Patent Application
20210083692
The present disclosure relates to the field of digital communications, and more specifically to a bit interleaver for a bit-interleaved coding and modulation system with quasi-cyclic low-density parity-check codes.
In recent years, bit-interleaved coding and modulation (hereinafter, BICM) systems have been used in the field of digital communications (see, for example, Non-Patent Literature 1).
BICM systems generally incorporate the following three steps.
(1) Encoding data blocks into codewords using, for example, quasi-cyclic low-density parity check (hereinafter, QC LDPC) code or similar.
(2) Performing bit interleaving on the bits of each codeword.
(3) Dividing each bit interleaved codeword into constellation words having a number of constellation bits, and mapping the constellation words to constellations.
ETSI EN 302 755 V1.2.1 (DVB-T2 Standards)
Typically, efficiency is desirable in interleaving applied to the codewords of quasi-cyclic low-density parity-check codes.
The present disclosure aims to provide an interleaving method enabling efficient interleaving to be applied to the codewords of quasi-cyclic low-density parity-check codes.
In order to achieve the above-stated aim, a bit interleaving method for a communication system using quasi-cyclic low-density parity check codes, comprises: a reception step of receiving a codeword of the quasi-cyclic low-density parity check codes made up of N cyclic blocks each including Q bits; a bit permutation step of applying a bit permutation process to the codeword so as to permute the bits in the codeword; a division step of dividing the codeword, after the bit permutation process, into a plurality of constellation words, each of the constellation words being made up of M bits and indicating one of 2M predetermined constellation points, and an intra-cyclic-block permutation step of applying an intra-cyclic-block permutation process to each of the cyclic blocks so as to permute the bits in each of the cyclic blocks, wherein the division step involves dividing the codeword, after the bit permutation process, into F×N/M sections, each including M/F of the cyclic blocks (F being a positive integer), and then dividing the codeword into the constellation words such that each of the constellation words is associated with one of the sections, and the bit permutation process is applied such that each of the constellation words includes F bits extracted from each of M/F different cyclic blocks subject to the permutation process, within a given section associated with a given one of the constellation words.
The bit interleaving method of the present invention enables effective interleaving to be applied to the codewords of the quasi-cyclic low-density parity-check codes.
(Background Information)
The input processor 110 converts an input bitstream into blocks of a predetermined length. The LDPC encoder 120 encodes the blocks into codewords using LDPC codes, and then transmits the codewords to the bit interleaver 130. The bit interleaver 130 applies an interleaving process to each LDPC codeword, then divides each interleaved codeword into a sequence of cell words (i.e., constellation words). The constellation mapper 140 maps each cell word (i.e., constellation word) to a sequence of constellations (e.g., using QAM). The generic modulator 150 at the output includes all processing blocks from the output of the BICM encoder to a radio frequency (hereinafter, RF) power amplifier.
An LDPC code is a linear error correcting code that is fully defined by a parity-check matrix (hereinafter, PCM). A PCM is a binary sparse matrix that represents the connection of codeword bits (hereinafter also termed variable nodes) to the parity checks (hereinafter also termed check nodes). The columns and the rows of the PCM respectively correspond to the variable nodes and the check nodes. In the PCM, a connection between a variable node and a check node is represented by a one-element.
Quasi-cyclic low-density parity check (hereinafter, QC LDPC) codes are one variety of LDPC codes. QC LDPC codes have a structure that is particularly suited to hardware implementation. In fact, most standards in use today employ QC LDPC codes. The PCM of a QC LDPC code has a special configuration made up of a plurality of circulant matrices. A circulant matrix is a square matrix in which each row is a cyclic shift of the elements in the previous row, and has one, two, or more folded diagonals. Each circulant matrix has a size of Q×Q. Here, Q represents the cyclic factor of the QC LDPC. The above-described quasi-cyclic configuration allows Q check nodes to be processed in parallel, which is clearly beneficial for efficient hardware implementation.
A special variety of QC LDPC codes are repeat-accumulate quasi-cyclic low-density parity check (hereinafter, RA QC LDPC) codes. RA QC LDPC codes are well known as being easy to encode, and are therefore used in a wide variety of standards (e.g., in second-generation DVB standards, including DVB-S2, DVB-T2, and DVB-C2). The right-hand side of the PCM corresponds to the parity bits. The one-elements therein are arranged in a staircase structure.
Above, and throughout, DVB-T is an abbreviation of Digital Video Broadcasting—Terrestrial, DVB-S2 is an abbreviation of Digital Video Broadcasting—Second Generation Satellite, DVB-T2 is an abbreviation of Digital Video Broadcasting—Second Generation Terrestrial, and DVB-C2 is an abbreviation of Digital Video Broadcasting—Second Generation Cable.
By applying a simple row permutation to the PCM shown in
The quasi-cyclic structure of the PCM parity portion is imparted by applying a suitable row permutation to only the parity bits of the PCM shown in
Typically, the bits of an LDPC codeword vary in importance, and the bits of a constellation vary in robustness level. Mapping the bits of an LDPC codeword to a constellation directly, i.e., without interleaving, leads to suboptimal performance. Thus, the bits of the LDPC codeword require interleaving prior to mapping onto constellations.
For this purpose, the bit interleaver 130 is provided between the LDPC encoder 120 and the constellation mapper 140, as shown in
The bits of the LDPC codeword differ in importance primarily because not all bits are necessarily involved in the same number of parity checks. The more parity checks (check nodes) a given codeword bit (variable node) is involved in, the more important the given codeword bit is in an iterative LDPC decoding process. A further reason is that the variable nodes each have different connectivity to the cycles of a Tanner graph representing the LDPC codes. Therefore, the codeword bits are likely to differ in importance despite being involved in the same number of parity checks. These ideas are well understood in the field. As a rule, the importance of the variable nodes increases as the number of check nodes connected therewith increases.
In the particular case of QC LDPC codes, all bits included in a cyclic block of Q bits have the same number of parity checks applied thereto, and have the same connectivity to the cycles of the Tanner graph. Thus, all bits have the same importance.
Similarly, the encoded bits of a constellation are widely known to have different levels of robustness. For example, a quadrature amplitude modulation (hereinafter, QAM) constellation is made up of two independent pulse amplitude modulation (hereinafter, PAM) symbols, one symbol corresponding to the real part and the other symbol corresponding to the imaginary part. The two PAM symbols each encode M bits.
Thus, a 16-QAM constellation encodes four bits and has two robustness levels. Likewise, a 64-QAM constellation encodes six bits and has three robustness levels. Also, a 256-QAM constellation encodes eight bits and has four robustness levels.
The following parameters are hereinafter used throughout the present description.
Cyclic factor: Q=8
Number of cyclic blocks per LDPC codeword: N=12
Number of bits per constellation: M=4 (i.e., 16-QAM)
Given the above parameters, the number of constellations to which each LDPC codeword is mapped is equal to Q×N/M=24. Typically, the parameters Q and N are selected such that Q×N is equal to a multiple of M for all constellations supported by the system.
A conventional bit interleaver is known from the DVB-T2 standard (see ETSI EN 302 755). The DVB-T2 standard is a television standard presenting improvements over the DVB-T standard, and describes a second-generation baseline transmission system for digital television broadcasting. The DVB-T2 standard gives the details of a channel coding and modulation system for broadcast television services and generic data.
The input processor 810 converts an input bitstream into blocks of a predetermined length. The BICM encoder 820 applies BICM processing to the input. The frame builder 830 uses input from the BICM encoder 820 and the like to generate a distribution frame structure in the DVB-T2 format. The OFDM generator 840 performs pilot addition, fast Fourier transform application, guard interval insertion, and the like on the distribution frame structure, then outputs a transmission signal in the DVB-T2 format.
The BICM used in the DVB-T2 standard is described in chapter 6 of the ETSI EN 302 755 standard. The aforementioned standard is incorporated herein by reference and explained below.
The BICM encoder 820 includes an LDPC encoder 821, a bit interleaver (in turn including a parity interleaver 822 and a column-row interleaver 823), a bit-to-cell demultiplexer 824, and a QAM mapper 825.
