Identification of symmetric/asymmetric quadratic nonlinear channels with cyclostationary inputs

A new family of blind nonlinear channel identification techniques are introduced that exploit the cyclostationary nature of most data communication signals to achieve significant advantages over conventional methods. These methods place less restrictions on the input signal, use lower order statistics, require less computation, are simple (direct solutions result for blind identification of most nonlinear models) and make use of the obvious relationship between spectral correlation and nonlinearity. Some simple frequency domain methods are proposed for blind identification of quadratic time-invariant nonlinear models that require computation of only the cyclic autocorrelation of the received sequence. These methods can be extended to the case of higher order nonlinear models. Linear subsystems in the models are allowed to be of nonminimum phase (NMP). A rational sampling scheme is suggested for estimation of spectral correlation in a bandlimited cyclostationary signal with high spectral correlation resolution. The results are supported by Monte-Carlo simulations