Optimal quality of service guarantees for noisy packet data networks

One of the major problems of multiplexing higher bit rate packet data on to the same network link is that the data gets crowded more closely, and this inevitably leads to the network becoming even more susceptible to noise and other channel impairments. We have carried out investigations of the effect of noise on the negotiated quality of service parameters-particularly on the end-to-end performance-such as channel throughput, and the probability of cell loss for ATM networks. To overcome the computational difficulties associated with using analytical models in the calculation of the temporal arrival of packets of multiplexed on-off sources to the same network access point, we have developed a simple, but cost effective method using probabilistic random access memory neural networks to achieve optimal performance of controlling declared usage parameters in the presence of channel and packetisation noise.