the cable itself. Testing includes time-domain reflectometer measurements, noise and crosstalk measurements, and impedance measurements. UTP cable is specified for use in two common network electrical standards, 10BaseT ethernet and 100BaseT ethernet. IBM Token Ring can be run over UTP cable if an appropriate impedance matching transformer (usually called a balun) is used at both ends of the UTP run. In an increasing number of systems, these matching transformers are being integrated into the network interface cards and network hubs themselves. Shielded twisted pair (STP) cable is identical to UTP cable with the addition of a shield around each twisted pair and usually another shield around the entire cable bundle. This shielding reduces susceptibility to electrical noise (such as that caused by heavy electrical machinery), provides a more predictable impedance, but also reduces the efficiency of the cable. A typical connection with STP cable can only run one third the distance of an equivalent UTP cable. This is because the capacitance of the STP is much higher than that of the UTP, resulting in three times the attenuation at a given frequency. STP cable is commonly used in IBM Token Ring network installations. IBM-Standard STP cable is available in two versions: type 1 cable contains two pairs of 22-gauge wire, each with an individual shield. It is a heavy, stiff cable that is difficult to handle and pull. Type 9 cable is a lighter, more flexible version that is made with 26-gauge wire. The lighter gauge wires result in a higher impedance, limiting the length of a cable run to two thirds that of a type 1 cable. © 1999 by CRC Press LLC 18-6 Section 18 IBM has defined a number of other cabling standards that include fiber optics, UTP, and mixed cable types within a single sheath. Coaxial cable was the original cable used for ethernet. It has largely been superseded by UTP, but is still in use in older installations. Coaxial cable can be used either in a bus or a star architecture. In a bus architecture, a heavy coaxial cable (often called thickwire) is run through the hallway, above the drop ceiling. Each end is terminated with a 50-Ω resistor to minimize electrical reflections. The coax is drilled wherever a connection to a computer is to be made, and an active tap is installed (Figure 18.2.6). The connection between the tap and the computer is via a 15-pin drop cable that carries signal and power to the active tap. FIGURE 18.2.6 Ethernet thickwire coax network. In a star architecture, a much lighter coaxial cable (often called “thinwire”) is used. A standard BNC connector is installed on each end, and the cable is plugged directly into the network interface card in the computer and into the network hub in the wiring closet. Thinwire networks can be “daisy-chained,” allowing several computers to share a single run of coaxial cable. Fiber Standards. Fiber optics are playing an increasingly important role in computer networks. The most common use at present is in backbone and wide area networks. As costs decrease, we can expect to see fiber to the desktop become the norm. How Fiber Optics Work. Fiber-optic cables are essentially “light pipes.” They are made up of two coaxial layers of glass, the inner called the core and the outer called the cladding (Figure 18.2.7). Each of the components has a different index of refraction, resulting in a reflective surface at their interface. FIGURE 18.2.7 Optical fiber construction. A modulated electrical signal drives a lightsource such as a laser diode, which injects a collimated beam of light into one end of the optical fiber. The lightbeam is reflected back and forth along the inner fiber and is coupled with an optical detector at the other end (Figure 18.2.8). The optical detector converts the lightbeam back to an electrical signal. While optical fibers can carry signals much greater distances than can electrical conductors operating at the same frequency, they do suffer from attenuation just as do electrical conductors. The mirror surface at the interface between the core and the cladding is not perfect, and photons can escape through this interface and be lost in the cladding. Also, not all lightbeam frequencies propagate at the same rate through the fiber, resulting in degradation of the signal over long distances. © 1999 by CRC Press LLC Communications and Information Systems 18-7 FIGURE 18.2.8 Optical transmission system. FIGURE 18.2.9 Multimode fiber. FIGURE 18.2.10 Single-mode fiber. Types of Fiber-Optic Cables. There are two kinds of fiber-optic cables in common use. The first, called multimode fiber, is less expensive to make and use, but has a more limited bandwidth. The second, called single-mode fiber, is more expensive to make and use, but can support very high bandwidths. Multimode fiber is used for distances of up to 2 km. It typically has a core diameter of 62.5 µm and a cladding diameter of 125 µm. Because of the large diameter of the core, there are many paths, or modes, between the two ends of the fiber. As demonstrated in Figure 18.2.9, a given photon may travel down the center of the core (mode 1), may reflect a few times at a shallow angle to the interface (mode 2), or may strike the interface at a sharp angle (mode 3) and reflect many times as it travels to the far end of the fiber. These many paths result in a high attenuation of the lightbeam as it travels down the fiber. This attenuation is caused by the large number of photons lost through the interface (for a given probability of reflection, the more reflections a photon must make, the higher the chance that the photon will pass through the interface rather than being reflected by it), as well as other attenuation mechanisms such as phase distortion. Single-mode fiber is designed for use in higher bandwidth and longer distance applications. It is made up of a much smaller core (typically 10 µm) in a somewhat smaller cladding (typically 100 µm) than is used in multimode fiber, resulting in a greatly reduced number of modes (Figure 18.2.10). As a result, single-mode fiber has a much lower attenuation than an equivalent multimode fiber. Single-mode fiber is much more difficult to splice and terminate than is multimode fiber, as there is very little margin for error in aligning the core. A 2-µm alignment offset in a multimode fiber is only about 3% of the total diameter, whereas it is 20% of the diameter of a single-mode fiber. Thus, a minor misalignment in single-mode fiber splice will result in a major increase in attenuation. Typical Fiber-Optic-Based Networks. One ubiquitous use of fiber in corporate networks is for fiber distributed data interface (FDDI) backbone networks. FDDI networks transfer data at 100 Mbps using a token-ring architecture. Rings of up to 200 km are supported, with a maximum distance between adjacent nodes of 2 km. © 1999 by CRC Press LLC 18-8 Section 18 Wide Area Networks Once out of the local network area, the characteristics of networks change. In general, connections between widely distributed locations are not handled by the installation of cable by the company itself. The company will generally turn to a telecom provider, such as the local telephone company, to provide the necessary connections. The telephone company does not generally sell cable connections. They sell bandwidth. Thus, the standards for wide area networks are based on fixed bandwidth allocations over telephone company facilities. The telephone company may user copper or fiber, at their discretion, to provide the requested service. The telephone company’s networks are usually implemented as rings or stars, as are local networks. There are three broad classes of digital service available. The first, traditionally copper based, is the DS-n service. The second, also copper based, is basic rate Integrated Services Digital Network (ISDN). The third, optical fiber based, is the OC-n service. DS-n. The DS series of services provide low- to medium-speed data connections. Table 18.2.2 summarizes the commonly available services and their customary uses. TABLE 18.2.2 DS-n Class Services Service Name DS-0 DS-1 DS-2 DS-3 Data Rate 56/64 Kbps 1.544 Mbps 6.2 Mbps 44.736 Mbps Common Use Voice, low speed data Multiple voice lines, data Data Data Pricing for these services is heavily dependent on the total mileage of the link. The links are usually leased for a minimum period (typically 1, 2, or 5 years). These leased data lines can support all network protocols. Leased lines are point-to-point. That is, they run from one location to another. A single line cannot be configured as a ring, a star, or a bus. To implement these architectures, several leased lines must be ordered, and connected together by the customer into the desired configuration. Basic Rate ISDN. An increasingly important wide area networking technology is the Integrated Services Digital Network (ISDN). The most commonly implemented version of this service is the basic rate service, which provides three data channels over a single pair of telephone cables. Two channels, known as the bearer (or “B”) channels, each provide a 64-Kbps dialup digital connection. The third channel, known as the data (or “D”) channel, is used for control information between the telephone switch and the ISDN terminal. In practice, an ISDN line is installed at a business or home location. The line is connected to a piece of equipment known as an ISDN terminal adapter. This device allows a variety of other devices to make use of the various channels provided by the ISDN line. One typical device has a connection for a voice telephone and an ethernet, and can be used to place normal telephone calls simultaneously with a digital dialup connection to a company’s ethernet. A higher-speed version of ISDN, known as primary rate ISDN, supports 24 B channels over a DS-1 leased line. One of these channels is typically used for control information, leaving 23 channels available for data transmission. OC-n. As demand for higher-speed connections continued to increase, standards for optical networks were established to meet these demands. The OC-n (Optical Carrier) standards (Table 18.2.3) provide data rates starting at 55 Mbps and going up to multiple gigabits per second. OC-based networks use the SONET (Synchronous Optical Network) signaling standards for transferring data. © 1999 by CRC Press LLC Communications and Information Systems 18-9 TABLE 18.2.3 OC-n Data Rates OC-n OC-1 OC-3 OC-12 OC-48 OC-192 Data Rate (Mbps) 51.84 155.52 622.08 2488.32 9953.28 OC/SONET networks are always implemented as rings. They can be configured to be fault-tolerant, so that no single break in the cable (or failed network node) will cause connections to fail. Logical connections are made across a SONET ring by the use of virtual circuits. These virtual circuits can either be permanent (PVC) or switched (SVC). A PVC is configured manually into the control electronics of the originating and terminating SONET control equipment. It will persist indefinitely until it is manually terminated. An SVC is established dynamically upon a request by a computer connected to the control equipment. The SVC persists until the requesting computer informs the control equipment that the circuit is no longer required. Conceptually, PVCs are like leased lines (permanent), and SVCs are like dial-up telephone lines (temporary). Asynchronous transfer mode (ATM) is an increasingly important protocol that is commonly implemented over SONET networks. ATM makes use of fixed-sized packets, called cells, to transfer data at very high speeds. Each of these cells is 53 bytes in size, with 5 bytes for addressing and cell management and 48 bytes for data. Because the cells are fixed in size and format, it is possible to build network switches that are very fast, as the processing and routing of the cell can be entirely done in hardware. To set up a virtual circuit between nodes, the originating node sends a request to its local ATM switch. In this request, the originating node specifies the bandwidth required for the link, the length of time the link is needed, and the quality of service required. The originating node will check its available network resources to see if it can grant the service requested. If it can, it will pass on the request to the next switch in the network which will repeat the check for available resources. This process continues until the destination node is reached and consents to the link request. An acknowledgment is relayed back to the originating node confirming the availability of the requested service. As this acknowledgment is returned through each switch, the switch commits the required resources to the virtual circuit. A key strength of ATM is that it can carry any type of information over the same link, rather than requiring separate links for voice, data, and video as is common today. This ability will likely reduce the overall cost of networking as separate voice, data, and video networks will not be required in the future. Defining Terms 100 The standard for running 10 Mb/sec ethernet over unshielded twisted pair network cabling. 100BaseT: The standard for running 100 Mb/sec ethernet over unshielded twisted pair network cabling. Active hub: A device that interconnects multiple network links. Coax: Coaxial cable. Local area network: A network that is within the local area, generally within an office area or building. STP: Shielded twisted pair cable. Token: A data packet that circulates around a token-ring network indicating that the network is available for data transmission. UTP: Unshielded twisted pair cable. Wide area network: A network that is outside the local area, generally between buildings, between cities, or between countries. © 1999 by CRC Press LLC 18-10 Section 18 References Acampora, A.S. 1994. An Introduction to Broadband Networks. Plenum Press, New York. Bates, R.J. 1992. Introduction to T1/T3 Networking. Artech House. Clark, M.P. 1991. Networks and Telecommunications: Design and Operation. John Wiley & Sons, New York. Conard, J.W. (Ed.) 1991. Handbook of Communications Systems Management. Auerbach, Boca Raton, FL. Davidson, R.P. 1994. Broadband Networking ABCs for Managers. John Wiley & Sons, New York. Kosiur, D.R. 1995. How Local Area Networks Work. Prentice-Hall, Englewood Cliffs, NJ. McElroy, M.W. 1993. The Corporate Cabling Guide. Artech House. McNamara, J.E. 1988. Technical Aspects of Data Communication. Digital Press. Minoli, D. 1993. Enterprise Networking. Artech House. Wireless Networks Introduction With increasing mobility of the workforce comes the need to provide network access to computers located in automobiles, in briefcases, on shipboard, and even in aircraft. Wireless networks provide the required connectivity. This section covers the basic radio frequency (RF) link types used in wireless networking, compares wireless networks to traditional wired networks, and discusses key trends in the field. The reference section provides pointers to sources of additional information. RF Technologies All mobile wireless networks make use of a radio frequency carrier*. This carrier may be fixed frequency (like an AM radio station signal) or spread spectrum (where the signal is spread over a wide frequency band). Section 18.4 provides additional information on radio frequency communications, including modulation methods. Point to Point. A point-to-point communications system relays the signal from the mobile user to a base station (Figure 18.2.11). The radio link is a single hop, that is, the signal is transferred directly from the mobile user to the base station without going through intermediate reception and retransmission steps. A point-to-point system generally requires a line of sight between the transmitter and the receiver. Because the curvature of the earth limits the line of sight, this type of system has a typical maximum range of tens of miles. FIGURE 18.2.11 Point-to-point link. Several point-to-point links can be put in series to relay a signal. Microwave relay stations are typically placed approximately every 30 mi to relay a signal from point to point over long distances. Each relay point receives the signal and retransmits it to the next station in line. * Some local area wireless networks use infrared light or laser carriers. These are not used in mobile applications because of their limited range. © 1999 by CRC Press LLC Communications and Information Systems 18-11 Cellular. To overcome the limitations of point-to-point systems for mobile telephony, a cellular radio system can be used. This system requires a large number of antennas, each at the center of a “cell”, as shown in Figure 18.2.12. As a ground-based vehicle approaches the edge of the coverage range for a particular antenna, the cellular control system communicates with the mobile data unit or telephone and assigns it to the next antenna in the direction in which the vehicle is heading. At a coordinated time, the in-progress conversation is transparently (to the user) switched to the newly assigned antenna without interruption. FIGURE 18.2.12 Coverage pattern for multiple cellular antennas. In congested downtown areas, cell antennas may be only a few hundred meters apart and use lowerpowered transmitters to reduce the size of the cell. This provides more channels for use by more simultaneous conversations, but requires more handoffs as the mobile unit moves through the city. In suburban and rural areas, the cell antennas may be several thousand meters apart, with higher-powered transmitters to provide a large cell. This minimizes the number of cell antennas that must be installed, reducing the cost of the system. A personal communications system (PCS) is a version of cellular technology that uses smaller “microcells” with lower power transmitters. This second-generation cellular system requires many more cell antennas (every few hundred meters), making it most effective in urban areas. PCS telephones are typically smaller and lighter than first-generation cellular telephones, because of their lower power and higher frequency operation. Satellite — Low Earth Orbit. Cellular and PCS systems require a very large number of antennas installed in a grid over a large area to provide coverage for mobile units. This limits the rate at which a system with complete coverage for a given area can be installed. Each cell antenna installation requires permission from the cell-site property owner, building permits, power, telecommunications connections, and regulatory approval before it can be built and made operational. In addition, cellular systems differ from country to country, requiring different mobile units for access in each country. One approach to addressing these problems is to stand the problem on its head. Rather than having a large set of fixed antennas which hand off the mobile unit as the mobile unit moves, use a number of satellite transceivers in low earth orbit (LEO). Establish the orbits of these transceivers so that there is at least one satellite in view of any place on the face of the earth at a given time. Then, as a given satellite begins to move out of range of the mobile unit, it hands off the call to another satellite that is just moving into range. This type of system can easily provide data and voice connectivity to any location on (or slightly above) the face of the Earth using a common mobile transceiver. For more information, see the subsection Satellite Applications. Comparing Wireless to Wired Networks It is clearly posible to provide data and voice access to mobile users with any of the above technologies. Yet there are significant limitations to these systems when compared to a hardwired network connection. As always, there are trade-offs that must be made. © 1999 by CRC Press LLC 18-12 Section 18 Bandwidth. A directly