CN113610216B - Multi-task neural network based on polarity conversion soft information assistance and multi-track detection method - Google Patents

Multi-task neural network based on polarity conversion soft information assistance and multi-track detection method Download PDF

Info

Publication number
CN113610216B
CN113610216B CN202110788397.5A CN202110788397A CN113610216B CN 113610216 B CN113610216 B CN 113610216B CN 202110788397 A CN202110788397 A CN 202110788397A CN 113610216 B CN113610216 B CN 113610216B
Authority
CN
China
Prior art keywords
soft information
neural network
track
bit
polarity
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202110788397.5A
Other languages
Chinese (zh)
Other versions
CN113610216A (en
Inventor
王遥
徐钰舒
文玉梅
李平
陈蕾
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shanghai Jiaotong University
Original Assignee
Shanghai Jiaotong University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shanghai Jiaotong University filed Critical Shanghai Jiaotong University
Priority to CN202110788397.5A priority Critical patent/CN113610216B/en
Publication of CN113610216A publication Critical patent/CN113610216A/en
Application granted granted Critical
Publication of CN113610216B publication Critical patent/CN113610216B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G11—INFORMATION STORAGE
    • G11C—STATIC STORES
    • G11C11/00—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor
    • G11C11/46—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor using thermoplastic elements
    • G—PHYSICS
    • G11—INFORMATION STORAGE
    • G11C—STATIC STORES
    • G11C11/00—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor
    • G11C11/48—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor using displaceable coupling elements, e.g. ferromagnetic cores, to produce change between different states of mutual or self-inductance

Abstract

The invention aims at a heat-assisted alternating magnetic recording system, and relates to a multi-task neural network and multi-track detection algorithm based on polarity transition soft information assistance, which mainly comprises the following steps: in the first iteration process, the read-back signals of the middle three tracks are input into a multitask neural network to obtain the equalized signals of the three tracks and bit soft information, and the equalized signals and the bit soft information are input into a modified BCJR detector and an LDPC decoder, wherein turbo equalization is formed by exchanging soft information between the BCJR detector and the LDPC decoder. In the second iteration process, the log-likelihood ratio output by the LDPC decoder is converted into two-dimensional polarity conversion soft information and is input into the multitask neural network together with the read-back signals of the three magnetic tracks to obtain updated balanced signals and bit soft information, and then the updated balanced signals and the bit soft information are sent into the improved BCJR detector and the LDPC decoder to obtain the error rate of the magnetic recording system.

