CN108322225A - LDPC interpretation methods based on mutual information and device - Google Patents
LDPC interpretation methods based on mutual information and device Download PDFInfo
- Publication number
- CN108322225A CN108322225A CN201710030720.6A CN201710030720A CN108322225A CN 108322225 A CN108322225 A CN 108322225A CN 201710030720 A CN201710030720 A CN 201710030720A CN 108322225 A CN108322225 A CN 108322225A
- Authority
- CN
- China
- Prior art keywords
- data
- processing procedure
- group
- combination
- mutual information
- 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.)
- Granted
Links
Classifications
-
- H—ELECTRICITY
- H03—ELECTRONIC CIRCUITRY
- H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
- H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
- H03M13/03—Error detection or forward error correction by redundancy in data representation, i.e. code words containing more digits than the source words
- H03M13/05—Error detection or forward error correction by redundancy in data representation, i.e. code words containing more digits than the source words using block codes, i.e. a predetermined number of check bits joined to a predetermined number of information bits
- H03M13/11—Error detection or forward error correction by redundancy in data representation, i.e. code words containing more digits than the source words using block codes, i.e. a predetermined number of check bits joined to a predetermined number of information bits using multiple parity bits
- H03M13/1102—Codes on graphs and decoding on graphs, e.g. low-density parity check [LDPC] codes
- H03M13/1105—Decoding
- H03M13/1111—Soft-decision decoding, e.g. by means of message passing or belief propagation algorithms
- H03M13/1125—Soft-decision decoding, e.g. by means of message passing or belief propagation algorithms using different domains for check node and bit node processing, wherein the different domains include probabilities, likelihood ratios, likelihood differences, log-likelihood ratios or log-likelihood difference pairs
Abstract
Provide LDPC interpretation methods and device based on mutual information.The disclosed method for obtaining the processing procedure combination for LDPC decodings, wherein the processing procedure combines the Soft Inform ation for obtaining error correction soft decoding, including：Using multiple processing procedures multi-group data is obtained from NVM chips；The mutual information for calculating the data stored in every group of reading data and NVM chips, obtains multiple mutual informations；Find the maximum mutual information with maximum value in the multiple mutual information；And record obtains the combination of the processing procedure of the maximum mutual information.
Description
Technical field
This application involves the decoding sides low density parity check code (Low Density Parity Check Code, LDPC)
Method and device, more specifically, this application involves the Soft Inform ations obtained based on mutual information for LDPC soft decodings.
Background technology
LDPC code is a kind of error-correcting code, and compared with other error corrections, under identical code check, LDPC code, which has, entangles
The characteristics of wrong ability is strong, fast convergence rate, therefore be SSD (solid state drive, Solid State Drive) controller at this stage
The best error-correcting code in field.
LDPC code is a kind of binary packet code, using supersparsity matrix as check matrix.Often row is (every in check matrix
Row) in nonzero element it is rare.For any legal code word V, the product of code word V and check matrix H is zero.
Common LDPC decoding algorithms include：Belief propagation algorithm (Belief Propagation, BP), minimum-sum algorithm
(Min-Sum) and bit reversal algorithm (Bit Flip) etc..And minimum-sum algorithm and based on the innovatory algorithm of minimum-sum algorithm by
It is simple in its hardware realization, and be widely adopted in SSD controller field with error-correcting performance good enough.
LDPC decodings include hard decoding and soft decoding, and decoder utilizes Soft Inform ation (also referred to as soft bit degree in soft decoding
Amount) it is iterated decoding.
To obtain Soft Inform ation, by using multiple and different judgement reading data of the voltage from storage unit, using multiple
The result for reading data obtains the Soft Inform ation of such as log-likelihood ratio (LLR, Logarithmic Likelihood Ratio).
In some NVM chips, it is possible to specify the judgement voltage used in read operation.Some flash chips provide " reading is reformed "
(Read-Retry) technology or order.By indicating different parameters for read operation, come used threshold value when selecting read operation
Voltage.
Invention content
To obtain Soft Inform ation, need to read data from storage unit using different judgement voltage.Which increases readings
The number of data data throughout, occupies flash memory bandwidth chahnel, increases the time for obtaining data and consume more work(
Rate.It needs to reduce the number for reading data from storage unit, and obtains optimal or preferably Soft Inform ation.Thus need to identify energy
It enough obtains the combination of the data playback mode of optimal or more excellent Soft Inform ation or adjudicates the combination of voltage.
The application is based on following basic ideas：
1, there is N number of processing procedure combination Ti；
2, stochastic variable X obtain stochastic variable Y by the combination Ti processing of some processing procedure, calculate stochastic variable X at this time
With the mutual information Mi of stochastic variable Y；
3, find maximum mutual information Mi and corresponding processing procedure combination Ti；
4, it is believed that Ti is most beneficial for the processing procedure combination for restoring stochastic variable X from stochastic variable Y.
According to the application in a first aspect, providing the side for obtaining processing procedure combination according to the first of the application first aspect
Method, wherein the processing procedure combines the Soft Inform ation for obtaining error correction soft decoding, including：Using multiple processing procedures from
NVM chips obtain multi-group data；The mutual information for calculating the data stored in every group of data and NVM chips, obtains multiple mutual informations；
Find the maximum mutual information with maximum value in the multiple mutual information；And record obtains the processed of the maximum mutual information
The combination of journey.
The method for obtaining processing procedure combination according to the first of the application first aspect, provides according to the application first party
The method that the second of face obtains processing procedure combination, wherein every group of data of multi-group data include one or more parts data, every part
Data are the data read from NVM chips by one of the multiple processing procedure.
According to the method that the first or second of the application first aspect obtains processing procedure combination, provide according to the application
The method that the third of first aspect obtains processing procedure combination, every group of data of wherein multi-group data include one or more parts number
According to every part of data are to carry out error correction to the data read from NVM chips by one of the multiple processing procedure to decode
The data arrived.
The method for obtaining processing procedure combination according to the first of the application first aspect, provides according to the application first party
The method that the 4th of face obtains processing procedure combination, every group of data of wherein multi-group data include one or more parts data, every part
Data are the data read from NVM chips by the read command with different parameters.
According to the method that the first of the application first aspect or the 4th obtain processing procedure combination, provide according to the application
The method that the 5th of first aspect obtains processing procedure combination, every group of data of wherein multi-group data include one or more parts number
According to every part of data are to carry out error correction to the data read from NVM chips by the read command with different parameters to decode
The data arrived.
