Performance Evaluation of SNR Prediction Schemes in Acoustic Communications using Variable-Rate Modulation

Variable-Rate Modulation (VRM) is a technique well known in wireless radio to potentially enhance the performance of digital transmissions over links of time-varying quality. VRM usually relies on feedback from the receiver in order to perform a timely adaptation of the modulation scheme in use. As underwater communications are subject to round-trip delays longer than the channel coherence time, this feedback may not be sufficiently representative of the channel quality at the time of the next transmission. Therefore, it may be more convenient to resort to a predictive processing of previous feedback on channel quality. In this paper, we explore the capability of several predictive techniques such as linear predictive filters, Kalman filters, Markov models as well as some heuristic schemes to infer future channel Signal-to-Noise Ratio (SNR) values based on current and past SNR samples, with the ultimate objective of reducing the probability of outage events (defined as the choice of a modulation scheme not supported by the current SNR). The SNR traces used in our evaluation are taken from real experiments.