CAREER: Timing Recovery in Digital Magnetic Recording Systems

Harvard University, Cambridge MA

Investigators

Abstract

As magnetic recording densities increase, powerful signal processing algorithms will need to be utilized to overcome a drop of up to 6dB in the signal-to-noise ratio. Recently, iterative decoding methods (applied to turbo codes and low-density parity-check codes) have shown to approach the channel capacity of additive white Gaussian noise channels. Simulation show that turbo codes routinely achieve coding gains up to 6dB over conventional maximum likelihood detection in magnetic recording partial response channels. A drawback of iterative decoding is a decision delay that is intolerably long for the timing recovery loop. Clearly, robust timing recovery algorithms that can operate at very low signal-to-noise ratios and can tolerate long decision delays are needed. This project is developing timing recovery methods for receivers employing iterative decoding strategies. Since iterative decoders introduce a delay of several thousand bits, it is suggested that the waveform be asynchronously oversampled, and that the decoding delay be used to refine the sampling in an iterative schedule that follows closely the schedule of the iterative decoder. This research developing a 'soft' decision-directed timing recovery method. During the decoding iterations of an iterative decoder, soft information is passed between the processing elements. This soft information is the reliability information regarding a specific bit (or a bit-sequence) and could be used to refine the sampling instants of the received waveform. This is precisely the strategy used here. Iterative decoding is well explained by passing conditional probability information on a code-graph. Recently, it has been shown that both equalization and decoding can be performed using the same message-passing strategy on a joint channel/code graph. The goal here is to expand this concept by formulating the entire receiver using a system graph. A suitable message-passing algorithm on the entire graph will then perform the basic receiver functions (synchronization, sampling, equalization, decoding).