The method for obtaining processing procedure combination according to the first of the application first aspect, provides according to the application first party
The method that the 6th of face obtains processing procedure combination, every group of data of wherein multi-group data include one or more parts data, every part
Data are the data read from NVM chips by the read command with different parameters；And the multi-group data includes by common
A data that read command is read from NVM chips.
According to the method that the first of the application first aspect or the 6th obtain processing procedure combination, provide according to the application
The method that the 7th of first aspect obtains processing procedure combination, every group of data of wherein multi-group data include one or more parts number
According to every part of data are to carry out error correction to the data read from NVM chips by the read command with different parameters to decode
The data arrived；And the multi-group data includes carrying out error correction to the data read from NVM chips by common read command
Decode obtained a data.
Any one for the method for obtaining processing procedure combination according to the first to the 7th of the application first aspect, provides root
According to the 8th method for obtaining processing procedure combination of the application first aspect, wherein every group of data of the multi-group data include
The data of specified number.
Any one for the method for obtaining processing procedure combination according to the first to the 8th of the application first aspect, provides root
According to the 9th method for obtaining processing procedure combination of the application first aspect, wherein calculating in every group of reading data and NVM chips
Storage data mutual information include：For first group of reading data in every group of reading data, statistics reads the every of data
The joint probability of a symbol and each symbol of storage data in NVM chips；Calculate the random change for representing the symbol for reading data
Measure the comentropy of Y；Calculate the comentropy for the stochastic variable X for representing the symbol that data are stored in NVM chips；Calculate stochastic variable X
With the united information entropy of stochastic variable Y；Calculate the comentropy of stochastic variable X and the comentropy of stochastic variable Y and, with random change
The difference for measuring the united information entropy of X and stochastic variable Y obtains first group of reading data and stores the mutual information of data in NVM chips.
Any one for the method for obtaining processing procedure combination according to the 9th of the application first aspect the, provides according to this Shen
Please first aspect the tenth method for obtaining processing procedure combination, wherein storing number in calculating every group of reading data and NVM chips
According to mutual information further include：It is general according to each symbol and combining for each symbol for storing data in NVM chips for reading data
Rate, calculate read data each symbol marginal probability and NVM chips in storage data each symbol edge it is general
Rate；According to the marginal probability for each symbol for reading data, the information for the stochastic variable Y for representing the symbol for reading data is calculated
Entropy；And the marginal probability according to each symbol for storing data in NVM chips, it calculates and represents storage data in NVM chips
The comentropy of the stochastic variable X of symbol.
Any one for the method for obtaining processing procedure combination according to the first to the tenth of the application first aspect, provides root
According to the 11st method for obtaining processing procedure combination of the application first aspect, wherein the data stored in the NVM chips, are
The data stored in the Physical Page of the specified type of NVM chips.
Any one for the method for obtaining processing procedure combination according to the first to the 11st of the application first aspect, provides
According to the 12nd of the application first aspect the obtain processing procedure combination method, recorded in processing procedure combination not
Including the use of the processing procedure of common read command.
According to the application second aspect, the first interpretation method according to the application second aspect is provided, including：Using institute
The combination of the processing procedure of record obtains first group of data from NVM chips；It is supplied to decoder to carry out first group of data
Soft decoding；Any one of the method that the combination of the wherein described processing procedure is provided according to the application first aspect obtains.
According to the third aspect of the application, the first interpretation method according to the application third aspect is provided, including：It utilizes
Common read command reads the first data from NVM chips；If to the error correction decoding failure of the first data, utilizes and recorded
The combination of processing procedure obtains first group of data from NVM chips；It is supplied to decoder to carry out soft decoding first group of data；
Any one of the method that the combination of the wherein described processing procedure is provided according to the application first aspect obtains.
According to the fourth aspect of the application, the first interpretation method according to the application fourth aspect is provided, including：It utilizes
Common read command reads the first data from NVM chips；If to the error correction decoding failure of the first data, utilizes and recorded
The combination of processing procedure obtains first group of data from NVM chips；First data are supplied to first group of data and are translated
Code device carries out soft decoding；Any one of the method that the combination of the wherein described processing procedure is provided according to the application first aspect
It obtains.
According to the first interpretation method of the fourth aspect of the application, provide according to the second of the application fourth aspect the decoding
Method, wherein the data that first group of data are not read including the use of common read command from NVM chips.
Interpretation method is provided according to the second aspect, the third aspect, fourth aspect of the application, is provided according to the application the
The third interpretation method of four aspects, including：The Soft Inform ation of data is stored from first group of data acquisition NVM chips, and right
The Soft Inform ation carries out soft decoding.
Interpretation method is provided according to the second aspect, the third aspect, fourth aspect of the application, is provided according to the application the
4th interpretation method of four aspects, wherein the decoder is ldpc decoder, Turbo code decoder or polarization code decoder.
According to the 5th of the application the aspect, provides and obtain processing procedure combination according to the first of the 5th aspect of the application
Device, wherein the processing procedure combines the Soft Inform ation for obtaining error correction soft decoding, described device includes：Data acquisition
Module, for obtaining multi-group data from NVM chips using multiple processing procedures；Mutual information computing module, for calculating every group of reading
The mutual information for going out the data stored in data and NVM chips, obtains multiple mutual informations；Maximum mutual information acquisition module, for looking for
To the maximum mutual information with maximum value in the multiple mutual information；And logging modle, obtain the maximum mutually for recording
The combination of the processing procedure of information.
According to the 6th of the application the aspect, the first code translator according to the 6th aspect of the application is provided, including：Data
Acquisition module, for utilizing the combination of recorded processing procedure to obtain first group of data from NVM chips；Soft decoding module is used
In by first group of data be supplied to decoder carry out soft decoding；The combination of the wherein described processing procedure is according to the application the 5th
The device of first acquisition processing procedure combination of aspect obtains.
According to the 7th of the application the aspect, the first code translator according to the 7th aspect of the application is provided, including：First
Data acquisition module, for reading the first data from NVM chips using common read command；Second data acquisition module, is used for
If to the error correction decoding failure of the first data, the combination of recorded processing procedure is utilized to obtain first group from NV M chips
Data；Soft decoding module, for being supplied to decoder to carry out soft decoding first group of data；The wherein described processing procedure
Combination is obtained according to the device of the first acquisition processing procedure combination of the 5th aspect of the application.