wired network connection can easily provide 100 Mbps. The systems discussed in the previous section are typically limited to tens of Kbps, four to five orders of magnitude below what is commonly available to directly wired users. This bandwidth limitation has profound implications for mobile users. For example, complex graphics and images require several minutes to download, rather than the several seconds that are required for a directly wired user. Any function that requires the transfer of large amounts of data will be greatly slowed by a wireless network link. The limitation is not likely to ever go away. There is a limited amount of RF spectrum available for mobile computing, while there is an essentially unlimited amount of network spectrum available for directly wired users (just add more fibers or wires!). While improvements in wireless bandwidth availability will come in time, it simply will never compare with the bandwidth available to a desktop computer user. Security. Wireless transmissions can be easily intercepted. The signal is broadcast through the atmosphere and can be received by anyone within range who has a properly tuned receiver. Thus, it is necessary to use encryption to protect the transmitted data. Encryption adds complexity and cost to any system. Securely exchanging a cryptographic key for a call can be difficult to do in a way that is difficult to compromise. Every system that needs to interoperate must agree on what cryptographic algorithm will be used and how it will be keyed. Costs. As can be inferred, the costs of establishing a mobile wireless network are very high. These costs must be recovered from the users of the system within its projected useful life. As an example, compare the cost of a typical local wired telephone call with the cost of a cellular telephone call. In many areas, the local wired call is offered at a fixed rate for an unlimited call length, while the cellular call is charged by the minute. Conclusions Wireless networking has clear usefulness where it is necessary to have access to data while mobile. Applications of wireless networking in such areas as police work or emergency services have already proven their value. The costs and limitations of wireless networking must be carefully considered before embarking on a major initiative. With the technology in this area developing rapidly, in a wide variety of incompatible directions, caution must be exercised. References Bates, R.J. 1994. Wireless Networked Communication. McGraw-Hill, New York. Breed, G. 1994. Wireless Communications Handbook. Cardiff. Calhoun, G. 1992. Wireless Access and the Local Telephone Network. Artech House. Davis, P.T. and McGuffin, C.R. 1995. Wireless Local Area Networks. McGraw-Hill, New York Lee, W.C.Y. 1995. Mobile Cellular Telecommunications. McGraw-Hill, New York Nemzow, M.A.W. 1995. Implementing Wireless Networks. McGraw-Hill, New York Satellite Communications Daniel F. DiFonzo Introduction The impact of satellites on world communications since commercial operations began in the mid-1960s is such that we now take for granted many services that were not available a few decades ago: worldwide TV, reliable communications with ships and aircraft, wide area data networks, communications to remote areas, direct TV broadcast to homes, position determination, and Earth observation (weather and map- © 1999 by CRC Press LLC Communications and Information Systems 18-13 ping). Future satellite-based global personal communications to hand-held portable telephones may usher in yet another new era. Satellites function as line-of-sight microwave relays in orbits high above the Earth which can “see” large areas of the Earth’s surface. This unique feature ensures the continued growth of satellites even as fiber-optic cables capture a larger market share of high-density point-to-point traffic. Satellites provide cost-effective access for areas with low (thin-route) communications traffic, because Earth terminals can be installed in locations where the high investment cost of terrestrial facilities might not be warranted. Satellites are particularly well suited to wide area coverage for broadcasting, mobile communications, and point-to-multipoint communications. Satellite Applications Figure 18.2.13 depicts several kinds of satellite links and orbits. The geostationary Earth orbit (GEO) is in the equatorial plane at an altitude of 36,000 km with a period of one sidereal day (23h 56m 4.09s). GEO satellites appear to be almost stationary from the ground (subject to small perturbations) and the Earth antennas pointing to these satellites may need only limited or no tracking capability. The orbits for which the highest altitude (apogee) is at or greater than GEO are sometimes referred to as high Earth orbits (HEO). Low Earth orbits (LEO) typically range from a few hundred kilometers to about 1000 km, and medium Earth orbits (MEO) are at intermediate altitudes. FIGURE 18.2.13 Satellite links and orbits. Initially, satellites were used primarily for point-to-point traffic in the GEO fixed satellite service (FSS), e.g., for telephony across the oceans and for point-to-multipoint TV distribution to cable head end stations. Large Earth station antennas with high-gain narrow beams and high uplink powers were needed to compensate for limited satellite power. This type of system, exemplified by the early global network of the International Telecommunications Satellite Consortium (Intelsat), used “Standard A” Earth antennas with 30-m diameters. Since the start of Intelsat, many other satellite organizations and consortia have been formed around the world to provide international, regional, and domestic services (Rees, 1990). As satellites have grown in power and sophistication, the average size of the Earth terminals has been reduced. High-gain satellite antennas and relatively high-power satellite transmitters have led to very small aperture Earth terminals (VSAT) with diameters of less than 2 m and modest powers of less than 10 W (Gagliardi, 1991). As depicted in Figure 18.2.13, VSAT terminals may be placed atop urban office buildings, permitting private networks of hundreds or thousands of terminals which bypass terrestrial lines. VSATs are usually incorporated into star networks where the small terminals communicate through the satellite with a larger Hub terminal. The Hub retransmits through the satellite to another small terminal. Therefore, VSAT-to-VSAT links require two hops with attendant time delays. With high-gain satellite antennas and relatively narrowband digital signals (e.g., compressed voice at ≤8 kbps), direct single-hop mesh interconnections of VSATs may be used. © 1999 by CRC Press LLC 18-14 Section 18 Satellite Functions The traditional function of a satellite is that of a “bent pipe” quasilinear repeater in space. As shown in Figure 18.2.13, uplink signals from Earth terminals directed at the satellite are received by the satellite’s antennas, amplified, translated to a different downlink frequency band, channelized into transponder channels, further amplified to relatively high power, and retransmitted toward the Earth. Transponder channels are generally rather broad (e.g., bandwidth of 36 MHz) and each may contain many individual or user channels. Multiple access techniques, to be discussed later, allow many users to share a satellite’s resources of bandwidth and power and to avoid interfering with each other and with other satellite or terrestrial systems. Multiple access systems segregate users by frequency, space, time, polarization, and signaling code orthogonality. Analog and digital modulations are both in widespread use. While frequency modulation (FM) has been prevalent, recent advances in digital voice and video compression will lead to the widespread use of digital modulation methods such as quartenary phase shift keying (QPSK) and quartenary amplitude modulation (QAM). Figure 18.2.14 depicts the functional diagram appropriate to a satellite using frequency division multiple access (FDMA) and reusing available frequencies by means of multiple antenna beams. Interference can result if the sidelobes of one beam receive or transmit substantial energy in the direction of the other beam. FIGURE 18.2.14 Satellite system block diagram. Newer satellite architectures, such the NASA Advanced Communications Technology (ACTS) and Motorola’s Iridium system, may use regenerative repeaters, which process the uplink signals by demodulating them to baseband. These baseband signals, which may be for individual users or may represent frequency division multiplexed (FDM) or time division multiplexed (TDM) signals from many users, are routed to downlink channels, modulated onto one or more radio frequency (RF) carriers, and transmitted to Earth. High-power direct broadcast satellites (DBS) operating at Ku-band (around 12 GHz) deliver TV directly to home receivers having antennas less than 1 m in size. Such systems using analog FM are operational in Japan and Europe. In the United States, DBS with digital modulation and compressed video will provide more than four NTSC TV channels per 24-MHz transponder channel. For the United States, where each DBS orbital location is allocated 32 transponder channels of 24 MHz each, more than 128 conventional TV channels can be provided from a single DBS orbital location. DBS is seen as an attractive medium for delivery of high-definition TV (HDTV) to a large number of homes. Mobile satellite services (MSS) operating at L-band around 1.6 GHz have revolutionized communications with ships and, more recently, with aircraft which would normally be out of reliable communi- © 1999 by CRC Press LLC Communications and Information Systems 18-15 cations range of terrestrial radio signals. The International Maritime Satellite Consortium (Inmarsat) operates the dominant system of this type. Links between LEO satellites (or the NASA shuttle) and GEO satellites are used for data relay, e.g., via the NASA Tracking and Data Relay Satellite System (TDRSS). Some systems will use intersatellite links (ISL) to improve the interconnectivity of a wide-area network. ISL systems would typically operate at frequencies above 20 GHz or even use optical links. An exciting new development is the prospective use of L-band frequencies with a large number (12 to 66) of LEO satellites for personal communications systems (PCS) directly with small hand-held portable telephones anywhere in the world. Access and Modulation Satellites act as central relay nodes which are visible to a large number of users who must efficiently use the limited power and bandwidth resources. A brief summary of issues specific to satellite systems is given below. Frequency division multiple access (FDMA) has been the most prevalent access for satellite systems until recently. Individual users assigned a particular frequency band may communicate at any time. Satellite filters subdivide a broad frequency band into a number of transponder channels, e.g., the 500MHz uplink FSS band from 5.925 to 6.425 GHz may be divided into 12 transponder channels of 36 MHz bandwidth plus guard bands. This limits the interference among adjacent channels in the corresponding downlink band of 3.7 to 4.2 GHz. FDMA implies that several individual carriers coexist in the transmit amplifiers. In order to operate the amplifiers in a quasilinear region relative to their saturated output power to limit intermodulation products, the amplifiers must be operated in a backed-off condition. For example, in order to limit thirdorder intermodulation power for two carriers in a conventional TWT (traveling wave tube) amplifier to ≈–20 dBc, its input power must be reduced (input backoff) by about 10 dB relative to the power that would drive it to saturation. The output power of the carriers is reduced by about 4 to 5 dB (output backoff). Amplifiers with fixed bias levels will consume power even if no carrier is present. Therefore, dc-to-RF efficiency degrades as the operating point is backed off. For amplifiers with many carriers, the intermodulation products have a noise-like spectrum and the noise power ratio is a better measure of multicarrier performance. Time division multiple access (TDMA) users share a common frequency band and are each assigned a unique time slot for their digital transmissions. At any instant there is only one carrier in the transmit amplifier, requiring little or no backoff from saturation. The dc-RF efficiency is high. A drawback is the system complexity required to synchronize widely dispersed users in order to avoid intersymbol interference caused by more than one user signal appearing in a given time slot. Also, the total transmission rate in a TDMA satellite channel must be essentially the sum of the users’ rates, including overhead bits such as for framing, synchronization and clock recovery, and source coding. At the present state of the art, Earth terminal hardware costs may be higher than for FDMA. Nevertheless, TDMA systems are gaining acceptance for some applications. Code division multiple access (CDMA) modulates each carrier with a unique pseudorandom code, usually by means of either a direct sequence or frequency hopping spread spectrum modulation. As the CDMA users occupy the same frequency band at the same time, the aggregate signal in the satellite amplifier is noise-like. Individual signals are extracted at the receiver by correlation processes. CDMA tolerates noise-like interference but does not tolerate large deviations from average loading conditions. One or more very strong carriers could violate the noise-like interference condition and generate strong intermodulation signals. User access is via assignments of a frequency, time slot, or code. Fixed assigned channels allow a user unlimited access. However, this may result in poor utilization efficiency for the satellite resources and may imply higher user costs (analogous to a leased terrestrial line). Other assignment schemes include demand assigned multiple access (DAMA) and random access (e.g., for the Aloha concept). DAMA systems require the user to first send a channel request over a common control channel. The © 1999 by CRC Press LLC 18-16 Section 18 network controller (at another earth station) seeks an empty channel and instructs the sending and receiving units to tune to it (either in frequency or time slot). A link is maintained for the call duration and then released to the system for other users to request. Random access is economical for lightly used burst traffic such as data. It relies on random time of arrival of data packets and protocols are in place for repeat requests in the event of collisions (Gagliardi, 1991). In practice, combinations of multiplexing and access techniques may be used. A broad band may be channelized or frequency division multiplexed (FDM) and FDMA may be used in each subband, e.g., FDM/FDMA. The traditional satellite modulation format has been FM. However, recent trends indicate that digital modulations such as M-ary PSK and QAM will become more prevalent for nearly all applications including voice, data, and TV. The efficiencies afforded by digital modulations arise partly because they allow signal processing for bandwidth compression. Compressed digital TV transmission allows a significant improvement in capacity compared with FM. Trends Satellite communications have approached a mature stage of development, and their competitiveness for point-to-point voice traffic, compared with fiber, has been questioned. However, as mentioned, satellites will continue to exploit their unique wide view of the Earth for such applications as broadcast and personal communications. The satellite industry’s maturity also presents another challenge. To date, satellite construction has resembled a craft industry with extensive custom design, long lead times, long test programs, and high cost. Satellites will benefit from modern “lean production” and “design-to-cost” concepts that could lead to systems having lower cost per unit of capacity and higher reliability. Technology advances that are being pursued include development of lightweight “lightsats” for economical provision of services at low cost, more sophisticated on-board processing to improve interconnectivity, intersatellite links, improved components such as batteries, and even such speculative concepts as providing satellite power from the ground via high-power laser beams using adaptive optics (Landis and Westerlund, 1992). Defining Terms Attitude: The angular orientation of a satellite in its orbit, characterized by roll (R), pitch (P), and yaw (Y). The roll axis points in the direction of flight, the yaw axis points toward the earth’s center, and pitch axis is perpendicular to the orbit plane such that R × P → Yb Backoff: Amplifiers are not linear devices when operated near saturation. To reduce intermodulation products for multiple carriers, the drive signal is reduced or backed off. Input backoff is the dB difference between the input power required for saturation and that employed. Output backoff refers to the reduction in output power relative to saturation. Bus: The satellite bus is the ensemble of all the subsystems that support the antennas and payload electronics. It includes subsystems for electrical power, attitude control, thermal control, TT&C, and structures. Frequency reuse: A way to increase the effective bandwidth of a satellite system when available spectrum is limited. Dual polarizations and multiple beams pointing to different Earth regions may utilize the same frequencies as long as, for example, the gain of one beam or polarization in the directions of the other beams or polarization (and vice versa) is low enough. Isolations of 27 to 35 dB are typical for reuse systems. Polarization isolation: Frequency reuse allocates the same bands to several independent satellite transponder channels. The only way these signals can be kept separate is to isolate the antenna response for one reuse channel in the direction or polarization of another. The beam isolation is the coupling factor for each interfering path (ideally 0 or –∞ dB). © 1999 by CRC Press LLC Communications and Information Systems 18-17 References Gagliardi, R.M. 1991. Satellite Communications. Van Nostrand Reinhold, New York. Griffin, M.D. and French, J.R. 1991. Space Vehicle Design. American Institute of Aeronautics and Astronautics, Washington, D.C. Long, M. 1990. The 1990 World Satellite Annual. MLE, Winter Beach, FL. Morgan, W.L. and Gordon, G.D. 1989. Communications Satellite Handbook. John Wiley & Sons, New York. Rees, D. 1990. Satellite Communications: The First Quarter Century of Service. John Wiley & Sons, New York. Richaria, M. 1995. Satellite Communications Systems: Design Principles, McGraw-Hill, New York. Roddy, D. 1995. Satellite Communications. Prentice-Hall, Englewood Cliffs, NJ. Wertz, J.R. and Larson, W.J., eds. 1991. Space Mission Analysis and Design. Kluwer Academic Publishers, Dordrecht, The Netherlands. Further Information For a brief history of satellite communications see Satellite Communications: The First Quarter Century of Service, by D. Rees (Wiley, 1990). Propagation issues are summarized in Propagation Effects Handbook for Satellite Systems Design, 1983, NASA Reference Publication 1082(03), November 1990; descriptions of the proposed LEO personal communications systems are in the FCC filings for Iridium (Motorola), Globalstar (SS/Loral), Odyssey (TRW), Ellipso (Ellipsat), and Aries (Constellation Communications), 1991 and 1992. Computer Communications Lloyd W. Taylor The previous three sections have discussed various communications links used in digital interconnections. This section discusses common protocols used to handle network addressing and delivery over these communication links. The OSI