The LDPC encoder 821 encodes the blocks into codewords using LDPC codes. The bit interleaver (which includes the parity interleaver 822 and the column-row interleaver 823) performs interleaving on the bits of the codewords. The bit-to-cell demultiplexer 824 demultiplexes the interleaved bits of the codewords into cell words (constellation words). The QAM mapper 825 maps the cell words (constellation words) to complex QAM symbols. The complex QAM symbols are also termed cells. In fact, the bit-to-cell demultiplexer 824 may also be considered a part of the bit interleaver. In such situations, the BICM encoder conforming to the DVB-T2 standard may be considered to have the basic structure shown in
The LDPC codes used in the DVB-T2 standard are RA QC LDPC codes having a cyclic factor of Q=360. Two codeword lengths are defined for the DVB-T2 standard, one being 16200 bits and the other being 64800 bits. In the present document, LDPC codes using a codeword length of 16200 bits are referred to as 16K codes (or as 16K LDPC codes), and LDPC codes having a codeword length of 64800 bits are referred to as 64K codes (or as 64K LDPC codes). The number of cyclic blocks per codeword is 45 for the 16K codes and 180 for the 64K codes. The available codes corresponding to each block length (codeword length) are given in Tables A1 through A6 of ETSI EN 302 755 for the DVB-T2 standard.
The bit interleaver is used only for constellations larger than quadrature phase-shift keying constellations (hereinafter, QPSK), and includes the parity interleaver 822, the column-row interleaver 823, and the bit-to-cell demultiplexer 824. According to the DVB-T2 standard, the bit interleaver does not include the bit-to-cell demultiplexer 824. However, the present document pertains to interleaving as applied to LDPC codes prior to constellation mapping. As such, the bit-to-cell demultiplexer 824 is treated as a part of the bit interleaver.
The parity interleaver 822 performs parity permutation on the parity bits of each codeword so as to clarify the quasi-cyclic structure thereof, as described above (see
Conceptually, the column-row interleaver 823 operates by writing the bits of each LDPC codeword column-wise in an interleaver matrix, then reading the bits row-wise. The first bit of the LDPC codeword is written first, and is read first. After writing and before reading the LDPC codeword bits, the column-row interleaver 823 cyclically shifts the columns of bits by a predetermined number of positions. This is termed column twisting in the DVB-T2 standard. The number of columns Nc and the number of rows Nr in the interleaver matrix are given in Table 1 for several constellation sizes, according to the two aforementioned LDPC codeword lengths.
The number of columns Nc is twice the number of bits per constellation, with the exception of 16K codes with a 256-QAM constellation. This exception occurs because the LDPC codeword length of 16200 is not a multiple of 16, i.e., is not twice the number of bits per 256-QAM constellation.
The codeword bit writing process and bit reading process for 16K codes performed by the column-row interleaver 823 is illustrated in
The arrows indicate the order in which the bits are written to and read from the interleaver matrix. For example, when the interleaver matrix has twelve columns, the codeword bits of the 16K code are written in the order given in
Prior to QAM mapping, the bit-to-cell demultiplexer 824 demultiplexes the LDPC codewords to obtain a plurality of parallel bit streams. The number of streams is twice the number of encoded bits M per QAM constellation, i.e., 2×M, with the exception of 16K LDPC codes with a 256-QAM constellation. For 16K LDPC codes with a 256-QAM constellation, the number of streams equal to the number of encoded bits M per QAM constellation. The M encoded bits per constellation are referred to as one cell word (constellation word). For the 16K LDPC codes, the number of cell words per codeword is 16200/M, as given below.
8100 cells for QPSK,
4050 cells for 16-QAM,
2700 cells for 64-QAM, and
2025 cells for 256-QAM.
According to Table 1, given above, the number of parallel streams is equal to the number of columns in the column-row interleaver for constellations larger than QPSK. The bit-to-cell demultiplexers corresponding to 16-QAM constellations, 64-QAM constellations, and 256-QAM constellations for 16K LDPC codes are respectively shown in
As shown in
In addition to having the simple demultiplexer 1110 (1210, 1310) simply demultiplex the LDPC codewords, to which interleaving has been applied, the bit-to-cell demultiplexer 824 also has the demux permutator 1120 (1220, 1320) perform a permutation on the demultiplexed parallel bit streams.
However, when the column-row interleaver is used (i.e., for 16-QAM constellations or larger), the permutation applied to the bit streams is identical to a permutation applied to the columns of the column-row interleaver due to the number of parallel bit streams being equal to the number of columns. Therefore, the permutation performed by the bit-to-cell demultiplexer 824 is regarded as a part of the bit interleaver.
The bit interleaver used in the DVB-T2 standard essentially has two problems.
The first problem is that parallelism is impaired when the number of cyclic blocks in the LDPC codeword is not a multiple of the number of columns in the bit interleaver matrix. Reduced parallelism leads to increased latency. This is especially problematic when iterative BICM decoding is used by the receiver. This situation occurs for several combinations of LDPC codeword length and constellation size in the DVB-T2 standard.
The second problem is that, in the DVB-T2 standard, the number of possible bit interleaver configurations is limited by the number of columns in the bit interleaver matrix.
A further problem of the DVB-T2 bit interleaver is that the regularity and parallelism of the permutation is impaired by the column twisting process.
Accordingly, a bit interleaver that reduces latency while improving parallelism is desired. These properties are particularly important in iterative BICM decoding.
(Inventor Discoveries)
The inventor has discovered, as the fruit of prolonged experimentation, that an interleaver satisfying the following two conditions is extremely efficient.
(Condition 1)
The M bits of each constellation are each mapped to one of M different cyclic blocks of the LDPC codeword. This is equivalent to mapping one bit from M different cyclic blocks of the LDPC codeword to a constellation word. This is schematically illustrated in
(Condition 2)
All constellation words mapped to the M cyclic blocks are mapped only to that particular cyclic block. This is equivalent to mapping all M×Q bits of the M different cyclic blocks each made up of Q bits to exactly Q constellations. This is schematically illustrated in
The above conditions imply that exactly Q constellations are mapped to each set of M cyclic blocks.
The following describes the details of a bit interleaver (i.e., a parallel bit interleaver) that satisfies conditions 1 and 2 given above. In the following description, processing and the units performing such processing are labeled with the same reference numbers wherever applicable.
In the present document, each group of M cyclic blocks and each group of Q constellation words is referred to as a section (or as an interleaver section).
In
A bit interleaver 2000 includes a bit permutator 2010, which in turn includes N/M (=3) section permutators 2021, 2022, and 2023, each operating independently. However, rather than providing three section permutators, a single section permutator may, for example, be provided so as to perform the three section permutation processes described below, switching therebetween over time.
The section permutators (2021, 2022, and 2023) each independently perform a section permutation on the 32 bits making up each of 4 cyclic blocks, such that one bit from every four cyclic blocks (i.e., QB1 through QB4, QB5 through QB8, and QB9 through QB12) is mapped to each group of eight constellation words (i.e., C1 through C8, C9 through C16, and C17 through C24).
Conditions 1 and 2, described above, ensure that the bit interleaver is divisible into N/M parallel sections. The section permutations applied to the parallel sections may all apply the same permutation rules, may each apply different permutation rules, or may involve a subset of the sections applying identical permutation rules while other differ.
For example, the section permutators may map the Q bits of a cyclic block (which each have the same importance in LDPC decoding) to bits having the same bit index (i.e., having the same robustness level) in the Q constellation words. For each cyclic block, the Q bits may be in sequential or in permutated order. The latter case is described with reference to
The section permutator 2101 includes intra-cyclic-block permutators 2111 through 2114 and a column-row permutator 2131. Rather than providing four intra-cyclic-block permutators, for example, a single intra-cyclic-block permutator may be provided and perform the four intra-cyclic-block permutation processes, described later, switching therebetween over time.
The intra-cyclic-block permutators (2111-2114) each perform an intra-cyclic-block permutation on the Q-bit (8-bit) cyclic blocks (QB1-QB4). The intra-cyclic-block permutations applied to the cyclic blocks in each section may all apply the same permutation rules, may each apply different permutation rules, or may involve a subset of the sections applying identical permutation rules while other differ.
The column-row permutator 2131 performs a column-row permutation on each group of M×Q(=32) bits. Specifically, the column-row permutator 2131 writes the M×Q bits row-wise into a M×Q(8×4) matrix, then reads the M×Q bits column-wise therefrom, thus applying the column-row permutation. The column-row permutation applied by the column-row permutator 2131 resembles the permutation applied to the 12×1350 matrix shown in
However, a variation in which the intra-cyclic-block permutation process is not part of the section permutation process is also plausible.
For example, a section permutation implemented without executing the intra-cyclic-block permutation and a structure of mapping by the section permutator are shown in
The section permutation described in
Advantageously, an additional cyclic block permutation may be applied to the N cyclic blocks before the bit interleaver performs the section permutation.
