A Novel HOS Approach for Blind Channel Equalization

We present in this paper a new equalization method with over 8 dB advantage in the residual ISI compared to the classical results presented by Godard and Shalvi/Weinstein (kurtosis criteria). A systematic derivation is presented for obtaining the conditional expectation for f(x/z) that does not rely on the knowledge of the convolutional noise power nor imposes any restrictions on the probability distribution of the (unobserved) input sequence. We use Edgeworth series which is directly related to quasi moments thus to cumulants, and the Laplace integral method. Although the Edgeworth expansion and the Laplace integral method are well known in non-linear optimal filtering theory, they have not yet been used in the field of blind equalization