No abstract

Noise and interference on communications channels can often deviate significantly from that predicted by Gaussian models. This unfortunate reality can degrade the system performance t o unacceptable levels. Empirical universal receivers have been developed which achieve asymptotic optimality over a very wide range of channels, and are thus useful in such situations. However, these receivers require training or pilot signals from the channel, which in turn reduces the effective bandwidth of the channel. In this paper, we demonstrate a "training-free" empirical receiver which is shown t o perform significantly better than reasonably trained universal receivers and nearly as well as the globally optimal receiver. Importantly, this receiver is capable of deriving the necessary training information directly from the transmitted communication signal, and thus requires no bandwidth overhead. Applications to QPSK reception are discussed thoroughly.

No abstract