Phase inference and error surface analysis of a blind non-minimum phase channel equalizer

Blind equalization of a non-minimum phase (NMP) channel can be achieved with an infinite impulse response (IIR) predictor and an all pass filter. The adaptive IIR predictor minimises the square of the prediction error with the simplified gradient least mean square algorithm which possesses the self-stabilizing property, and the all pass filter is adapted to minimise the fourth moment of its output. This paper analyses the error surface of the all pass filter output for an NMP channel and proposes a new scheme to initialise the all pass filter from the IIR predictor coefficients. A method used to achieve blind equalization with only second order statistics which has immunity to local minima is also proposed. Simulation results are used to support the work