According to the eighth aspect of the application, the first code translator according to the application eighth aspect is provided, including：First
Data acquisition module, for reading the first data from NVM chips using common read command；Second data acquisition module, is used for
If to the error correction decoding failure of the first data, the combination of recorded processing procedure is utilized to obtain first group from NV M chips
Data；Soft decoding module, for being supplied to decoder to carry out soft decoding first data and first group of data；Wherein
The combination of the processing procedure is obtained according to the device of the first acquisition processing procedure combination of the 5th aspect of the application.
According to the 9th of the application the aspect, the first solid storage device according to the 9th aspect of the application is provided, including,
Controller, decoder and nonvolatile storage；In response to the read command from information processing equipment, the controller is for utilizing
The combination of the processing procedure recorded obtains first group of data from nonvolatile storage；And the controller is used for described the
One group of data is supplied to the decoder to carry out soft decoding.
According to the tenth of the application the aspect, the first solid storage device according to the tenth aspect of the application is provided, including,
Controller, decoder and nonvolatile storage；In response to the read command from information processing equipment, the controller is for utilizing
Common read command reads the first data from nonvolatile storage；In response to decoder instruction to the wrong school of the first data
Positive decoding failure, the controller are used to utilize the combination of recorded processing procedure to obtain first group of data from NVM chips；With
And the controller is used to be supplied to the decoder to carry out soft decoding first group of data.
Description of the drawings
When being read together with attached drawing, by reference to the detailed description of illustrative embodiment, will be best understood below
The present invention and preferred use pattern and its further objects and advantages, wherein attached drawing include：
Fig. 1 is the schematic diagram according to the system of the embodiment of the present application；
Fig. 2 is the flow chart combined according to the acquisition preferred process process of the embodiment of the present application；
Fig. 3 is the schematic diagram according to the system of the another embodiment of the application；And
Fig. 4 is the flow chart according to the decoding process of the embodiment of the present application.
Specific implementation mode
Fig. 1 is the schematic diagram according to the system of the embodiment of the present application.
Data x is written in NVM chips 110.By one or more processing procedures, quilt is read from NVM chips 110
The data of write-in.In Fig. 1, as an example, three kinds of read procedures (read procedure 141, read procedure 142 and read procedure 143) are shown,
And it is denoted as y1 by the data that read procedure 141 is read, the data that read procedure 142 is read are denoted as y2, the number that read procedure 143 is read
According to being denoted as y3.In the example in fig 1, the difference between each read procedure is to have used different judgement voltage, or uses
Order is reformed in reading with different parameters.The data (y1, y2, y3) read by three kinds of read procedures are provided to ECC (mistakes
Correcting code) decoding unit 120, carry out error correction decoding for generating Soft Inform ation, and based on Soft Inform ation.It is (low close including LDPC
Spend parity check code), Turbo code, a variety of error corrections codings such as Polar (polarization) code can be using Soft Inform ation into row decoding.This
The embodiment of application can be applied to these and be encoded using the error correction that Soft Inform ation decodes.Although and the embodiment in Fig. 1
In, obtain 3 kinds of reading data using 3 kinds of read procedures, and for obtaining Soft Inform ation, it is possible to understand that ground, can use a kind, 2 kinds, 4
Kind and the read procedure of other quantity obtain Soft Inform ation.
In Fig. 1, three kinds of read procedures for reading data (y1, y1 and y1) constitute the combination of a read procedure.And
Also calculate the mutual information of the data and the data obtained by the combined treatment of read procedure that are written into NVM chips 110.
It is written into the data x of NVM chips 110 according to known to, and is combined and is read by read procedure from NVM chips 110
The data (for example, y1, y2 and y3) gone out calculate mutual information (MI, Mutual- in mutual information computing module 130
Information).And calculate the mutual information of the combination of other read procedures.And compare by the multiple combinations acquisition of read procedure
Multiple mutual informations.The maximum read procedure combination of selection mutual information is used for obtaining Soft Inform ation, to improve error correction decoding at
Power.
Fig. 2 is the flow chart according to the processing procedure of the embodiment of the present application.One purpose of embodiment shown in Fig. 2 is
Obtain the combination of the processing procedure with maximum mutual information.In step 210, obtained from NVM chips using a variety of processing procedures
Multi-group data.To read data instance from NVM chips, there are many processing procedures for reading data.As an example, parameter
For r reading reform be from NVM chips read data one of processing procedure.Reading the parameter r that reforms has such as 3 kinds different to take
It is worth (such as r1, r2, r3), then illustrate 7 kinds of the processing procedure for reading data combinations in table 1, each combination can be by
The used combination of parameter value for reading to reform indicates.Also, in each combination of processing procedure, there can be different parameters
And/or the processing procedure of different number.By every group of combination of processing procedure, multi-group data has been read from NVM chips.Referring to
Table 1, the data read with y instructions, the subscript of parameter r that subscript and the reading of y are reformed are corresponding.Under be designated as the reading data yi of i
It is to reform reading by parameter for the reading with same index i.
In step 220, calculates every group of reading data and store the mutual information of data in NVM chips, obtain multiple mutual informations.
As an example, for one group of reading data y1, the mutual information of the group and storage data x in NVM chips that include data y1 is calculated.
In another example, for one group of reading data y1 and y2, calculating includes that the group of reading data y1 and y2 are stored with NVM chips
The mutual information of data x.
Hereinafter, example by way of example is described to the computational methods of mutual information.
To obtain data x, the data being written into NVM chips can be known data x.As another example, i.e.,
Keep data x unknown, after NVM chips 110 read data y1, y2 and/or y3, using ECC components 120 (referring to Fig. 1) to reading
Data y1, y2 and/or y3 carry out error correction decoding, to obtain the data x being written into NVM chips.
In step 230, to being based on every group of calculated mutual information of reading data, the value found in multiple mutual informations is maximum
Mutual information (maximum mutual information).And in step 240, determine the processing procedure combination for obtaining maximum mutual information (for example, parameter point
Not Wei r1, the combination that the reading of r2, r3 is reformed).In general, including the combination of more processing procedures, the mutual of bigger will be obtained
Information.In an embodiment according to the present invention, for the combination of processing procedure, by the quantity point of processing procedure wherein included
Class, and to each quantity classification, determine that (for example, in table 1, processing procedure combines the processing procedure combination with maximum mutual information
There are 3 classes, respectively include 1 processing procedure, 2 processing procedures and 3 processing procedures).
