Adaptive Neural Congestion Controller for ATM Network with Heavy Traffic

Abstract

This paper presents an adaptive control scheme using a newly developed Minimal Resource Allocation Network (MRAN) to solve the traffic congestion problem in ATM networks. MRAN generates a minimal radial basis function neural network by adding and pruning hidden neurons based on the input data and is ideal for on-line adaptive control for fast time varying nonlinear systems. The ATM traffic modeling is carried out using the well-known network simulation software OPNET for multiplexed traffic (combining both speech and video signals). Performance of MRAN controller is compared with conventional method and Back-Propagation (BP) neural network controller with the aim of minimizing the congestion episodes and maintaining the quality. Simulation results indicate that MRAN controller performs better than both conventional and BP controller in reducing the congestion and maintaining a better quality of the traffic.

ATM congestion control using Minimal Resource Allocation Networks (MRAN)

“…The traffic model consists of three types of input sources (bursty, VBR and custom traffic), and their traffic model parameters are selected and multiplexed to result in a heavy traffic scenario. A preliminary study using two bursty and VBR traffic sources, along with a custom traffic source, has indicated the suitability of MRAN for congestion control [13]. In this paper, the performance of MRAN is evaluated for a multiplexed traffic consisting of four bursty and four VBR sources, along with a custom source to create a heavily congested traffic situation.…”