Description

Multi-task neural network based on polarity conversion soft information assistance and multi-track detection method
Technical Field
The invention relates to the technical field of signal processing in a heat-assisted alternating magnetic recording system, in particular to a multi-task neural network and a multi-track detection method based on polarity transition soft information assistance.
Background
The predominant magnetic recording technology currently used in the market is based primarily on perpendicular magnetic recording technology, but from magnetic recording mediaThe storage density is further increased by the 'triple dilemma' caused by three mutually-restricted factors of thermal stability, writing capability of a write head and signal-to-noise ratio. To further increase storage density, the corresponding bit size needs to be reduced, while the magnetic recording medium grain volume needs to be reduced to maintain a sufficient signal-to-noise ratio. On the other hand, to ensure the thermal stability of the magnetic recording medium grains, the anisotropy needs to be significantly improved, which poses a serious challenge to the magnetic write field that can be generated by the current magnetic write head. In order to further increase the storage density, a series of new magnetic recording techniques are proposed in the field of magnetic recording research, which are respectively: bitmap magnetic recording technology, heat-assisted magnetic recording technology, microwave-assisted magnetic recording technology, and shingle recording technology. The next generation magnetic recording technology can reach 1Tb/in2The above storage densities far exceed the storage density limit of conventional perpendicular magnetic recording systems. Recently proposed heat-assisted alternate magnetic recording technique [1 ] in contrast to conventional heat-assisted magnetic recording technique]The storage density can be further improved. In the thermally-assisted alternating magnetic recording system, the magnetic tracks are divided into high-temperature write magnetic tracks and low-temperature write magnetic tracks, the top layer is the low-temperature magnetic tracks, the bottom layer is the high-temperature magnetic tracks, and the high-temperature magnetic tracks and the low-temperature magnetic tracks are arranged in a crossed manner. The writing temperature of the high and low temperature magnetic tracks is different, but the writing sequence is fixed, and the specific sequence is that the low temperature magnetic tracks are written firstly and then the low temperature magnetic tracks are written on two sides of the low temperature magnetic tracks to cut the low temperature magnetic tracks. Although the thermally-assisted alternating magnetic recording technique can significantly increase the track density during writing, it introduces two-dimensional nonlinear displacement caused by two-dimensional demagnetization field and curved information bit effect caused by circular hot spot during information writing [2 ]]Therefore, serious nonlinear distortion, two-dimensional intersymbol interference and medium noise are brought to a read-back signal in the information reading process. For a conventional magnetic recording system, due to the low storage density, the system noise is mainly determined by white noise; while the noise in a thermally-assisted alternating magnetic recording system is dominated by thermal noise and storage medium noise, in addition to white noise. Therefore, it is necessary to research advanced two-dimensional signal processing algorithms to eliminate the above mentioned effects of non-linear distortion, two-dimensional inter-code interference and noise, thereby reducingThe bit error rate of the low magnetic recording system is reduced, and the storage density of a user is improved finally.
Researchers in the field of magnetic recording have studied various equalization and detection algorithms in order to mitigate two-dimensional inter-symbol interference and media noise in high-density magnetic recording systems. Nabavi [3] proposes a two-dimensional linear equalizer based on the minimum mean square error criterion to mitigate inter-track interference. Yamashita [4] studied the use of two-dimensional neural network equalizers to mitigate the effects of inter-track interference and media noise on error rate performance in two-dimensional magnetic recording systems. Recently, Wang [5] and Xu [6] have investigated a two-dimensional hybrid linear equalizer and improved Viterbi detector to mitigate the effects of inter-track interference in bit-mapped magnetic recording systems. However, the current research on the heat-assisted alternating magnetic recording system mainly focuses on the design of a novel read-write system and the optimization of read-write performance, and there are few corresponding signal processing algorithm researches and reports. In order to further increase the user storage density, there is an urgent need to develop signal processing algorithms suitable for use in heat-assisted alternating magnetic recording systems.
Reference to the literature
[1]S.Granz,W.Zhu,E.Seng,U.H.Kan,C.Rea,G.Ju,J.U.Thiele,T.Rausch,E.C.Gage,“Heat-assisted interlaced magnetic recording,”in IEEE Trans.Magn.,vol.54,pp.1-4,2018.
[2]J.Zhu,H.Li,“Correcting transition curvature in heat-assisted magnetic recording,”in IEEE Trans.Magn.,vol.53,pp.1-7,2017.
[3]S.Nabavi,B.V.K.V.Kumar,“Two-dimensional generalized partial response equalizer for bit-patterned media,”in Proc.IEEE Int.Conf.Commun.,pp.6249–6254,2007.
[4]M.Yamashita,H.Osawa,Y.Okamoto,Y.Nakamura,Y.Suzuki,K.Miura,H.Muraoka,“Read/write channel modeling and two-dimensional neural network equalization for two dimensional magnetic recording,”in IEEE Trans.Magn.,vol.47,pp.3558-3561,2011.
[5]Y.Wang,B.V.K.V.Kumar,“Improved multitrack detection with hybrid 2-D equalizer and modified viterbi detector,”in IEEE Trans.Magn.,vol.53,pp.1-10,2017.
[6]Y.Xu,Y.Wang,Y.Li,L.Chen,Y.Wen,P.Li,"Multitrack detection with two-dimensional hybrid equalizer for high-density bit-patterned media recording,”in IEEE Magnetics Letters,vol.11,pp.1-5,2020.
