Research On Data Read/Write Channel Techniques For Ultra-High Density Two-Dimensional Magnetic Recording

Hard disk storage based on magnetic recording technology will continue to dominate the information storage market for longer due to its obvious advantage of capacity-to-price ratio.As a key way for the future development of magnetic storage technology,two-dimensional magnetic recording(TDMR)acquires readback signals and recovers data from interfered waveforms along and across the magnetic tracks by two-dimensional(2-D)signal process-ing methods.Since high-density magnetic storage requires smaller storage cells,each bit is inevitably affected by the signal interference from adjacent bits,media noise,and sys-tem noise.The main challenge of high-density magnetic storage is to effectively mitigate2-D signal interference including down-track intersymbol interference(ISI)and cross-track intertrack interference(ITI).To address the interference highlighted by ultra-high-density TDMR,this thesis conducts in-depth research on media and read-write channel models,2-D signal equalization/detection,and data coding,respectively:a high-density recording grain model and a read-write model reflecting the actual media characteristics are constructed to reproduce the media noise and signal interference?The 2-D neural network equalization and detection algorithms based on disk data blocks are proposed to effectively eliminate 2-D non-linear interference and achieve data detection of this kind of channels?a constrained coding method to eliminate harmful continuous transitions is proposed,which provides theoretical support to improve the reliability of data access and magnetic storage density.For the characterization of media noise and signal interference in ultra-high-density magnetic recording,the Voronoi media grain model and the read-write process model based on 2-D random points are proposed.An iterative generation algorithm based on 2-D random distributed points is designed to construct a Voronoi grain model that reflects the actual me-dia characteristics,i.e.,the standard deviation to the average size ratio of grain areas which conforms to the real media parameters.Compared with the existing Voronoi grain model based on a standard grid,the proposed construction method of the media model reduces the long-range correlation of particle positions by 15.3%which more realistically characterizes the randomness distribution of the recording media.Based on the proposed media model,the write error(WE)rates for different parameters of media particles and recording bits are analyzed.The corresponding WE characteristics are introduced into the write process,and the readback model of 2-D magnetic recording is established based on the written data and the effective readback response area.The constructed read/write channel model accurately includes the WE characteristics that occur in the write process and the 2-D interference char-acteristics in the read process.For the 2-D interference problem in the data read process,a neural network-based 2-D signal equalization algorithm is proposed.In the readback process of ultra-high-density magnetic recording,the readback signal of a recording bit is affected by its adjacent multiple bits at the same time,in addition to the influence of a variety of system noise from the circuit system,mechanical vibration,etc.Therefore,the TDMR reader induction signal is a nonlinear superposition of multiple sources.The proposed neural network equalization algorithm can employ the nonlinear processing capability of the neural networks to estimate the written data corresponding to the related readback signal data block and other system noise,which effectively suppresses the signal crosstalk in the TDMR data recovery process.The experimental results show the conventional equalization detection algorithms are unable to achieve the raw bit error rate(BER)required for effective data recovery under low SNR conditions while the proposed neural network equalizer can achieve a gain of more than 8d B compared with conventional 2-D equalizers,and the raw BER of the system is reduced by more than 30%.For the sequence detection of 2-D inference channels of ultra-high-density magnetic recording,a neural network-based 2-D signal detection method is proposed.The conven-tional sequence detector only performs detection based on a limited 1-D ISI response,ignores the inter-track signal response,and cannot detect the target channel signal according to the nonlinear correlation among adjacent channel signals.The proposed neural network detector does not rely on any partial response target and directly calculates the magnetization state of the corresponding bit in the centroid position of the multi-track readback data block,which improves the reliability of the information detection.The detector shows the capability of both signal equalization and detection,which enables easier and more efficient data recov-ery.Experimental results show that the neural network detector with an input sample size of 3_×3 reduces the raw BER by nearly 50%compared with the conventional equalization detection method under most signal-to-noise ratio conditions.To curb the interference of the magnetization transition noise within the recording pat-terns in ultra-high-density magnetic storage systems,a maximum transition run(MTR)code to limit the continuous magnetization switching of written data is proposed.For higher-density magnetic storage,the sizes of recording media particles and recording bits are de-creasing,the recording bits are getting closer to each other,and the system reliability is more obviously affected by the transition noise.By eliminating the continuous magnetization tran-sitions in the recording pattern,the MTR length is 1,which effectively reduces the influence of transition noise in the written data and significantly improves the reliability of the system detection results.The experimental results show that the BER of the constrained coding with an MTR length of 1 is reduced by about 50%compared with that of the unconstrained coding system.When the signal-to-noise ratio is 12 d B,this coding method can reduce the BER by about 30%and 60%compared with MTR codes with lengths of 2 and 3,respectively.In summary,this thesis systematically investigates the equalization,detection,and con-straint code techniques for the 2-D magnetic recording data write-read process,establishes theoretical models of ultra-high-density 2-D magnetic recording media and read/write chan-nels,proposes a neural network-based 2-D signal equalization algorithm and detection al-gorithm,and a constrained coding method to effectively eliminate the transition noise.The experimental results show that the proposed neural network signal processing algorithm ef-fectively suppresses the nonlinear interference in the ultra-high-density magnetic storage system and reduces the complexity of data reading.Meanwhile,the proposed constrained coding method eliminates the continuous transition noise of the written data and effectively improves the reliability of the magnetic storage system.This thesis possesses important theory significance and application value for remarkably increasing the magnetic storage density and capacity.