UC San Diego

Two dimensional magnetic recording (TDMR) is one of the leading technologies proposed to achieve ultra high storage density in the next generation hard disk drives. A typical read channel of TDMR can be approximated by a multihead multitrack (MHMT) model which is characterized by the intersymbol interference (ISI) in the downtrack direction and intertrack interference (ITI) in the crosstrack direction. In this dissertation, we aim to address two challenging problems that can potentially make the maximum likelihood (ML) detection for the MHMT channel impractical.

One problem associated with the conventional ML detector is the ITI dependency of the ML trellis, which makes it inefficient to adapt to a time-varying ITI environment. We propose a novel detection method, called the weighted sum-subtract joint detection (WSSJD), to solve this issue. We show that the ITI interfering channels in MHMT system can be transformed to several parallel subchannels, whose joint trellis is independent of ITI. For the case when the receiver lacks the knowledge of ITI, we propose a gain loop structure which can be incorporated into WSSJD to efficiently track the ITI estimates. We present the implementation of WSSJD and simulation results for several widely-studied MHMT models including the 2H2T channel, the 3H3T channel, and the 3H2T channel.

Another challenging problem is the exponentially increased computational complexity of the ML detector due to its multitrack processing scheme. We propose to use the reduced-state sequence estimation (RSSE) to mitigate this issue. The underlying idea is to drop less likely paths at early state during the detection by using a heavily reduced subset trellis. To apply RSSE to the MHMT channel, we find that the channel transformation developed in WSSJD offers a natural set partition principle on the input constellation that is necessary to successfully implement RSSE. The theoretical error event analysis shows a good consistency with the simulation results.

Flash memory has faster access speed, lower power consumption, and better data integrity, which make it more attractive than the magnetic recording in many applications. However, the lifetime of flash memory is limited by the program/erase cycles. Write-once memory (WOM) codes are proposed to extend the flash memory lifetime by constraining the unidirectional changes of cell levels. In this work, we study the expected performance of WOM codes. Dynamic programming based algorithm is proposed to construct the optimal WOM updating function of a given labeling function.