The bit interleaver 2300 shown in
The cyclic block permutator 2310 performs cyclic block permutations 2311-2318 on the cyclic blocks QB1-QB12. Here, the cyclic block permutations 2311-2318 all follow the same permutation rules.
The cyclic block permutation performed on the N cyclic blocks is particularly advantageous in enabling optimized mapping of the LDPC codeword bits onto the constellation bits, resulting in optimized performance.
Stage A: (inter) cyclic block permutation
Stage B: intra-cyclic-block permutation
Stage C: column-row permutation
The (inter) cyclic block permutation is applied to the N cyclic blocks making up the codeword, the intra-cyclic-block permutation is applied to the Q bits of each cyclic block, and the column-row permutation is applied to the M×Q sections.
The bit interleaver 2400 shown in
In the bit interleaver 2400, the (inter) cyclic block permutation is performed by the cyclic block permutator 2310 (stage A), the intra-cyclic-block permutation is performed by the intra-cyclic-block permutators 2111-2122 (stage B), and the column-row permutation is performed by the column-row permutators 2131-2133 (stage C).
The intra-cyclic-block permutators 2111-2122 may be removed from the bit interleaver 2400 shown in
The intra-cyclic-block permutators may have similar structures. This is advantageous in that the intra-cyclic-block permutators are thus implementable using identical resources (e.g., hardware blocks). Alternatively, the intra-cyclic-block permutations may consist of cyclical shifts, which allow for efficient hardware implementation using barrel shifters. An implementation using the barrel shifters in the LDPC decoder is also possible.
The following describes a transmitter that includes the bit interleaver performing a bit interleaving process that satisfies Conditions 1 and 2, with reference to
The LDPC encoder 2510 encodes input blocks into codewords using QC-LDPC codes, and then transmits the codewords to the bit interleaver 2520.
The bit interleaver 2520 receives the codeword of QC-LDPC code from the LDPC encoder 2510. The codeword is made up of N=12 cyclic blocks, each cyclic block including Q=8 bits. The bit interleaver 2520 performs interleaving on the bits of the codewords so as to permute the bits of each of the codewords. The bit interleaver 2520 divides the interleaved codeword into a plurality of constellation words, each made up of M=4 bits and indicating one of 2M=16 constellation points, then outputs the constellation words to the constellation mapper 2530. The bit interleaver 2520 may apply the bit interleaving process discussed with reference to
The constellation mapper 2530 receives the constellation words from the bit interleaver 2520 and performs constellation mapping on the constellation words so received.
The modulator 2740 generates a transmission signal using orthogonal frequency division multiplexing (hereinafter, OFDM) or similar.
The following describes a BICM encoder that includes the bit interleaver performing a bit interleaving process that satisfies Conditions 1 and 2, with reference to
The BICM encoder 2600 shown in
In
The main memory 2601 receives a sequence of bits for transmission from, for example, the (non-diagrammed) input processor, and stores the received bit sequence.
The LDPC controller 2611 outputs a read address to the main memory 2601. The main memory 2601 accordingly outputs the bit sequence, eight bits at a time beginning with the lead bit, to the rotator 2612. The rotator 2612 is controlled by the LDPC controller 2611 to perform a predetermined number of cyclical shifts on the eight bits supplied thereto by the main memory 2601, and then outputs the eight cyclically-shifted bits to the check node processors of the check node processor group 2613, one bit at a time, the bits and the check node processors being in one-to-one correspondence. Each check node processer of the check node processor group 2613 is controlled by the LDPC controller 2611 to perform check node processing on each bit input thereto, then outputs the results to the de-rotator 2614. The de-rotator 2614 is controlled by the LDPC controller 2611 to perform a predetermined number of cyclic shifts on the eight bits received from the check node processor group 2613 so as to cancel the cyclic shift applied by the rotator 2612, and then outputs the eight shifted bits to the main memory 2601. The LDPC controller 2611 outputs a write address to the main memory 2601. The main memory 2601 accordingly stores the eight bits supplied thereto by the de-rotator 2614. The LDPC controller 2611, the rotator 2612, the check node processor group 2613, and the de-rotator 2614 make up the BICM encoder in the LDPC encoder 2510 shown in
The QB counter 2631 counts from 0 to 11 and outputs the counter value to table A 2632. The count operation of the QB counter 2631 is defined in consideration of N=12.
Table A 2632 is a simple look-up table in which the cyclic block permutation rules are stored. That is, Table A 2632 stores N=12 pieces of cyclic block read order information (information associating a different cyclic block with each of the 12 counter values from the QB counter 2631). Table A 2632 outputs a read address to the main memory 2601 such that the bits of one cyclic block (i.e., Q=8 bits) corresponding to the counter value supplied by the QB counter 2631 are supplied from the main memory 2601 to interleaver B 2633. Thus, the main memory 2601 outputs the bits of one cyclic block corresponding to the counter value of the QB counter 2631 to interleaver B 2633. The processing using table A 2632 is executed as the cyclic block permutation process (stage A).
Interleaver B 2633 performs a predetermined number of cyclical shifts on the bits of the cyclic block supplied by the main memory 2601, and outputs the results to a first tier register of the register group 2634. The processing by interleaver B 2633 is executed as the intra-cyclic-block permutation process (stage B). Each register in the register group 2634 stores one cyclic block of bits with timing matching the reception of a control pulse, and outputs the cyclic block of bits before receiving the next control pulse.
When the QB counter 2631 performs the aforementioned process for counter values 0 through 3, the bits of four cyclic blocks (i.e., 32 bits) are input to interleaver C 2635. At this time, interleaver C 2635 interleaves the bits of the four cyclic blocks input thereto, and the mappers of the mapper group 2651 output one constellation word of bits (i.e., M=4 bits). Through the interleaving process, four bits, i.e., one from each of the four registers in the register group 2634, are supplied to each mapper. This processing by interleaver C 2635 is executed as the column-row permutation process (stage C).
The QB counter 2631, table A 2632, interleaver B 2633, the register group 2634, and interleaver C 2635 make up the bit interleaver 2520 of the BICM encoder shown in
The mappers of the mapper group 2651 each map four bits supplied thereto from interleaver C 2635 to a constellation, then output the results. The mapper group 2651 makes up the constellation mapper 2530 of the BICM encoder shown in
For each codeword, the above-described set of processes are applied three times, once each for counter values 0-3, 4-7, and 8-11 of the QB counter 2631.
The Embodiment depicted in
The following describes a receiver receiving signals from a transmitter that includes the bit interleaver performing a bit interleaving process that satisfies Conditions 1 and 2, with reference to
The receiver 2700 shown in
The demodulator 2710 performs a demodulation process through OFDM, for example, and outputs the demodulated results.
The constellation demapper 2720 of the non-iterative BICM decoder generates a soft bit sequence by applying a demapping process to the input from the demodulator 2710, and outputs the soft bit sequence so generated to the constellation demapper 2730. The soft bits are a measure of probability that a given bit is a zero-bit or a one-bit. Typically, the soft bits are represented as log-likelihood ratios (hereinafter, LLRs), defined as follows.
LLR(b)=ln[p(b=0)/p(b=1)]
where p(b=0) indicates the probability of the given bit b being a zero-bit, and p(b=1) represents the probability of the given bit b being a one-bit. Of course, p(b=0)+p(b=1)=1.
The bit deinterleaver 2730 performs an interleaving process on a soft bit sequence output by the constellation demapper (i.e., a bit de-interleaving process) so as to cancel the bit interleaving process applied to the bit sequence by the bit interleaver in the transmitter illustrated in
The LDPC decoder 2740 receives the soft bit sequence deinterleaved by the bit deinterleaver 2730, and performs an LDPC decoding process using the soft bit sequence so received.
One improved technique offering significant performance gains is iterative BICM decoding.
As shown in
The receiver 2800 of
After one or more LDPC decoding iterations, extrinsic information, obtained by the subtractor 2760 subtracting the input to the LDPC decoder 2740 from the output of the LDPC decoder 2740, is output to the bit interleaver 2750. The bit interleaver 2750 performs an interleaving process on the extrinsic information using the same interleaving rules as those applied to the bit sequence by the bit interleaver of the transmitter depicted in
As shown in
As shown, the BICM decoder 2900 includes a main LLR memory 2901, a buffer LLR memory 2902, an LDPC controller 2911, a rotator 2912, a check node processor group 2913, a de-rotator 2914, a QB counter 2931, table A 2932, a subtractor group 2933, interleaver B 2934, register group 2935, interleaver C 2936, a demapper group 2937, deinterleaver C 2938, register group 2939, deinterleaver B 2940, and a delayer 2941.
