Classifications

• • H—ELECTRICITY
  • H03—ELECTRONIC CIRCUITRY
  • H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
  • H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
  • H03M13/29—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes combining two or more codes or code structures, e.g. product codes, generalised product codes, concatenated codes, inner and outer codes
  • H03M13/2957—Turbo codes and decoding
  • H03M13/2975—Judging correct decoding, e.g. iteration stopping criteria
• • H—ELECTRICITY
  • H03—ELECTRONIC CIRCUITRY
  • H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
  • H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
  • H03M13/03—Error detection or forward error correction by redundancy in data representation, i.e. code words containing more digits than the source words
  • H03M13/05—Error detection or forward error correction by redundancy in data representation, i.e. code words containing more digits than the source words using block codes, i.e. a predetermined number of check bits joined to a predetermined number of information bits
  • H03M13/09—Error detection only, e.g. using cyclic redundancy check [CRC] codes or single parity bit
• • H—ELECTRICITY
  • H03—ELECTRONIC CIRCUITRY
  • H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
  • H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
  • H03M13/29—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes combining two or more codes or code structures, e.g. product codes, generalised product codes, concatenated codes, inner and outer codes
• • H—ELECTRICITY
  • H03—ELECTRONIC CIRCUITRY
  • H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
  • H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
  • H03M13/37—Decoding methods or techniques, not specific to the particular type of coding provided for in groups H03M13/03 - H03M13/35
  • H03M13/3707—Adaptive decoding and hybrid decoding, e.g. decoding methods or techniques providing more than one decoding algorithm for one code
• • H—ELECTRICITY
  • H03—ELECTRONIC CIRCUITRY
  • H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
  • H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
  • H03M13/37—Decoding methods or techniques, not specific to the particular type of coding provided for in groups H03M13/03 - H03M13/35
  • H03M13/3738—Decoding methods or techniques, not specific to the particular type of coding provided for in groups H03M13/03 - H03M13/35 with judging correct decoding
• • H—ELECTRICITY
  • H03—ELECTRONIC CIRCUITRY
  • H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
  • H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
  • H03M13/37—Decoding methods or techniques, not specific to the particular type of coding provided for in groups H03M13/03 - H03M13/35
  • H03M13/3746—Decoding methods or techniques, not specific to the particular type of coding provided for in groups H03M13/03 - H03M13/35 with iterative decoding
• • H—ELECTRICITY
  • H03—ELECTRONIC CIRCUITRY
  • H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
  • H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
  • H03M13/63—Joint error correction and other techniques
  • H03M13/635—Error control coding in combination with rate matching
  • H03M13/6362—Error control coding in combination with rate matching by puncturing
  • H03M13/6368—Error control coding in combination with rate matching by puncturing using rate compatible puncturing or complementary puncturing
• • H—ELECTRICITY
  • H03—ELECTRONIC CIRCUITRY
  • H03M—CODING; DECODING; CODE CONVERSION IN GENERAL
  • H03M13/00—Coding, decoding or code conversion, for error detection or error correction; Coding theory basic assumptions; Coding bounds; Error probability evaluation methods; Channel models; Simulation or testing of codes
  • H03M13/65—Purpose and implementation aspects
  • H03M13/6508—Flexibility, adaptability, parametrability and configurability of the implementation

