Wireless Ad Hoc and Sensor Networks

Predictive Congestion Control for Wireless Sensor Networks 457Queue level (packets)throughput (packets/iteration)2015105q iestimated f outf out00 5 10 15 20 25Time (iteration)FIGURE 9.12Queue utilization and estimation of the outgoing flow.10 packets, the controller parameters k vb = 0.1 and l = 0.001. The actualoutgoing flow is set to 0 packets per iteration at the beginning of simulation,then increased to 2, and later to 6 packets per iteration. Next, theoutgoing flow is decreased to 4 packets per iteration. These outgoing flowvariations can also be viewed as MAC data rate changes, thus providingan indication how the proposed protocol performs in networks that supportmultiple modulation rates.Figure 9.12 illustrates the actual and estimated value of the outgoingflow, together with the queue utilization. The outgoing flow estimationis able to track the actual value, f out . The queue utilization, q i , variesfrom the ideal value when the sudden change in the outgoing flowoccurs, because the outgoing traffic estimation could not predict suchabrupt changes. However, after the sudden change, in just a few iterationsof the algorithm, the queue level has converged to the idealvalue as the estimation scheme quickly detected and accommodatedthe changed bandwidth. Additionally, Figure 9.13 presents the errorin estimation of the outgoing flow, e f , and the error of queue utilization,e bi . The errors are bounded and quickly converge to zero because thescheme adapts to the changed outgoing flow rate. Moreover, in agradual change in the outgoing traffic rate, for example during timeinterval of 10 to 14 iterations, the outgoing flow estimation quicklyadapts to the change, and the estimation error decreases even thoughthe outgoing flow rate continues to increase.