Performance of the hybrid MLPNN based VE (hMLPNN-VE) for the nonlinear PMR channels

Abstract

This paper proposes a hybrid of multilayer perceptron neural network (MLPNN) and Volterra equalizer (VE) denoted hMLPNN-VE in nonlinear perpendicular magnetic recording (PMR) channels. The proposed detector integrates the nonlinear product terms of the delayed readback signals generated from the VE into the nonlinear processing of the MLPNN. The detection performance comparison is evaluated in terms of the tradeoff between the bit error rate (BER), complexity and reliability for a nonlinear Volterra channel at high normalized recording density. The proposed hMLPNN-VE outperforms MLPNN based equalizer (MLPNNE), VE and the conventional partial response maximum likelihood (PRML) detector.

Fuzzy logic‐based adaptive equaliser for non‐linear perpendicular magnetic recording channels

“…The k th input symbol a k ∈{±1} with channel bit period T passes through the delay differentiator (1– D )/2 to form the transition sequence b k ∈{0, ±1}. The noise‐free signal is then obtained as the convolution between b k and the transition response g ( t ) given by [14] where …”

“…A VE is basically performing an inverse of the channel and analogous to the VS model (3). The delayed output of L VE th ‐order of VE( N , L VE ) is [14] where is a (2 N –1)‐element column vector of weight coefficients and is a (2 N –1)‐element column vector of non‐linear product terms of S k generated from the non‐linear combiner (Fig. 2).…”

“…Through the optimisation of VE( N , L VE ) [14], the (2 N –1)‐dimension of S is reduced to N 1 ‐dimension of , , where N C i represents the number of i combinations from a given number N . Therefore, VE( N , L VE ) is replaced by VE( N , Q ) of the Q th order.…”

“…Their performances in terms of the BER are similar, but the h MLPNN‐VE is much less complex. Furthermore, it is evident from [14] that the h MLPNN‐VE can significantly reduce BER. However, this equaliser and the others related to NN with a number of system parameters maybe over fitted and not suitable in practice.…”

“…This approach outperforms PRML and VE in terms of BER at the expense of complexity. Recently, the hybrid of MLPNN and VE called h MLPNN‐VE [14] has been proposed and implemented in the non‐linear PMR channels which are modelled by Volterra series (VS) model. By means of double non‐linear transformation, the non‐linear product terms of delayed readback signal of VE scheme are first formulated and transformed into a second non‐linear processing of MLPNN at hidden layer.…”