BRAVE

Abstract

Rate selection in a wireless network is the problem of estimating the current channel conditions and determining the best physical layer bit rate for the outgoing frames in order to maximize the current throughput. All rate adaptation algorithms in literature arrive at an estimate of the current channel conditions by considering the recent history often in the order of seconds. In vehicular WiFi access networks, the constantly changing wireless channel conditions make the channel history quickly irrelevant. We develop BRAVE - an SNR-based rate adaptation algorithm, which only considers short history (500 ms) to make rate selection decisions. We show that a coarse-grained training approach is sufficient to estimate the SNR thresholds for rate selection as opposed to previous approaches that train on a per environment or a per AP basis. We study three frame-based rate adaptation algorithms and a popular SNR-based rate adaptation algorithm along with BRAVE and highlight their shortcomings in the rapidly changing vehicular WiFi access environment. In order to compare the algorithms under repeatable channel conditions, we also develop and use a novel emulation methodology where a software radio-based programmable noise generator is used to emulate varying link quality under vehicular mobility. We show that BRAVE performs significantly better than several prominent frame-based and the SNR-based rate adaptation algorithms.