In
The demappers in the demapper group 2937 each perform a demapping process on the output of a demodulator (not diagrammed), then outputs the LLR values so obtained to deinterleaver C 2938. The demapper group 2937 makes up the constellation demapper 2720 of the iterative BICM decoder shown in
Deinterleaver C 2938 applies a deinterleaving process to the LLR values (i.e., a new interleaving process cancelling the interleaving process applied by the transmitter during stage C), then outputs the deinterleaved LLR values to the registers of register group 2939. Each register stores one cyclic block of LLR values (i.e., eight LLR values). In register group 2939, the cyclic block of LLR values stored by each register is sequentially output to a later tier such that the content of each register is sequentially updated. Deinterleaver B 2940 applies a deinterleaving process to the cyclic block of (eight) LLR values (i.e., a new interleaving process cancelling the interleaving process applied by the transmitter during stage B), then writes the results to the main LLR memory 2901 and the buffer LLR memory 2902 in accordance with table A 2932 (discussed later). An interleaving process cancelling the interleaving process applied by the transmitter during stage A is achieved by this writing to the main LLR memory 2901 and the buffer LLR memory 2902 in accordance with the content of Table A 2932.
Thus, the main LLR memory 2901 stores the post-deinterleaving LLR values, and is also used by the LDPC decoder (i.e., the LDPC controller 2911, the rotator 2912, the check node processor group 2913, and the de-rotator 2914). The LDPC decoding process is an iterative process involving one or more iterations. In each LDPC decoding iteration, the LLR values in the main LLR memory 2901 are updated. In order to compute the extrinsic information needed for iterative BICM decoding, the old LLR values are saved in the buffer LLR memory 2902.
The following describes the LDPC decoder operations.
The LDPC controller 2911 outputs the read address to the main LLR memory 2901 in accordance with the parity-check matrix of the LDPC codes. Thus, the main LLR memory 2901 sequentially outputs one cyclic block of LLR values to the rotator 2912. The rotator 2912 is controlled by the LDPC controller 2911 to perform a predetermined number of cyclical shifts on the cyclic block of LLR values supplied sequentially by the main LLR memory 2901, then outputs the LLR values so shifted to the check node processors of the check node processor group 2913 one at a time. The check node processors of the check node processor group 2913 are controlled by the LDPC controller 2911 to perform a check node process on the sequence of LLR values sequentially input thereto. Next, the check node processors of the check node processor group 2913 are controlled by the LDPC controller 2911 to sequentially output the LLR values resulting from the check node process. The de-rotator 2914 is controlled by the LDPC controller 2911 to performs a predetermined number of cyclic shifts cancelling the cyclic shift applied to the cyclic block sequentially received from the check node processor group 2913 by the rotator 2912, then sequentially outputs the shifted results to the main LLR memory 2901. The LDPC controller 2911 outputs the write address to the main LLR memory 2901 in accordance with the parity-check matrix of the LDPC codes. Thus, the main LLR memory 2901 stores the cyclic block of results sequentially supplied thereto by the de-rotator 2914. The LDPC controller 2911 repeatedly executes the above-described processing in accordance with the parity-check matrix of the LDPC codes.
After a predetermined number of LDPC iterations, a BICM iteration is performed. The LDPC and BICM iterations are also respectively referred to as inner and outer iterations. These two types of iterative may also overlap in some implementations. This enables the speed of convergence to be increased. The BICM and LDPC decoding processes are well known in the field, and the details thereof are thus omitted.
The QB counter 2931 counts from 0 to 11 and outputs the counter value to table A 2932. The count operation of the QB counter 2931 is defined in consideration of N=12.
Table A 2932 is a simple look-up table in which the cyclic block permutation rules are stored. That is, Table A 2932 stores N=12 pieces of cyclic block read (and write) order information (i.e., with information associating a different cyclic block with each of the 12 counter values from the QB counter 2631). Table A 2932 outputs the read address to the main LLR memory 2901 and to the buffer LLR memory 2902 such that one cyclic block of LLR values corresponding to the counter value supplied by the QB counter 2931 are supplied to the subtractor group 2933 by the main LLR memory 2901 and to the buffer LLR memory 2902. Thus, the main LLR memory 2901 and the buffer LLR memory 2902 each output a cyclic block of LLR values corresponding to the counter value of the QB counter 2931 to the subtractor 2933. The delayer 2941 makes a delay adjustment such that the position of the LLR value read from the main LLR memory 2901 and the buffer LLR memory 2902 match the write position of the same LLR values to the main LLR memory 2901 and the buffer LLR memory 2902. The processing using table A 2932 is executed as the cyclic block permutation process (stage A).
The subtractor 2933 in the subtractor group subtracts the output of the buffer LLR memory 2902 from the output of the main LLR memory 2901, then outputs the extrinsic information for one cyclic block thus obtained (i.e., eight pieces of extrinsic information) to interleaver B 2934.
Interleaver B 2634 performs a predetermined number of cyclical shifts on the pieces of extrinsic information for one of the cyclic blocks supplied by the subtractor 2933, and outputs the results to a first tier register of the register group 2935. The processing performed by interleaver B 2934 corresponds to the intra-cyclic-block permutation (stage B). Each register in the register group 2935 stores eight bits with timing matching the reception of a control pulse, and outputs the eight bits before receiving the next control pulse.
When the QB counter 2631 performs the aforementioned process for counter values 0 through 3, the extrinsic information for four cyclic blocks (i.e., 32 pieces of extrinsic information) are input to interleaver C 2936. At this time, interleaver C 2936 performs an interleaving process on the extrinsic information input thereto for four cyclic blocks, then outputs one constellation word of extrinsic information (i.e., M=4 pieces of extrinsic information) to each demapper of the demapper group 2937. Through the interleaving process, the four pieces of extrinsic information are supplied to the demappers of the demapper group 2951 from the four registers in register group 2935, one at a time. This processing by interleaver C 2936 is executed as the column-row permutation process (stage C).
The QB counter 2931, table A 2932, interleaver B 2934, the register group 2935, and interleaver C 2936 make up the bit interleaver 2750 of the BICM decoder shown in
The demappers of the demapper group 2937 uses the four pieces of extrinsic information supplied by interleaver C 2936 as a-priori information to perform a demapping process, then output the resulting LLR values to deinterleaver C 2938.
Deinterleaver C 2938 applies a deinterleaving process to the LLR values (i.e., a new interleaving process cancelling the interleaving process applied by the transmitter during stage C), then outputs the deinterleaved LLR values to the registers of the register group 2939. Each register stores one cyclic block of LLR values (i.e., eight LLR values). In register group 2939, the cyclic block of LLR values stored by each register is sequentially output to a later tier such that the content of each register is sequentially updated. Deinterleaver B 2940 applies a deinterleaving process to the cyclic block of (eight) LLR values (i.e., a new interleaving process cancelling the interleaving process applied by the transmitter during stage B), then writes the results to the main LLR memory 2901 and the buffer LLR memory 2902. The main LLR memory 2901 and the buffer LLR memory 2902 receive the write address from table A 2932 via the delayer 2941, then store one cyclic block of LLR values (i.e., eight LLR values) received from the deinterleaver 2940 in accordance with the received write address. An interleaving process cancelling the interleaving process applied by the transmitter during stage A is achieved by this writing in accordance with the content of table A 2932.
For each codeword, the above-described set of processes is applied three times, once each for counter values 0-3, 4-7, and 8-11 of the QB counter 2931.
The QB counter 2931, table A 2932, deinterleaver B 26938, the register group 2939, and interleaver C 2940 make up the bit interleaver 2730 of the BICM decoder shown in
Interleaver B 2934 and deinterleaver B 2940 are reconfigurable. This requires a certain hardware cost, but this cost is minimized by attentive design. Interleaver C 2936 and deinterleaver 2938 implement the column-row permutation. This permutation is uniform for a predetermined constellation size. Thus, the cost of implementation is reduced.
The Embodiment depicted in
(Further Experimenter Discoveries)
The interleavers satisfying Conditions 1 and 2 given above (i.e., parallel interleavers) assume that the number of bits per constellation word M is a divisor of the number of cyclic blocks N. However, M is not typically a divisor of N. For example, the 16K LDPC codes used in the DVB-T2 standard has N=45 cyclic blocks per 16K LDPC codeword. When M is not a divisor of N, the mapping of square constellations, such as QAM constellations where M is even, is not straightforward.
