TCP performance over indoor bursty wireless channel models

It is known that in indoor environments errors tend to happen in bursts, severely damaging the performance of upper layer protocols and, in particular, TCP. This work compares the performance of this protocol over two channel models (by means of simulation) that are able to replicate the presence of long error bursts: the first one is based on a Hidden Markov Process, while the second one uses an auto-regressive filter to emulate the received signal level. The study is conducted by means of extensive simulations using the ns-3 framework. The results show that both alternatives actually mimic the behavior shown by TCP over real bursty channels, although the former shows a more predictable performance (for a single configuration). Both of them clearly outperforms other legacy approaches, and yet do not introduce a relevant overhead in terms of simulation time.