Dynamic Power-Conscious Routing for MANETs An Initial Approach by gdf57j


									   Dynamic Power-Conscious Routing for MANETs: An Initial Approach
                              Madhavi W. Subbarao, Member, IEEE
                         National Institute of Standards and Technology
                                  100 Bureau Drive Stop 8920
                              Gaithersburg, MD, USA 20899-8920
                Phone: 301-975-4974 FAX: 301-590-0932 Email: subbarao@nist.gov
                October 12, 1999                           use just that amount needed to maintain an ac-
                                                           ceptable signal-to-noise ratio SNR at the re-
   Abstract | We develop an initial dynamic power-         ceiver. Reducing the transmitter power allows
conscious routing scheme MPR that incorporates           spatial reuse of the channel and thus, increases
physical layer and link layer statistics to conserve       network throughput 1 . Altering the transmis-
power, while compensating for the channel condi-           sion power also reduces the amount of interfer-
tions and interference environment at the intended         ence caused to other networks operating on ad-
receiver. The aim of MPR is to route a packet on           jacent radio frequency channels. In networks
a path that will require the least amount of total         where nodes operate on battery power, conserv-
power expended and for each node to transmit with          ing power is crucial since battery life determines
just enough power to ensure reliable communication.        whether a network is operational or not. Mili-
We evaluate the performance of MPR and present             tary networks desire to maintain a low probabil-
our preliminary results.                                   ity of intercept and or a low probability of detec-
                                                           tion 4 . Hence, nodes prefer to radiate as little
               I. Introduction                             power as necessary and transmit as infrequently
   A mobile ad hoc network MANET is an au-               as possible, thus decreasing the probability of
tonomous collection of mobile nodes that com-              detection or interception.
municate over relatively bandwidth-constrained                The bene ts of power conservation control
wireless links. Signi cant examples of MANETs              for MANETs prompt the important question:
include establishing survivable, dynamic com-              What is the most power e cient way to route
munication for emergency rescue operations,                a packet from a source to a destination such
disaster relief e orts, and military networks.             that the packet is received with an acceptable
MANETs need e cient distributed algorithms                 packet success rate 5 ? Since channel conditions
to determine network organization connectiv-              and multiuser interference levels are constantly
ity, link scheduling, and routing. Message rout-          changing with time, the transmitter power nec-
ing in a decentralized environment where net-              essary on a particular link must be determined
work topology uctuates is not a well-de ned                dynamically. In 7 , Wieselthier, Nguyen, and
problem. Factors such as variable wireless link            Ephremides address this problem in the context
quality, propagation path loss, fading, multiuser          of wireless multicasting, and in 3 , Pursley, Rus-
interference, and topological changes, become              sell, and Wysocarski consider this problem in a
relevant issues.                                           frequency-hopping ad-hoc network.
   In addition to the characteristics mentioned,              In this paper, we conduct an initial inves-
an important issue in network routing for                  tigation on the e ects of energy-e cient wire-
MANETs is to conserve power while still achiev-            less routing in MANETs. We develop an initial
ing a high packet success rate. This can be ac-            dynamic power-conscious routing scheme min-
complished by altering the transmitter power to            imum power routing -MPR that incorporates
                                                           physical layer and link layer statistics to con-
                                                           serve power, while compensating for the propa-
gation path loss, shadowing and fading e ects,           power expended and for each node to transmit
and interference environment at the intended             with just enough power to ensure that the trans-
receiver. The main idea of MPR is to select              mission is received with an acceptable bit error
the path between a given source and destina-             rate . Threshold  is a design parameter and
tion that will require the least amount of total         may be selected according to the network perfor-
power expended, while still maintaining an ac-           mance desired. Let E be the bit-energy-to-noise-
ceptable SNR at each receiver. A cost" function          density ratio, Eb =N0ef f , necessary at a node to
is assigned to every link re ecting the transmit-        achieve .
