Abstract: An adaptive blind equalization method and device uses self-decisions both in the convergence phase and the steady-state phase, making it possible to obtain a good and smooth adaptation in the convergence phase and a low error probability for the decided symbols in the steady state phase. By using the self-decided error and a reference error, a binary consent function is constructed which tells when it is allowed to proceed with the adaptation and when the adaptation should not be used. The binary consent function compares the self-decided error sign with the reference error sign and is based on the discovery that in certain areas of the symbol constellation plane, when the signs of the previous errors are equal, there is a high probability that the self-decided error signs and true error signs (unknown) will also be equal.

Abstract: The adaptive auto-orthogonalizing equalizer in the discrete frequency domain comprises a filter with N cells operating by the overlap-save method, a decision element which assigns one of the possible levels to the filtered samples, and a control circuit which at each iteration varies the cell gains on the basis of the input samples, the output samples and the decided symbols, and which comprises an estimator (S.sub.A) for estimating the gradient of the mean square error, an estimator (S.sub.B) for estimating a matrix which modifies said gradient such that its direction always passes as close as possible to the minimum to be sought, a projector (P.sub.S) for projecting said corrected gradient into a constraint region, and a circuit for adjusting the filter cell gain vector at each iteration.