Network Model The International Standards Organization (ISO) developed a model of network communications called the Open Standards Interconnect (OSI model). This is an abstract model that does not necessarily imply any particular implementation. In fact, it is most commonly used as a reference model into which various network protocols are mapped. The OSI model (Table 18.2.4) defines seven layers. The topmost layer defines the interface between the user-interface software and the network, while the bottommost layer defines the electrical characteristics of the network itself. Not all protocols implement all seven layers, as will be seen later in this section. TABLE 18.2.4 The OSI Network Model Layer Name Application Presentation Session Transport Network Data link Physical Function Interface between user applications and network Interface between network applications and network Destination definition Internet addressing Switching, routing, management Network access control, local addressing Electrical or optical cable, signaling, and transmission standards © 1999 by CRC Press LLC 18-18 Section 18 The seven layers have the following functions: The application layer is the layer that the user or the user’s application interfaces with. This layer provides such capabilities as file transfer, electronic mail transfer, and videoteleconferencing transfer. The presentation layer defines the format of the information to be transferred. It handles the mapping between the network representation of data and the local representation of data. This layer is typically where network encryption and compression will occur. The session layer is responsible for setting up connections between processes on separate computers. It connects the processes, manages the connection, and terminates the connection when the data transfer is complete. The transport layer guarantees error-free communication for the connection. It handles error checking and retransmission of lost packets, and manages the traffic flow between the end points of the connection. The network layer manages the creation, maintenance, and termination of connections between computers. It handles the routing of the data across the intervening networks. It interfaces between the logical abstractions of the higher layers with the technical specifics of the lower. The data link layer handles data at the packet level. It enables the receiving computer to check that these blocks of data have been received reliably. The physical layer implements the standards for sending bits across a particular type of network. It includes standards for the wire or fiber itself, standards for the signal characteristics, and standards for the connection of the network to the computer. These layers work together to ensure reliable communications between processes on different computer systems. Note that each computer must implement the same set of protocols to be able to interoperate. Even if both systems use protocols which are “OSI Standard,” there is no guarantee that they will talk to each other. Common Network Protocols A typical corporate network has many different protocols in use at any one time. This is a legacy of the days when every vendor implemented his own network standards, which resulted in serious interoperability problems. In the recent past, there has been a significant move toward the use of TCP/IP as a standard protocol for all network communications. Most vendors today offer TCP/IP as an option, and many have adopted it as a standard. If all systems on a corporate network “speak the same language,” it is much easier to get them to talk to each other. It is also much simpler to configure and operate the computers on that network, as they need speak only one language, rather than several, to interconnect with all other computers. The three protocols selected for discussion in this section are among the most common. They are provided here as a set of examples of how various protocols are implemented, to provide an introduction to the complexities of internetworking. TCP/IP. The Transmission Control Protocol/Internet Protocol (TCP/IP) was developed by two engineers in 1974. They built on the work of a Ph.D. student at Harvard, who had developed the idea for ethernet as his doctoral thesis. By 1982, these ideas had been further developed by others and were published as standards by the U.S. Department of Defense. TCP/IP is a very simple set of protocols. It implements only those pieces that are necessary for reliable and efficient network communications and leaves the more complex tasks (such as network routing) to other applications. As a result, TCP/IP is relatively simple to implement and therefore has been implemented on every computer operating system. It has become the lingua franca of networking. © 1999 by CRC Press LLC Communications and Information Systems 18-19 The mapping of TCP/IP into the OSI model is shown in Table 18.2.5. Note that TCP/IP does not implement all layers of the OSI model explicitly, but combines several layers into a single implementation piece. TABLE 18.2.5 Mapping of TCP/IP into the OSI Model OSI Layer Name Application Presentation Session Transport TCP/IP Implementation File transfer protocol (FTP) Simple mail transport protocol (SMTP) Virtual terminal protocol (Telnet) Transmission control protocol (TCP) User datagram protocol (UDP) Internet control message protocol (ICMP) Internet protocol (IP) Address resolution protocol (ARP) Ethernet (coax, twisted pair), token ring, FDDI, ATM, X.25, SONET, LocalTalk, ... Network Data link Physical The protocols listed perform the following functions: File transfer protocol implements the messaging necessary to transfer files from one computer to another. It handles such things as authentication, directory listings, file selection, and datastream format. Most systems implement an FTP command that allows the user to interact with the FTP system through the use of either text commands or a graphical user interface. Simple mail transport protocol implements a standard for the transfer of electronic mail messages between computer systems. It handles such things as system identification, addressing, the separation of the body of the message from the headers of the message, and the termination of the transfer. A typical computer will implement a mail user interface that completely isolates the user from this protocol. Virtual terminal protocol implements a standard for the negotiation and connection of a terminal emulator to a remote computer. Telnet negotiates such things as data representation standards (ASCII, EBCDIC, Latin1), datastream format (8-bit or 7-bit), and control characters to be used for interrupt and flow control. Typical computers implement a user interface that provides interpretation of terminal control codes sent by the remote computer (e.g., “clear screen”), as well as controls for establishing and terminating a terminal emulation session. Transmission control protocol underlies the above protocols and provides a reliable, full-duplex, datastream service between two computers. It establishes a virtual circuit (or connection) between the two systems, then sends and receives data across that connection until the communicating processes request that the circuit be disestablished. TCP depends on IP for sending and receiving the packets that are transferred across the virtual circuit. User datagram protocol is used by processes that do not require guaranteed delivery of data. No virtual circuit is established. The sending process simply addresses the packet to a particular process on the remote computer and sends it across the network. It is up to the processes at either end to ensure that the packet is received properly, as the UDP protocol makes no guarantees. Internet control message protocol is used to send control messages concerning network functions to TCP/IP clients. Instructions such as “reduce your transmission rate” and “your requested destination is unreachable” are transferred at this level. Internet protocol defines the standard datagram that transports information across the network. It does not guarantee delivery, but expects higher-level protocols (such as TCP) to handle that function. It handles packet-level flow control and error checking. © 1999 by CRC Press LLC 18-20 Section 18 Address resolution protocol handles the mapping of a logical IP address to a physical network address. It enables the local computer to determine where the packet must be sent for the next step of its journey across the network. IPX. The Internet packet exchange (IPX) protocol is commonly used in Novell NetWare filesharing systems. It is based on the XNS suite of protocols designed by Xerox. IPX is a lightweight protocol. It is designed to be very efficient (easy to process) and to provide maximum performance on a local network. The downside of lightweight protocols is that they tend not to work as well in large internets. Because they are less complex, they lack the necessary components to work well in complex network environments. The differences between IPX and TCP/IP can be seen by comparing Tables 18.2.6 and 18.2.5. Notice the significantly fewer number of layers in the IPX set of protocols. This simplification is the source of the efficiency of IPX. TABLE 18.2.6 Mapping of IPX into the OSI Model OSI Layer Name Application Presentation Session Transport Network Data link Physical IPX Implementation Netware core protocol (NCP) Sequenced packet exchange (SPX) Internet packet exchange (IPX) Ethernet (coax, twisted pair), token ring, FDDI, ATM, X.25, SONET, LocalTalk, … The protocols listed perform the following functions: Netware core protocol implements directory services, connection services, security, and similar functions. It is essentially the operating system for NetWare servers and, as such, implements far more functionality than the TCP/IP protocol, where the top of the OSI stack is underneath the operating system. Core services can be added to NetWare servers by adding netware loadable modules (NLM), which can be used to add things such as data base engines and gateways to TCP/IP networks. Sequenced packet exchange builds on the services offered by IPX. SPX adds packet acknowledgment to the functions in IPX, so that the sending computer can know for certain that the packet was received at the far end. Note that the use of SPX is not required in NetWare networks. Internetwork packet exchange implements an efficient packet transfer service. The packet contains only address information, data, and a error-checking code, resulting in a packet with little overhead. It should be noted that current NetWare networks can be implemented using TCP/IP protocols. This is appropriate where the network is large (or expected to grow), or where there is already a requirement for TCP/IP, such as in networks connected to the Internet. AppleTalk. AppleTalk was designed by Apple Computer to provide a simple-to-install, simple-tomanage network. It has an extensive set of protocols to provide such services as automatic address selection (for workstations and servers), printer sharing, routing, and security. Anyone who has installed an AppleTalk network has benefited from its simple “plug and play” characteristics. This simplicity for the user comes at the cost of a significant complexity in the protocol. Since AppleTalk must discover on its own what most other protocols require the network administrator to manually configure, it follows that there must be many additional functions, resulting in a necessarily more complex protocol suite. © 1999 by CRC Press LLC Communications and Information Systems 18-21 The major protocols used in AppleTalk are listed in Table 18.2.7. Notice the large number of protocols at the session and transport layers. These additional protocols are used to self-configure the AppleTalk network. TABLE 18.2.7 Mapping of AppleTalk into the OSI Model OSI Layer Name Application Presentation Session Transport Network Data link Physical AppleShare AppleTalk filing protocol (AFP), PostScript AppleTalk session protocol (ASP), AppleTalk data stream protocol (ADSP), zone information protocol (ZIP), printer access protocol (PAP) AppleTalk transaction protocol (ATP), name binding protocol (NBP), AppleTalk echo protocol (AEP), routing table maintenance protocol (RTMP) Datagram delivery protocol (DDP) Ethernet (coax, twisted pair), token ring, FDDI, ATM, X.25, SONET, LocalTalk, … AppleTalk Implementation The protocols listed perform the following functions: AppleShare provides file sharing services. AppleTalk filing protocol handles the communications between the user’s computer and the AppleShare fileserver. It allows users to share files and applications, handle security, and ensure that each user has a current view of the shared file service. PostScript defines the format of documents printed from Macintosh applications. It is a page description language, developed by Adobe. AppleTalk session protocol manages the creation, operation, and destruction of sessions between computers. It organizes the sequencing of function requests (who gets access to what service in what order). AppleTalk data stream protocol is a connection-oriented protocol (like TCP) that reliably transfers a stream of bytes between two computers. Zone information protocol handles the discovery and mapping of the entire AppleTalk network. It finds all connected networks and builds a table that is used by AppleTalk to find other computers, and to route packets appropriately. Printer access protocol sets up and manages connections between the user’s computer and printers or print servers. AppleTalk transaction protocol is used to pass requests and responses reliably between computers. It ensures that the receiving computer accurately gets the request and also ensures that the answer returns reliably to the requester. Name binding protocol translates between the name of a computer and its address. AppleTalk echo protocol responds to a request by a remote computer by simply acknowledging that an echo request was received. This function is used to be sure that a remote computer is active. Routing table maintenance protocol is used by AppleTalk routers to discover and keep track of the configuration of the AppleTalk internet. It is used to maintain routing tables or internal network maps. Datagram delivery protocol is the core of the AppleTalk protocol suite. It moves packets of data from one computer to another, but does not guarantee reliable delivery. © 1999 by CRC Press LLC 18-22 Section 18 Defining Terms AppleTalk: A set of protocols developed by Apple computer to interconnect Macintosh computers and peripherals. FTP: File transfer protocol. IP: Internet protocol. IPX: Internetwork packet exchange. ISO: International Standards Organization. Layer: One part of a protocol stack. OSI model: The model of network interfaces as a seven-layer entity. Protocol: A standard way of doing things. SPX: Sequenced packet exchange. TCP: Transmission control protocol. Telnet: Virtual terminal protocol. References Comer, D. 1988. Internetworking with TCP/IP. Prentice-Hall, Englewood Cliffs, NJ. Cypser, R.J. 1991. Communications for Cooperating Systems: OST, SNA, and TCP/IP. Addison-Wesley, Reading, MA. Sheldon, T. 1993. Novell NetWare 4. McGraw-Hill, New York. Sidhu, G.S. et al. 1990. Inside AppleTalk. Apple Computer. Tittle, E. and Robbins, M. 1994. Network Design Essentials. Academic Press, New York. Wilder, F. 1993. A Guide to the TCP/IP Protocol Suite. Amacom, New York. © 1999 by CRC Press LLC Communications and Information Systems 18-23 18.3 Communications and Information Theory A. Brinton Cooper, III Communications Theory Communications science and technology have been experiencing unprecedented growth and impact in the last decade of the 20th century. The reasons are twofold. Advances in making electronic devices smaller and in reducing their requirement for electric power permit the use of complex communication processing, coding, and decoding functions that until recently languished in the research literature. On the other hand, spurred by the demands of society for more and better communication functions, scholars and their students continue to bring forth more efficient algorithms for using communication channels and to advance networking from the traditional techniques used by the “telephone company” to a plethora of methods for networks of portable and mobile terminals. Yet, these advances could not have occurred without a firm theoretical basis. Reliance on intuitive and untested ideas in communications often leads down the wrong path. Shannon’s idea of error-free communications at nonzero rates through noisy channels and Nyquist’s notion that all the information in a continuous band-limited signal is contained in a string of samples that are taken at a uniform, finite rate ran counter to the collective intuition of practicing communications engineers of 1948 and 1928, respectively. In what follows we introduce communications theory and information theory, believing that an introduction to the principles upon which a discipline is based provides a firm foundation for exploring the technology which it underlies. Structure and Functions of a Communication System The purpose of a communications system is to convey information. In order that the information be transmitted efficiently and economically and be received reliably and with fidelity, several important signal processing operations are performed prior to sending the information over the (usually noisy) channel. Complementary functions are performed at the receiver. The center of focus of a communications system (Figure 18.3.1) is the channel. A channel is a medium of communication by which information from a source is conveyed to a destination (or sink). The source can be a microphone, a sensor, a video camera, a compact disc player — anything that produces an electrical signal which represents information. The destination is the recipient of the source signal. It can be a piece of magnetic tape, a computer file, a meter, or a loudspeaker. For the moment, we consider “information” to have its intuitive meaning. Information is placed on the channel (or sent over the channel) by a transmitter and obtained from the channel by a receiver. The physical nature of most communication channels renders them unsuitable for conveying information in the form produced by the source. That is, the signal produced by a channel transmitter is quite different from that produced by a source. The modulator bridges this gap by modifying one or more parameters of the transmitter output signal in accordance with the output of the source. The corresponding demodulator in the receiver recovers the information signal from the received channel waveform. FIGURE 18.3.1 A simple communication system. © 1999 by CRC Press LLC 18-24 Section 18 Consider the example shown in Figure 18.3.2. The source is an optical sensor that produces a continuous electrical waveform representing a picture or image of something. This waveform is processed in order to make it more suitable for transmission over a binary communication channel. A source encoder converts the analog waveform into a string of binary digits (analog to digital conversion) and removes redundant elements (compression) that can be reconstructed at the destination. If the information is sensitive, there may be a stage of encryption in order to prevent eavesdroppers from viewing the image. A channel encoder adds to this digit string some extra bits which are used by a channel decoder in the receiver to correct patterns of errors that can be caused by noise, interference, or distortion in the channel. A transmitter converts the digital signal to one that can be carried over the channel. At the receiving