Disclosure of Invention
The invention aims to provide a method for realizing multi-track signal equalization and bit information prediction by utilizing a multitask neural network aiming at serious nonlinear distortion and two-dimensional intersymbol interference in a heat-assisted alternating magnetic recording system. The predicted bit soft information is then embedded as an estimate of the inter-track interference into the branch weights of the modified BCJR detector to improve multi-track detection performance.
The purpose of the invention is realized by the following technical scheme:
the invention provides a multi-task neural network and multi-track detection method based on polarity transition soft information assistance, which mainly comprises the following steps:
the method comprises the following steps: the randomly generated user bit data is written to the magnetic recording medium.
Step two: the read-back signals of the middle three tracks are obtained simultaneously by the read head array.
Step three: and establishing a multitask neural network to simultaneously perform multi-track signal equalization and bit prediction. The multitask neural network in the method is a fully-connected neural network, comprises an input layer, two hidden layers and an output layer, and can simultaneously output balanced signals and bit soft information. And taking the read-back signal obtained in the step two as the input of the multitask neural network to obtain an equalization signal and bit soft information. The joint loss function used in the multitask neural network training process is as follows:
Figure BDA0003160049630000021
wherein
Figure BDA0003160049630000022
And
Figure BDA0003160049630000023
and respectively representing the equalized signal and bit soft information of the ith track at the k moment output by the multitask neural network. M represents the length of the signal.
Figure BDA0003160049630000024
And ai[k]Respectively representing the ideal equalized signal and the real bits of the ith track at time k.
Figure BDA0003160049630000025
Is an ideal equalized signal obtained by convolving user bits with a two-dimensional partial response target, where the expression of the two-dimensional partial response target is shown in the following formula:
Figure BDA0003160049630000031
wherein
Figure BDA0003160049630000032
gi[1],…,gi[I]]Indicates the partial response target for the ith track, and I indicates the length of the intersymbol interference. According to the constraint, there is g when the ith track is detectedi[1]＝1。
Step four: and inputting the balanced signals of the three tracks and the bit soft information obtained by the multitask neural network into the improved BCJR detector to obtain the bit detection soft information. Taking the middle low-temperature track as an example, the calculation formula of the branch weight of the improved BCJR detector is as follows:
Figure BDA0003160049630000033
wherein Sk＝(a2[k],…,a2[k-I+1]) Indicating the state at the kth instant of the modified BCJR detector, alpha2[k]And y2[k]Respectively representing the real bit and equalized signal at time k for the intermediate cryogenic track.
Figure BDA0003160049630000034
Bit soft information of the ith track at the k moment representing the multitask neural network prediction. σ represents the standard deviation of gaussian white noise.
Step five: the output of the modified BCJR detector is fed into the LDPC decoder and turbo equalization is constructed by exchanging soft information between the modified BCJR detector and the LDPC decoder. And then converting the log-likelihood ratio output by the LDPC decoder into two-dimensional polarity transition information. When a two-dimensional polarity transition mode is defined, only a central bit and four adjacent bits around the central bit are considered, and if the polarities of the two adjacent bits are opposite, polarity transition is considered to exist; if the polarities of two adjacent bits are the same, it is determined that there is no polarity transition. In order to avoid error propagation possibly caused by polarity transition hard information, the log-likelihood ratio output by the LDPC decoder is converted into polarity transition soft information, and the conversion formula is as follows:
ST(a,b)＝P(a＝1)·P(b＝-1)+P(a＝-1)·P(b＝1)
where a, b represent two adjacent bits, P (-) represents a probability,
Figure BDA0003160049630000035
Figure BDA0003160049630000036
LLPDCrepresenting the log-likelihood ratio of the LDPC decoder output.
Step six: and inputting the two-dimensional polarity conversion soft information and the read-back signal into a multitask neural network together to obtain an updated equalization signal and bit soft information, and sending the updated equalization signal and bit soft information into an improved BCJR detector and an LDPC decoder.
Step seven: and carrying out hard decision on the log-likelihood ratio output by the LDPC decoder to obtain a decoded bit sequence and a final error rate.
Compared with the prior art, the invention has the beneficial effects that:
nonlinear distortion caused by arc-shaped hot light spots in the writing process and two-dimensional intersymbol interference caused by storage density increase in a heat-assisted alternating magnetic recording system can be well relieved, and therefore the performance of multi-track detection is improved. Compared with the traditional linear equalizer, the multitask neural network equalizer has good nonlinear fitting capacity, so that nonlinear distortion in a read-back signal can be reduced. Compared with the traditional single-task neural network equalizer, the multi-task neural network provided by the invention can simultaneously output the equalization signal and the bit soft information, thereby obtaining the estimation value of the inter-track interference. In addition, the multitask neural network exchanges and shares information among different tasks through the sharing layer, so that the learning performance of each task is improved. The two-dimensional polarity transition soft information assisted multitask neural network can learn the complex nonlinear relation between different polarity transition types and read-back signals, so that better balance and prediction effects are achieved.
