Non-Bayesian Rate-Adaptive Wireless Communication Using ARQ-Feedback

To combat the detrimental effects of the variability in wireless channels, we consider cross-layer rate adaptation. We study communication systems that utilize the limited feedback in the form of linklayer Automatic Repeat-reQuest (ARQ) to maximize the physical-layer transmission rate, subject to a certain upper bound on the expected packet error rate. We assume that there is no knowledge of the prior distribution of the channel state at the transmitter. We first analyze the fundamental limitations of such systems and derive an upper bound on the achievable rate for signalling schemes based on uncoded modulation with QAM and random-coded modulation with Gaussian ensembles. We show that, for channel estimation based on binary ARQ feedback, it may be preferable to use a separate training sequence at high error rates, rather than to exploit low-error-rate data packets themselves. We also develop an adaptive recursive estimator, which is provably asymptotically optimal and asymptotically efficient.