In step 240, the processing procedure combination that can obtain maximum mutual information is obtained.Next, needing to utilize soft letter
When breath carries out soft decoding, combined come from NVM chips using the processing procedure with maximum mutual information obtained in step 240
One group of reading data is obtained, is advantageous.It is processed with maximum mutual information using what is obtained in step 240 in step 250
Cheng Zuhe (for example, parameter is respectively r1, combination that the reading of r2, r3 is reformed) obtains one group of reading data from NVM chips.With
And in step 260, obtained in step 250 one group of reading data is supplied to ldpc decoder, is gone forward side by side for obtaining Soft Inform ation
Row soft decoding.It is to be appreciated that step 250 applies also for the energy profit such as Turbo code, Polar (polarization) code with step 260
Process is decoded with the error correction of Soft Inform ation.
Table 1
| Processing procedure group | Read data |
| r1 | y1 |
| r2 | y2 |
| r3 | y3 |
| r1,r2 | y1,y2 |
| r1,r3 | y1,y3 |
| r2,r3 | y2,y3 |
| r1,r2,r3 | y1,y2,y3 |
According to an embodiment of the present application, the implementation of step 260 is arrived in the implementation of step 210 to step 240 with step 250
It can occur at different times and/or place.For example, to step 240, in product, (solid state disk, controls NVM chips step 210
Device chip, ECC decoders etc.) the research/development stage, and step 250 is carried out to step 260 when using the product by the user.
The obtained processing procedure combination with maximum mutual information of step 240, may be recorded in the NVM chips of solid storage device
In.Or make the control module of solid storage device in soft decoding to be implemented by such as program code or circuit, using step
The rapid 240 obtained processing procedure combinations with maximum mutual information.
As another embodiment, step 210 to step 260 occurs in the neighbouring time.To solid storage device
It is combined to obtain the processing procedure with maximum mutual information to step 240 by executing step 210, and then needed at runtime
Acquired processing procedure is used to combine with step 260 by executing step 250 when carrying out soft decoding.
The acquisition process of mutual information is described below.According to an embodiment of the present application, pass through a variety of processing procedure T processing
Stochastic variable X, the result of processing calculate according to formula (1) and (2) the mutual information I (X, Y) of X and Y as stochastic variable Y.
H (X)=- ∑x∈Xp(x)log2p(x)(1)
I (X, Y)=H (X)+H (Y)-H (X, Y) (2)
Stochastic variable X indicates the data in initial data, such as write-in NVM.
Stochastic variable Y indicates to pass through the data that processing procedure T is obtained to stochastic variable X, such as the number read from NVM chips
According to.
As an example, X represent write-in NVM chips initial data, and processing procedure T be parameter be r reading reform, Y generations
The data that table is read from NVM chips.
H (X) is the comentropy of stochastic variable X, and (formula (1)) is calculated by the marginal probability of X；H (Y) is stochastic variable Y
Comentropy calculates (formula (1)) by the marginal probability of Y, and H (X, Y) is the united information entropy of stochastic variable X and Y, and H (X, Y)=-
∑(x, y) ∈ (X, Y)P (x, y) log2P (x, y).
X is a possible values (also referred to as symbol) of stochastic variable X.Y be stochastic variable Y a possible values (
Referred to as symbol).P (x) is the marginal probability of x, and x ∈ X, p (y) are the marginal probability of y, y ∈ Y.P (x, y) is x and y while occurring
Probability (joint probability of x and y).As an example, it if X represents the initial data of write-in NVM chips, is stored in NVM chips
Data possibility value be " 0 " and " 1 ", and from NVM chips read data possibility value be " 0 " and " 1 ".To
{ " 0 ", " 1 " } be x/y value set.
Table 2
In table 2, x0 and x1 is two possible values of stochastic variable X, and y0, y1 ..., yn are respectively by multiple processing
One of process T handles an obtained value of value of stochastic variable X.And p (x0, y0) indicates that x0 and y0 occurs simultaneously in table 2
Probability (joint probability of x0 and y0).By the statistics to stochastic variable X and Y, construct table 2, and according to formula (1) with
(2) the mutual information I (X, Y) of X and Y is calculated.
Table 3 illustrates in one example to the statistics of stochastic variable X and Y.The value x of stochastic variable X is represented in NVM cores
1 bit information of piece storage, the value y of careless variable Y are represented according to specific processing procedure (for example, reading weight of the parameter as 0
Do) the processing obtained results of x.The information of NVM chips storage is reformed with the reading that parameter is 0 and is repeatedly read, with structure
Table 3.By many experiments, obtains when storing data " 0 " in NVM chips (x values are 0), read the number that data are " 0 "
Have 999 times, when storing data " 0 " in NVM chips, reading the number that data are " 1 " has 1 time, is stored when in NVM chips
When data " 1 " (x values are 1), reading the number that data are " 0 " has 2 times, when storing data " 1 " in NVM chips, reads
Data, which are the number of " 1 ", to be had 998 times.And it is that 2000 (999+1+998+2) are secondary to amount to experiment number.
Table 3
| y\x | 0 | 1 |
| 0 | 999 | 2 |
| 1 | 1 | 998 |
Experimental result from table 3, the Probability p (x, y) (joint probability of x and y) that x and y can be obtained while being occurred, p
(0,0)=999/2000, p (0,1)=2/2000, p (1,0)=1/2000, p (1,1)=998/2000.
Next, calculating p (x) and H (X)：
P (0)=p (0,0)+p (0,1),
P (1)=p (1,0)+p (1,1), and
H (X)=- p (0) log2p(0)-p(1)log2p(1)。
Next, calculating p (y) and H (Y) and H (X, Y)：
P (0)=p (0,0)+p (1,0),
P (1)=p (0,1)+p (1,1),
H (Y)=- p (0) log2p(0)-p(1)log2P (1), and
H (X, Y)=- p (0,0) log2p(0,0)-p(0,1)log2p(0,1)-p(1,0)log2p(1,0)-p(1,1)
log2p(1,1).And can calculate with parameter be 0 reading recast handled when, mutual information I (X, Y)=H (X) of X and Y
+H(Y)-H(X,Y)。
Each reading is reformed as processing procedure, the data x stored in NVM chips is read out, obtains reading data
Y, and calculate in the case where reading to reform processing procedure Ti, the mutual information I of X and YTi(X,Y)。
Statistical experiment needed for structure table 3 can in the lab carry out before solid storage device is delivered for use,
It can be carried out in the use of solid storage device.Optionally, table 3 is being obtained, after calculating mutual information I (X, Y), can continued
X and Y are counted, to update table 3 and corresponding mutual information I (X, Y).
