Blind equalization in the presence of bounded errors with unknown bound

We consider the blind identification of a FIR channel when the errors on the observations (received signal) are bounded. We show that, using the fact that the input signal is in a finite alphabet, the channel parameters and the bound can be estimated on-line. However, since the input sequence is unknown, the number of estimators to be run in parallel grows exponentially fast. A sub-optimal approach is thus considered, which consists in retaining only the K best candidate sequences for the L/sub /spl infin//-norm.