PERFORMANCE STUDY ON DETECTION METHODS FOR DIGITAL MAGNETIC RECORDING CHANNELS

Abstract

Advanced signal processing techniques are playing more important roles than ever in the digital magnetic recording products, Sampling detection schemes are essential to provide reliable and high-speed recovery of data that is packed in the media at very high density. The detection performance of several sampling detection schemes in digital magnetic recording channel are studied with both theoretical analysis and computer simulations. Among these schemes are partial response equalization with maximum likelihood detection (PRML) family, including PR4ML and EPR4ML, conventional decision feedback equalization (DFE), multilevel decision feedback equalization (MDFE) and trellis coded partial response (TCPR). These schemes are either being used currently or considered as popular candidates for next generation products. Methods for designing essential parts of the read channel, i.e. the equalizer and the detector, are presented. The performances of these detection schemes are examined in a variety of channel models covering wide range of characteristics of the digital magnetic recording channel. Starting from a simple linear time-invariant system with additive white Gaussian noise (AWGN), other impairment exclusive to digital magnetic recording channel are incorporated. While taking into account additional noise sources such as transition noise, the channel is modeled by a linear time-varying system. This model is then extended to accommodate nonlinear distortions clue to the interactions of consecutive transitions such as transition broadening, and partial erasure. Magneto-resistive (MR) head nonlinearity is modeled by a memoryless nonlinear device characterized by its transfer function. With the incorporation of off-track interference, a comprehensive channel model is established. Extensive studies on the detection performance of the detection schemes for various channel models are conducted. For the linear channel model with AWGN, the theoretical performance is derived and verified by simulation. For other channel models, extensive simulation results are presented. The signal to noise ratio (SNR) definitions throughout the work are consistent and independent of the coding scheme and recording density. The performances of various detection schemes are compared under different circumstances. The studies show that PR4ML and TCPR have good performance at moderate density but may loss much at high density due to noise enhancement. EPR4ML has superior performance in linear channel at high density owing to close match between the equalization target and the channel response. DFE has robust performance over a wide range of densities, but the error propagation problem is remarkable. All these four schemes are vulnerable to nonlinearities in the media due to the interactions of closely written transitions. MDFE has inferior performance at low to moderate density because of the penalty in coding rate. But the run length constraint in the input sequence helps MDFE to recover much of the coding rate loss at high density and keeps the read back signal free from nonlinearities in media. MDFE has more tolerance than PR4ML in the presence of MR head nonlinearity. The error propagation in MDFE is not as severe as in DFE.