end, each of these functions is reversed, in sequence, to reproduce the original optical picture or image. FIGURE 18.3.2 Example transmission system. Signals The contents of a communication are carried by a signal, typically a time-varying voltage, current, or electromagnetic wave. Information is actually conveyed in the time variations. Important distinctions among types of signals permit the development and use of methods and tools for analyzing and processing signals. Characteristics of Signals. A signal s(t) is said to be periodic if and only if s(t + T) = s(t), –∞ < t < ∞, where T is said to be the period of s(t). Any signal which is not periodic is said to be aperiodic. Familiar periodic signals include the sinusoid sin(2πft + φ), which has period t = 1/f. This is an important signal model in communication theory. A signal which is a completely specified function of time at every instant of time is said to be deterministic. By contrast, the precise values of a random signal cannot be predicted in advance. Noise in a radio communication system often is modeled as a random signal. In many cases of interest, the output of a noisy communication channel can be modeled as the sum of a deterministic and a random part, r(t) = s(t) + n(t), where r(t) is the received waveform, s(t) is the transmitted signal, and n(t) is the noise induced by the channel. A signal can be a continuous function of a continuous-time random variable, such as the sinusoid used above. On the other hand, in many systems, signals are sensed and measured only at discrete values of the time variable. This practice is the foundation of digital communications. Further, the signal s(t) may not be a continuous function of time, but rather it may assume values from a finite set only. Such a signal is said to be a quantized or a discrete-valued signal. Signal Representations. Analyzing the effects on communication signals of processing circuits, propagation paths, and noise and interference requires accessible mathematical representations of those signals. What follows introduces widely used and powerful representations for deterministic signals, both periodic and aperiodic. More complete treatments can be found in many excellent texts. Several are mentioned at the end of this section. Suppose the signal s(t) represents the waveform produced by someone singing. A microphone can be placed to capture the sound for display on an oscilloscope, thus permitting the viewing of s(t) as a function of time. Now we know, for example, that men and women sound fundamentally different when they sing. Women are said to have voices that are “higher” than those of men. Yet, examining on the oscilloscope the waveforms produced by male and female singers may fail to reveal these differences to all but a trained observer. However, they would be captured easily by a graphical display of the frequency spectrum, or simply the spectrum of the singer’s voice. The usual spectral representation is given by the Fourier transform S(f) of the signal: © 1999 by CRC Press LLC Communications and Information Systems 18-25 S( f ) = ∫ ∞ −∞ s(t ) exp( j 2π ft ) dt (18.3.1) We also say that S( f ) is the frequency domain representation of signal s(t) and that it specifies the spectral composition of the signal. Such a tool makes it possible to determine easily the effects on a signal of filters, noisy and bandwidth-limited communication channels, antennas, and other devices through which it may pass. Bandwidth. The breadth of spectral occupancy of a signal is called its bandwidth. This term makes specific a notion of the ‘width’ of the signal. As an important parameter of the communication channel, bandwidth measures the range of frequencies over which the channel passes energy with relatively little attenuation. (In this regard, a channel behaves as does any filter, a device which permits one or more bands of frequencies to pass relatively unattenuated while deeply attenuating all other frequencies.) A signal whose spectrum is limited to a range of frequencies near the origin is said to be a low pass signal (Figure 18.3.3), while one whose spectrum is centered about some frequency away from the origin is called a band pass signal (Figure 18.3.4). Channels (and filters) have similar designations. FIGURE 18.3.3 Low-pass filter. FIGURE 18.3.4 Band-pass filter. It is not possible for physical channels and real signals to assume perfectly rectangular functional forms since neither mathematical discontinuities nor infinite slopes can occur in physical signals. Therefore, the definition of bandwidth must specify the level of attenuation at which spectral width is measured. Two definitions are commonly used. The half power or three-decibel (dB) bandwidth is defined by the maximum and minimum frequencies in the signal where its amplitude has dropped to 1 / 2 of its peak value. The definition comes from the decibel, the logarithmic expression of the ratio of power loss or gain: Power ratio (decibels) = log10 Pout Pin © 1999 by CRC Press LLC 18-26 Section 18 Notice that –3 dB corresponds to a power ratio of one half. A major virtue of the 3-dB bandwidth is the ease with which it can be measured in the laboratory or in the field. Noise in Communication Systems. Whenever a complex communication signal is transmitted through a channel, the channel output is not likely to be an exact replica of what was transmitted. If the channel is linear and time-invariant*, two types of signal modification can occur: (1) distortion, caused by a nonuniform attenuation of the signal spectrum or by a nonconstant time delay of all parts of the signal spectrum, and (2) noise. Sources of noise include thermal vibrations in the receiving circuits, nearby rotating machinery or automotive ignition systems, and natural processes in the propagation path. Most commonly experienced in communication systems is white Gaussian noise, so called because its spectrum is constant over all frequencies and because its amplitude follows a Gaussian probability law. Noise is, of course, a random phenomenon. Otherwise, the communications receiver could simply subtract the (known) noise waveform from the channel output, leaving a noise-free signal that is proportional to what was transmitted. Some Communication Functions Modulation. The spectra of signals such as speech, video, or computer data are concentrated at the low frequencies and are examples of baseband signals. The transmission of such signals by wire or cable for short distances, perhaps on the order of the dimensions of a room or office, is not difficult. An example is the wire or cable connection of loudspeakers to the amplifier of a home entertainment system. However, speech or music transmitted over 10, 50, or 100 mi of cable would experience so much attenuation that whatever is received would be quite useless. By contrast, bandpass signals, having spectra concentrated at much higher frequencies, are useful for such applications because they can be transmitted by the propagation of electromagnetic energy from an antenna (e.g., by radio). Radio transmission requires an antenna having physical dimensions which are approximately the wavelength of the signals to be transmitted. Typical wavelengths for mobile radio signals are approximately 1.0 cm to 1.0 m, while the wavelength of a 3000-Hz audio signal is 105 m, quite an impractical size for a transmitting antenna. Either baseband or bandpass signals can be transmitted via radio if suitable modulation of a sinusoidal carrier is used. Modulation is the process of combining an information-bearing baseband signal s(t) with a bandpass signal so that the combination is suitable for transmission over a specified communication medium. For analog modulation processes, the bandpass signal is usually a simple sinusoid, Asin(2πft + φ), the frequency of which is known as the carrier frequency f, the amplitude A, and the phase angle φ. A modulator modifies one or more of these parameters in accordance with the time variation of s(t) as discussed below. Pulse or digital modulation requires first that s(t) be digitized. The resulting string of binary digits modulates a parameter of the carrier in accordance with the sequence of its values. Demodulation/Detection. The communications receiver must recover s(t) from the received modulated signal for presentation to the destination. This demodulation process is more complicated than modulation because the signal has most likely been corrupted by noise in the transmission. In an analog receiver, the demodulator performs an estimation process by which it tries to determine the time-varying value of some parameter of the received carrier, e.g., its amplitude or frequency. The detector is concerned with whether or not a signal is present in noise. In the detection problem, the receiver is trying to decide which of a finite number (in this case, two) of signals was actually sent. Again, this is not a trivial problem, as the received signal is accompanied by channel noise. Processing of the Information. Certain operations can be performed on s(t) to prepare it for more efficient transmission. These operations include compression, a form of source coding which can reduce the bandwidth occupied by the signal and make the modulation, transmission, and demodulation processes more efficient. In a digital communication system, the binary representation of s(t), can be further * A linear, time-invariant channel transforms a signal s(t) into As(t+T) + n(t) where A is constant with time. © 1999 by CRC Press LLC Communications and Information Systems 18-27 encoded in order to protect against channel errors. These operations are described in some detail under the subsection on “Information Theory.” Communications Techniques Analog Modulation Amplitude Modulation and Its Variants. Amplitude modulation (AM) continues to be widely used in everyday communications including standard broadcast radio, the Citizens’ Radio Service, international shortwave broadcasting, and certain segments of the Amateur Radio Service. Its popularity stems from the relatively simple transmitters and receivers required. Let s(t) be the baseband signal to be transmitted. Then the amplitude-modulated carrier is v(t) = A[1 + Ks(t)]cos2πfct, where the carrier amplitude is A, the baseband signal amplitude is K, and the carrier frequency is fc. We have omitted the random phase angle φ since it is of no significance in AM. Now, consider the modulating signal s(t) = cos2πfmt. v(t ) = A(1 + K cos 2 π fm t ) cos 2 π fc t Trigonometric expansion gives v(t ) = A cos 2 π fc t + The Fourier transform of v(t) is V( f ) = A KA δ( f − fc ) + δ( f + fc ) + δ( f − fc − f m ) + δ( f + f c + f m ) 2 4 KA δ( f − fc + f m ) + δ( f + fc − f m ) + 4 where δ(t), the dirac delta function, is defined by δ(t ) = 0, t ≠ 0, and KA KA cos 2 π( fc + fm ) t + cos( fc − fm ) t 2 2 (18.3.2) [ ] [ ] (18.3.3) [ ] δ(t ) dt = 1 ∫ ∞ −∞ This function provides a convenient way to represent sampling and time-shifting operations. Equation (18.3.2) shows that the transmitted waveform is the sum of three terms: a carrier term and two sidebands, with each of the latter carrying the information-bearing modulation. For K ≤ 1, the power in the carrier term is proportional to A2, while the power in each sideband is proportional to A2/4. Hence, two thirds of the power in the transmitted signal carries no useful information. It is merely “along for the ride.” Further, each sideband carries the same information, so the signal occupies twice the bandwidth that is necessary to represent the modulation faithfully. Finally, on noisy channels, it is important to measure the ratio of signal power to noise power at the receiver output, (SNR)O. Let (SNR)I be the signalto-noise ratio at the receiver input. The quotient of these two ratios, a “figure of merit” for modulation schemes, (SNR)o K2 = (SNR) I 2 + K 2 is never greater than one third for AM. Compare this with other methods given below. Each of the following three variants on AM is designed to provide savings in power and bandwidth. © 1999 by CRC Press LLC 18-28 Section 18 Double sideband-suppressed carrier (DSB-SC) suppresses the transmission of the carrier itself, thus obtaining more efficient use of the power transmitted. However, it requires the same channel bandwidth as ordinary AM. The modulation process is represented as follows: v(t ) = As(t ) cos2π fc t The Fourier transform of the DSB-SC signal is V( f ) = 1 A S( f − fc ) + S( f + fc ) 2 (18.3.4) [ ] (18.3.5) Now notice that all components of the signal carry the information being conveyed. No part of the signal is along for a “free ride”. In the receiver, the demodulator first multiplies the received signal by Acos2πfct: v0 (t ) = AA′s(t ) cos(2 π fc t + φ) cos(2 π fc t ) = ( AA′ 2) cos(4 π fc t + φ) + ( AA′ 2) s(t ) cos φ (18.3.6) Notice that the first term can be removed by a simple lowpass filter.* This leaves a term which is proportional to the information s(t). However, as the value of φ (the random phase difference between the carrier and the locally generated copy of the carrier) approaches π/2, the amplitude of the coefficient of s(t) will be small, and the output signal may be too weak to be useful. Even worse, φ may vary randomly with time, thus distorting the information s(t). In practice, a phase tracking system is commonly used to keep the local oscillator “in phase” with the received signal. When such tracking is used, the receiver is said to perform coherent detection. The figure of merit for the DSB-SC is (SNR)O/(SNR)I = 1. This improvement over AM is a consequence of not having a carrier-only component in the spectrum. Vestigial sideband (VSB) modulation is used in commercial television to convey video information. One entire sideband and a very small portion of the other are transmitted. Although carrier may or may not be sent, depending upon the application, the low cost of envelope detectors has dictated that commercial TV use VSB with a small carrier component. This is helpful because television video signals have large bandwidths and carry significant amounts of information in the low frequencies. One can think of generating single sideband (SSB) modulation by generating an ordinary AM signal, then filtering out the carrier and one of the sidebands, so that the transmitted signal is either the upper sideband (USB modulation) or the lower sideband (LSB modulation) only. The Fourier transform shows that this is equivalent to a linear translation of the modulating signal s(t) from baseband to frequencies near fc. Since the transmitted signal is essentially a replica of the information-bearing signal s(t), SSB is a very efficient modulation technique, placing all the transmitted power into transmitting the information and doing so with a minimal use of bandwidth. As with DSB-SC, the SSB receiver does not degrade the signal-to-noise ratio of the received signal in noise. Frequency Modulation. The use of frequency modulation (FM) in standard radio broadcast communications has surpassed that of AM due in part to the higher fidelity afforded by the larger channel bandwidths and in part to the inherent resistance of FM receivers to propagation disturbances, electrical storms, and human-induced interference as well as the lack of many atmospheric propagation disturbances in the FM broadcast band.** FM is also used in a variety of mobile and public safety applications, in certain segments of the Amateur Radio Service and in military communications. A low-pass filter attenuates all frequencies above its cut-off frequency while permitting frequencies at or below that frequency to pass unhindered. In the example of DSB-SC demodulation, the cut-off frequency should be greater than fc but less than 4fc. © 1999 by CRC Press LLC * Communications and Information Systems 18-29 Let s(t) be the information-bearing baseband signal. The frequency modulated signal is  v(t ) = A cos2 π fc t + 2 πK  ∫ s(τ) dτ t 0  (18.3.7) Observe that the envelope of v(t) is constant, independent of the message signal, s(t). As we did for AM, let us exhibit important properties of FM by studying a carrier that is frequency modulated with a single sinusoid. Let s(t) = Amcos2πfmt. Substituting into Equation (18.3.7), differentiating and dividing by 2π give the frequency at any instant of time as fi (t ) = fc + KAm cos 2 π fm t = fc + ∆f cos 2 π fm t where ∆f = KAm. The quantity ∆f is known as the frequency deviation and indicates the largest difference between the actual, instantaneous frequency of the FM signal and the carrier frequency fc. Note that, while ∆f is proportional to the amplitude of s(t), it is independent of the frequency of s(t). The derivative of the instantaneous frequency gives the phase angle of the FM signal as a function of time: φ i (t ) = 2 π f c t + ∆f sin 2 π fm t fm The quantity β = ∆f/fm is called the modulation index of the FM signal. It is the maximum difference between the instantaneous value of the time-varying phase of the signal and the phase of the unmodulated carrier fc. The modulation index β will be used to distinguish between two types of FM systems. Write the FM signal as a function of time: v(t ) = Am cos(2 π fc t + β sin 2 π fm t ) (18.3.8) Expanding Equation 18.3.8 using the trigonometric formula for the cosine of the sum of two variables gives: v(t ) = A cos 2 π fc t cos(β sin 2 π fm t ) − A sin 2 π fc t sin(β sin 2 π fm t ) When β is much smaller than one, this quickly simplifies to v(t ) = A cos 2 π fc t − Aβ sin 2 π fc t sin 2 π fm t (18.3.9) (18.3.10) The resistance of FM broadcast signals to nighttime fading and other propagation anomalies is actually due to the use by FM of carrier frequencies between 88 and 108 MHz (in the VHF band) where signals travel through the atmosphere in “direct line of sight” from transmitter to receiver. Contrast this with the signals between 0.540 and 1.600 MHz (used by standard AM broadcast radio) which travel through the ionosphere where they are reflected back to earth, often hundreds or thousands of miles from the transmitter. Even so, FM is far less vulnerable to additive noise and interference (such as lightning) because it carries information in the argument of a sinusoid. Further, most FM receivers enhance this effect by passing the received signal through a limiter circuit which prevents the amplitude of the received signal from varying above a set value. ** © 1999 by CRC Press LLC 18-30 Section 18 which represents the sum of two signals at frequency fc, one of which has constant amplitude and the other of which has an amplitude which is proportional to the modulation. Thus, the narrowband FM signal seems to exhibit some amplitude modulation and will not have a constant amplitude. After trigonometric expansion, Equation (18.3.10) shows v(t ) = A cos 2 π fc t + 0.5 Aβ cos 2 π( fc + fm ) t − 0.5 Aβ cos 2 π( fc − fm ) t (18.3.11) Equation (18.3.11), which represents a narrowband FM signal, has the appearance of the AM signal shown in Equation (18.3.1) except for a change of sign in the last term. Thus, one might conclude that a narrowband FM signal has the same bandwidth as an AM signal. This approximation can be shown to hold so long as β is less than one. On the other hand, when β is larger than one, the small angle approximations used in narrowband FM do not offer a valid representation of the signal. Careful analysis of the mathematics shows that the wideband FM signal with sinusoidal modulation has the following characteristics: • Its spectrum consists of a carrier at frequency fc and sidebands at all integer multiples of fm above and below fc. • The amplitude of the carrier component of the spectrum is a function of the modulation index β. This occurs because the envelope of v(t) is constant for an FM signal, so any power in the sidebands must be