Blind equalization attempts to remove the interference caused by a communication channel without using any known training sequences. Blind equalizers may be implemented with linear prediction-error filters (PEFs). For many practical channel types, a suitable delay at the output of the equalizer allows for achieving a small estimation error. The delay cannot be controlled with one-step predictors. Consequently, multistep PEF-based algorithms have been suggested as a solution to the problem. The derivation of the existing algorithms is based on the assumption of a noiseless channel, which results in zero-forcing equalization. We consider the effects of additive noise at the output of the multistep PEF. Analytical error bounds for two PEF-based blind equalizers in the presence of noise are derived. The obtained results are verified with simulations. The effect of energy concentration in the channel impulse response on the error bound is also addressed.

Blind equalization of a communication channel using a prediction-based lattice blind equalizer (LBE) is considered. Second order cyclostationary statistics and a single-input multiple-output model arising from fractional sampling of the received data are used. The performance of the LBE algorithm is studied in extensive simulations where commonly used example channels are employed. Convergence in the mean square error (MSE) and symbol error rate (SER) as well as the number of symbols required to open the eye are studied at different SNRs. Robustness in the face of channel order mismatch and channels with common subchannel zeros is considered. The simulation results are compared to the results obtained by the fractionally spaced constant modulus algorithm, the cyclic-RLS algorithm and the subspace method by Moulines et al. (see IEEE Trans. Signal Proc., vol.43, no.2, p.516-25, 1995).