Drawings
The drawings of the present invention are described below.
FIG. 1 is a block flow diagram of the process
FIG. 2 is a system block diagram of a thermally-assisted alternating magnetic recording system
FIG. 3 is a block diagram of a multitasking neural network
FIG. 4 schematic diagram of two-dimensional polarity transitions
Detailed Description
The invention will be further explained with reference to the drawings. The invention provides a multi-task neural network and multi-track detection method based on polarity transition soft information assistance, which mainly comprises the following steps:
the method comprises the following steps: the randomly generated user bit data is written to the magnetic recording medium as shown in fig. 2. A total of two high temperature written tracks HT1 and HT3 and three low temperature written tracks LT0, LT2, LT4 are included in the thermally-assisted alternating magnetic recording system.
Step two: the read-back signals of the middle three tracks HT1, LT2 and HT3 are simultaneously acquired by the read head array, and in order to reduce the interference of the outermost tracks LT0 and LT4, the read head positions corresponding to the high-temperature tracks HT1 and HT3 on the two sides are moved to the middle when the read-back signals are acquired.
Step three: the read-back signals for the middle three tracks are input into a multitasking neural network as shown in FIG. 3. The multitask neural network is a fully connected neural network and comprises an input layer, two hidden layers and an output layer. The multi-task neural network simultaneously outputs the equalization signal and the bit soft information, wherein the equalization process belongs to a regression problem, and the process of predicting the bit soft information belongs to a classification problem, which is equivalent to simultaneously optimizing two different tasks. Compared with a single-task neural network, the multi-task neural network improves the learning performance of each task through information interaction among different tasks and reduces the computational complexity at the same time. The joint loss function used by the multitask neural network in the training process is as follows:
Figure BDA0003160049630000041
wherein
Figure BDA0003160049630000042
And
Figure BDA0003160049630000043
and respectively representing the equalized signal and bit soft information of the ith track at the k moment output by the multitask neural network. M represents the length of the signal.
Figure BDA0003160049630000044
And
Figure BDA0003160049630000045
respectively representing the ideal equalized signal and the true bit values of the ith track at time k.
Figure BDA0003160049630000046
Is an ideal equalized signal obtained by convolving user bits with a two-dimensional partial response target, where the expression of the two-dimensional partial response target is shown in the following formula:
Figure BDA0003160049630000047
wherein
Figure BDA0003160049630000048
The partial response target of the ith track is indicated, and I-2 indicates the length of the intersymbol interference. According to the constraint, there is g when the ith track is detectedi[1]＝1。
Step four: and inputting the equalized signals and the bit soft information output by the multitask neural network into the improved BCJR detector. Wherein the predicted bit soft information is used for inter-track interference prediction and is embedded into branch weights of the modified BCJR detector. Taking the middle low temperature track LT2 as an example, the branch weight calculation formula of the improved BCJR detector is as follows:
Figure BDA0003160049630000049
wherein
Figure BDA0003160049630000051
Representing the state at the kth instant in the BCJR detector,
Figure BDA0003160049630000052
and
Figure BDA0003160049630000053
respectively representing the true bit and equalized signal of the intermediate low temperature track LT2 at time k.
Figure BDA0003160049630000054
Bit soft information of the predicted ith track at the time k is represented. σ represents the standard deviation of gaussian white noise.
Step five: the output of the modified BCJR detector is fed into the LDPC decoder and turbo equalization is constructed by exchanging soft information between the modified BCJR detector and the LDPC decoder. And then converting the log-likelihood ratio output by the LDPC decoder into two-dimensional polarity transition information. When defining a two-dimensional polarity transition pattern we consider the center bit and its four surrounding neighboring bits, as shown in fig. 4. In consideration of symmetry, the two-dimensional polarity transition patterns include 5 types in total, and when the polarities of two adjacent bits are opposite, it is determined that there is a polarity transition, i.e., T is 1; conversely, when the polarities of two adjacent bits are the same, it is determined that there is no polarity transition, i.e., T is 0. In order to avoid error propagation possibly caused by polarity transition hard information, the log-likelihood ratio output by the LDPC decoder is converted into polarity transition soft information, and the conversion formula is as follows:
ST(a，b)＝P(a＝1)·P(b＝-1)+P(a＝-1)·P(b＝1)
where a, b represent two adjacent bits, P (-) represents a probability,
Figure BDA0003160049630000055
Figure BDA0003160049630000056
LLPDCrepresenting the log-likelihood ratio of the LDPC decoder output.
Step six: and inputting the two-dimensional polarity conversion soft information and the read-back signal into a multitask neural network together to obtain an equalization signal and bit soft information, and inputting the equalization signal and the bit soft information into an improved BCJR detector and a subsequent LDPC decoder.
Step seven: and carrying out hard decision on the log-likelihood ratio output by the LDPC decoder to obtain a decoded bit sequence and a final error rate.