Thus, a particular solution is proposed that involves excluding one or more of the N cyclic blocks and implementing the interleaver discussed in Embodiment 1 (i.e. the parallel interleaver), above, only to the remaining cyclic blocks.
In other words, N′ cyclic blocks are selected from among the N cyclic blocks such that N′ is a multiple of the number of bits per constellation words Min the bit interleaver. The bit interleaver divides the N′ selected cyclic blocks into N′/M sections such that each section includes M cyclic blocks, then performs a section permutation on each section. The bits of the excluded (i.e., not selected) cyclic block may or may not be interleaved.
Specifically, a bit interleaving method pertaining to the disclosure is intended for a communication system using quasi-cyclic low-density parity check codes, and comprises: a reception step of receiving a codeword of the quasi-cyclic low-density parity check codes made up of N cyclic blocks each including Q bits; a bit permutation step of applying a bit permutation process to the codeword so as to permute the bits in the codeword; a division step of dividing the codeword, after the bit permutation process, into a plurality of constellation words, each of the constellation words being made up of M bits and indicating one of 2M predetermined constellation points; and a selection step of selecting a subset of N′ cyclic blocks from among the N cyclic blocks such that N′ is a multiple of M, wherein prior to the bit permutation process, the codeword is divided into N′/M sections, each of the sections including M of the cyclic blocks, and each of the constellation words being associated with one of the N′/M sections, and in the bit permutation step, the bit permutation process is applied such that the M bits in each of the constellation words include one bit from each of M different cyclic blocks in a given section associated with a given constellation word, and such that all bits in the given section are mapped to only Q of the constellation words associated with the given section.
Similarly, a bit interleaver pertaining to the disclosure is intended for use in a communication system using quasi-cyclic low-density parity check codes, and comprises: a bit permutation unit receiving a codeword of the quasi-cyclic low-density parity check codes made up of N cyclic blocks each including Q bits, applying a bit permutation process to the codeword so as to permute the bits in the codeword, and dividing the codeword, for output after the bit permutation process, into a plurality of constellation words, each of the constellation words being made up of M bits and indicating one of 2M predetermined constellation points; and a selection unit selecting a subset of N′ cyclic blocks from among the N cyclic blocks such that N′ is a multiple of M, wherein prior to the bit permutation process, the codeword is divided into N′/M sections, each of the sections including M of the cyclic blocks, and each of the constellation words being associated with one of the N′/M sections, and the bit permutation unit applies the bit permutation process such that the M bits in each of the constellation words include one bit from each of M different cyclic blocks in a given section associated with a given constellation word, and such that all bits in the given section are mapped to only Q of the constellation words associated with the given section.
Also, the bits in the codeword may include a bit group that is one of: not included in the selected subset of N′ cyclic blocks and left as-is without rearranging; and not included in the selected subset of N′ cyclic blocks, and subject to a rearranging process that is independent of the bit permutation process and that is applied only to any unselected cyclic blocks.
According to this configuration, bit interleaving is applied to a maximised number of cyclic blocks.
For example, the excluded cyclic blocks may be the cyclic blocks having a variable node of the lowest weight. For the example RA QC LDPC codes (see
Moreover, the selecting step is also usable for selecting the cyclic blocks according to the importance of the bits included therein.
The importance of the bits included in the cyclic blocks is determined according to the number of parity bits associated therewith.
When the codeword is for repeat-accumulate quasi-cyclic low-density parity-check codes, the unselected cyclic blocks correspond to the parity section of the codeword.
According to this configuration, most of the bits corresponding to the codeword are involved in optimal interleaving. Reducing the number of bits involved in interleaving results in greatly improved overall efficacy.
Further, the subset of N selected cyclic blocks may be formed from N continuous cyclic blocks beginning with the first bit of the codeword.
According to this configuration, hardware implementation is further simplified. Particularly, for the example of the parity-check matrix in repeat-accumulate quasi-cyclic low-density parity-check codes, the last bit of the codeword is the bit having the least associativity.
In general, the number of interleaver sections (each made up of M cyclic blocks) is given by floor(N/M), and the number of excluded cyclic blocks is given by rem(N, M). Here, floor(N/M) is a function that returns the largest integer smaller than or equal to N/M, and rem(N, M) is a function that returns the remainder of dividing N by M.
Table 2 indicates the number of sections and the number of excluded cyclic blocks for various constellation sizes (i.e., number of bits per constellation M), for the 16K LDPC codes of the DVB-T2 standard (where N=45 cyclic blocks are used).
According to the interleaving method satisfying Conditions 1 and 2 as described above, each constellation word is mapped onto M cyclic blocks. However, for large constellations (i.e., constellations having many constellation points), an interleaving method satisfying Conditions 1 and 2 requires an extremely large amount of delay registers (see the implementation described with reference to
By mapping two or more bits of each constellation word onto the same cyclic block, the number of cyclic blocks onto which each constellation word is mapped is reduced. The number of constellation word bits mapped onto the same cyclic block is termed the folding factor and is denoted F. For example, when F=2 for a 16-QAM constellation, each constellation word is mapped onto four cyclic blocks rather than two. The only constraint is that the folding factor F (which is an integer greater than one) must be a divisor of both M and Q. When F=1, no folding is involved, i.e., the situation corresponds to Embodiment 1, described above.
A complex QAM constellation symbol is decomposed into two identical real PAM symbols. Thus, the M bits of the QAM constellation are divided into two identical real PAM symbols, which are sets of M/2 bits. The bits of each constellation word are then mapped onto the same number M/2 of cyclic blocks. A folding factor of F=2 is advantageous for QAM constellation.
For complex constellations that cannot be decomposed into real numbers, such as 8-PSK (phase shift keying), 16-APSK (amplitude phase shift keying), 32-APSK and so on in the DVB-S2 standard, the folding method is not easily applicable. However, the folding method is usable when F is a divisor of M. Unfortunately, this prevents any guarantee that each cyclic block has only bits of the same robustness level from the constellations mapped thereto.
Beneficially, the folding method is, for example, applied such that only bits of the same robustness level in the constellations are mapped to the cyclic blocks.
Folding is also beneficial in reducing the number of excluded cyclic blocks, or even eliminating the need to exclude any cyclic blocks. As described above, a certain number of cyclic blocks making up the codeword must be excluded when the interleaver described in Embodiment 1 (i.e., a parallel interleaver) is used.
Without folding (i.e., when F=1), the number of groups of M cyclic blocks (i.e., the number of sections) is floor(N/M), and the number of excluded cyclic blocks is rem(N, M). With folding, the number of groups of M/F cyclic blocks is floor(N/(M/F)), and the number of excluded cyclic blocks is rem(N, M/F). Specific example of these numbers are given in Table 3, for the LDPC codes used in the DVB-T2 standard.
The inventor discovered that, in order to perform folding (where F is an integer equal to or greater than two), Conditions 1 and 2 must be modified into Conditions 1A and 2A, as given below.
(Condition 1A)
The M bits of each constellation word are each mapped to one of M/F different cyclic blocks of the LDPC codeword. This is equivalent to mapping one bit from M/F different cyclic blocks of the LDPC codeword to a constellation word. This is schematically illustrated in
(Condition 2A)
All constellation words mapped to the M/F cyclic blocks are mapped only to that particular cyclic block. This is equivalent to mapping all M×Q/F bits of the M/F different cyclic blocks each made up of Q bits to exactly Q/F constellations. This is schematically illustrated in
When F=1, no folding is involved, and thus Conditions 1A and 2A are equivalent to Conditions 1 and 2.
The following describes the details of a bit interleaver (i.e., a parallel bit interleaver) that satisfies conditions 1A and 2A, given above. In the following description, processing and the units performing such processing are labeled with the same reference numbers wherever applicable.
In the present document, each group of M/F cyclic blocks or of Q/F constellation words is referred to as a folding section (or as a folding interleaver section).
When F=1 (i.e., no folding), the folding interleaver sections match the interleaver sections and the bit interleaver is configured identically to the bit interleaver from Embodiment 1.
In
The bit interleaver 2000A includes a bit permutator 2010A. The bit permutator 2010A includes F×N/M=6 folding section permutators 2021A-2026A each operating independently. Rather than providing six folding section permutators, a single folding section permutator may be provided so as to performs six (later described) folding section permutation processes, switching therebetween over time.