ter power required to reliably communicate on              Without loss of generality, consider a trans-
that link. As an initial approach, the distributed       mission from node i to node j , where i 6= j ,
Bellman-Ford algorithm can be used to perform            and i; j 2 f1; : : :; N g, where N is the number of
  shortest" path routing with the cost functions         nodes in the network. The received Eb =N0ef f is
as the link distances. The resulting shortest            given by
path" is the MPR path from a given source to                     "            
a destination. We compare the performance of                          Eb                  PRij =D
                                                                                     = N + P =W ;       1
MPR to that of shortest distance routing with                        N0ef f       ij    0      Iij
power control SD-PC and minimum hop rout-
ing with power control MH-PC, and present         where D is the data rate in bits per second, W
our preliminary results.                            is the system bandwidth in Hertz, N0 =2 is the
                                                    power spectral density of the thermal noise, PIij
       II. Power-Conscious Routing                  is the power of the interference at node j due
A. System Model                                     to all nodes excluding node i, and PRij is the
                                                    received power at node j due to node i. From
   Consider a transmitter communicating with a the description in Section II-A, it follows that
receiver at a distance of r0 in a MANET. As the the received power is given by
transmitted signal propagates to the receiver, it                                   ,
is subject to the e ects of shadowing and multi-                 PRij = KFij PTij rij ;          2
path fading, and its power decays with distance,
i.e., PR KFPT r0  , where K is a constant, where PTij is the transmitter power used at
F is a non-negative random attenuation for the node i to communicate with node j , Fij is a non-
e ects of shadowing and fading, PT is the trans- negative random attenuation for the e ects of
mitter power, and  is the path loss exponent. shadowing and fading on link ij , and rij is the
At the receiver, the desired signal is corrupted by distance between node i and node j . Substitut-
interference from other active nodes in the net- ing 2 into 1, we obtain
work. We assume that nodes know the identity                   "       
of all other nodes in the network and the dis-                     Eb                  ,
tances to their immediate neighbors, i.e., nodes                 N0ef f ij = Sij PTij rij ;       3
that are within transmission range. Interfering
nodes use the same modulation scheme as the where
transmitter and nodes can vary their transmit
power up to a maximum power Pmax . We as-                      Sij = DN KFij =W  ;              4
                                                                           0 + PIij
sume that the multiuser interference is a Gaus-
sian random process. At the receiver, the de- may be interpreted as a dynamic link scale factor
coder maintains an estimate of the average SNR. re ecting the current channel characteristics and
                                                    interference on link ij . These scale factors re-
B. Minimum Power Routing Protocol                     ect a link's most recent reception environment.
   The aim of MPR is to route a packet on a Note that Sij 6= Sji since channel conditions are
path that will require the least amount of total not symmetric.
  It is desirable for Eb =N0eff ij to equal the en-                                                 ^
                                                          mission interval, and hence the value for Sij is
ergy ratio E , since this is the minimum Eb =N0ef f       valid for many packet transmissions.
necessary to achieve the bit error rate . Hence,           For every pair of nodes i and j , a cost Cij
with knowledge of scale factor Sij , node i can           given by
easily determine the power PTij necessary to                      
achieve this goal using Eq. 3, i.e.,                    Cij =       PTij 1 +    if PTij 1 +   Pmax ; 7
                                                                      1              otherwise;
               PTij =      E :                 5
                        Sij rij                          is assigned, where  is a dampening constant
                                                          to inhibit oscillations. The inequality in 7 is
   Let Eb=N0eff ij be an estimate of the re-              necessary since the transmitter power is limited
ceived bit energy ratio at the output of the de-          by Pmax . The cost Cij is the power necessary
coder at node j . Many methods may be used                to communicate from node i to node j to com-
to determine Eb =N0eff ij , e.g., using side infor-       pensate for channel conditions and interference.
mation by embedding known test symbols in                 Since nodes only know estimates of the link scale
packet transmissions 2 . Although PTij was se-            factors, the power required on a link must be
lected to achieve energy ratio E at the receiver,         overplayed. Thus,  provides an extra margin
since network conditions are changing, the ac-            for the transmission power and is a design pa-
tual received Eb =N0eff ij may di er from E . If          rameter that must be selected. As an initial ap-
node j has knowledge of the transmitter power             proach, the distributed Bellman-Ford algorithm
PTij which can be accomplished by including              can be used to perform shortest" path routing
PTij in the packet header, it can update its es-         with the Cij s as the link distances. The resulting
timated scale factor using a smoothing function             shortest path" is the MPR path from a given
as follows,                                               source to a destination. If there is more than
                                                          one path with the same minimum total cost, the
   ^              Eb=N0eff ij  ^                          MPR path is chosen as the one with the small-
   Sij = 1 ,          , +  Sij ;           6        est maximum cost on any one link. MPR avoids
                   PTij rij
                                                          congested areas and is also minimax optimal,
which mitigates the uctuations due to mul-                i.e., given some uncertainty in the link scale fac-
tiuser interference and is a smoothing fac-              tors, it minimizes the worse case total path cost.
