On the Impact of Ignoring Markovian Channel Memory on the Analysis of Wireless Systems

Recent wireless measurement studies have revealed the presence of high-order memory in wireless bit-error channels. However, most wireless studies continue to employ the memory-less or 1st order Gilbert bit-error channels to design, analyze and verify wireless protocols and systems. The inaccuracies incurred by ignoring high-order channel memory are largely unexplored. This paper quantifies inaccuracies incurred by ignoring bit-level wireless channel memory in the context of two simple and commonly-used protocol metrics: (i) packet good put of an abstract unreliable protocol and (ii) number of retransmissions per packet for an abstract reliable protocol. We analytically derive expected values of these metrics in terms of the parameters of four models with varying levels of memory. We then train the models using actual 802.11b bit-error traces and use the analytical expressions to estimate the metrics. Comparison of the model-based estimates with actual values of the metrics derived from the traces shows that the memory-less and 1st order models incur significant inaccuracies. The remaining two models that capture high-order memory are very accurate in their estimates of the crucial good put and retransmission metrics.