Signal/Noise Modeling and Detector Design for Media Noise Dominated Magnetic Recording Systems

该项目的目标是开发必要的工具和技术，以设计和实现在信号依赖的相关噪声环境中运行的磁记录系统的信号检测器，即媒体噪声主导的环境。根据最近的研究，预计目前的磁记录技术将在大约五年内达到超顺磁的极限。当磁性介质颗粒变得非常小，以至于室温下的热能足以通过随机改变其磁化方向来破坏颗粒的稳定时，就会出现超顺磁极限。为了推回即将到来的超顺磁极限，信号处理电路将不得不应对一个彻底改变的信号/噪声环境：信噪比将下降，磁介质噪声将主导电子白噪声，由于磁介质中的符号间相互作用和读头引入的着色，总体噪声将变得极其相关和依赖于信号。本项目将发展对非平稳随机建模的基本理解，并利用它来开发一种磁记录检测器，理想地调谐到相关的非平稳介质噪声。该项目将由哈佛大学和IBM阿尔马登研究中心合作进行。IBM是公认的全球磁记录技术领导者，在提供建立项目理论和实验部分之间联系所需的尖端实验支持方面处于独特地位。该项目包括三项主要努力。(i)将发展对依赖信号的噪声建模问题的理解。这是广泛应用于平稳过程的自回归建模方法在非平稳过程中的推广。特别注意将给予客观模型大小估计通过赤池的信息理论标准或Rissanen的最小描述长度原则的适应。（ii）接收机将通过复杂性约束优化程序进行设计。由于码间干扰长度和噪声记忆长度对接收机复杂度的贡献是相等的，因此约束复杂度检测器将在一定的性能指标下最小化这两个长度的总和。（iii）最终目标是探测器在VLSI技术中的实现。为了实现这一目标，将先开发软件检测器，然后再开发硬件检测器。这两种探测器将在IBM阿尔马登研究中心的真实信道环境中进行测试。

The object of the project is the development of the necessary tools and know-how to design and implementable signal detector for magnetic recording systems operating in signal-dependent correlated noise environments, i.e., media noise dominated environments. According to recent studies, it is anticipated that the current magnetic recording technology will hit the superparamagnetic limit in approximately five years. The superparamagnetic limit occurs when the magnetic media particles become so small that thermal energy at room temperature is enough to destabilize the particle by randomly switching its magnetization orientation. To push back the approaching superparamagnetic limit, signal processing circuits will have to cope with a radically changed signal/noise environment: the signal-to-noise ratio will drop, the magnetic media noise will dominate over the electronics white noise, and the overall noise will become extremely correlated and signal-dependent due to the intersymbol interactions in the magnetic media and due to the coloring introduced by the read head. This project will develop a fundamental understanding of nonstationary stochastic modeling and use it to develop a magnetic recording detector ideally tuned to correlated nonstationary media noise.The project will be conducted through collaboration between Harvard University and IBM Almaden Research Center. IBM, a recognized global magnetic recording technology leader, is in a unique position to provide the cutting edge experimental support required establishing the link between the theoretical and experimental parts of the project. The project combines three major efforts. (i) A understanding of signal-dependent noise modeling issues will be developed. This is an extension of widely used autoregressive modeling methods in stationary processes to nonstationary processes. Particular attention will be given to objective model-size estimation through the adaptation of Akaike's information theoretic criterion or Rissanen's minimum description length principle. (ii) The receiver will be designed through a complexity-constrained optimization procedure. Since both the intersymbol interference length and the noise memory length contribute equally to the receiver complexity, the constrained-complexity detector will minimize the sum of the two lengths under a performance specification. (iii) The final goal is the implementation of the detector in VLSI technology. Towards this goal, a software detector and then a hardware detector will be developed. Both detectors will be tested in real channel environments at IBM Almaden Research Center.