taken from power in the unmodulated carrier. Although the infinite number of sidebands suggests that the transmission bandwidth of the wideband FM signal must also be infinite, it is found that, at frequencies sufficiently far from the carrier, the sideband amplitudes are of insignificant magnitude to cause noticeable distortion in the demodulated signal. It has been found that the effective transmission bandwidth of an FM signal with sinusoidal modulation is WT = 2∆f + 2fm = 2∆f (1 + 1/β). A more accurate estimate of the required transmission bandwidth can be made by choosing to retain all sidebands whose amplitudes exceed some specified fraction of the carrier amplitude. Determination of the transmission bandwidth required to support this criterion then becomes a numerical exercise. Digital Communications. Most modern communications advances are occurring in the rapidly expanding field of digital communications. In digital communication systems, all information is represented as strings of symbols that take values from a set of finite size. The most common is the binary representation, in which all information is represented as strings of binary (0,1) symbols. More generally, however, “digital” can imply a finite set of any size. In a quartenary system, for example, information is represented as strings of symbols that take values from an alphabet of four symbols. Industry is working on digital television, implementing what is essentially an analog function using digital techniques and circuits. Why is this happening? In communication channels, digital signals are fundamentally more robust and resistant to corruption than are analog signals. Digital signals do not experience intermodulation; they do not exhibit crosstalk; if the amplitude is abruptly cut off by a faulty amplifier (clipping), no signal distortion results. When errors occur in a digital bitstream, error control coding techniques can be used to seek out and reverse the errors. Typically, there is no corresponding way to reverse the distortion of an analog signal. To communicate over very long distances, it is quite easy to regenerate digital pulses at sufficiently frequent intervals to avoid error or loss. The analog counterpart is a repeater that must exhibit a linear relationship between output and input signals over a wide range of input signal amplitudes. In manufacturing, the cost of digital hardware is lower than that of analog circuitry, and the reliability is higher. Digital bitstreams from a wide variety of sources can be transmitted over the same channels. They can even be intermingled and stored easily on magnetic media for later transmission, thus giving impetus to multimedia communications. Digital communications are easily encrypted to protect them from eavesdropping; this is quite difficult for analog signals and usually results in their being digitized when such protection is required. Finally, with the advent of modern, computer-controlled © 1999 by CRC Press LLC Communications and Information Systems 18-31 telephone switching systems, the exclusive use of digital communications signals affords great ease in switching and multiplexing operations. On the other hand, digital communications poses certain challenges. If analog information is to be transmitted via a digital channel, it must first be converted into a digital format. This requires two basic steps: Sampling: Values of the analog signal are measured (“sampled”) at equally spaced time intervals. Nothing that occurs between the sample values will affect the transmitted signal. In fact, according to Nyquist’s sampling theorem, all the information contained in the original, continuous-time signal is contained in the samples, so long as they are taken at least 2B times per second, where B is the highest frequency component in the signal. Quantization: By itself, sampling is not adequate to prepare an analog signal for digital transmission. Within the limits of resolution of the sampling circuitry, each sample is a value of a continuous variable. In order to transmit the sample values in finite time, they must be represented by a finite string of symbols. Therefore, each sample is rounded or truncated to its “nearest neighbor” in the finite set; the process is known as quantization. For example, if the amplitude of the continuous signal remains within the interval (–Amax, Amax), the quantizer divides this interval into L levels and the quantized value is taken as the midpoint of the interval in which the continuous value falls. Each sample value, therefore, could be in error by plus or minus half the value of the sample interval. These sample values are expressed as binary numbers which are transmitted over the channel. In the receiver, they are recovered and a stepwise approximation to the original signal is built. If this stepwise approximation is completely correct, it still differs from the actual continuous signal because of the loss of amplitude information caused by the quantization process. While smoothing circuits can remove the discontinuities, they cannot assure recovery of the correct value of the analog signal. Thus, there is a residual quantization error that can be minimized in the design but never eliminated. For example, assume the use of uniform quantization (all quantization levels are the same size). If the average power in the original signal is S, and if L = 2R where R is the number of bits used to encode each sample, then the signal-to-noise ratio at the quantizer output is  3S  SNRQ =  2  2 2 R  Amax  Additional functions such as compression and error control may be required prior to transmission over the channel. These are discussed in the subsection on “Information Theory.” Baseband Digital Modulation. Baseband digital modulation schemes are suitable for transmission of digital information over wire or cable, for example. A typical baseband system is shown in Figure 18.3.5. FIGURE 18.3.5 Typical digital baseband transmitter. For each symbol to be transmitted, a waveform is chosen as its representation on the channel. Examples of waveform selection schemes include the following. • Nonreturn-to-zero (NRZ) schemes represent a binary ONE as +V volts and ZERO as –V volts for some design value of V. • In return-to-zero (RZ) systems, ONE is represented by +V volts and ZERO by a voltage of value zero. © 1999 by CRC Press LLC 18-32 Section 18 • In phase-coded representations, a voltage transition occurs during each symbol interval, whether or not the symbol has changed value from the previous interval. Such a property is useful in magnetic recording systems and optical communications where arbitrarily long, constant voltages (which approximate direct current) cannot be physically transmitted. For example, in the popular Manchester code, ONE is represented by a positive pulse for one half the symbol period and a negative pulse for the other half, while ZERO uses the same scheme with the polarities reversed. Bandpass Digital Modulation Schemes. The most general form of bandpass-modulated signal is v(t) = A(t)cosφ(t). Note that the amplitude A(t) or phase φ(t) of the sinusoid, or both, can vary with time and, hence, carry information via digital modulation. More specific to our uses, we can also write ϕ(t ) = 2 πf0 (t ) + φ 0 (t ) We now consider a few important examples of digital modulation. We assume that the phase of the carrier is used in the demodulation process, so that the receivers are coherent. In general, coherent modulation gives a lower error probability for a given signal-to-noise ratio than does noncoherent modulation. We consider channels in which additive, zero-mean, stationary, white Gaussian noise is added to the signal. Thus, the binary data recovered at the receiver are not guaranteed to be free of errors. If a binary communication signal is received with power P (watts) at a speed of B bits per second, each binary symbol (or bit) contains Eb = P/B joules of energy. We call Eb the signal energy per bit. A characteristic of white Gaussian noise is a uniform spectrum having constant noise power in every unit of bandwidth. If this power is N0 watts per Hz of bandwidth, then a receiver of bandwidth W will intercept N0W watts of channel noise power. In such cases, it is useful to consider the ratio Eb /N0, a dimensionless signal-to-noise ratio often called the ratio of signal energy per bit-to-noise spectral density. Digital modulation schemes are compared by plotting the probability of error per bit, Pb, as a function of Eb /N0. Phase-shift-keying (PSK) carries digital information in discrete shifts of the carrier phase. The modulated waveform is v(t ) = 2E 2 πi  cos ω 0 t +  T M i = 1,K, M, 0≤t≤T where E is the signal energy and T is the duration of the transmission of a symbol. When M = 2, this is called binary PSK (or simply PSK). The general case (M ≠ 2) is often denoted MPSK. The average probability of error in the binary case is Pe = 1 π ∫ ∞ Eb N0 exp − x 2 dx ( ) Of all the modulation techniques considered, PSK offers the minimum probability of error for fixed Eb /N0. Binary frequency-shift-keying (FSK) is one of the oldest digital bandpass modulation forms in existence. It was used extensively in early dial-up computer modems at speeds up to 1200 bits/sec, beyond which more sophisticated modulation methods are needed in order to signal faster in the same bandwidth. Each discrete source symbol is represented by one of two frequencies, fi: v(t ) = 2A cos(2 π fi t + φ) T i = 1, 2 0 ≤ t ≤ Tb where Tb = the length of a transmitted bit. © 1999 by CRC Press LLC Communications and Information Systems 18-33 The arbitrary, constant phase term is represented by φ, and the general (nonbinary) case is usually denoted MFSK. The case of interest here is coherent FSK and its performance is given by 1 π Pe = ∫ ∞ Eb 2 N0 exp − x 2 dx ( ) It is interesting to note that the signal-to-noise ratio for coherent binary FSK must be twice that for coherent binary PSK to give the same probability of error. Amplitude-shift-keying (ASK) carries the information in the discrete-valued amplitude of the signal. 2 Ei (t ) cos(ω 0 t + φ) T v(t ) = i = 1, 2 When one of the amplitude values is 0, this gives rise to “on-off” keying, surely the oldest form of digital modulation. It was commonly used in radiotelegraphy from the earliest days of radio and continues in use by amateur radio operators. In mid-1995, the U.S. Navy retired its remaining radiotelegraph operators because of the very slow rate of information exchange afforded by ASK. Information Theory Information is the commodity in which communication systems deal. Information is conveyed to the commuter listening to a news broadcast on the car radio during morning rush hour; it is the response to a customer’s inquiry about his or her bank account; it is what the patient learns following a physical examination. In each case, the content of the information cannot be completely predicted. The information is “news”; it is something of a surprise. Information theory provides mathematical bounds on the performance of communication systems. Specifically, it affords: • A lower bound on the number of discrete symbols (or on the bandwidth of a continuous signal) necessary to represent a source without loss of information • An upper bound on the rate at which information can be reliably transmitted over a noisy channel It relies heavily on concepts of uncertainty, by which the nature of information is represented. Therefore, probabilistic concepts are needed. Information theory provides a formal definition of information but does not provide a subjective meaning. Information is provided by a source and is transmitted over a channel where it is accepted by a destination or a sink. Sources (as well as channels and sinks) can be discrete or continuous depending upon the representation of the information. Sources and Source Coding A common model of an information source is something that periodically emits one symbol X from a known alphabet, A = {x0, x1, …, xK–1}, according to the probabilities: pj = P X = x j ( ) j = 0,1,K K − 1 If successive source symbols are statistically independent, this is said to be a discrete memoryless source. From the event xj, an observer gains an amount of information defined by i(xj) = log2(1/pj), from which we conclude: © 1999 by CRC Press LLC 18-34 Section 18 1. A certain event produces no information: i(xj) = 0, pj = 1. 2. The occurrence of an event cannot take away information that the observer already has: 0 ≤ p j ≤1 ⇒ i(x j) ≥ 0; that is, information is always positive. 3. Unlikely events produce more information than do likely events: pj < pi ⇒ i(xj) > i(xi). 4. The contributions of successive symbols to information are additive: i(xi, xj, xk) = i(xi) + i(xj) + i(xk). Because the logarithm is taken to the base 2, we call the unit of information the bit, a contraction of binary digit. Example Consider a fair coin (one for which the chances of a head and a tail are equal). We say that one bit of information is conveyed each time the coin is tossed. Example A modem which communicates to another modem across telephone lines by sending one of four audio tones every second is sending information at the rate of two bits per second. The average value of the information per source symbol is an important quantity. It is given by H ( A) = E i x j   [ ( )] = ∑ p i( x ) = ∑ p log  p1  K −1 K −1 j j j 2 j =0 j =0 j and is called the entropy* of the discrete memoryless source with source alphabet A. From this definition, it can be shown that 0 ≤ H(A) < log2K, where K is the size of alphabet A. In addition, H(A) = 0 if and only if pj = 1 for some j and all other probabilities are zero. The upper bound on entropy is given by H(A) < log2K if and only if all symbols in the alphabet are equiprobable, that is, pj = 1/K for all j. The output of the typical information source is rarely a sequence of equiprobable symbols, so its entropy is rarely maximum, and its transmission is not as efficient (in terms of bits of information per symbol) as is possible. In order to improve this efficiency source coding is employed. Typically, a source code assigns short binary sequences (or code words) to more probable source symbols and long binary sequences to less probable symbols**. Such a variable length code should also be uniquely decodable so that the original source sequence can be recovered without ambiguity from the encoded sequence. We confine our attention to binary sequences. Figure 18.3.6 shows a source coding scheme in which the output symbol sj of the discrete memoryless source is encoded into a binary string bj = (b0 , b1 , K, bl j ). FIGURE 18.3.6 Source encoding. If the string bj has length lj, then the average codeword length L is simply L = Σ K=−01 p j l j , and we j say that the source code produces L (average) bits per source symbol. Now, let Lmin be the smallest possible value of L. Then the coding efficiency of the source code is defined to be η = Lmin / L . The Mechanical engineers are likely to be familiar with the entropy (k lnΩ) from statistical mechanics. Here, k is Boltzmann’s constant, Ω is the number of states in the system, and ln is the natural logarithm. This entropy measures quantitatively the randomness of the system and, therefore, is highly suggestive of information theoretic entropy. The relationship between the entropies has been the object of study by scientists and philosophers for years. ** The English language performs a heuristic form of source coding. Frequently used words such as “a”, “is”, “he”, “it”, “we”, “I”, “me”, etc. are composed of a few letters while less-used words such as “subcutaneous” are assigned much longer strings. © 1999 by CRC Press LLC * Communications and Information Systems 18-35 fundamental bound on source codeword length is given by Shannon’s source coding theorem: The average codeword length L for a discrete memoryless source of entropy H(A) is lower-bounded by that entropy — L ≥ H(A). Lossless Compression. The discrete memoryless source provides a valid model for many physical sources, including sampled speech and the outputs of various environmental sensors. Typically, much of the information from such sources is redundant; that is, the same underlying information may be represented in the values of several output symbols. It is prudent to remove this redundancy in order to use less bandwidth and/or time for transmission. Source coding algorithms which remove such redundancy in such a manner that the original source data can be reconstructed exactly are said to perform lossless data compression. They work by assigning short sequences to the most probable source outputs and longer sequences to less probable source outputs. A (variable length) source code in which no codeword is a prefix of another codeword is said to be a prefix code. Prefix codes are always uniquely decodable, and their average lengths obey the inequalities: H(A) ≤ L ≤ H(A) + 1. Asymptotically, the efficiency of a prefix code approaches 1. Because of their large decoding complexities, however, we look for other classes of lossless compression algorithms. The Huffman code tries to assign to each source symbol a binary sequence, the length of which is roughly equal to the amount of information carried by that symbol. The algorithm is straightforward and transparently easy to decode. The average codeword length approaches the source entropy H(A). However, the Huffman code requires knowledge of the probability distribution of source symbols, and this often is not available. The Lempel-Ziv algorithm, by contrast, adapts to the statistics of the source as they are revealed in the source text. A binary source sequence is read left to right and parsed into segments, each of which is the shortest subsequence not encountered previously. Each of these sequences is encoded into a fixedlength binary code sequence. Thus, in contrast to Huffman codes and others, a variable number of source symbols is encoded into a fixed number of code symbols. Lempel-Ziv achieves, on average, a reduction of 55% in the number of symbols representing a source file; Huffman coding typically achieves 43%. This accounts for the enormous popularity of the former. Rate-Distortion Theory. If the source output is a continuous variable x, the number of bits required for an exact representation of a sample value is infinite. If a finite number is used, the representation is not ˆ exact, but is often useful in many applications. The degree to which the representation x is inexact is called distortion and can be thought of as the cost associated with representing an exact source by an ˆ approximation. We write distortion as D( x, x ). A typical distortion function is the familiar squared-error criterion: ˆ ˆ D( x, x ) = ( x − x ) 2 In a typical application, we think of a source as presenting a number n of samples per second. If each sample is, on average, represented by R bits, then we say that the source produces information at the rate of Rn bits per second. It is common, however, to normalize the source rate to the sample rate, in which case we say that we have a source of rate R (bits per sample). In principle, as R increases, we should be able to make distortion decrease. The mapping of source symbols (or sample values) to strings of bits is called a source code. Naturally, it is the aim of source code designers to minimize the distortion for a given rate. Let D be the expected distortion between two sequences of source values. We call (R,D) a rate distortion pair and say that a given (R,D) is achievable if there exists a source code having rate R and expected distortion D. The minimum rate R for which a source code exists having expected distortion D is called the rate distortion function R(D). The average amount of uncertainty about a random variable X provided