Wireless Ad Hoc and Sensor Networks

Congestion Control in ATM Networks and the Internet 913.3.1.2 Implementation of the Proposed SchemeTwo types of ATM cells, namely, flow-data cells and RM cells, on anABR connection are considered. A source/virtual receives a regularsequence of RM cells that provide feedback to enable it to adjust itsrate of cell transmission. The bulk of the RM cells are initiated by thesource, which transmit one forward RM (FRM) cell for every (Nrm –1)data cells, where Nrm is a preset parameter. As each FRM is receivedat the destination, it is turned around and transmitted back to thesource as a backward RM (BRM) cell. Each FRM cell contains a congestionindication (CI), no increase (NI), and explicit cell rate (ER)fields, besides current cell rate (CCR) and minimum cell rate (minCR)fields. Any intermediate switch can set CI or NI fields, the source willset the CCR and MinCR field, whereas the intermediate or destinationswitches set the ER field. Our proposed scheme at each intermediate/destination switch will provide the feedback, uk ( ), to be used in the ERfield for its source/virtual source and, accordingly, the sources/virtualsources will alter their rate based on the ER field. Therefore, the implementationis straightforward. The next step is to determine the weightupdates so that the tracking performance of the closed-loop trackingerror dynamics is guaranteed.3.3.2 Weight UpdatesIt is required to demonstrate that the performance criterion in terms ofcell losses, ck ( ), transmission delay, and buffer utilization monitored usingthe error in buffer occupancy, ek ( ), is suitably small, source rates adjustedfairly, and that the NN weights, Wk ˆ ( ), Vk ˆ ( ), remain bounded for the trafficrate, and uk ( ), is bounded and finite. In the following theorem, discretetimeweight-tuning algorithms based on the error in buffer occupancy aregiven, which guarantee that both the error in buffer occupancy and weightestimates are bounded.THEOREM 3.3.1 (TRAFFIC CONTROLLER DESIGN)Let the desired buffer length, x d , be finite and the NN traffic approximation errorbound, ε N , which is equal to the target CLR, and the disturbance bound, d M , beknown constants. Take the control input for Equation 3.5 as Equation 3.19 withweight tuning provided byˆ ( ) ˆ ( ) ˆTVk+ = Vk− αϕ( k)[ˆ y( k) + Bkek ( )] −Γ I− αϕˆ T( )ˆ ϕ ( ) ˆ Tk k V ( k),1 1 1 1 1v1 1(3.22)ˆ ( ) ˆ ( ) ˆTWk+ 1 = Wk− αϕ 2 2( ke ) ( k+ 1) −ΓI−αϕ 2ˆ 2( k)ˆϕ2 T ( k) Wˆ ( k ),(3.23)