The folding section permutators (2021A, 2022A, 2023A, 2024A, 2025A, 2026A) are independent and each apply a folding section permutation process to the 16 bits of two cyclic blocks such that F=2 bits from each of M/F=2 cyclic blocks (QB1-QB2, QB3-QB4, QB5-QB6, QB7-QB8, QB9-QB10, QB11-QB12) are mapped to a given set of four constellation words (C1-C4, C5-C8, C9-C12, C13-C16, C17-C20, C21-C24).
Conditions 1A and 2A, described above, simply ensure that the bit interleaver is divisible into F×N/M parallel folding sections. The folding section permutations applied to the parallel folding sections may all apply the same permutation rules, may each apply different permutation rules, or may involve a subset of the sections applying identical permutation rules while other differ.
For example, the folding section permutators may map Q bits from each cyclic block to bits of Q/F constellation words having the same robustness level. This is illustrated in
However, in the example of
The folding section permutators 2201A (and 2202A) each include a column-row permutator 2131A (or 2132A).
The column-row permutators 2131A (and 2132A) each perform a column-row permutation process on Q×M/F=16 cyclic blocks QB1 and QB2 (or QB3 and QB4). To be exact, the column-row permutators 2131A (and 2132A) write Q×M/F=16 bits row-wise into a Q×M/F (8×2) matrix, then read the 16 bits so written column-wise to perform the column-row permutation process. The column-row permutation applied by the column-row permutators 2131A and 2132A resembles the permutation applied to the 12×1350 matrix shown in
Folding with a folding factor of F reduces the number of cyclic blocks mapped to a single constellation word. Thus, the matrix used in the column-row permutation is reduced in size, from M rows to M/F rows.
As shown in
The folding section permutation described in
The folding section permutators of
Advantageously, an additional cyclic block permutation may be applied to the N cyclic blocks before the bit interleaver performs the folding section permutation.
The bit interleaver 2300A shown in
The bit interleaver 2400A shown in
The folding section permutators 2201A-2206A each include a column-row permutator 2131A-2136A. The folding section permutators 2133A-2136A each perform substantially identical permutations with the column-row permutators 2131A-2132A.
The bit interleavers shown in
The following describes a transmitter that includes the bit interleaver performing a bit interleaving process that satisfies Conditions 1A and 2A, with reference to
The bit interleaver 2520A receives the codeword in QC-LDPC code from the LDPC encoder 2510. The codeword is made up of N=12 cyclic blocks, each cyclic block including Q=8 bits. The bit interleaver 2520A performs interleaving on the bits of the codewords. The bit interleaver 2520A divides the interleaved codeword into a plurality of constellation words, each made up of M=4 bits and indicating one of 2M=16 constellation points, then outputs the constellation words to the constellation mapper 2530. The bit interleaver 2520A may apply the bit interleaving process discussed with reference to
The following describes a receiver receiving signals from a transmitter that includes the bit interleaver performing a bit interleaving process that satisfies Conditions 1A and 2A.
The bit deinterleaver 2730A performs an interleaving process on the soft bit sequence output from the constellation demapper 2720 so as to cancel the bit interleaving process applied to the bit sequence by the bit interleaver 2520A in the transmitter 2500A.
The bit interleaver 2750A performs an interleaving process on the extrinsic information using the same interleaving rules as the interleaving process applied to the bit sequence by the bit interleaver 2520A in the transmitter 2500A.
From a hardware implementation perspective, folding is desirable in that, for example, the bits of a constellation are thereby located in fewer LLR memory locations. Typically, the LLR memory in the decoder includes G×N addressable memory locations, each location being capable of storing Q/G LLR values. Here, G is an implementation parameter that is a divisor of Q and is hereinafter referred to as memory granularity. The LLR memory locations in the decoder and the LLR values of the first constellation are shown in
The number of LLR values per memory location, i.e., the value of Q/G, is necessarily a multiple of F. The LLR values of each constellation are thus stored at the same position in all memory locations. This ensures that the LLR values of any constellation word are stored in M/F memory locations. A counter-example is given in
In addition to simple QAM constellations where folding with F=2 is applicable, folding is even more useful when two or more constellation symbols are jointly decoded. Joint decoding is necessary, for instance, for the maximum-likelihood decoding of block codes (e.g., space-time codes, space-frequency codes, and the like), and for rotated constellations in two or more dimensions.
In general, a block code encodes two or more input symbols (e.g., x1, . . . , xk) onto two or more output symbols (e.g., y1, . . . , yL). Here, L is at least equal to K. The block codes are modeled on an L×K generator matrix. Here, the output signal vector Y is obtained as a result of left-multiplying the input signal vector Xby the generator matrix G (i.e., Y=GX).
The elements of the input signal vector X and the output signal vector Y, as well as the elements of the generator matrix G, may be real or complex. Depending on the type of code, the output signal vector Y may be transmitted in different time slots or in different frequency slots, may be transmitted over different antennas, or may be transmitted using a variety of different time slots, frequency slots, and antennas.
In the receiver, maximum-likelihood decoding is required in order to decode all elements of the input signal vector X Examples of block codes for multiple-input multiple-output (hereinafter, MIMO) systems include Alamouti codes, Golden codes, and spatial multiplexing.
When K symbols are encoded in the same block, a folding factor of up to K is obviously useable. Further, provided that the symbols are QAM symbols (including two divisible PAM symbols), then the folding factor may be increased to 2K.
According to a further aspect of the present disclosure, when constellations of different sizes, i.e., hybrid constellations, are jointly encoded, the two constellations have different robustness levels. Thus, for example, the cyclic block to which the bits of one constellation word are mapped is distinct from the cyclic block to which the bits of the other constellation word are mapped.
The following describes an example of a coded spatial-multiplexing MIMO system using two transmit antennas. The complex signal prior to coding is X=[x1 x2]. Here, x1 is a signal to which QPSK has been applied, and x2 is a signal to which 16-QAM has been applied.
The complex signal after decoding is Y=[y1 y2]. Here, y1 and y2 are signals respectively transmitted by a first antenna and a second antenna. Y is obtained through a left-multiplication of X with a 2×2 generator matrix G (where the elements of G may be either real or complex) (i.e., Y=GX).
x1 is a QPSK symbol having a real part b1 and an imaginary part b2.
x2 is a 16-QAM symbol having real part b3 and b4 and having imaginary parts b5 and b6.
The two symbols are jointly decoded by the receiver and thus form a constellation block or generated block.
The entire 6-bit constellation block has three robustness levels.
Level 1: 1: QPSK bits b1 and b2 are mapped to QB1
Level 2: 16-QAM bits b3 and b5 are mapped to QB2
Level 3: 16-QAM bits b4 and b6 are mapped to QB3
When one of the constellations has M1 bits and the other constellation has M2 bits, the N cyclic groups are divided into one or more groups of M1 cyclic block and one or more groups of M2 cyclic blocks in order to perform the bit interleaving process.
The following describes an example of an interleaver performing folding in a situation where Nis not a multiple of M.
The number of folding sections is floor(N/(M/F))=6, and the number of excluded cyclic blocks is rem(N, M/F)=1.
The bit interleaver 2000B selects 13−1=12 cyclic blocks from among cyclic blocks QB1-QB13, such that the selected cyclic blocks QB1-QB12 satisfy conditions A1 and A2 and are subject to interleaving. The bit permutator 2010A in the bit interleaver 2000B performs the permutation process described with reference to
As an example of an interleaver not performing folding when N is not a multiple of M, the bit permutator 2010A shown in
The following describes a specific example of a section permutation performed on LDPC codes using the DVB-T2 standard as described in Table 3.
(Ex. 1A) When N=45, Q=360, M=4, and without folding (i.e., F=1)
The bit interleaver 4400A selects a subset of 44 cyclic blocks QB1-QB44 from among 45 cyclic blocks QB1-QB45.
The subset is then divided into eleven sections 1-11 each including four cyclic blocks. For example, section 1 includes cyclic blocks QB1-QB4, and section 11 includes cyclic blocks QB41-QB44.
Eleven section permutators (4401-4411) in the bit interleaver 4400A each perform a permutation process on the four respective cyclic blocks, as described with reference to
The bits of cyclic block QB45 are here mapped to a constellation word without interleaving.
(Ex. 1B) When N=45, Q=360, M=4, and with folding (i.e., F=2)
The bit interleaver 4400B selects a subset of 44 cyclic blocks QB1-QB44 from among 45 cyclic blocks QB1-QB45.
The subset is then divided into eleven sections 1-11 each including two cyclic blocks.
Here, 22 section permutators (4421, 4422, . . . , 4442) in the bit interleaver 4400B each perform a permutation process on the two respective cyclic blocks, as described with reference to
The bits of cyclic block QB45 are here mapped to a constellation word without interleaving.