It is to be appreciated that for structure table 3, and mutual information I (X, Y) is obtained, it needs, with regard to processing procedure Ti, to obtain (x, y)
The statistical information (joint probability of x and y) of be possible to value.
In conjunction with table 4, the mutual information acquisition process according to the another example of the application is described.Table 4 illustrates in one example
To the statistics of stochastic variable X and Y.The value x of stochastic variable X represents 1 bit information stored in NVM chips, careless variable Y
Value y represent according to parameter as 0 readings reform with parameter as 1 reading reform twice processing x obtained results (respectively by r0 and
R1 is indicated).The reading for being 1 with parameter is reformed to the information of NVM chips storage with the reading that parameter is 0 repeatedly to be read again, with structure
Build table 4.By many experiments, obtain, when storing data " 0 " (x values are 0) in NVM chips, reading being reformed by reading twice
Going out the number that data are " 00 " has 999 times, when storing data " 0 " in NVM chips, is by reading to reform reading data twice
The number of " 01 " has 1 time, reads the number that data are " 10 " by reading to reform twice and has a times, reading number is reformed by reading twice
Have b times according to the number for " 11 ".It is also shown in table 4 when storing data " 1 " in NVM chips, reading is reformed by reading twice
The number of the different y values gone out.And the sum that total metering number is numerical part items in table 3.
And from the statistical data in table 4, capable of calculating the Probability p (x, y) that x and y occurs simultaneously, (x combines gai with y's
Rate).And the marginal probability p (x) of x is obtained, and the comentropy H (X) of stochastic variable X, the marginal probability p (y) of y, the letter of stochastic variable Y
Entropy H (Y) is ceased, and the processing procedure for obtaining reforming reading data twice with the reading that the reading that parameter is 0 is reformed with parameter is 1 combines Tr0r1
Under, the mutual information I of X and YTr0r1(X,Y)。
The place that reading data twice are reformed in the reading for being 2 with parameter is reformed for other processing procedures, such as with the reading that parameter is 1
Reason process combines Tr1r2, the reading for being 0 with parameter is reformed with parameter is 2 reading is reformed reads the processing procedures of data and combines T twice01r2,
Processing procedure (combination) T of single reading evidence is reformed in the reading for being 1 with parameterr1Deng each processing procedure group can be calculated separately out
Under conjunction, the mutual information of X and Y.
Table 4
Mutual information embodies the correlation of stochastic variable X and stochastic variable Y.Mutual information is higher, represents correlations of the Y with X
It is stronger.It selects in multiple processing procedures, obtains the processing procedure combination of maximum mutual information, used as when obtaining Soft Inform ation
Processing procedure combination.More information can be obtained from processing procedure combination, and improve the success rate of soft decoding.
Optionally, it is selected in multiple possible processing procedures combinations, and abandons that complexity is excessively high, cost is excessive
Processing procedure combines.Such as the processing procedure that data are read from NVM chips is combined, when carrying out soft decoding, will use
The number for reading to reform reading data is limited to small be equal to 3 times.To in the case where reading to have reformed m kind parameters, optionally handle
Process combination has(from the reading of m kind parameters is reformed, selecting arbitrary 3 kinds of readings data)It (is reformed from the reading of m kind parameters
In, select arbitrary 2 kinds of readings data)(from the reading of m kind parameters is reformed, selecting arbitrary a kind of reading data) kind.Another
In a example, the specified processing procedure combination selected includes that the mistake of (common read command) reading data is reformed with the reading that parameter is 0
Journey (further including optionally, that the process for reading data is reformed with the reading of other parameters value), to be utilized when carrying out soft decoding
Data acquired in the hard decoding stage.
In the embodiment according to the application Fig. 3, processing procedure includes reading data from NVM chips, and to reading
Data carry out error correction decoding.
In the fig. 3 embodiment, data x is written in NVM chips 310.By one or more processing procedures, from
The data being written into are read in NVM chips 310.In figure 3, as an example, show that three kinds of read procedures (read by read procedure 341
Journey 342 and read procedure 343), and it is denoted as y4 by the data that read procedure 341 is read, the data that read procedure 342 is read are denoted as
Y5, the data that read procedure 343 is read are denoted as y6.In the example in figure 3, the difference between each read procedure is to have used difference
Judgement voltage, or the reading with different parameters has been used to reform order.
The data (y4, y5, y6) read by three kinds of read procedures are supplied to ECC (error-correcting code) decoding unit
320, it is decoded by error correction, respectively obtains output data y4 ', y5 ' and y6 '.It should be understood that carry out ECC decodings, carry
The input for supplying ECC components 320 is (including multiple bits) code word, and not 1 bit information.And in the reality according to the application
It applies in example, for purposes of clarity, data x, data y are comprising such as 1 bit information in the codeword.By from NVM cores
Piece 310 reads code word c, by ECC module 320 to code word c into row decoding, and obtains each bit (x) of code word c after decoding
As a result (y).According to the application embodiment, ECC module 320 may fail to the decoding of code word c, at this time ECC module
320 can still provide the decoding result c ' of code word c.Although decoding in result c ', there are mistakes, and each bit therein is still
It is the handling result (y) obtained according to the application optional embodiment.
Foundation is written into the data x of NVM chips 310, and reads and pass through by different read procedures from NVM chips 310
ECC module 320 decodes obtained data (y4 ', y5 ' and y6 '), is calculated in mutual information computing module 330 and is written into NVM chips
310 data and mutual information (MI, the Mutual- that obtained data are read and decoded through ECC module 320 from NVM chips 310
Information)。
In figure 3, three kinds of read procedures for reading data (y4, y5 and y6) constitute the combination of a read procedure.And
The data for being written into NVM chips 310 are also calculated with the combination by other read procedures (different from the group of three kinds of read procedures in Fig. 3
Close) the obtained mutual information of data of processing.And compare by multiple mutual informations of the multiple combinations acquisition of read procedure.Select mutual information
Maximum read procedure combines to be used to obtain Soft Inform ation, to improve the success rate of error correction decoding.
