The asymptotic behavior of the best achievable error probability over the memoryless Gaussian channel has received much attention over the years both in the absence and in the presence of feedback. While much of the earlier work on feedback has focused on ideal feedback, there has recently been growing interest in imperfect feedback. In this talk, we consider the scenario of an average-power-limited Gaussian channel with intermittent feedback, i.e., where each channel output is fed back with some given probability ρ. We present converse bounds and coding schemes for the two cases where the receiver either does or does not know which symbols are fed back. In particular, we show that the asymptotic decay of the probability of error is double-exponential in the blocklength in some cases but only exponential in other cases. Our proof techniques will also provide insight into the exponential decay of the error probabilities for different imperfect feedback models.     «

The asymptotic behavior of the best achievable error probability over the memoryless Gaussian channel has received much attention over the years both in the absence and in the presence of feedback. While much of the earlier work on feedback has focused on ideal feedback, there has recently been growing interest in imperfect feedback. In this talk, we consider the scenario of an average-power-limited Gaussian channel with intermittent feedback, i.e., where each channel output is fed back with...     »