tor. An initial value for Sij may be computed
as described in Section II-C. The estimated link          C. Network Implementation
scale factor Sij accounts for variable channel               Initially, nodes transmit using power Pmax ,
conditions and for all types of Gaussian inter-           and the cost of every link is set to a constant d,
ference, e.g., multiuser interference and partial-        where d = Pmax 1+ . This will result in nodes
band jamming. If the received bit error rate              initially routing packets according to the mini-
ij on link ij is less than threshold , the ef-          mum number of hops to the destination. The
fect of 6 is that node j decreases its link Sij          rst time node j for j 2 f1; : : :; N g, receives a
value, indicating an increase in its interference         transmission from another node, say node i, it
noisy channel level, and thus, an increase in                                               ^
                                                          will compute its link scale factor Sij , i.e,
the power necessary to communicate on link ij
as computed by 5. The opposite behavior oc-                            ^     Eb=N0ef f ij
                                                                         Sij =        , :                8
curs when ij is greater than .                                               Pmax rij 
   Each time node j receives a packet from a
node i, it computes and stores a value for                The link costs will be computed as described in
Sij that accurately re ects its current SNR on            Section II-B and propagated throughout the net-
link ij . We assume that the rate of change of            work. If the cost of a particular link has not
the network is much slower than a packet trans-           yet been computed within a speci ed amount of
time because no data packet was transmitted on             put, end-to-end delay, e ciency, and average
that link, a boost" packet is transmitted on the           power expended are used to analyze the per-
link and the link cost is computed. Once all of            formance of the routing protocols. End-to-end
the link costs have been computed, the routing             throughput is de ned as the number of pack-
protocol is now MPR.                                       ets that successfully reach their nal destination
   The MPR path costs must be periodically cir-            per unit time. End-to-end delay is based only
culated around the network. This information               on successful packets and is de ned as the av-
can be passed around via data packets, acknowl-            erage time required for a packet to arrive at its
edgments, and special control packets known as             destination. E ciency is the number of received
packet radio organization packets PROPs 6 .              data packets divided by the total number of data
For this initial implemenation, we assume an un-           packets and control packets transmitted. Aver-
derlying information dissemination scheme.                 age power expended is the average power con-
   A dynamic routing table is maintained by each           sumed in the network relaying successful packets
node. For each destination, a node stores the              including necessary control packets from their
outgoing link for the most power-e cient route             source to their nal destination per unit time.
and the corresponding path cost, distance to                  First, we consider a 16 node static network
the destination, and the necessary transmitter             with packet generation rate = 10 pack-
power. Since network conditions are changing,              ets second node and a total of 10; 000 packets
routing tables are continually updated based on            being exchanged. The routing table update in-
an update interval, and the transmission power             terval is 10s, and the shadowing parameters are
is altered on a per packet basis according to Eq.             = 0:8 and TS = 5s. From Table II, we see
5. Before an update, if a link cost is deemed            that MPR achieves approximately double the
out-dated, i.e., the cost has not been recomputed          throughput for similar power consumption lev-
within a speci ed interval before an update, a             els, or alternatively, requires approximately 2:5
  boost" packet is transmitted on that link in or-         times less power for similar throughput levels.
der to compute a current link cost.                        The overall end-to-end delay is comparable for
                                                           all schemes. While MPR does not optimize on
 III. Performance of Power Conscious                       the number of hops, it routes around undesir-
               Routing                                     able links and hence, requires overall lower power
                                                           consumption.        Next, with the same network
   We compare the performance of MPR to that
of SD-PC and MH-PC, and present our prelimi-
nary results. The transmission power for SD-PC                    Parameter                Value
and MH-PC is altered to overcome the distance                    Network area         900 m x 600 m
between the transmitter and intended receiver.                     Data rate              1 Mbps
We use the modeling and simulation tool OP-                   Max TX power range      500 mW 250 m
                                                                 Min frequency           2.4 GHz
NET to build a network prototype and execute                      Bandwidth               83 MHz
the simulations. We assume a MANET using                          Modulation     Direct-Sequence BPSK
the ALOHA random access protocol. We con-                       Processing Gain            20 dB
sider a slow fading log-normal shadowing en-                   Packet length            100 bits
vironment, and vary the random attenuation ef-                    Shadowing      10 log F  N 0; 64dB 2 
                                                                   , , ,        3 x 10,4 ; 2:6; 0:8; 0:2
fects on a link every TS seconds according to a
  correlation factor. We assume that a node has                Table I: Network simulation parameters.