by observing another random variable Y is called the conditional entropy H(X|Y). Thus, if the true source entropy is H(X), then the average uncertainty provided by source coding © 1999 by CRC Press LLC 18-36 Section 18 ˆ can be written H ( X | X ). The average amount of information provided by one random variable X about ˆ another random variable X is called the average mutual information between the two and can be written: ˆ ˆ I X, X = H( X ) − H X | X ( ) ( ) ˆ Then R(D) = min I ( X , X ). Shannon showed that it is possible to construct source codes that can achieve distortion D for any rate R > R(D) and that it is impossible to construct source codes that can achieve arbitrary distortion for any rate below the rate distortion bound. There is great interest in studying rate distortion bounds for various coding schemes. One of the oldest applications is in a speech compression application where 8000 samples per second are taken and quantized as eight-bit numbers to produce a 64-kb/sec uncompressed signal. Simply using the correlation between adjacent samples reduces the number of bits (hence the required channel bandwidth) by factors of two to four with little additional perceived distortion. Of greater modern interest, however, are the efforts to compress images and video. While lossless compression methods can compress an image by a factor of three, methods employing (guaranteed loss) quantization can compress an image by a factor of 50 with what is claimed subjectively to be little loss in picture quality. A major outstanding problem is that the mean square error as a measure of distortion correlates very poorly with picture quality as judged subjectively by human viewers. A better, quantitative method of evaluating distortion in images and video is sorely needed. Alas, this is a difficult problem, and solutions do not seem immediately forthcoming. Channels, Capacity, and Coding On many channels, satisfactory communication is limited by noise. Whether traveling by coaxial cable or radio, as a signal gets farther from its source it grows weaker. In the cable, this attenuation is caused by the cable’s nonzero resistance, which dissipates energy in the signal. In radio propagation, a variety of phenomena cause an apparent weakening of the signal, but in every case, radio waves spread out in space spherically just as circular waves spread out when a stone is dropped into a still pond. A receiving antenna can be thought of as a window of constant size which intercepts a fraction of the power in the signal. At distance R from the source, all of the signal power P (watts) passes through a sphere of area 4πR2. An antenna of effective area A will, therefore, intercept (P/4πR2)A watts. Thus, radio signals attenuate as the square of the distance over which they travel. In either case, when the signal has been sufficiently attenuated, the noise resident in receiving equipment and the propagation medium can distort the signal or, in fact, cause bit errors. Prior to 1948, communications engineers assumed that the constraints of noise would forever limit not only communications distance, or range, but signal quality as well. In that year, however, Claude Shannon published a remarkable theorem which continues to shape communications research and technology. Shannon’s Noisy Channel Coding Theorem. For a large and interesting class of communications channels, noise limits the rate at which information can be transmitted. In 1948, Shannon showed how to determine the channel capacity, the maximum rate at which information can be sent through the channel at an arbitrarily small error probability. He explained how noise induces uncertainty into the channel output. So, for example, if information from a source is fed to the channel at rate H(X), it exits the channel with rate H(X|Y), the amount of information that the receiver knows about source X when observing Y. Clearly, H(X) ≥ H(X|Y). So, the average amount of information passed through the channel is I(X,Y) = H(X) – H(X|Y). The channel capacity then is defined as the maximum overall values of input distributions of H(X) – H(X|Y); that is, C = max I ( X , Y ) p( x ) © 1999 by CRC Press LLC Communications and Information Systems 18-37 Shannon’s Channel Coding Theorem. There exist channel codes that permit communication at any rate less than capacity with as small an error probability as desired. It is not possible to achieve arbitrarily small error probability with any code at a rate greater than capacity. For example, for a channel of bandwidth W perturbed by additive white Gaussian noise having average power N, when the average received signal power is S, the capacity C of the channel is given by: S C = W log 2 1 +   N This example suggests that information can be encoded so that its transmission over a channel of finite bandwidth with finite signal-to-noise ratio can be received with a probability of error that is arbitrarily close to zero. Prior to this discovery, communication engineers believed that the only way to combat channel noise was to use very narrow bandwidths and send information at very low rates. They felt that errors would be inevitable in the received signal as long as the channel noise had finite power. Although Shannon’s noisy channel theorem is an impressive result, it does not show how to encode the transmitted signals in order to achieve capacity. This fact has provided employment opportunities for many coding theorists and designers for nearly a half century. Further, while most well-designed and correctly used error control codes can provide significant reductions in error rate, most do not provide an arbitrarily small error probability while sending information at capacity. How Coding Improves Communications. While error control coding can provide a reduced error rate at the receiver output, its contribution to system performance is far more profound. Coding works by introducing redundancy into a stream of symbols being sent over the noisy channel. In order to transmit k binary symbols using an n-bit binary block code, n code symbols must be sent during the same amount of time that the uncoded system would send k information symbols. To support this increased transmission rate requires a larger channel bandwidth which admits proportionally more noise power into the receiver (assuming white noise) and which causes, therefore, a greater error rate in the received bit stream. Thus, the code not only has to correct errors at the original uncoded error rate, but also must actually correct errors at a higher error rate caused by the increased noise. For example, suppose a source is transmitting noncoherent binary frequency-shift keying (FSK) at a rate of B bits per second using a transmitted power of P (watts) on a channel corrupted by white Gaussian noise. The energy per bit Eb is given by P/B, and the power spectral density of the noise is N0 (watts/Hz). For binary FSK, the error probability is given by Pe = 1/2 e −2 Eb / N0 . If a code of rate R (bits/symbol) is used, then the energy per bit is reduced to R Eb , and the error probability is actually increased to Pe = 1/ e −2 REb / N0 . Appropriate choice of error control code, however, will reduce the error probability not 2 merely back to Pe, but rather to a significantly smaller number, Pd. Of course, it is always possible to achieve the same reduction in error probability merely by increasing the transmitter power from P to some larger value, PH. The logarithmic ratio, 10logPH/P, of these two values is called the coding gain. Thus, for a given modulation format and channel noise, an important feature afforded by coding is the use of less transmitter power than if coding is not used. Error Control Codes. Perhaps the most widely known scheme for the control of channel errors is the single parity check that is appended to every fixed-length block of binary digits so that the number of ONEs in an augmented block is, for example, always even (or odd; the choice is arbitrary). Suppose, for example, that data are to be transmitted in four-bit blocks (b0, b1, b2, b3). A single parity check position p4 is set to 0 if the number of ONEs in the 4-tuple is even and to 1 if odd. (This is known as even parity. A similar rule for odd parity can be used as well.) A single error in a block of even parity will cause the number of ONEs to be odd. This odd parity condition can be detected, triggering an automatic retransmission of that portion of the data containing the block with the error condition. If the channel error rate is fairly low, this method is convenient and efficient. However, if the channel error rate is not quite so low, more powerful error control is needed. Consider computing parity checks on subsets of the bi. For example, using even parity, compute parity bits p0 on © 1999 by CRC Press LLC 18-38 Section 18 data bits b1, b2, and b3; p1 on b0, b2, and b3; and p2 on b0, b1, and b2. Instead of transmitting 4 bits of information in five binary symbols (as above), we now transmit 4 bits of information using seven binary symbols. This is somewhat more inefficient, but if any single binary digit in the channel is corrupted, the location of that error in the codeword is uniquely determined by noting which parity checks fail. One way to decode such a code is to recompute the parity checks p0, p1, and p2. An error in b0 (only) will cause p1, and p2 to fail. No other single error will cause p1 and p2, and any other parity check to fail. Similarly, a single error in b1 will cause p0 and p2 to fail, etc. This code is called a single error-correcting code because it can correct any single bit error in a block. Comparing this code with the single parity-check code that detects any odd number of errors, we see that the latter adds one redundant bit for every four data bits, while the single error-correcting code adds three redundant bits for every four data bits. We say that the single parity-check code has a rate R = 4/5 = 0.8 bits/symbol, while the single error-correcting code has R = 4/7 = 0.57 bits/symbol. Therefore, the price for error correction is, to a degree, decreased transmission rate. The preceding single error-correcting code is a particular example (the Hamming code) of linear block code (LBC). The parity check computations used in encoding are succinctly expressed in its generator matrix G: 1000011  0100101  G=  0010111   0001110 Encoding of a 4-bit information vector b = (b0, b1, b2, b3) is performed by matrix multiplication: v = bG = (b0 , b1 , b2 , b3 , p0 , p1 , p2 ) where the {pj} are as given above and v is the codeword produced by information vector. Every LBC has a G-matrix having k rows and n columns and is said to be an (n,k) code; k is called the dimension of the code and n is its blocklength. For the foregoing example, (n,k) = (7,4) and the code rate is R= k 4 = = 0.57 bits symbol n 7 The Hamming distance d between two words of an LBC is the number of positions in which they differ. The minimum Hamming distance dmin of an LBC is the smallest Hamming distance for all pairs of codewords. An LBC is guaranteed to correct t = [(dmin – 1)/2] or fewer errors in any word received from a noisy channel. Generally, codes with large values of dmin have small values of code rate R and vice versa, so the communication engineer faces a serious trade-off between attempted transmission speed (high rate) and error correction required to achieve a required fidelity. For example, the VarsharmovGilbert bound tells us that codes with rates greater than some number exist (if only we can find them): d  d  d k d    d  ≥ 1 −  min log 2  min  + 1 − min  log 2 1 − min   = 1 − H  min   n     n  n n  n   n In the receiver, the error correction process is performed by a decoder, a digital (usually) circuit that implements an appropriate decoding algorithm. Of course, any LBC can be decoded by comparing the received word with every code word, choosing the codeword most likely to have been transmitted. In most cases, this maximum likelihood decoding procedure can be done only by looking up all the © 1999 by CRC Press LLC Communications and Information Systems 18-39 codewords in a table, a process which is too complex and time-consuming for most practical communication situations. Fortunately, mathematical decoding algorithms exist for most popular families of LBCs and can decode with much less complexity (but with a higher error rate) than maximum likelihood decoding. Applications of Coding. Coding is found everywhere that digital communication is found. It has been said that if a digital communications system does not use coding, it is probably overdesigned. We find trellis codes in popular computer dial-up modems; Reed-Solomon codes in military communications equipment and in the compact disc player; convolutional codes in outer space. Error control codes will play an important part in the emerging digital cellular telephone systems and in any digital communication application where transmitter (or battery) power is at a premium. Defining Terms Bandpass: The characterization of a signal or a channel, the spectrum of which is centered around a frequency far from zero. Contrast with baseband. Bandwidth: The nominal width of the spectrum of a signal or of the bandpass characteristic of an electronic filter or a communication channel. Baseband: The characterization of a signal or a channel, the spectrum of which is concentrated at low frequencies, typically in the audio and video ranges. Carrier: The nominal frequency at which a bandpass signal exists. For some formats, such as amplitude modulation, the carrier is the center frequency of the spectrum. It is the frequency remaining in the spectrum as the level of modulation is gradually reduced to zero. Channel: A medium of transmission of a communication system. Typically, it refers to the physical medium and includes fiber-optic channels, voice-grade telephone channels, mobile wireless channels, and satellite communication channels. Channel code: A code (q.v.) which is designed to control channel noise-induced errors in a transmission. Channel codes use redundant symbols to detect the presence of and to locate and correct errors. Coherent demodulation: Any demodulation process in which the phase of the arriving signal is known, measured, or estimated with sufficient accuracy to improve the signal-to-noise ratio at the receiver output. Code: A mapping from a set of symbols or messages into strings of symbols. Codes often, but not always, map into strings of binary symbols. Compression: The technique of removing redundancy from a signal so that it can be transmitted in less time or use less bandwidth or so that it occupies a smaller space on storage media. Crosstalk: Leakage of a signal being carried on one communication channel to another. It is an undesirable phenomenon that is often caused when the output of a channel is inadvertently coupled into another. Decibel: A logarithmic expression of power ratio computed by multiplying the common logarithm of the ratio by 10. Demodulate: In a radio receiver, to remove the information-bearing signal from the received signal. Digital communications: The field of communications in which all information is represented as strings of symbols drawn from a set of finite size. Most commonly, the term implies binary communications, but, in fact, it refers equally well to nonbinary communications as well. Distortion: A measure of the difference between two signals. The various forms of distortion usually arise from nonlinear phenomena. Entropy: A measure of the average uncertainty of a random variable. For a random variable with probability distribution p(x), the entropy H(X) is defined as –∑x p(x)log p(x). Huffman coding: A procedure that constructs the source code of minimum average length for a random variable. © 1999 by CRC Press LLC 18-40 Section 18 Intermodulation: The unintentional modulation of one communication signal by another. It is an undesirable effect of having elements of two communication systems in too close proximity to one another. Lempel-Ziv coding: A dictionary-based procedure for source coding that does not use the probability distribution of the source and is nonetheless asymptotically optimal. Linear block code: A channel code with dimension k, blocklength n, and k by n generator matrix G which produces a code word when multiplied by an information block of length k. Modulate: To combine source information with a bandpass signal at some carrier frequency. Noise: A signal n(t), the value of which at any time t is a random variable having a probability distribution that governs its values. Quantization: A process by which the output of a continuous source is represented by one of a set of discrete values. Rate-distortion function: The minimum rate at which a source can be described to within a given value of average distortion. Sample: The value of a continuously varying function of time measured at a single instant. Signal: a time-varying voltage, current, or electromagnetic wave that conveys or represents information. Source: Anything that produces an electrical signal that represents information. Source code: A code (q.v.) that compresses a source signal, reducing its data rate and attempting to maximize its entropy. Spectrum: The representation of a time-varying signal in the frequency domain, usually obtained by taking the Fourier transform of the signal. References Cover, T.M. and Thomas, J.A. 1991. Elements of Information Theory. John Wiley & Sons, New York. Gallager, R. 1968. Information Theory and Reliable Communication. John Wiley & Sons, New York. Haykin, S. 1994. Communication Systems, 3rd ed., John Wiley & Sons, New York. Michelson, A.M. and Levesque, A.H. 1985. Error Control Techniques for Digital Communication. John Wiley & Sons, New York. Proakis, J.G. 1995. Digital Communications, 3rd ed. McGraw-Hill, New York. Sklar, B. 1988. Digital Communications: Fundamentals and Applications. Prentice-Hall, Englewood Cliffs, NJ. Sloane, N.J.A. and Wyner, A.D., Eds. 1993. Claude Elwood Shannon: Collected Papers. IEEE Press, New York. Wicker, S.B. 1995. Error Control Systems for Digital Communication and Storage. Prentice-Hall, Englewood Cliffs, NJ. Further Information Information theory as presented here treats only the case where one user transmits one message over one channel at a time. During the past 15 years, a multiuser information theory has been emerging. This theory provides bounds on the information rates of more than one user transmitting over the same channel simultaneously. The beginnings of the theory are documented in the reference by Cover and Thomas. New results appear frequently in the IEEE Transactions on Information Theory. Source and channel coding remain fertile areas of research and development. In addition to the information theory transactions, important results and some applications can be found in the IEEE Transactions on Communications. Special areas are given emphasis in the IEEE Journal on Selected Areas in Communications, nearly all the issues of which are “special” issues. Another accessible text on digital communications is Digital Communications, by E. A. Lee and D. G. Messerschmitt (Kluwer, 1988). © 1999 by CRC Press LLC Communications and Information Systems 18-41 18.4 Applications Accessing the Internet Lloyd W. Taylor The Internet can provide access to a wide variety of information. Understanding what the Internet is, how it works, and what tools you can use will help in finding that fact or that person that is needed. What Is the Internet? The Internet is a voluntary, worldwide association of networks. There is no one “in charge” of the Internet, and anyone can belong. There have been many attempts to define the Internet, all of which fall short in one way or another. Probably the most accurate definition