The data (for example, y1, y2, y3) read by three kinds of read procedures are provided to ECC (error correction coding) component
120, it is additionally operable to generate Soft Inform ation, and error correction decoding is carried out based on Soft Inform ation.
Although and in the fig. 3 embodiment, obtain 3 kinds of reading data using 3 kinds of read procedures, and for obtaining mutual information
With Soft Inform ation, it is possible to understand that ground can obtain soft letter using the combination of a kind, 2 kinds, 4 kinds and the read procedure of other quantity
Breath.
Table 5
| Y=r0r1r2 x | 0 | 1 |
| 000 | 999 | 2 |
| 001 | 1 | 998 |
| 010 | C(0,010) | C(1,010) |
| 100 | C(0,100) | C(1,100) |
| 101 | C(0,101) | C(1,101) |
| 110 | C(0,110) | C(1,110) |
| 111 | C(0,111) | C(1,111) |
In conjunction with table 5, the mutual information acquisition process according to the another example of the application is described.Table 5 illustrates in one example
To the statistics of stochastic variable X and Y.The value x of stochastic variable X represents 1 bit information stored in NVM chips, careless variable Y
Value y represent according to parameter as 0 readings reform, parameter as 1 reading weight and parameter as 2 reading reform three times processing x obtained by
The result (respectively by r0, r1 and r2 indicate) that is decoded again through ECC respectively of result.To the information ginseng of NVM chips storage
The reading that number is 0 is reformed, the reading weight that parameter is 1 and the reading that parameter is 2 are reformed and repeatedly read, and is decoded by ECC, with structure
Build table 5.In table 5, table entries C (0,010) is indicated in many experiments, when storing (1 bit) data " 0 " (x in NVM chips
When value is 0), data are read to x and the number for decoding to obtain " 010 " through ECC has C (0,010) secondary by reading to reform three times.With
And the sum that the experiment number amounted to is numerical part items in table 5.
And from the statistical data in table 5, the Probability p (x, y) that x and y occurs simultaneously can be calculated.And obtain the edge of x
Probability p (x), the comentropy H (X) of stochastic variable X, the comentropy H (Y) of the marginal probability p (y) of y, stochastic variable Y, and obtain with
The place read data three times and decoded respectively through ECC is reformed in the reading that it is 2 with parameter that the reading that the reading that parameter is 0 is reformed, parameter is 1, which is reformed,
Reason process combines Tr0r1r2Under, the mutual information I of X and YTr0r1r2(X,Y)。
Other processing procedures are combined, such as the reading for being 2 with parameter reformed with the reading that parameter is 1 reforms and read data twice
And the processing procedure respectively through ECC decodings combines Tr1r2, the reading for being 0 with parameter is reformed the reading for being 2 with parameter and reformed to be read twice
According to and respectively through ECC decoding processing procedure combine T01r2, readings for being 1 with parameter reform single reading evidence and decode through ECC
Processing procedure combines Tr1Deng can calculate separately out under the combination of each processing procedure, the mutual information of X and Y.
In the another embodiment according to the application, under different scenes, obtains and handle NVM by the multiple combinations of processing procedure
The mutual information of the result that data in chip obtain and the data stored in NVM chips.For example, for MLC (Multi-Level
Cell) the NVM chips or Physical Page of type distinguishes two bits in each storage unit (Cell) (for example, MSB (Most
Significant Bit, most significant bit) and LSB (Least Significant Bit, least significant bit)).It is mutual obtaining
When information, stochastic variable X only represents the data of data or LSB from MSB, and it both includes the data packet from MSB to avoid
Include the data from LSB.Similarly, it for the NVM chips or Physical Page of TLC (Triple-Level Cell) type, is obtaining
When mutual information, 3 bits in each storage unit (Cell) are distinguished.
According to an embodiment of the present application, identification has obtained the combination of the processing procedure of maximum mutual information, such as handles
Process combines T.Alternatively, in the set of the combination of processing procedure, identification has obtained the processing procedure combination of maximum mutual information, example
As processing procedure combines Ts.Record process combination T or processing procedure combine Ts in solid storage device.
According to the another embodiment of the application, issued admittedly in using solid storage device, such as in response to receiving host
The read command of state storage device, solid storage device handle read command.
Fig. 4 is the flow chart of solid storage device processing I/O command according to the ... of the embodiment of the present invention.
To handle read command, the reading that the solid-state storage application parameter first according to the embodiment of the present application is 0 is reformed (also referred to as
" common read command ") from the specified address read-outing data (410) of NVM chips.ECC decodings are carried out to the data of reading.If ECC is hard
Successfully decoded (420) obtain correctly reading data for responding read command.If the hard decoding failures of ECC (420), using being remembered
The processing procedure combination T or processing procedure combination Ts with maximum mutual information of record, obtain data (430) from NVM.As act
Example, processing procedure combination T or processing procedure combination Ts are that the reading for being for example 1 with parameter is reformed the reading for being 2 with parameter and reformed twice
Data are read from NVM.In another example, processing procedure combination Ts be the reading for being 0 with parameter reform, the reading that parameter is 1
The reading for being 2 with parameter is reformed to reform and read data from NVM chips three times.Due to the common read command used in step 410
(that is, parameter be 0 reading reform), in step 430, the reading for being only 1 with parameter reform the reading for being 2 with parameter reform twice from
Data are read in NVM chips, and reuse the data read in step 410 from NVM chips using common read command.And
The data read in step 410 are combined into the data read twice in step 430, Ts is combined from NVM cores by processing procedure
The data read three times in piece, to save the process for once reading data from NVM chips.And advantageously, obtaining
Processing procedure combines in Ts, and it includes being reformed from NVM chips to read from data with the reading that parameter is 0 to make processing procedure combination Ts
Reason process, to which during decoding, the data acquired in step 410 can be utilized.
ECC is carried out next, the data that T or processing procedure combination Ts are read will be closed with processing process group and give ECC module
Soft decoding.