knowledge of the transmitter power used to com-            con guration, we vary the packet generation rate
municate with it and hence, uses 6 to update               and plot the e ciency and average power ex-
the estimate of its link scale factor. A list of the       pended in Figures 1 and 2 respectively. We
simulation parameters is given in Table I.                 see that as increases, the e ciency increases
   Performance measures of end-to-end through-             until the point where further packet generation
    Measure     MPR SD-PC SD-PC MH-PC MH-PC
      Hops      30682 24945 15321 25075 17485
    Overhead    0.0077  0     0     0      0
  Pk delay*s  28.5  24.5   26   24.8  27.6
 Pk pwr*mW     305   660   279   702   266
 Hop pwr*mW 91.3     244   94.1  255   91.3
    E ciency     0.95  0.92  0.51  0.92   0.6
 Thruput pk s 9.58   9.2   5.15  9.13   5.7
Table II: Simulation results for a 16 node static
network. * mean value of three trials

causes excess levels of network tra c, and thus,
a decrease in e ciency. MPR achieves approxi-
mately double the e ciency as SD-PC and MH-
PC for low values of and approximately a strik-
ing 4:5 times higher e ciency for larger values
of , since MPR adapts to changing interference           Figure 1: E ciency vs. Packet generation rate
levels. For low values of , MPR utilizes from             .
30 , 50 less power relaying successful packets
than SD-PC and MH-PC. For higher values of                              IV. Conclusion
  , although MPR utilizes approximately 50mW
more power than SD-PC and MH-PC, since both                 We conducted an intitial investigation of
MH-PC and SD-PC achieve low e ciency, most               energy-e cient wireless routing in MANETs.
of the total power expended in those schemes is          We presented our preliminary results and con-
on unsuccessful transmissions.                           clude that MPR shows promise as a power
    Finally, we introduce mobility into the net-         conscious routing scheme for MANETs. MPR
work with nodes moving at a speed of 4m=s and            adapts to the changing channel conditions and
investigate the e ect of di erent routing table          interference environment of a node. The power-
update intervals on MPR. The packet generation           conscious concepts developed herein can be
rate is = 10 packets second node. In Figure 3,           adopted in other MANET routing algorithms.
we plot the network e ciency verses update in-                       Acknowledgements
terval frequency s. We consider the e ciency
of only data transmissions, and the global ef-              I would like to thank Jean-Sebastien Pegon
  ciency of both data and control packets, i.e.,         for his hard-work and diligent e orts in creating
data packets received divided by total commu-            the simulation environment in OPNET, execut-
nication packets - both data and control. We             ing the simulations, and producing the plots.
see that as the update interval decreases, the
data e ciency increases since the routing infor-                          References
mation utilized is more current. However, the            1 L. Kleinrock and J. Silvester, Spatial reuse
global e ciency increases until it reaches a point         in mutlihop packet radio networks," Proc.
where further updates cause too much overhead              IEEE, vol. 75, pp. 156 167, Jan. 1987.
communication, and hence, a decrease in net-
work e ciency. Clearly, there is a trade-o be-           2 M. B. Pursley, The derivation and use of
tween utilizing current routing information and            side                            information
the communication overhead generated. It is our            in frequency-hop spread spectrum commu-
conjecture, that the optimum update interval is            nications," IEICE Trans. Commun., Special
the same as the slow fading duration TS .                  Issue on Spread Spectrum Techniques and
                                                       Applications, vol. E76-B, pp. 814 24, Aug.
                                                   3   M. B. Pursley, H. B. Russell, and J. S.
                                                       Wysocarski, Energy-e cient routing in
                                                       frequency-hop packet networks with adap-
                                                       tive transmission," in Proc. IEEE MILCOM,
                                                   4   F. J. Ricci and D. Schutzer, U.S. Military
                                                       Communications. Rockville, MD.: Com-
                                                       puter Science Press, 1986.
                                                   5   M. W. Subbarao, On Optimizing Perfor-
                                                       mance in Mobile Packet Radio Networks.
                                                       PhD thesis, The Johns Hopkins Univer-
                                                       sity, Dept. Electrical Engineering, Baltimore,
                                                       MD, Mar. 1998.
                                                   6   L. Westcott and J. Jubin, A distributed
                                                       routing design for a broadcast environment,"
Figure 2: Average power expended vs. Packet            in Proc. IEEE MILCOM, vol. 3, Boston,
generation rate .                                      MA, pp. 10.4.1 5, Oct. 1982.
                                                   7   J. Wieselthier,          G. D. Nguyen,
                                                       and A. Ephremides, Multicasting in energy-
                                                       limited ad-hoc wireless networks," in Proc.
                                                       IEEE MILCOM, Bedford, MA, Oct. 1998.

Figure 3: MPR: E ciency vs. Update frequency


To top