Signal processing system design for two-dimensional magnetic recording disk drives

Abstract

As the global digital data volume keeps exploding with an annual growth rate of over 60%, reliable and ever more affordable mass data storage hard disk drives (HDDs) are indispensable in our digital universe. However, the approaching superparamagnetic limit dictates that conventional magnetic recording technology scaling methods will no longer assure the areal density growth beyond 1 Tbit per square inch. There is an emerging opinion that two-dimensional magnetic recording is the most viable and economic option to sustain the historical areal density growth and enable future low-cost mass data storage systems. This thesis presents a variety of read channel signal processing system design solutions for the emerging two-dimensional magnetic recording HDDs.

Finally, this thesis presents a virtual sector design strategy to customize very high density HDDs for big data infrastructure. It interleaves a few consecutive physical sectors to form a virtual sector, and benefits the read channel performance by averaging bit error rates and enabling message passing among different codewords. Several techniques based upon the concept of journaling and caching are further developed to reduce the speed performance impact of virtual sector on HDDs.

Furthermore, this thesis introduces a self-directed equalization method for multi-sensor read head. It exploits the spatial diversity of read-back signals from different read sensors to accurately estimate run-time read head offsets, based upon which the equalizer coefficients are accordingly self-configured in order to improve the equalization performance. Simulation results show that it outperforms the traditional least mean square algorithm, at minimal implementation overhead.

Next, this thesis formulates the 2-D pattern dependent noise prediction to explore the noise characteristics of different data patterns in very high bit density scenario. It confirms the effectiveness of noise prediction in 2-D detection and the superiority of 2-D noise prediction over 1-D noise prediction.

performance. Its silicon implementation cost is also evaluated, together with a pipelining strategy to reduce the hardware consumption.

Under the framework of 2-D read channel signal detection, this thesis further develops two low-complexity 2-D soft-output Viterbi algorithms, representing different complexity vs. performance trade-offs, to reduce the complexity of the soft information generation operation that dominates the complexity of the whole detector. Compared with the original symbol-based 2-D soft-output detection, the developed design solutions can achieve almost the same detection performance and meanwhile reduce the silicon area at least by 40%.

It first presents a multi-track joint 2-D read channel signal detection algorithm to simultaneously detect multiple tracks with the aid of multi-sensor read head. Compared with conventional 1-D signal detection, this design solution can achieve significantly superior