According to an embodiment of the present application, due to obtaining processing procedure combination T or processing with maximum mutual information
Process combines Ts, has in soft decoding with the obtained data of processing process group conjunction T or processing procedure combination Ts higher or highest
Success rate.Thus in the hard decoding failures of ECC (420), no longer trial is reformed with the reading with parameters and read from NVM chips
Data, but data are directly read from NVM chips using processing procedure combination T or processing procedure combination Ts, and carry out soft translate
Code.To eliminate the process that the reading that the data searched for and there is the reading of parameters to reform middle reading can be properly decoded is reformed,
The error handle time after the hard decoding failures of ECC (420) is reduced, the performance of solid storage device processing read command is improved,
Also reduce power consumption.
In the another embodiment according to the application, identifying that will to encounter higher decoding failure using common read command several
In the case of rate, step 410 and step 420 are saved, and in response to the read command from host or user, directly using processed
Cheng Zuhe T or processing procedure combine Ts and obtain data from NVM chips, and carry out soft decoding to the data of acquisition.Further to drop
Reduction process delay and raising efficiency.
To carry out ECC soft decodings, by by processing procedure combine T or processing procedure combination Ts to data give ECC and translate
Code component (for example, ECC decoding units 320 of Fig. 3), its log-likelihood ratio (LLR, Log are calculated according to input data
Likelihood Ratio).LLR table is provided in ECC decoding units.It uses input data as index, inquires and obtain in LLR table
Log-likelihood ratio corresponding with input data, and as Soft Inform ation and carry out soft decoding according to log-likelihood ratio.
Except the scheme according to the prior art obtains LLR table, according to an embodiment of the present application, additionally provides and obtain LLR
The method of table.As an example, processing procedure combination Ts includes with the reading that parameter is a is reformed, the reading that parameter is b reforms and is with parameter
The reading of c is reformed reads data from NVM chips.For 1 bit stored in NVM chips, read to reform 3 ratios of reading respectively three times
It is special.In table 6, respectively represent 3 bits of reading with " bit 2 ", " bit 1 " and " bit 0 ", value be respectively " RRb ",
" RRa " and " RRc ".
Table 6
| Bit 2 | Bit 1 | Bit 0 |
| RRb | RRa | RRc |
Table 7, which is illustrated, closes the data (x) stored in NVM chips with processing process group in the example according to the application
The statistics for the data (y) that Ts is read.Data item p (x, y) in table 7 be x and y and meanwhile occur probability (x and combining for y are general
Rate).And the LLR table arrived calculated with p (x, y) is also illustrated in table 7.
In table 7, the value of LLR (011) and LLR (100) is 0, and LLR (n)=ln (p (0, n)/p (1, n)).Wherein n be than
The decimal representation of special sequence (RRb, RRa, RRc) or bit sequence (RRb, RRa, RRc).
And optionally, upper limit value and lower limiting value are provided for the value of LLR (n).If calculated ln (p (0, n1)/p (1,
N1)) it is more than upper limit value, then sets LLR (n1) to upper limit value.And if calculated ln (p (0, n2)/p (1, n2)) is less than
Lower limiting value then sets LLR (n2) to lower limiting value.
Table 7
It can be by software, hardware, firmware, FPGA (field programmable gate array, Field Programmable Gate
Array) and/or ASIC (application specific integrated circuit, Application Specific Integrated Circuit) etc. is real
Existing LDPC interpretation methods according to the ... of the embodiment of the present invention.And LDPC interpretation methods according to the ... of the embodiment of the present invention can be applied to base
In the solid storage device of NVM chips, including but not limited to solid state disk, USB flash disk, SD card, mobile phone, tablet electricity can also be applied to
The portable electronic devices such as brain and other a variety of electronics for needing to store information using NVM chips (such as nand flash memory)
Equipment.Interpretation method according to the ... of the embodiment of the present invention applies also for communication, magnetic storage, optical storage etc. and uses error-correcting code
Equipment or service.
Although the example of present invention reference is described, it is intended merely to the purpose explained rather than to the limit of the present invention
System, the change to embodiment, increase and/or deletion can be made without departing from the scope of the present invention.
In field involved by these embodiments, benefiting from the description above with the introduction presented in associated attached drawing
Technical staff will be recognized that record here the present invention it is many change and other embodiment.It should therefore be understood that this hair
It is bright to be not limited to disclosed specific implementation mode, it is intended to will to change and other embodiment includes in the scope of the appended claims
It is interior.Although using specific term herein, them are used only on general significance and describing significance and not is
The purpose of limitation and use.
Claims (10)
1. a kind of method obtaining processing procedure combination, wherein processing procedure combination is for obtaining error correction soft decoding
Soft Inform ation, including：
Using multiple processing procedures multi-group data is obtained from NVM chips；
The mutual information for calculating the data stored in every group of data and NVM chips, obtains multiple mutual informations；
Find the maximum mutual information with maximum value in the multiple mutual information；And
Record obtains the combination of the processing procedure of the maximum mutual information.
2. according to the method for claim 1；
Wherein, every group of data of multi-group data include one or more parts data, every part of data be by the multiple processing procedure it
One data read from NVM chips.
3. according to the method described in one of claim 1-2, every group of data of wherein multi-group data include one or more parts data,
Every part of data are to carry out error correction to the data read from NVM chips by one of the multiple processing procedure to decode
Data.
4. according to the method described in one of claim 1-3, wherein calculating every group of reading data and storage data in NVM chips
Mutual information includes：
For first group of reading data in every group of reading data,
Statistics reads the joint probability of each symbol and each symbol of storage data in NVM chips of data；
Calculate the comentropy for the stochastic variable Y for representing the symbol for reading data；
Calculate the comentropy for the stochastic variable X for representing the symbol that data are stored in NVM chips；
Calculate the united information entropy of stochastic variable X and stochastic variable Y；
Calculate the comentropy of stochastic variable X and the comentropy of stochastic variable Y and, with combining for stochastic variable X and stochastic variable Y
The difference of comentropy obtains first group of reading data and stores the mutual information of data in NVM chips.
5. according to the method described in one of claim 1-4, wherein
The data stored in the NVM chips, the data stored in the Physical Page for being the specified type of NVM chips.
6. a kind of interpretation method, including：
The combination of recorded processing procedure is utilized to obtain first group of data from NVM chips；
It is supplied to decoder to carry out soft decoding first group of data；
The combination of the wherein described processing procedure is obtained according to claim 1-5 any one of them methods.
7. a kind of interpretation method, including：
The first data are read from NVM chips using common read command；
If to the error correction decoding failure of the first data, the combination of recorded processing procedure is utilized to obtain the from NVM chips
One group of data；
It is supplied to decoder to carry out soft decoding first data and first group of data；
The combination of the wherein described processing procedure is obtained according to claim 1-5 any one of them methods.