Claims (3)

1. A multi-task neural network and multi-track detection method based on polarity transition soft information assistance comprises the following steps:
the method comprises the following steps: writing randomly generated user data into a magnetic recording medium, wherein a heat-assisted alternating magnetic recording system adopts a staggered arrangement method in the writing process, the bottom layer is provided with two high-temperature writing magnetic tracks, and the top layer is provided with three low-temperature writing magnetic tracks;
step two: simultaneously acquiring read-back signals of the middle three magnetic tracks through the read head array, namely the read-back signals of the high-temperature written magnetic track, the low-temperature written magnetic track and the high-temperature written magnetic track;
step three: constructing a multitask neural network for obtaining an equalization signal and bit soft information according to the read-back signal; the multi-task neural network is constructed and comprises an input layer, two hidden layers and an output layer, a joint loss function combining mean square error and cross entropy loss function is adopted in the multi-task neural network training process, the two tasks of balance learning and bit prediction are learned at the same time, and the learning performance of each task is improved through information exchange and sharing between the two different tasks;
step four: inputting the equalization signal and bit soft information into an improved BCJR detector, wherein the predicted bit soft information is used for predicting inter-track interference and is embedded into branch weights of the improved BCJR detector, and finally obtaining a log-likelihood ratio of decoded bits through an LDPC decoder;
the calculation formula of the branch weight of the improved BCJR detector is as follows:
Figure FDA0003500387960000011
wherein Sk＝(a2[k]，...，a2[k-I+1]) Represents the state of the modified BCJR detector at the k-th time, a2[k]And y2[k]Respectively representing the real bit and equalized signal at time k of the intermediate cryogenic track,
Figure FDA0003500387960000012
bit soft information of an ith magnetic track predicted by the multitask neural network at the k moment is represented, and sigma represents the standard deviation of Gaussian white noise;
converting the log-likelihood ratio into two-dimensional polarity conversion soft information, and inputting the two-dimensional polarity conversion soft information and the read-back signal into the multitask neural network again to obtain an equalization signal and bit soft information;
and step six, sending the output equalization signals and bit soft information of the multitask neural network into an improved BCJR detector and an LDPC decoder for detection and error correction, and finally obtaining the error rate by carrying out hard decision on the decoding result of the LDPC decoder.
2. The polarity-transition-soft-information-assist-based multitask neural network and multi-track detection method as claimed in claim 1, characterized in that: the joint loss function is expressed as follows:
Figure FDA0003500387960000013
wherein
Figure FDA0003500387960000014
And
Figure FDA0003500387960000015
respectively representing the equalized signal and bit soft information of the ith track at the k moment of the output of the multitask neural network, M represents the length of the signal,
Figure FDA0003500387960000016
and ai[k]Respectively representing the ideal equalized signal and the real bits of the ith track at time k,
Figure FDA0003500387960000017
is an ideal equalized signal obtained by convolving user bits with a two-dimensional partial response target, where the expression of the two-dimensional partial response target is shown in the following formula:
Figure FDA0003500387960000018
wherein
Figure FDA0003500387960000019
Representing the partial response target of the ith track, I representing the length of intersymbol interference, and g when detecting the ith track according to a constrainti[1]＝1。
3. The polarity-transition-soft-information-assist-based multitask neural network and multi-track detection method as claimed in claim 1, characterized in that: converting the log-likelihood ratio into two-dimensional polarity transition soft information in the fifth step, specifically, the two-dimensional polarity transition modes include 5 types in total, and when the polarities of two adjacent bits are opposite, determining that polarity transition exists, that is, T is 1; conversely, when the polarities of two adjacent bits are the same, it is determined that there is no polarity transition, i.e., T is 0; in order to avoid error propagation possibly caused by polarity transition hard information, the log-likelihood ratio output by the LDPC decoder is converted into polarity transition soft information, and the conversion formula is as follows:
ST(a，b)＝P(a＝1)·P(b＝-1)+P(a＝-1)·P(b＝1)
where a, b represent two adjacent bits, P (-) represents a probability,
Figure FDA0003500387960000021
Figure FDA0003500387960000022
LLPDCrepresenting the log-likelihood ratio of the LDPC decoder output.
CN202110788397.5A 2021-07-13 2021-07-13 Multi-task neural network based on polarity conversion soft information assistance and multi-track detection method Active CN113610216B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110788397.5A CN113610216B (en) 2021-07-13 2021-07-13 Multi-task neural network based on polarity conversion soft information assistance and multi-track detection method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110788397.5A CN113610216B (en) 2021-07-13 2021-07-13 Multi-task neural network based on polarity conversion soft information assistance and multi-track detection method

