At high linear recording densities, coping with electronics and media noise becomes increasingly difficult. We are developing advanced data detection schemes that shape the magnetic recording channel impulse response to a polynomial target with a few non-integer coefficients to incorporate the noise characteristics. This sequence detection approach is termed noise predictive maximum likelihood (NPML) detection [2010-10], [2009-3], [2009-4], [2011-8].

The transition from conventional PR4 and EPR4 detection schemes with fixed target polynomial to advanced noise-predictive schemes with targets that are self-adapting to the channel conditions provides, together with the adaptive digital front-end functions, a fully adaptive design that approximates maximum-likelihood detection.

Noise prediction for stationary Gaussian noise sources extends to the case where the noise characteristics depend on local data patterns such as transition noise in magnetic recording. In this case, maximum likelihood sequence estimation can be implemented by using an NPML sequence detection algorithm with branch-metric computations involving noise prediction conditional on the local data pattern.

Depending on the nature of the noise process, data-dependent NPML (DD-NPML) could provide further improvements in detector performance at the cost of higher complexity, as indicated in Fig. 1.

Data detection

Figure 1. Sequence detection techniques

Ask the expert

Mark A. Lantz

Mark A. Lantz

IBM Research scientist