8. a kind of device obtaining processing procedure combination, wherein processing procedure combination is for obtaining error correction soft decoding
Soft Inform ation, described device include：
Data acquisition module, for obtaining multi-group data from NVM chips using multiple processing procedures；
Mutual information computing module, the mutual information for calculating the data stored in every group of reading data and NVM chips, obtains multiple
Mutual information；
Maximum mutual information acquisition module, for finding the maximum mutual information with maximum value in the multiple mutual information；And
Logging modle, the combination for recording the processing procedure for obtaining the maximum mutual information.
9. a kind of code translator, including：
Data acquisition module, for utilizing the combination of recorded processing procedure to obtain first group of data from NVM chips；
Soft decoding module, for being supplied to decoder to carry out soft decoding first group of data；
The combination of the wherein described processing procedure device according to claim 8 for obtaining processing procedure combination obtains.
10. a kind of solid storage device, including, controller, decoder and nonvolatile storage；
In response to the read command from information processing equipment, the controller be used to utilize the combination of recorded processing procedure from
Nonvolatile storage obtains first group of data；And
The controller is used to be supplied to the decoder to carry out soft decoding first group of data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710030720.6A CN108322225B (en) | 2017-01-16 | 2017-01-16 | LDPC decoding method and device based on mutual information |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710030720.6A CN108322225B (en) | 2017-01-16 | 2017-01-16 | LDPC decoding method and device based on mutual information |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108322225A true CN108322225A (en) | 2018-07-24 |
| CN108322225B CN108322225B (en) | 2023-05-16 |
Family
ID=62891790
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710030720.6A Active CN108322225B (en) | 2017-01-16 | 2017-01-16 | LDPC decoding method and device based on mutual information |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108322225B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102412843A (en) * | 2011-07-28 | 2012-04-11 | 清华大学 | Adaptive normalized minimum sum LDPC (Low Density Parity Check Code) decoding method and decoder |
| US20130094286A1 (en) * | 2011-10-18 | 2013-04-18 | Seagate Technology Llc | Determining optimal read reference and programming voltages for non-volatile memory using mutual information |
| US20130107611A1 (en) * | 2011-10-27 | 2013-05-02 | Agency For Science, Technology And Research | Memory Device with Soft-Decision Decoding |
| CN104464822A (en) * | 2014-11-21 | 2015-03-25 | 湖南大学 | LDPC error correction encoding method based on flash memory error section |
| US20160092301A1 (en) * | 2014-09-28 | 2016-03-31 | Apple Inc. | Correcting soft reliability measures of storage values read from memory cells |
-
2017
- 2017-01-16 CN CN201710030720.6A patent/CN108322225B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102412843A (en) * | 2011-07-28 | 2012-04-11 | 清华大学 | Adaptive normalized minimum sum LDPC (Low Density Parity Check Code) decoding method and decoder |
| US20130094286A1 (en) * | 2011-10-18 | 2013-04-18 | Seagate Technology Llc | Determining optimal read reference and programming voltages for non-volatile memory using mutual information |
| US20130107611A1 (en) * | 2011-10-27 | 2013-05-02 | Agency For Science, Technology And Research | Memory Device with Soft-Decision Decoding |
| US20160092301A1 (en) * | 2014-09-28 | 2016-03-31 | Apple Inc. | Correcting soft reliability measures of storage values read from memory cells |
| CN104464822A (en) * | 2014-11-21 | 2015-03-25 | 湖南大学 | LDPC error correction encoding method based on flash memory error section |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108322225B (en) | 2023-05-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11599773B2 (en) | Neural networks and systems for decoding encoded data | |
| US9583217B2 (en) | Decoding method, memory storage device and memory control circuit unit | |
| TWI353521B (en) | Soft-input soft-output decoder for nonvolatile mem | |
| TWI511146B (en) | Method and apparatus for optimized threshold search in memory cells | |
| US9543983B2 (en) | Decoding method, memory storage device and memory control circuit unit | |
| US20150293811A1 (en) | Decoding method, memory storage device and memory controlling circuit unit | |
| CN106169308A (en) | Memory Controller and operational approach thereof | |
| CN106997777A (en) | VSS LDPC decoders with improved hard decoder handling capacity | |
| CN104572334B (en) | Coding/decoding method, memory storage apparatus and memorizer control circuit unit | |
| US9274891B2 (en) | Decoding method, memory storage device, and memory controlling circuit unit | |
| US11768732B2 (en) | Soft decoding method using LLR conversion table | |
| US20160011934A1 (en) | Decoding method, memory control circuit unit and memory storage device | |
| US8892986B2 (en) | Apparatuses and methods for combining error coding and modulation schemes | |
| US20190132010A1 (en) | Method of operating decoder for reducing computational complexity and method of operating data storage device including the decoder | |
| CN102017425A (en) | System and method for performing concatenated error correction | |
| CN110415753B (en) | Error correction circuit and method for operating the same | |
| US10193569B2 (en) | Decoding method, memory storage device and memory control circuit unit | |
| JP2018160064A (en) | Memory controller, memory system, and control method | |
| WO2023059517A1 (en) | Systems for estimating bit error rate (ber) of encoded data using neural networks | |
| TWI536749B (en) | Decoding method, memory storage device and memory controlling circuit unit | |
| CN105304143B (en) | Coding/decoding method, memorizer control circuit unit and memory storage apparatus | |
| CN108628695A (en) | Mutual information computational methods and device | |
| CN109427401A (en) | Coding method and the memory storage apparatus for using the method | |
| CN103137213A (en) | Storage control device with low density parity check code coding capacity and method | |
| CN103973316B (en) | Coding/decoding method with the variable node renovator using scaling constant conciliates code system |
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 | ||
| CB02 | Change of applicant information |
Address after: 100192 room A302, building B-2, Dongsheng Science Park, Zhongguancun, 66 xixiaokou Road, Haidian District, Beijing Applicant after: Beijing yihengchuangyuan Technology Co.,Ltd. Address before: 100192 Room 302, 3 / F, building B-2, Dongsheng Science Park, 66 xixiaokou Road, Haidian District, Beijing Applicant before: BEIJING MEMBLAZE TECHNOLOGY Co.,Ltd. |
|
| CB02 | Change of applicant information | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |