METHOD AND DEVICE FOR LAYERED DECODING OF A SUCCESSION OF BLOCKS ENCODED WITH AN LDPC CODE

Abstract

The metrics matrix may include at least one particular layer including at least one particular column having several metrics cues, respectively, situated in different rows. For the particular layers the updating of the channel cue is associated with the particular column involving at each iteration one updated metric cue selected from all the metrics cues of the particular column. The row of the selected metric cues may change at each iteration.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

Other advantages and features of the present disclosure will become apparent on examining the detailed description of embodiments and modes of implementation, which are in no way limiting, and the appended drawings, in which:

FIG. 1 schematically illustrates the internal structure of a wireless apparatus incorporating a decoder according to the invention;

FIG. 2 illustrates in greater detail, but still schematically, an embodiment of an LDPC decoder according to the invention;

FIG. 3 schematically illustrates an exemplary embodiment of an input memory of an LDPC decoder according to the prior art;

FIGS. 4 and 5 respectively illustrate exemplary parity and metrics matrices subdivided into layers, according to the invention;

FIGS. 6 and 7 illustrate an exemplary construction of a parity matrix in the case of the DVB-S2 standard according to the invention;

FIGS. 8 and 9 illustrate a mode of implementation of a method according to the invention;

FIG. 10 illustrates in greater detail a hardware architecture of processing means or a processor according to the invention; and

FIGS. 11 and 12 represent comparative plots of the bit error rate and the block error rate, respectively, as a function of signal-to-noise ration between the invention and the prior art, wherein curve C1 represents the prior art and curve C2 represents the invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

Embodiments and modes of implementation will now be described in greater detail in the case of a DVB-S2 application for which the length of the encoded blocks is 64,800 bits, although the embodiments disclosed are valid for all sizes of blocks.

In FIG. 1, the reference WAP denotes a wireless apparatus that may be used in a communication system complying with the DVB-S2 standard. The wireless apparatus WAP comprises a transmission chain TXCH capable of transmitting coded and modulated information over a transmission channel CH, for example, the air. Furthermore, this apparatus WAP also comprises a reception chain RXCH capable of receiving and decoding information originating from the transmission channel CH.

The transmission chain TXCH comprises, in a known manner, source coding means or a source coder SCM that receive application data APP so as, in particular, to compress them in order to reduce the data bit rate. The transmission chain also comprises channel coding means or a channel coder, whose function is, in particular, to add redundancy so as to be able to subsequently correct the potential reception errors due to the noise in the transmission channel. These channel coding means comprise, for example, LDPC coding means or a LDPC coder.

The transmission chain also comprises modulation means or a modulator MDM so as to adapt the signal to the transmission channel (satellite channel or radio channel, for example). The reception chain RXCH comprises similar means for performing the inverse functions. More precisely, there are demodulation means or a demodulator DMDM, followed by channel decoding means of a channel decoder CHDCM that comprise, for example, an LDPC decoder, followed by source decoding means or a source decoder SDCM delivering to the user the user data DUT corresponding to the application data APP.

Referring more particularly now to FIG. 2, it may be seen that an exemplary LDPC decoder, referenced EC, schematically comprises input memory MMCH, decoding means or a decoder MDCD, and hand output memory MMHD. The input memory MMCH receives a succession of blocks BLC_i encoded with an LDPC code. Each block comprises N coded cues, N being equal to 64,800 in the DVB-S2 standard, for example. These cues are probability ratios and are generally known to the person skilled in the art as Log Likelihood Ratio (LLR). These cues are coded on n bits. The sign of each cue is representative of its logic value while its absolute value is representative of the confidence that the logic value of this cue is correct.

Each received block BLC_i is decoded in the decoding means MDCD comprising, in this example, F processors operating in parallel. The value F is equal to 360 in one application of the DVB-S2 standard, for example. The battery of F processors performs the updating of the check nodes and of the bit nodes. A metrics memory MMT contains the internal metrics (equal in number to the number of “1s” in the parity matrix). A mixing device, which is known by those skilled in the art (which is a shifting device in the case of an LDPC coding applied to the DVB-S2 standard), makes it possible to place the corresponding data opposite the corresponding processors. Finally, on completion of the decoding, the processors deliver, into the memory MMHD, the N hard decisions corresponding to the N decoded logic values of the block BLC_i.

In a prior-art approach illustrated in FIG. 3, the input memory comprises two buffer memories BFA, BFB, each capable of storing a block. The architecture of the decoding means and also the characteristics of the memory are chosen in such a way that the duration of decoding of an entire block is less than the duration for receiving and storing an entire block. In practice, these two durations are relatively similar. This is the reason why a time constraint analysis uses two buffer memories in which the blocks are stored alternately. One buffer memory is used for decoding a block while the subsequent block is stored in the other buffer memory, and vice versa.

FIG. 4 illustrates an exemplary parity matrix H of an LDPC code, this matrix H being subdivided here into two layers LY1 and LY2. To decode each block, the metrics memory MMT comprises metrics cues or metrics E_ij corresponding to the “1s” of the matrix H. The metrics matrix MMT additionally comprises “0s” corresponding to the “0s” of the matrix H. For the decoding of a block, all the metrics cues E_ij are set to zero (initialization).

Additionally, the buffer memory BFCH, which is either the buffer memory BFA or the buffer memory BFB, stores, on initializing the decoding, the cues C1-CS (LLR1-LLR8) of the specific block to be decoded. The metrics matrix MMT has a similar structure to parity matrix H, and therefore comprises, in this example illustrated in FIG. 5, two layers LY1 and LY2. The block will be decoded using a layered belief propagation algorithm.

More precisely, in a conventional layered BP-type algorithm, each iteration begins with the calculation of the check nodes, that is to say, the updating of the metrics cues for all the rows of a first layer of the matrix and for all the columns of this first layer. More precisely, if E′_ij denotes an updated metric cue, this updated metric cue is given by the following formula:

E′ _ij =g(C_k−E_ik) with k belonging to W_i and k≠j  (I)

in which W_i denotes the whole set of positions of the “1s” in row i of the relevant layer and g denotes a conventional function used in updating the check nodes, this function may comprise a hyperbolic tangent function.

Next, the channel cues C_j are updated for all the columns of the specific layer, using the following formula:

C′ _j =C _j −E _ij +E′ _ij

Once these operations have been performed, the next layer is then updated, and in this way, all the columns are processed in succession before going to the next iteration, where all the operations just described are repeated. After a determined number of iterations, the decoding is stopped and the hard decisions, that is to say, the logic values of the decoded bits of the block, are simply the signs of the channel cues C_j.

In practice, the LDPC codes that can be realized in terms of hardware are, for example, based on pseudo-random matrices. This implies that a parity matrix H comprises several regular sub-matrices. This offers parallelism in the calculation of the metrics cues. For example, the matrices used in the DVB-S2 standard have a parallelism of 360. The parity matrix H comprises matrices D that are interleaved, D being a diagonal matrix cyclically shifted by any value.

FIG. 6 illustrates examples of such matrices D while FIG. 7 illustrates an example of a parity matrix H that can be used in the DVB-S2 standard. Such a matrix H comprises two parts. The first part corresponds to the information bits and is formed with 360×360 sub-matrices. The other part corresponds to the redundancy bits and comprises an N-K×N−K square matrix with a double diagonal of “1s”. All the rows of a sub-matrix are separated by q rows in the parity matrix H so that the matrix H has 360 q rows (or N-K rows).

To perform the decoding calculations, the data is read in groups of 360 so as to harness the matrix parallelism and reach the desired decoding speed. The application of a layered decoding algorithm to such matrices involves the parity matrix H being subdivided into a certain number of layers, the convergence of the decoding being faster as the number of layers increases.

One chooses to subdivide H into q layers of 360 rows each, this making it possible to have the largest possible number of layers while preserving the parallelism of 360. Also, each row of a layer is separated from the next row of the layer by q rows. Stated otherwise, the first layer comprises rows 1, q+1, 2q+1, and so forth, while the second layer comprises rows 2, q+2, 2 q+2, and so forth.

It is noted in FIG. 7, the matrix H comprises, for a given layer, several “1s” in one and the same column. This may pose hardware complexity and wire routing problems for the implementation of conventional layered decoding algorithms. Specifically, assuming by way of example and for simplicity that a metric matrix MMT, such as that illustrated in FIG. 8, comprises three “1s” in column j of layer LY1. These three “1s” of the parity matrix are represented here by the metrics cues E_1j, E_2j and E_3j, where C_j denotes the corresponding channel cue.

In an implementation of a conventional layered decoding algorithm, the step of updating the check nodes comprises updating the metric cues E′_1j, E′_2j, and E′_3j. Next, the updating of the channel cues C′) is then performed according to the following formula:

C′ _j =C _j −E _1j −E _2j −E _3j +E′ _1j +E′ _2j +E′ _3j.

Stated otherwise, it uses three subtractions and three additions to update the channel cue C_j.

According to the mode of implementation of the method according to an embodiment, as illustrated in FIG. 9, when several metrics cues belong to the same column of a layer, the layered decoding algorithm is simplified. Specifically, instead of updating the channel cue with all the values of the metrics of the column, only a single one is used. On the other hand, the metric selected for the updating is changed at each iteration. This is explained in detail in the flow chart of FIG. 9.

More precisely, the processing of a layer 9 comprises a column processing 90, within which a calculation of C_j-E_ij is firstly performed for every value of i (step 900). Next, a calculation of the new metrics cues E′_ij is performed for every value of i (step 901) by applying the abovementioned formula (I).

Next, the channel cue C_j is, in a step 902, updated using a single updated metric cue, namely in the present case, the metric cue E′_pj where p is equal to k modulo N_mj where k denotes the index number of the current iteration, and N_mj denotes the number of metrics cues in column j. Stated otherwise, the updated channel cue C′_j is equal to C_j−E_pj+E′_pj. Next, in step 903, the metrics matrix MMT is updated with the updated metric cue E′_pj. The method moves to the next column, and when the whole layer has been processed, the method moves to the next layer.

Returning now to the example of FIG. 8, in which there are three “1s” in column j, step 901 will provide the updated metrics cues E′_1j, E′_2j and E′_3j. Assuming that the current iteration is iteration No. k, the new updated channel cue C′_jk is then given by the following formula:

C′ _jk =C _jk −E _1jk +E′ _jk.

It is noted that during iteration k, the first metric cue is used for the updating of the channel cue.

At the next iteration k+1, the new channel cue C′_j(k+1) is given by the following formula:

C′ _j(k+1) =C _j(k+1) −E _2j(k+1) +E′ _2j(k+1)

It is noted at this iteration that the second metric cue that is used.

During the next iteration, it will be the third metric cue that will be used and then the first again, and so on and so forth. The hardware architecture of the decoding means or a decoder MDCD may be simplified, as illustrated in FIG. 10. The decoding means comprise processing means or a processor MT and control means or a controller MCTL, for example, ones that may be readily embodied with the help of logic circuits, suitable for activating them in such a way as to implement the decoding method, just described in the above embodiments.

The processing means or a processor MT may have a particularly simple architecture and comprise subtraction means or a subtractor SUB connected to the output of the metrics memory MMT and to the output of the channel memory BFCH. Additionally, calculation means or a calculator MC performs, in particular, the updating of the check nodes (metric cues) and provides the processors and also the shifting means or a shift register, which are connected to the output of the subtraction means SUB and deliver the metrics cues that have been updated with a latency T.

The processing means or a processor may comprise summation means or a summer ADD connected to the output of the calculation means and to the output of the subtraction means SUB by way of a delay means or a buffer MRD producing a delay equal to T. The output of the adder ADD is connected to the input of the channel memory. Additionally, the output of the calculation means MC is connected to the input of the metric memory MMT.

