A faster Arimoto-Blahut algorithm via squeezing

We present a “squeezing” strategy which aims at improving the convergence rate of the classical Arimoto-Blahut algorithm while preserving its simplicity and monotonic convergence properties.