(Ex. 2A) When N=45, Q=360, M=6, and without folding (i.e., F=1)
The bit interleaver 4500A selects a subset of 42 cyclic blocks QB1-QB42 from among 45 cyclic blocks QB1-QB45.
The subset is then divided into seven sections 1-7 each including six cyclic blocks. For example, section 1 includes cyclic blocks QB1-QB6, and section 7 includes cyclic blocks QB37-QB42.
Seven section permutators (4501 through 4507) in the bit interleaver 4500A each perform a permutation process on the six respective cyclic blocks, as described with reference to
Here, cyclic blocks QB43-QB45 are not included in any subset.
The bits of cyclic blocks QB43 and QB44 are here mapped to constellation words without interleaving.
In contrast, the bits of cyclic block QB45 are subject to intra-cyclic-block permutation by the intra-cyclic-block permutator 4545.
That is, the bits of cyclic blocks QB43 and QB44 are not included in any subset, and are not subject to permutation. However, while the bits of cyclic block QB45 are likewise not included in any subset, these bits are subject to permutation by an intra-cyclic-block permutator 4545 that is separate from the section permutators (4501 through 4507).
In the example of
(Ex. 2B) When N=45, Q=360, M=6, and with folding (i.e., F=2)
The bit interleaver 4500B selects all 45 cyclic blocks QB1-QB45 from among the 45 cyclic blocks QB1-QB45.
The subset is then divided into fifteen sections 1-15 each including three cyclic blocks.
Here, 15 section permutators (4511, . . . , 4526) in the bit interleaver 4500B each perform a permutation process on the three respective cyclic blocks, as described with reference to
The following describes an effective bit interleaving method. In the LDPC decoding process, the error correction capability can be decreased by the presence of ineffective check nodes. The inventors have discovered a further method of preventing the occurrence of ineffective check nodes, using the method mentioned above. The following discusses the means by which ineffective check nodes occur, and the means of cancelling such nodes.
Ineffective check nodes occur when two or more LDPC variable nodes that are connected to the same check node have bits from the same constellation mapped thereto. Supposing that such a constellation undergoes deep fading, the corresponding LLR values output by the constellation demapper are then very small or are zero.
Hypothetically, when two or more variable nodes are connected to the same check node, and the check node has LLR values that are either zero or extremely small, then the check node is unable to employ the error correction process. Thus, as a result, the value convergence of the iterative process for a first iteration by a LDPC decoder is slowed. Such check nodes are termed ineffective check nodes.
The following describes the manner in which ineffective check nodes occur, using a specific example. Consider the third cyclic block of check nodes shown in
These connections are also described as shown in
As described above, a correspondence relationship exists between
As for the eight cyclic blocks connected to check nodes 17 through 24,
The following describes two examples in which ineffective check nodes occur. These examples are a selected sample.
The occurrence of ineffective check nodes is preventable by avoiding mapping a variable node onto a constellation connected to the same check node. For the parallel bit interleaver, this is realized by performing a further permutation on the bits of each cyclic block. This further permutation is termed an intra-cyclic-block permutation, below. The intra-cyclic-block permutation is typically applied differently according to the cyclic block in question.
Implementation is simplified by having the intra-cyclic-block permutation by using one or more cyclic shifts. When there is only one cyclic shift, a (reconfigurable) rotator and de-rotator included in the LDPC decoder are reusable, thus reducing implementation complexity in the circuit.
Reference to
When a cyclic block requiring no cyclic shift is input, the shift value is set to zero and the bit sequence is output as-is, without cyclic shifting.
Hence, the number of ineffective check nodes in the LDPC codes is minimized by selecting an appropriate permutation for each cyclic block. Of course, the intra-cyclic-block permutations must be adapted whenever the LDPC codes are changed, such as when the PCM of the employed code is changed. The intra-cyclic-block permutation may be optimally realized using a set of (limited) PCM including a plurality of predetermined PCM, the set being stored in advance (alternatively, shift values or other parameters may be stored, the parameters defining the actual permutation method). Table B for
As shown in
In contrast to
The QB counter 5531 notifies the QB permutation table 5532 of the number of the target cyclic block.
The QB permutation table 5532 is a lookup table similar to table A 2632 of
The QB shift table 5536 stores shift values for performing a cyclic shift on a bit sequence for each of the cyclic blocks. The QB shift table 5536 determines the shift value according to the number of the cyclic block in the notification from the QB permutation table 5532, then notifies the rotator (interleaver B) B) 5533 thereof. The QB shift table 5536 is equivalent to table B (5101A, 5101B) in
The rotator (interleaver B) 5533 performs a rightward cyclic shift by the shift value on the bit sequence input thereto in accordance with the shift value in the notification, then outputs the results to the register group 5534. The rotator (interleaver B) 5533 performs an intra-cyclic-block permutation in the BICM encoder, i.e., corresponds to the intra-cyclic-block permutation 5410 of
As shown, the iterative BICM decoder 5600 includes a main LLR memory 5601, a buffer LLR memory 5602, an LDPC controller 5611, a rotator 5612, a check node processor group 5613, a de-rotator 5614, a QB counter 5631, a table 5632, a subtractor group 5633, interleaver B 5634, register group 5635, interleaver C 5636, a demapper group 5637, de-interleaver C 5638, register group 5639, de-interleaver B 5640, a delayer 5641, and a QB shift table 5642.
As comparison to
The QB counter 5631 notifies the QB permutation table 5632 of the number of the target cyclic block.
The QB permutation table 5632 is a lookup table similar to table A 2932 of
The QB shift table 5642 stores shift values for performing a cyclic shift on a bit sequence for each of the cyclic blocks. The QB shift table 5642 determines the shift value according to the number of the cyclic block in the notification from the QB permutation table 5632, then notifies the rotator (interleaver B) 5634 accordingly. Also, the de-rotator (deinterleaver B) 5640 is notified of the shift value via one of the delayers 5641, in order to be able to undo the interleaving process for the intra-cyclic-block permutation. The QB shift table 5542 is equivalent to table B (5101A, 5101B) in
The rotator (interleaver B) 5634 performs a cyclic shift on the bit sequence input thereto according to the shift value in the notification from the QB shift table 5642, then outputs the results to the register group 5635. The rotator (interleaver B) 5634 performs an intra-cyclic-block permutation in the iterative BICM decoder 5600.
Also, in accordance with the shift value in the notification from the QB shift table 5642, the de-rotator (deinterleaver B) 5640 performs a cyclic shift on the bit sequence input thereto from the register group 5639 that is the inverse of the shift applied by the rotator (interleaver B) 5634, then outputs the results to the main LLR memory 5601.
The column-row interleaver (interleaver C) 5636 corresponds to interleaver C of
According to the above configuration, the BICM encoder is able to simply execute a intra-cyclic-block permutation, and situations where the variable nodes associated with a check node are mapped to the same constellation in plurality is thus avoided. Accordingly, the occurrence probability of ineffective check nodes, which cannot be used in error correction, is reduced.
(Supplement 1)
The present disclosure is not limited to the Embodiments described above. Provided that the aims of the invention and accompanying aims are achieved, other variations are also possible, such as the following.
(1) Embodiment 1 is described above using the parameters N=12, Q=8, and M=4. However, no limitation to the parameters N, M, and Q is intended. Here, N may be any multiple of M. When N is two or more times M, the processing by the bit interleaver is divisible into a plurality of sections.
(2) In Embodiment 2, when folding is used, i.e., when F is two or greater, the parameters given for the example are N=12, Q=8, M=4, and a folding factor of F=2. However, no limitation to the parameters N, M, Q, and F is intended. Here, F is a divisor of M and Q, and N is a multiple of M/F.
(3) In Embodiment 2, when folding is used, the value of F is given as two, which is the number of bits having the same robustness level in a single 16-QAM constellation. However, no limitation is intended. The value of F need not be equal to the number of bits having the same robustness level in a constellation, and may indeed be other than the number of bits having the same robustness level in a constellation.
(4) In Embodiment 2, when folding is used, the example describes a folding factor of F=2, and QAM constellations being 16-QAM constellations. However, no limitation is intended. When F=2, the QAM constellations may be other than 16-QAM constellations (e.g., 64-QAM constellations or 256-QAM constellations).
(5) In Embodiment 4, table B and the QB shift table store cyclic shift values for a rightward shift of a bit sequence. However, these tables may also store values for a leftward shift, and the shift values need not be absolute minimums. No limitation is intended provided that the tables enable the prevention of a situation where the variable nodes associated with a check node are mapped in plurality to the same constellation.