FIG. 11 illustrates the plot of the bit error rate (BER) as a function of the signal-to-noise ratio for parity matrices of the DVB-S2 type with block lengths (N) equal to 64,800 bits, QPSK modulation and a coding rate of ½. Curve C1 shows the plot of the BER rate for a conventional layered decoding algorithm while C2 shows the plot of the BER rate for decoding according to the embodiments disclosed herein. Curve C1 of FIG. 12 illustrates, under the same conditions, the block error rate FER for a decoding according to the prior art while curve C2 of FIG. 12 relates to the plot of the FER rate for decoding according to the embodiments disclosed herein. Hence, in both cases a reduction of the degradation in performance is obtained. Specifically, the maximum discrepancy between the two curves is around 0.005 dB.

These above described embodiments are not limited to the modes of implementation just described, but encompasses all variants thereof. More precisely, these embodiments are completely compatible with an input memory structure MMCH, such as that described in the French patent application filed in the name of the applicant on the same day as the present patent application and entitled “Procédé et dispositif de décodage de blocs encodés avec un code LDPC” [Method and device for decoding blocks encoded with an LDPC code], in which the input memory is capable of storing more than two blocks and comprises for example p+q elementary memories each capable of storing N/p cues (LLRs) and a transfer memory, the transfer memory playing the role here of the buffer memory BFCH.

Claims

1-10. (canceled)

11. A method for successively and individually decoding a plurality of blocks encoded with an LDPC code, each block comprising N cues, the method comprising: initializing N channel cues, respectively, with the N cues of the block and metrics cues distributed within a metrics matrix including N columns and subdivided into a plurality of layers, each layer comprising a plurality of rows; and decoding each block iteratively by at least updating, layer-by-layer, the metrics cuesdistributed over the N columns of the plurality of layers, andupdating the N channel cues respectively associated with the N columns of each layer;the metrics cues being situated in different rows for each layer;the updating of the N channel cues respectively associated with the N columns involving at each iteration a single updated metric cue selected from all the metrics cues of a current column for a current iteration, a respective row of the selected single updated metric cue changing at each iteration.

12. The method according to claim 11 wherein the metrics cues are stored in a metrics memory; wherein the updating of the metrics cues distributed over the N columns of the plurality of layers comprises, at each iteration, a write to the metric memory of the selected single updated metric cue for the current iteration.

13. The method according to claim 11 wherein an index number of a respective row of the selected single updated metric cue at the current iteration is equal to k modulo Nm, where k denotes the index number of the current iteration, and Nm denotes the number of metrics cues distributed over the N columns.

14. The method according to claim 11 wherein the encoded blocks comply with a DVB-S2 standard.

15. A method for successively and individually decoding a plurality of blocks encoded with an LDPC code, each block comprising N cues, the method comprising: initializing N channel cues, respectively, with the N cues of each block and metrics cues distributed within a metrics matrix including N columns and subdivided into a plurality of layers, each layer comprising a plurality of rows, the metrics cues being in different rows; andprocessing each layer in column-by-column manner to decode each block by at least updating respective metrics cues for each row, andupdating a respective channel cue associated with a current column and the metrics matrix using a single updated metric cue selected from the metrics cues of the current column, a respective row of the selected single updated metric cue changing at each iteration.

16. The method according to claim 15 wherein the metrics cues are stored in a metrics memory; wherein the updating of the respective metrics cues distributed over the N columns of each layer comprises, at each iteration, a write to the metric memory of the selected single updated metric cue for a current iteration.

17. The method according to claim 15 wherein an index number of the respective row of the selected single updated metric cue selected at a current iteration is equal to k modulo Nm, where k denotes the index number of the current iteration, and Nm denotes the number of metrics cues distributed over the N columns.

18. The method according to claim 15 wherein the encoded blocks comply with a DVB-S2 standard.

19. A decoder apparatus comprising: an input for receiving a succession of blocks encoded with an LDPC code, each block comprising N cues;a channel memory for storing N channel cues;a metrics memory for storing metrics cues distributed within a metrics matrix comprising N columns and subdivided into a plurality of layers, each comprising a plurality of rows, the metrics cues being in different rows; anda decoder for successively decoding each block and comprising a processor, and a controller for initializing the N channel cues respectively with the N cues of each block and the metrics cues, and for iteratively activating said processor for performing an update, layer-by-layer, of the metrics cues distributed over the plurality of rows of a current layer, andperforming an update of the N channel cues respectively associated with the N columns of each layer, andsaid processor for updating a respective channel cue associated with a current column while involving at each iteration a single updated metric cue selected from all the metrics cues of the current column, a row of the single selected update metric cue changing at each iteration.

20. The decoder apparatus according to claim 19 wherein said processor is for updating the metrics cues distributed over the plurality of rows of a respective layer by at least writing, at each iteration, to said metric memory, the selected single updated metric cue for a current iteration.

21. The decoder apparatus according to claim 19 wherein an index number of a respective row of the selected single updated metric cue at a current iteration is equal to k modulo Nm, where k denotes the index number of the current iteration, and Nm denotes the number of metrics cues distributed over the current column.

22. The decoder apparatus according to claim 19 wherein said processor comprises a subtractor connected to an output of said metrics memory and to an output of said channel memory; and further comprising: a calculator connected to an output of said subtractor and for delivering the metrics cues updated with a latency T; andan adder connected to an output of said calculator and to the output of said subtractor by way of at least a buffer producing a delay equal to T, an output of the buffer being connected to an input of said channel memory, the output of said calculator being connected to an input of said metrics memory.

23. The decoder apparatus according to claim 19 wherein the encoded blocks comply with a DVB-S2 standard.

24. A decoder apparatus for decoding a succession of blocks encoded with an LDPC code, each block comprising N cues, and comprising: a channel memory for storing N channel cues;a metrics memory for storing metrics cues distributed within a metrics matrix comprising N columns and subdivided into a plurality of layers, each comprising a plurality of rows; anda decoder for successively decoding each block and comprising a processor, and a controller for initializing the N channel cues respectively with the N cues of each block and the metrics cues, and for iteratively activating said processor for processing each layer in column-by-column manner to decode each block by at least updating respective metrics cues for each row, andupdating a respective channel cue associated with a current column and the metrics matrix using a single updated metric cue selected from the metrics cues of the current column, a respective row of the selected single updated metric cue changing at each iteration.

25. The decoder apparatus according to claim 24 wherein said processor is for updating the metrics cues distributed over the plurality of rows of a respective layer by at least writing, at each iteration, to said metric memory, the selected single updated metric cue for a current iteration.

26. The decoder apparatus according to claim 24 wherein an index number of a respective row of the selected single updated metric cue at a current iteration is equal to k modulo Nm, where k denotes the index number of the current iteration, and Nm denotes the number of metrics cues distributed over the current column.

27. The decoder apparatus according to claim 24 wherein said processor comprises a subtractor connected to an output of said metrics memory and to an output of said channel memory; and further comprising: a calculator connected to an output of said subtractor and for delivering the metrics cues updated with a latency T; andan adder connected to an output of said calculator and to the output of said subtractor by way of at least a buffer producing a delay equal to T, an output of the buffer being connected to an input of said channel memory, the output of said calculator being connected to an input of said metrics memory.

28. A signal receiver complying with a DVB-S2 standard and including a decoder for decoding a succession of blocks encoded with an LDPC code, each block comprising N cues, the decoder comprising: a channel memory for storing N channel cues;a metrics memory for storing metrics cues distributed within a metrics matrix comprising N columns and subdivided into a plurality of layers, each comprising a plurality of rows; anda decoder for successively decoding each block and comprising a processor, and a controller for initializing the N channel cues respectively with the N cues of each block and the metrics cues, and for iteratively activating said processor for processing each layer in column-by-column manner to decode each block by at least updating respective metrics cues for each row, andupdating a respective channel cue associated with a current column and the metrics matrix using a single updated metric cue selected from the metrics cues of the current column, a respective row of the selected single updated metric cue changing at each iteration.

29. The signal receiver according to claim 28 wherein said processor is for updating the metrics cues distributed over the plurality of rows of a respective layer by at least writing, at each iteration, to said metric memory, the selected single updated metric cue for a current iteration.

30. The signal receiver according to claim 28 wherein an index number of a respective row of the selected single updated metric cue at a current iteration is equal to k modulo Nm, where k denotes the index number of the current iteration, and Nm denotes the number of metrics cues distributed over the current column.