is that the Internet is any interconnected network that uses the TCP/IP protocol suite. A Brief History of the Internet The Internet got its start as a project of the Defense Advanced Projects Research Agency (DARPA) in 1962. The plan was to create a military network that could survive nuclear attack by automatically rerouting traffic around network nodes that had been destroyed. The first nodes were installed in 1969. By 1971 there were 15 nodes spread across the United States. In 1983, responsibility for the ARPANet (as it was then called) was split. The military portion of the Internet was transferred to the Department of Defense and merged with the Defense Data Network (DDN). The civilian portion continued to be operated as ARPANet. In 1986, the NSFNet was established, which superseded the ARPANet by 1990. By this time there were over 313,000 connected computers. Use of the Internet continued to grow. In 1993, the NSF decided to get out of the business of running the Internet, and the entire system was privatized. In the meantime, a number of private Internet access providers (IAPs) had come to exist, largely eliminating the need for any government subsidy or operation of the Internet. By the beginning of 1994, there were over 2.2 million connected computers. Today, there are dozens of IAPs and thousands of Internet service providers (ISPs) around the United States, and many more internationally. The deregulation of telephone companies in the United States has resulted in explosive growth, as the costs of long distance leased lines have dropped precipitously. Anyone with a few thousand dollars can set up an ISP by simply purchasing a Unix system and several modems and signing up for Internet access with an IAP. The Internet works by routing TCP/IP packets from one router to another until the packet reaches its destination (Figure 18.4.1). Each router determines the best path to the next router by the use of routing tables it maintains, based on conversations with adjacent routers. FIGURE 18.4.1 Router network. © 1999 by CRC Press LLC 18-42 Section 18 Typically, a company will contract with an Internet access provider (IAP) for a fixed bandwidth connection to that IAP’s nearest point of presence (POP). The IAP will usually interconnect its routers via high speed leased data lines. The IAP will install a leased line between its POP and a router located at the company’s site, and then connect the router to the company’s local network. Once the router is programmed and enabled, your company’s network is part of the Internet. Individuals can also gain access to the Internet for a monthly fee by contracting with an Internet service provider (ISP). The ISP will typically offer either a “shell account,” which provides you with dialup access to a Unix system via a terminal emulator, or a “SLIP/PPP account,” which allows you to connect your personal computer directly to the Internet via a dialup modem. Internet Tools There are several basic tools that can be used to access information across the Internet. Each has its particular uses and shortcomings. This section provides a brief introduction to each. Telnet. Telnet is a tool that allows a computer to emulate a simple terminal. It provides a mechanism by which the user can connect to another computer as if that computer were physically attached to the user’s workstation. A typical telnet application will emulate one or more terminal types. The most commonly emulated terminal type is the Digital Equipment Corporation’s VT100 terminal, with the IBM 3270 terminal a close second. The VT100 terminal is commonly used to communicate with Unix and VMS systems, while the IBM3270 is used to communicate with IBM mainframe systems based on the VM or MVS operating systems. Telnet provides a text-only interface and is incapable of displaying graphics. It is a good choice when a simple, efficient connection to a remote computer is desired, or when a text-only interface is sufficient. If a graphical interface is required for an interactive session, the X-Window System is a better choice. X-Window System. The X-Window System (“X”) was developed as part of MIT’s Project Athena. It is capable of full color graphics display and requires a mouse on the workstation for window and command selection. The X-Window server runs on the user’s workstation. It receives drawing commands from the remote computer and creates the requested text and graphics. Mouse clicks and keyboard input are sent to the remote computer, which responds by sending commands back to the workstation to fulfill the user’s request. In general, a window manager must be running either on the remote computer or on the user’s workstation, but not both. The window manager is responsible for controlling the placement and movement of windows, for creating and destroying windows, and for starting applications on the remote computer at the request of the user. X requires significant processing power on the local workstation and can demand significant network bandwidth for proper operation. In general, it should be run only on high-end workstations that are directly connected to an ethernet. Attempting to run X-Windows over a dialup line will result in very slow performance. FTP. The file transfer protocol (FTP) is used to move files from one computer to another. It is capable of moving all types of files (text or binary; images, sounds, software, etc.). FTP provides a collection of commands that allow the user to connect to remote systems, to manipulate files and directories on the remote system, and to move files back and forth between local and remote systems. These commands vary somewhat between implementations of FTP. Many systems on the Internet offer anonymous FTP. This de facto standard allows remote users to connect to a public portion of the computer and to retrieve (and sometimes place) files. The standard way to use anonymous FTP is to connect to the remote computer, log in as anonymous, and use your email address as the password. You will generally be granted read-only access to a set of directories and files which you can then download and use. © 1999 by CRC Press LLC Communications and Information Systems 18-43 Email. Electronic mail (“email”) is traditionally the most commonly used Internet application. Communicating with colleagues and friends around the world essentially instantaneously is a powerful capability. There are two parts to email. The user agent (UA) is the part that the user directly interacts with. This is the program on the workstation that provides the necessary commands and capabilities to create and send messages. The UA may include features such as spell checking, attachments, and address books. There are many different implementations of user agents for every kind of operating system. All of them use the same transport mechanism. The second part of email is the transport mechanism, or message transfer agent (MTA). It is the responsibility of the MTA to receive the message from the UA, to parse the mailing instructions (e.g., To:, Cc:), and to send the message on to the intended recipients. On the Internet, all MTAs use the simple mail transport protocol (SMTP) to carry messages between systems. At this writing, the ability to send attachments and binary files between systems remains problematic. There are several different standards for encapsulating attachments and graphics within SMTP messages. Unfortunately, these standards are not interoperable. The most commonly used encapsulation standards are MIME (multipurpose Internet mail extensions) and X.400. To successfully transfer attachments and graphics via email, you must ensure that the recipients of your message use a UA that uses a compatible encapsulation standard. Worldwide Web. The Worldwide Web (WWW or “Web”) first appeared on the Internet in 1992. It was developed as a way of sharing scientific information among researchers, but quickly was adopted by millions of people as the way to access information on the Internet. By 1995, half of all the traffic on the Internet was Web traffic. A typical Web browser runs on the local workstation and passes requests for information to the remote server. The browser may use a variety of transport mechanisms to obtain the information (e.g., ftp, http — hypertext transport protocol) as directed by the server. The unifying concept behind the power of the Web is the universal resource locator (URL). A URL is a unique address for a specific piece of information on the Web. It is made up of four parts: a protocol identifier, a host name, a path name, and an item name. Figure 18.4.2 shows a typical URL, where the protocol identifier is http, the host name is www.asmenet.org, the path name is /techaff/, and the item name is techprog.html. FIGURE 18.4.2 A typical URL. To retrieve this item, the Web browser will use the http protocol to connect to the www.asmenet.org server. Then it will change to the techaff directory and retrieve the techprog.html item. Because the item ends in html, the browser knows that this is a hypertext document and will parse and directly display the item. If it were a different type of item (Table 18.4.1), the browser would either directly display the item or call an external program to process it. In the case where there is no external program identified for the particular extension, the browser will typically offer the user the option of saving the item to disk or of canceling the download. TABLE 18.4.1 Typical Item Extensions Extension .gif .au .mpeg .html File Type Compressed image Audio file Digital movie Hypertext Typical External Program None required SoundPlayer MPEGPlay None required © 1999 by CRC Press LLC 18-44 Section 18 Web browsers are capable of transferring information using all major transport mechanisms (e.g., ftp, gopher, http). Because of this, they largely eliminate the need to use these other tools. For those just getting started on the Internet, a good Web browser should be the first tool learned. Finding Things on the Internet The Internet is largely an anarchy. There is no central authority that provides an organizing structure to the information available. Anyone with an Internet connection can become an information provider by simply setting up a Webserver or an FTP server. An inescapable consequence of this is that it can be very difficult to find specific information on the Internet. There is no rhyme or reason to where information is placed, and similar types of information may be located thousands of miles apart. A number of organizations have taken upon themselves to try to provide some organization. They typically take one of two approaches: either a content-based approach or an automated indexing approach. Content-based tools make use of discipline specialists to discover and index information manually. These discipline specialists comb the Internet for relevant information. When they find such information, they add a link to it from their content-based page. Thus, if you wish to find information of a specific type, you may start from a general definition of information type, and then further refine your query by clicking on the appropriate keywords. Table 18.4.2* lists a few content-based indexes current as of the date of publication of this book. A directory of content-based sites is maintained at http://home.netscape.com/home/internet-directory.html TABLE 18.4.2 Content-Based Information Indexes Name Yahoo Directory McKinley Internet Directory Lycos Virtual Tourist URL http://www.yahoo.com/ http://www.mckinley.com/ http://www.lycos.com/ http://www.vtourist.com/ Discipline All All All Webservers sorted by geographic region Content-based searching is most effective when you know the area in which you are interested and want to find a resource that is relevant to that area. It has a couple of drawbacks: first, if the content specialist has not yet indexed an information resource, there will be no way for you to find it; second, links across disciplines may not exist, limiting the breadth of the information you may be able to locate. Automated indexing tools are based on software agents that visit every Web site on a periodic basis and index the content of those sites by keyword. A variety of indexing strategies are used, resulting in a wide range of usefulness for a particular purpose. A selection of indexing tool sites are listed in Table 18.4.3. A directory of these sites is available at http://home.netscape.com/home/internet-search.html TABLE 18.4.3 Index-Based Information Sites Name InfoSeek Search WebCrawler Search W3 Search Engine Altavista Search Engine URL http://www.infoseek.com/ http://webcrawler.com/ http://cuiwww.unige.ch/meta-index.html http://altavista.digital.com URLs and search engines come and go with surprising frequency. These specific addresses may well cease to operate during the lifetime of this book. Similar services are likely to be available - check with a local Internet specialist or reference librarian for assistance. © 1999 by CRC Press LLC * Communications and Information Systems 18-45 Automated indexing tools are useful when you are looking for information by keyword or keyphrase. Their major disadvantage is the volume of information that will be returned on simple queries. For example, a simple keyword search on “Mechanical Engineering” returned over 12,000 pointers to information. It is therefore important to make your keywords as specific as possible. It is also important to read the guidelines for keywords on each of the servers you use. They each have their own specific search rules and syntax. Internet Security Issues The Internet is an inherently insecure environment. Anything that you send over the Internet can be read by someone with sufficient motivation. Because of this, it is critical to think twice about what you send. In general, the best rule of thumb is to ask yourself the question “Would I be upset if this message were printed on the front page of The New York Times?” If the answer is yes, you should seriously consider using encryption to protect your message. One popular freeware package that implements email encryption, and is available for all major operating systems, is PGP. Check with your system administrator to see if it is available on your system. The Internet also provides a path for people outside your organization to access computers and information within. If your network staff has not installed an Internet firewall, your system is directly accessible to anyone. You are the only one who can ensure that your system is secure. If you are using a multiuser workstation (like a Unix or VMS system), your system administrator is responsible for ensuring that known security holes have been closed. Your responsibility is to pick strong passwords (ones that do not appear in the dictionary and are hard to guess) so that someone else will not be able to compromise your account. You should change your password periodically (every 3 months or so) to help protect yourself. Defining Terms IAP: Internet access provider — a company that sells access to the Internet. ISP: Internet service provider — a company that sells Internet services. PPP: Point-to-point protocol — a standard for encapsulating multiple protocols (including TCP/IP) over dialup telephone lines. Routing: Moving data packets from one router to the next until they reach their destination. SLIP: Serial line internet protocol — a standard for encapsulating TCP/IP and transmitting it over dialup lines. TCP/IP: Transmission control protocol/Internet protocol. References Crumlish, C. 1994. A Guided Tour of the Internet, Sybex, Inc. Eddings, J. and Wattenmaker, P. 1994. How the Internet Works. Ziff-Davis. Gaffin, A. 1994. Everybody’s Guide to the Internet. Baker, Kehoe, B.P. 1994. Zen and the Art of the Internet. West. Otte, P. 1994. The Information Superhighway: Beyond the Internet. QUE. Randall, N. 1994. Teach Yourself the Internet: Around the World in 21 Days. SAMS. Salus, P.H. 1995. Casting the Net: From Arpanet to Internet and Beyond. Tittle, E. and Robbins, M. 1994. Internet Access Essentials. Academic Press, New York. © 1999 by CRC Press LLC 18-46 Section 18 Data Acquisition Dhammika Kurumbalapitiya and S. Ratnajeevan H. Hoole Data acquisition includes everything from gathering data, to transporting it, to storing it. The term data acquisition is described as the “phase of data handling that begins with sensing of variables and ends with a magnetic recording of raw data, may include a complete telemetering link” (McGraw-Hill, Dictionary of Scientific and Technical Terms, 2nd ed., 1978). Here the term variables refers to those physical quantities that are associated with a natural or artificial process. A data acquisition phase involves a real-time computing environment where the computer must be keyed to the time scale of the process. Figure 18.4.3 shows a block diagram of a data acquisition system which gives a simplified block diagram of a data acquisition system current in the early 1990s. FIGURE 18.4.3 Block diagram of a data acquisition system. The path the data travel through the system is called the data acquisition channel. Data are first captured and subsequently translated into usable signals using transducers. In this discussion, usable signals are assumed to be electrical voltages, either unipolar (that is, single ended, with a common ground so that we need just one lead wire to carry the signal) or bipolar (that is, common mode, with the signal carried by a wire pair, so that the reference of the rest of the system is not part of the output). These voltages can be either analog or digital, depending on the nature of the measurand (the quantity being captured). When there is more than one analog input, they are subsequently sent to an analog multiplexer (MUX). Both the analog and the digital signals are then conditioned using signal conditioners. There are two additional steps for those conditioned analog signals. First they must be sampled and next converted to digital data. This conversion is done by analog-to-digital converters (ADC). © 1999 by CRC Press LLC Communications and Information Systems 18-47 Once the analog-to-digital conversion is done, the rest of the steps have to deal with digital data only. The calendar/clock block shown in Figure 18.4.3 is used to add the time-of-date information, an important parameter of a real-time processing environment, into the half-processed data. The digital processor performs the overall system control tasks using a software program, which is usually called system software. These control tasks also include display, printer, data recorder, and communication interface management. A well-regulated power supply unit (PSU) and a stable clock are essential components in many data acquisition systems. There are systems where massive amounts of data points are produced within a very short period of time, and they are equipped with on-board memory so that a considerable amount of data points can be stored locally. Data are transmitted to the host computer once the local storage has reached its full capacity. Historically, data acquisition evolved in modular form, until monolithic silicon came along and reduced the size of the modules. The analysis and design of data acquisition systems are a discipline that has roots in the following subject areas: signal theory, transducers, analog signal processing, noise, sampling theory, quantizing and encoding theory, analog-to-digital conversion theory, analog and digital electronics, data communication, and systems engineering. Cost, accuracy, bit resolution, speed of operation, on-board memory, power consumption, stability of operation under various operating conditions, number of input channels and their ranges, on-board space, supply voltage requirements, compatibility with