Publications (2)

Publication Number Publication Date
CN113610216A CN113610216A (en) 2021-11-05
CN113610216B true CN113610216B (en) 2022-04-01

Family

ID=78304513

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110788397.5A Active CN113610216B (en) 2021-07-13 2021-07-13 Multi-task neural network based on polarity conversion soft information assistance and multi-track detection method

Country Status (1)

Country Link
CN (1) CN113610216B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117692101A (en) * 2023-12-04 2024-03-12 北京科技大学 Burst error code detection method and device based on convolutional neural network

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5978782A (en) * 1996-07-05 1999-11-02 National Semiconductor Corporation Neural network signal processor for magnetic storage channels
US20020012185A1 (en) * 2000-06-08 2002-01-31 Hideki Sawaguchi Device and signal processing circuit for magnetic recording, magnetic recording apparatus
CN102436834A (en) * 2011-11-14 2012-05-02 中兴通讯股份有限公司 Method and device for reading magnetic disk information
CN104601514A (en) * 2015-01-27 2015-05-06 中国民航大学 Subspace projection and CLEAN combined interference suppression method
CN108650201A (en) * 2018-05-10 2018-10-12 东南大学 Channel equalization method, interpretation method based on neural network and corresponding equipment
US20200389188A1 (en) * 2019-05-31 2020-12-10 Washington State University Deep neural network a posteriori probability detectors and media noise predictors for one-and two-dimensional magnetic recording

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP4324316B2 (en) * 2000-10-23 2009-09-02 株式会社日立グローバルストレージテクノロジーズ Perpendicular magnetic recording / reproducing device
US9786316B2 (en) * 2015-12-02 2017-10-10 Avago Technologies General Ip (Singapore) Pte. Ltd. Multi-track data readers with iterative inter-track interference cancellation
CN110278001B (en) * 2019-06-19 2020-12-01 北京交通大学 Polarization code partition decoding method based on deep learning