Also, the cyclic shift may be replaced by a intra-cyclic-block permutation with no regularity such that, in a given constellation, no check node is associated with a plurality of variable nodes. However, when an intra-cyclic-block permutation without regularity is used, then either the transmitter must communicate the intra-cyclic-block permutation method to the receiver, or the method for implementing the permutation must be determined in advance between the transmitter and receiver.
(6) In the above-described Embodiments, the constellations are described as 16-QAM (i.e., M=4). However, the constellation may be specified by other modulation methods such as QPSK and QAM, such as the circular constellations employed in the DVB-S2 standard, higher-dimensional constellations, and so on.
(7) The methods and devices discussed in the above Embodiments may be implemented as software or as hardware. No particular limitation is intended in this regard. Specifically, the above-described Embodiments may be implemented as a computer-readable medium having embodied thereon computer-executable instructions that are adapted for allowing a computer, a microprocessor, a microcontroller, and the like to execute the above-described methods. Also, the above-described Embodiments may be implemented as an Application-Specific Integrated Circuit (ASIC) or as an Field Programmable Gate Array (FPGA).
(Supplement 2)
The interleaving method, interleaver, deinterleaving method, deinterleaver, and decoder of the present disclosure, and the effects thereof, are described below.
In one aspect, a first bit interleaving method for a communication system using quasi-cyclic low-density parity check codes comprises: a reception step of receiving a codeword of the quasi-cyclic low-density parity check codes made up of N cyclic blocks each including Q bits; a bit permutation step of applying a bit permutation process to the codeword so as to permute the bits in the codeword; a division step of dividing the codeword, after the bit permutation process, into a plurality of constellation words, each of the constellation words being made up of M bits and indicating one of 2M predetermined constellation points, and an intra-cyclic-block permutation step of applying an intra-cyclic-block permutation process to each of the cyclic blocks so as to permute the bits in each of the cyclic blocks, wherein the division step involves dividing the codeword, after the bit permutation process, into F×N/M sections, each including M/F of the cyclic blocks (F being a positive integer), and then dividing the codeword into the constellation words such that each of the constellation words is associated with one of the sections, and the bit permutation process is applied such that each of the constellation words includes F bits extracted from each of M/F different cyclic blocks subject to the permutation process, within a given section associated with a given one of the constellation words.
In another aspect, a first bit interleaver for a communication system using quasi-cyclic low-density parity check codes comprises: a bit permutation unit receiving a codeword of the quasi-cyclic low-density parity check codes made up of N cyclic blocks each including Q bits, and applying a bit permutation process to the codeword so as to permute the bits in the codeword; a division unit dividing the codeword, after the bit permutation process, into a plurality of constellation words, each of the constellation words being made up of M bits and indicating one of 2M predetermined constellation points; and an intra-cyclic-block permutation unit applying an intra-cyclic-block permutation process to each of the cyclic blocks so as to permute the bits in each of the cyclic blocks, wherein the division unit divides the codeword, after the bit permutation process, into F×N/M sections, each including M/F of the cyclic blocks (F being a positive integer), and then dividing the codeword into the constellation words such that each of the constellation words is associated with one of the sections, and the bit permutation process is applied such that each of the constellation words includes F bits extracted from each of M/F different cyclic blocks subject to the permutation process, within a given section associated with a given one of the constellation words.
Here, the division corresponds to that performed by the BICM encoder and the BICM decoder in the above-described Embodiments, and to reading the sequence of cyclic block bits from the main memory and the main LLR memory.
Accordingly, reductions in circuit surface area and in electric power consumption are achieved, in addition to enabling the realization of a bit interleaving process having high parallelism. In addition, by executing the intra-cyclic-block permutation, the check nodes are made more likely to be able to decrease the probability of ineffective check nodes occurring and being unusable by the receiver.
In a second aspect of a bit interleaving method, the intra-cyclic-block permutation process is performed such that bits in the codeword that are linked to a common check node of the QC-LDPC codes are each mapped to a different one of the constellation words.
Also, in a second aspect of a bit interleaver, the intra-cyclic-block permutation unit performs the intra-cyclic-block permutation process such that bits in the codeword that are linked to a common check node of the QC-LDPC codes are each mapped to a different one of the constellation words.
This ensures that the check nodes are able to decrease the probability of ineffective check nodes occurring and being unusable by the receiver.
In a third aspect of a bit interleaving method, at least one instance of the intra-cyclic-block permutation process applied to each of the cyclic blocks is a cyclic shift performed on at least a subset of a bit sequence in each of the cyclic blocks.
Also, in a third aspect of a bit interleaver, at least one instance of the intra-cyclic-block permutation process applied to each of the cyclic blocks is a cyclic shift performed on at least a subset of a bit sequence in each of the cyclic blocks.
Accordingly, the cyclic shift has a simple configuration enabling the realization of the intra-cyclic-block permutation process.
In a fourth aspect of a bit interleaving method, a column-row permutation step of applying a column-row permutation process to each of a plurality of sections of Q×M/F bits, the column-row permutation process being realized by writing the Q×M/F bits row-wise into a matrix having M/F rows and Q columns, then reading the Q×M/F bits from the matrix column-wise.
Also, in a fourth aspect of a bit interleaver, a column-row permutation unit applying a column-row permutation process to each of a plurality of sections of Q×M/F bits, the column-row permutation process being realized by writing the Q×M/F bits row-wise into a matrix having M/F rows and Q columns, then reading the Q×M/F bits from the matrix column-wise.
In a fifth aspect of a bit interleaving method, the bit interleaving method further comprises a selection step of selecting one intra-cyclic-block permutation method to be applied to the cyclic blocks from among a predetermined number of intra-cyclic-block permutation methods, the selection being made according to the QC-LDPC codes specified for use in the communication system.
Also, in a fifth aspect of a bit interleaver, the bit interleaver of the first aspect further comprises a selection unit selecting one intra-cyclic-block permutation method to be applied to the cyclic blocks from among a predetermined number of intra-cyclic-block permutation methods, the selection being made according to the QC-LDPC codes specified for use in the communication system.
Accordingly, the intra-cyclic-block permutation method defined according to the PCM is deliberately specified and is usable for encoding.
In a further aspect, a bit deinterleaving method for a bit stream in a communication system using QC-LDPC codes comprises: a reception step of receiving a bit sequence made up of N×Q bits; and a reverse bit permutation step of applying the bit interleaving method of the first aspect in a reversed order to the received bit sequence, so as to restore a codeword of the QC-LDPC codes.
Also, in still another aspect, a bit deinterleaver for a bit stream in a communication system using QC-LDPC codes comprises a reverse bit permutation unit receiving a bit sequence made up of N×Q bits and applying the bit permutation process applied by the bit interleaver of the first aspect in a reversed order to the received bit sequence so as to restore a codeword of the QC-LDPC codes.
In another aspect, a decoder for a bit interleaving and demodulating system using quasi-cyclic low-density parity check codes, comprising: a constellation demapper generating a soft bit sequence indicating a probability of a corresponding bit being one of a zero-bit and a one-bit; the bit deinterleaver of the other aspect deinterleaving the soft bit sequence; and a low-density check parity check decoder decoding the deinterleaved soft bit sequence.
In a second aspect, a decoder of the other aspect further comprises: a subtraction unit calculating a difference between input and output of the low-density parity check decoder; and the bit interleaver of the first aspect, providing the difference to the constellation demapper as feedback.
Accordingly, a bit interleaving process having high parallelism is realizable.
The present invention is applicable to a bit interleaver in a bit-interleaved coding and modulation system used for quasi-cyclic low-density parity codes, and to a bit deinterleaver corresponding to such a bit interleaver.
| Number | Date | Country | Kind |
|---|---|---|---|
| 11004125.8 | May 2011 | EP | regional |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 16437307 | Jun 2019 | US |
| Child | 17102739 | US | |
| Parent | 16117625 | Aug 2018 | US |
| Child | 16437307 | US | |
| Parent | 15581148 | Apr 2017 | US |
| Child | 16117625 | US | |
| Parent | 15335501 | Oct 2016 | US |
| Child | 15581148 | US | |
| Parent | 15070290 | Mar 2016 | US |
| Child | 15335501 | US | |
| Parent | 14804466 | Jul 2015 | US |
| Child | 15070290 | US | |
| Parent | 14116632 | Nov 2013 | US |
| Child | 14804466 | US |