existing bus interfaces, and the types of data recording instruments involved are some of the prime factors that must be considered when designing or buying a data acquisition system. Data acquisition systems are involved in a wide range of applications, such as machine control, robot control, medical and analytical instrumentation, vibration analysis, spectral analysis, correlation analysis, transient analysis, digital audio and video, seismic analysis, test equipment, machine monitoring, and environmental monitoring. The Analog and Digital Signal Interface The data acquisition system must be designed to match the process being measured as well as the enduser requirements. The nature of the process is mainly characterized by its speed and number of measuring points, whereas the end-user requirement is mainly the flexibility in control. Certain processes require data acquisition with no interruption where computers are used in controlling. On the other hand, there are cases where the acquisition starts at a certain instance and continues for a definite period. In this case the acquisition cycle is repeated in a periodic manner, and it can be controlled manually or by software. Controllers access the process via the analog and digital interface submodules, which are sometimes called analog and digital front ends. Many applications require information capturing from more than one channel. The use of the analog MUX in Figure 18.4.3 is to cater to multiple analog inputs. A detailed diagram of this input circuitry is shown in Figure 18.4.4 and the functional description is as follows. When the MUX is addressed to select an input, say xi(t), the same address will be decoded by the decoding logic to generate another address, which is used in addressing the programmable register. The programmable register contains further information regarding how to handle xi(t). The outcome of the register is then used in subsequent tuning of the signal conditioner. Complex programmable control tasks might include automatic gain selection for each channel, and hence the contents of this register are known as the channel gain list. The MUX address generator could be programmed in many ways, and one simple way is to scan the input channels in a cyclic fashion where the address can be generated by means of a binary counter. Microprocessors are also used in addressing MUXs in applications where complex channel selection tasks are involved. Multiplexers are available in integrated circuit form, though relay MUXs are widely used because they minimize errors due to cross talk and bias currents. Relay MUX modules are usually designed as plugged-in units and can be connected according to the requirements. There are applications where the data acquisition cycle is triggered by the process itself. In this case an analog or digital trigger signal is sent to the unit by the process, and a separate external trigger interface circuitry is supplied. The internal controller assumes its duties once it has been triggered. It takes a finite time to settle the signal xi(t) through the MUX up to the signal conditioner once it is addressed. Therefore, it is possible to process x i–1(t) during the selection time of xi(t) for greater speeds. © 1999 by CRC Press LLC 18-48 Section 18 FIGURE 18.4.4 Input circuitry. This function is known as pipelining and will be illustrated in the subsection 4, on “Analog Signal Conditioning.” In some data acquisition applications the data acquisition module is a plugged-in card in a computer, which is installed far away from the process. In such cases, transducers — the process sensing elements — are connected to the data acquisition module using transmission lines or a radio link. In the latter case a complete demodulating unit is required at the input. When transmission lines are used in the interconnection, care must be taken to minimize electromagnetic interference since transmission lines pick up noise easily. In the case of a single-ended transducer output configuration, a single wire is adequate for the signal transmission, but a common ground must be established between the two ends as given in Figure 18.4.5(a). For the transducers that have common mode outputs, a shielded twisted pair of wires will carry the signal. In this case, the shield, the transducer’s encasing chassis, and the data acquisition module’s reference may be connected to the same ground as shown in Figure 18.4.5(c). In high-speed applications the transmission line impedance should be matched with the output impedance of the transducer in order to prevent reflected traveling waves. If the transducer output is not strong enough to transmit for a long distance, it is best to amplify it before transmission. FIGURE 18.4.5 Sensor/acquisition module interconnections. Transducers that produce digital outputs may be first connected to Schmitt trigger circuits for pulse shaping purposes, and this can be considered a form of digital signal conditioning. This becomes an essential requirement when such inputs are connected through long transmission lines where the line capacitance significantly affects the rising and falling edges of the incoming wave. Opto-isolators are sometimes used in coupling when the voltage levels of the two sides of the transducer and the input circuit of the data acquisition unit do not match each other. Special kinds of connectors are designed and widely used in interconnecting transmission lines and data acquisition equipment in order to screen © 1999 by CRC Press LLC Communications and Information Systems 18-49 the signals from noise. Analog and digital signal grounds should be kept separate where possible to prevent digital signals from flowing in the analog ground circuit and including spurious analog signal noise. Analog Signal Conditioning The objective of an analog signal conditioner is to increase the quality of the transducer output to a desired level before analog-to-digital conversion. A signal conditioner mainly consists of a preamplifier, which is either an instrumentation amplifier or an operational amplifier, and/or a filter. Coupling more and more circuits to the data acquisition channel has to be done, taking great care that these signal conditioning circuits do not add more noise or unstable behavior to the data acquisition channel. Generalpurpose signal conditioner modules are commercially available for applications. Some details were given in the previous section about programmable signal conditioners and the discussion is continued here. Figure 18.4.6 shows an instrumentation amplifier with programmable gain where the programs are stored in the channel-gain list. The reason for having such sophistication is to match transducer outputs with the maximum allowable input range of the ADC. This is very important in improving accuracy in cases where transducer output voltage ranges are much smaller than the full-scale input range of an ADC, as is usually the case. Indeed, this is equally true for signals that are larger than the full-scale range, and in such cases the amplifier functions as an attenuator. Furthermore, the instrumentation amplifier converts a bipolar voltage signal into a unipolar voltage with respect to the system ground. This action will reduce a major control task as far as the ADC is concerned; that is, the ADC is always sent unipolar voltages, and hence it is possible to maintain unchanged the mode control input which toggles the ADC between the unipolar and bipolar modes of an ADC. FIGURE 18.4.6 Programmable gain instrumentation amplifier. Values of the signal-to-noise ratio (SNR) RMS signal  SNR =   RMS noise    2 (18.4.1) at the input and the output of the instrumentation amplifier are related to its common-mode rejection ratio (CMRR) given by SNR output SNR input CMRR = (18.4.2) © 1999 by CRC Press LLC 18-50 Section 18 Hence, higher values of SNRoutput indicate low noise power. Therefore, instrumentation amplifiers are designed to have very high CMRR figures. The existence of noise will result in an error in the ADC output. The allowable error is normally expressed as a fraction of the least significant bit (LSB) of the code such as ±(1/X)LSB. The amount of error voltage (Verror) corresponding to this figure can be found considering the bit resolution ( N) and the ADC’s maximum analog input voltage (Vmax) as given in 1  V Verror = ± Nmax ×  volt  2 −1 X (18.4.3) Other specifications of amplifiers include the temperature dependence of the input offset voltage (Voffset, µV/°C) and the current (Voffset, pA/°C) associated with the operational amplifiers in use. High slew rate (V/µs) amplifiers are recommended in high-speed applications. Generally, the higher the bandwidth, the better the performance. Cascading a filter with the preamplifier will result in better performance by eliminating noise. Active filters are commonly used because of their compact design, but passive filters are still in use. The cutoff frequency, f c, is one of the important performance indices of a filter that has to be designed to match the channel’s requirements. The value fc is a function of the preamplifier bandwidth, its output SNR, and the output SNR of the filter. Sample-and-Hold and A/D Techniques in Data Acquisition Sample-and-hold systems are primarily used to maintain a constant magnitude representing the input, across the input of the ADC throughout a precisely known period of time. Such systems are called sample-and-hold amplifiers (SHA), and their characteristics are crucial to the overall system accuracy and reliability of digital data. The SHA is not an essential item in applications where the analog input does not vary more than ±(1/2)LSB of voltage. As the name indicates, an SHA operates in two different modes, which are digitally controlled. In the sampling mode it acts as an input voltage follower, where, once it is triggered into its hold mode, it should ideally retain the signal voltage level at the time of the trigger. When it is brought back into the sampling mode, it instantly assumes the voltage level at the input. Figure 18.4.7 shows the simplified circuit diagram of a monolithic sampling-and-hold circuit and the associated switching waveforms. The differential amplifiers function as input and output buffers, and the capacitor acts as the storage mechanism. When the mode control switch is at its on position, the two buffers are connected in series and the capacitor follows the input with minimum time delay, if it is small. Now, if the mode control is switched off, the feedback loop is interrupted, and the capacitor ideally retains its terminal voltage until the next sampling signal occurs. Leakage and bias currents usually cause the capacitor to discharge and/or charge in the hold mode and the fluctuation of the hold voltage is called droop, which could be minimized by having a large capacitor. Therefore, the capacitance has to be selected such that the circuit performs well in both modes. Several time intervals are defined relative to the switching waveform of SHAs. The acquisition time (ta) is the time taken by the device to reach its final value after the sample command has been given. The setting time (ts) is the time taken to settle the output. The aperture uncertainty or aperture jitter (tus) is the range of variation of the aperture time. It is important to note here that the sampling techniques have a well-formulated theoretical background. ADCs perform a key function in the data acquisition process. The application of various ADC technologies in a data acquisition system depends mainly on the cost, bit resolution, and speed. Successive approximation types are more common at high resolution at moderate speeds (<1 MHz). This kind of ADC offers the best trade-offs among bit resolution, accuracy, speed, and cost. Flash converters, on the other hand, are best suited for high-speed applications. Integrating-type converters are suitable for highresolution and -accuracy applications. Many techniques have been developed in coupling sample-hold circuits and ADCs in data acquisition systems because no single ADC or sampling technology is able to satisfy the ever-increasing requirements of data acquisition applications. Figure 18.4.8 illustrates the various sampling and ADC configurations © 1999 by CRC Press LLC Communications and Information Systems 18-51 FIGURE 18.4.7 Monolithic sample-and-hold circuit. FIGURE 18.4.8 ADC and sampling configurations. used in practice. It can be seen that the sampling frequencies are increased because of pipelining, parallelism, or concurrent architecture. The increase in the sampling frequency improves the bandwidth, improving, in turn, the SNR in the channel. The Communication Interface of a Data Acquisition System The communication interface is the module through which the acquired data are sent; as well, other control tasks are established between the data acquisition module and the host computer (Figure 18.4.3). There are basically two different ways of establishing a data link between the two. One way is to use © 1999 by CRC Press LLC 18-52 Section 18 interrupts and the other is through direct memory access (DMA). In the case of an interrupt-driven mode, an interrupt-request signal is sent to the computer. Upon receiving it, the computer will first finish the execution of the current instruction, suspend the next, and then send an interrupt-acknowledge signal asking the module to send data. The operation is asynchronous since the sender sends data when it wants to do so. Getting the computer ready to receive data is known as handshaking. In the case of a DMA transfer, the DMA controller is given the starting address of the memory location where the data have to be written. The DMA controller asks the computer to freeze its operations until it has finished writing data directly into the memory. The operation does not need any waiting time and therefore it is fast. Data acquisition systems are usually designed to couple with existing computer systems, and many computer systems provide standard bus architecture, allowing users to connect various peripherals that are compatible with its bus. Data acquisition systems are computer peripherals that follow the above description. Since ADCs produce parallel data, many data acquisition systems provide outputs compatible with parallel instrument buses such as the IEEE-488 (HP-IB or GPIB) or the VMEbus. Personal computerbased data acquisition boards must have communication interfaces compatible with the computer bus in order to share resources. The RS-232 standard communication interfaces are widely used in serial data transfer. Communication interfaces for data acquisition systems are normally designed to satisfy the electrical, mechanical, and protocol standards of the interface bus. Electrical standards include power supply requirements, methods of supply, the data transfer rate (baud rate), the width of the address, and the line terminating impedance. Mechanical requirements are the type, size, and the pin assignments of the connectors. The data transfer protocol determines the procedure of data transfer between the two systems. A definition of the timing and input–output philosophy—whether the transfer is in synchronous, asynchronous, or quasi-synchronous mode and how errors are detected and handled — is an important factor to be considered. Data Recording It is important to provide storage media to cater to large streams of data being produced. Data acquisition systems use graph paper, paper tapes, magnetic tapes, magnetic floppy disks, hard disks, or any combination of these as their data recorders. Paper and magnetic tape storage schemes are known as sequential access storage, whereas disk storage is called direct access storage. Tapes are cost-effective media compared to disk drives and are still in wide use. In many laboratory situations it will be much more cost effective to network a number of systems to a single, high-capacity hard drive, which acts as a file server. This adoption of digital recording provides the ultimate in signal-to-noise ratio, accuracy of signal waveform, and freedom from tape transfer flutter. Data storage capacity, access time, transfer rate, and error rate are some of the performance indices that are associated with these devices. Software Aspects So far the discussion has been mainly on the hardware side of the data acquisition system. The other most important part is the software system associated with a data acquisition system, which can generally be divided into two — the system software and the user-interface program. Both must be designed properly in order to achieve the maximum use of the system. The system software is mainly written in assembly language with many lines of code, whereas the user interface is built using a high-level software development tool. One main part of system software is written to handle the input–output (I/O) operations. The use of assembly language results in the fast execution of I/O commands. The I/O software has to deal with how the basic input–output programming tasks such as interrupt and DMA handling are done. The other aspects of system software are to perform the internal control tasks such as providing trigger pulses for the ADC and SHA, addressing the input multiplexer, the accessing and editing of the channelgain list, transferring data into the on-board memory, and the addition of the clock/calendar information into data. Multitasking software programs are best suited for many data acquisition systems because it may be necessary to read data from the data acquisition module and display and print it at the same time. Menu-driven user interfaces are common and have a variety of functions built into them. © 1999 by CRC Press LLC Communications and Information Systems 18-53 Defining Terms Analog-to-digital converter (ADC): A device that converts analog input voltage signals into digital form. Common-mode rejection ratio (CMRR): A measure of quality of an amplifier with differential inputs and the ratio between the common-mode gain and the differential gain. Direct memory access (DMA): The process of sending data from an external device into the computer memory with no involvement of the computer’s central processing unit. Least significant bit (LSB): The 20th bit in a digital word. Multiplexer (MUX): A combinational logic device with many input channels and usually just one output. The function performed by the device is connecting one and only one input channel at a time to the output. The required input channel is selected by sending the channel address to the MUX. Power supply unit (PSU): The unit that generates the necessary voltage levels required by a system. Sample-and-hold amplifier (SHA): A unity gain amplifier with a mode control switch where the input of the amplifier is connected to a time-varying voltage signal. A trigger pulse at the mode control switch causes it to read the input at the instance of the trigger and maintain that value until the next trigger pulse. Signal-to-noise ratio (SNR): The ratio between the signal power and the noise power at a point in the signal traveling path. References Feucht, D.L. 1990. Handbook of Analog Circuit Design. Academic Press, San Diego. Fink, D.G. and Christiansen, D., Eds. 1989. Electronic Engineers’ Handbook, 3rd ed. McGraw-Hill, New York. Frizell, K.W. et. al. 1993. Guidelines for PC-Based Data Acquisition Systems for Hydraulic Engineering. American Society for Civil Engineering. Holloway, P. 1990.Technology focus interview. Electronic Eng. December. Rigby, W.H. and Dalby, T. 1994. Computer Interfacing: A Practical Approach to Data Acquisition and Control. Prentice-Hall, Englewood Cliffs, NJ. Tatkow, M. and Turner, J. 1990. New techniques for high-speed data acquisition. Electronic Eng. September. © 1999 by CRC Press LLC