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5978782A (en) * 1996-07-05 1999-11-02 National Semiconductor Corporation Neural network signal processor for magnetic storage channels
US20020012185A1 (en) * 2000-06-08 2002-01-31 Hideki Sawaguchi Device and signal processing circuit for magnetic recording, magnetic recording apparatus
CN102436834A (en) * 2011-11-14 2012-05-02 中兴通讯股份有限公司 Method and device for reading magnetic disk information
CN104601514A (en) * 2015-01-27 2015-05-06 中国民航大学 Subspace projection and CLEAN combined interference suppression method
CN108650201A (en) * 2018-05-10 2018-10-12 东南大学 Channel equalization method, interpretation method based on neural network and corresponding equipment
US20200389188A1 (en) * 2019-05-31 2020-12-10 Washington State University Deep neural network a posteriori probability detectors and media noise predictors for one-and two-dimensional magnetic recording

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
"Multitrack Detection With a Two-Dimensional Soft-Transition Assisted Multitask Neural Network for Heat-Assisted Interlaced Magnetic Recording";Yuan Li et al.;《IEEE TRANSACTIONS ON MAGNETICS》;20210331;第57卷(第3期);第3100508-1-3100508-8页 *
"基于卷积神经网络的信道均衡算法";陈敏华 等;《计算机应用与软件》;20170930;第34卷(第9期);第257-261页 *
"硬盘读写通道的FIR噪声预测算法与仿真模型";刘春 等;《计算机科学》;20071231;第34卷(第11期);第286-289页 *

Also Published As

Publication number Publication date
CN113610216A (en) 2021-11-05

Similar Documents

Publication Publication Date Title
Karakulak et al. Joint-track equalization and detection for bit patterned media recording
Sayyafan et al. Deep neural network based media noise predictors for use in high-density magnetic recording turbo-detectors
CN113610216B (en) Multi-task neural network based on polarity conversion soft information assistance and multi-track detection method
Sayyafan et al. Deep neural network media noise predictor turbo-detection system for 1-D and 2-D high-density magnetic recording
Vasić et al. A study of TDMR signal processing opportunities based on quasi-micromagnetic simulations
Yao et al. Two-track joint detection for two-dimensional magnetic recording (TDMR)
Buajong et al. Improvement in bit error rate with a combination of a rate-3/4 modulation code and intertrack interference subtraction for array-reader-based magnetic recording
Nishikawa et al. A Study on Performance Evaluation With Neural Network Detector in SMR System
Li et al. Multitrack detection with 2-D iterative soft estimate aided neural network equalizer for heat-assisted interlaced magnetic recording
Keskinoz Two-dimensional equalization/detection for patterned media storage
Sayyafan et al. Turbo-detection for multilayer magnetic recording using deep neural network-based equalizer and media noise predictor
Aboutaleb et al. A perspective on deep neural network-based detection for multilayer magnetic recording
Busyatras et al. A simple 2D modulation code in single-reader two-track reading BPMR systems
Jeong et al. Bit-flipping scheme using k-means algorithm for bit-patterned media recording
Myint et al. Reduced complexity multi-track joint detector for sidetrack data estimation in high areal density BPMR
Qin et al. CNN-based machine learning channel on TDMR drive data
Buajong et al. A simple inter-track interference subtraction technique in bit-patterned media recording (BPMR) systems
CN110211611B (en) Two-dimensional channel equalization model training method and two-dimensional channel equalization method
Hwang Asynchronous inter-track interference cancellation (A-ITIC) for interlaced magnetic recording (IMR)
Wang et al. Bidirectional decision feedback modified Viterbi detection (BD-DFMV) for shingled bit-patterned magnetic recording (BPMR) with 2D sectors and alternating track widths
Wang et al. Joint Four-Reader Array Equalization and Detection for a Single Track in TDMR
Warisarn et al. Mitigation of TMR using energy ratio and bit-flipping techniques in multitrack multihead BPMR systems
Xu et al. Multitrack detection with a two-dimensional soft-transition assisted multitask neural network for heat-assisted interlaced magnetic recording
Buajong et al. Multilayer perceptron-based detection for a coded two-dimensional magnetic recording
Koonkarnkhai et al. A simple 2-head 2-track detection method for bit-patterned magnetic recording systems

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant