This is straight from the lietrature. A Bussgang LMS algorithm. Don't expect the true values to be estimnated. The non-linearity used is Tanh(.)
which corresponds to a Laplace driving noise (though you can compare with Guassian if you so choose).
The object of teh exercise is just to minimise the mean square error. Only the output of an unknown system is available to measure and not the input (hence blind identification).
If you know more about these techniques then let me know.