Definitions

• TDMA time divisional multiple access or when the latency should be low generally use short codes or medium long aforementioned (eg MPEG size).
• turbo codes in the general sense of the term, including turbo codes or TPC Turbo Product Codes
• SCCC International Concatenated Convolutional Code
• SCCCC Parallel Concatenated convolutional convolutional code
• LDPC abbreviation Anglo-Saxon low density parity check
• the size of the code word limits the decoupling between the two dimensions. A smaller size also amplifies the statistical effect of a bad local encoding, which is less well compensated by the rest of the code.
• concatenation methods for example LPCD and turbo code concatenations, are known from the prior art to reduce the error floor (error floor) of the turbo code, due to a free distance (in the sense of Hamming) too low.
• Another possibility is not to reduce the coding rate of the internal code. Since the coding rate is smaller, the operating point (expressed in Eb / NO with Eb Energy per bit and NO noise spectral density and not bits coded by the external code) is not compensated by the gain of the external code. .
• the invention relates to a method of improving iterative decoding of short codes within a demodulator, characterized in that it comprises at least the following steps: • Decode the metric vector of the bits considered at the output of the demodulator,
• the value of the CRC is defined, for example, so that the overall encoding rate is not changed by inserting the CRC, by perforating it in the decoded code iteratively.
• An existing CRC can be used, starting from the same initialization vector provided by the demodulator.
• the second decoding is performed, for example, starting from an initialization vector different from the initialization vector supplied by the demodulator, when the corrective vector, representing the difference between the initial vector and the vector on which the new vector is attempted.
• decoding is obtained by an anti-gradient of the last half-iterations.
• the anti-gradient consists of a hollow matrix where only the N correct non-zero coefficients correspond to the least reliable decoded values during the last half-iteration, the penultimate (or the two last-to-last) half-iterations turbo code.
• N_correct is typically 3/4 * d-free, where d_free is the distance of the code, and the coefficients of the matrix are each of opposite sign to the decoded value when the penultimate iteration.
• the second decoding can be carried out starting from an initialization vector different from the initialization vector supplied by the demodulator, when the corrective vector representing the difference between the initial vector and the vector on which the new decoding is attempted is provided. by a low amplitude noise corresponding to a low temperature in a simulated annealing method.
• noise is injected according to the steps described above on a reduced number of points at the input of the decoder.
• This reduced number of points is of the order of the free distance of the code, and the points correspond to the least reliable values coming from the penultimate or the second last iteration.
• the second super iteration is repeated, for example, for different bit positions.
• the position of the inverted bit is taken a priori (as for a simulated annealing) or a posteriori thanks to the decoding of the first super-iteration (as for an inverse gradient).
• the new successive input vectors therefore consist of the vector initially received, but in which a value was forced to bit 0 (thus to the minimum LLR) or bit 1, ie to the maximum metric.
• the position of the inverted bit (with respect to the decoded value) during the different successive super-iterations is random or follows the order of increasing reliability, ie starting with the least reliable of the decoded bits.
• the number of super iterations is greater than two and the starting point of the new super iteration consists, for example, of the initial point at the output of the demodulator, and a correction vector of size (number of non-zero coordinates) or of weight (Euclidean norm) variable depending on the iteration (typically increasing) or fixed.
• the number of super iterations is greater than 2 and the starting point of the new super iteration is, for example, iteratively constituted, therefore, from the entry point of the previous superiteration, and a corrective vector of fixed size or weight. or not.
• the number of super-iterations is limited by a maximum fixed a priori. Super iterations stop as soon as the CRC found matches.
• FIG. 1 the implementation diagram of the invention, indicating the added blocks: CRC (2), CRC check (8) and feedback function (6) and the action point of the decision feedback used by the invention,
• FIG. 2 shows the gain of the invention in the case of an ATM cell (asynchronous transfer mode)
• Figure 3 shows the gain of the invention in the case of an ATM cell.
• the basic idea of the method according to the invention is notably to use a CRC control code adapted to indicate, in particular at the end of decoding, that the result obtained must be considered as valid. , Exact CRC, or on the contrary that the decoded result can not be in accordance with the message sent.
• the decoder restarts a decoding since the first iteration, but by modifying the initial information received from the demodulator.
• FIG. 1 shows in solid lines the steps and the means implemented by the invention in the case where the decoding method already includes a CRC. In dotted lines, means for introducing a CRC at the encoder is shown.
• the source 1 transmits signals composed of bits which are encoded by an iterative coder 3.
• the coded bits are then transmitted to a modulator 4.
• the modulated bits pass through a propagation channel and are received by a receiver 5,6,7 8.
• the receiver comprises, for example, a demodulator 5 receiving the signal from the propagation channel.
• the demodulated signals are then transmitted to a super-iterative decoder 6,7,8 having the characteristics of the invention and comprising in particular an anti-gradient device 6, an iterative decoder 7 and a CRC control device 8.
• the function of the controller 8 is notably to control the value of the CRC control code present in the signal at a predetermined value.
• the decoding function is particularly suitable for:
• the decoder is particularly suitable for detecting the end of code convergence.
• Various criteria known to those skilled in the art can be used.
• the decoding convergence criterion of a TPC is extremely simple, since the decoding can stop as soon as the word corresponding to the output (raw soft output) of a half-iteration is a code word for the other iteration. Methods exist in the same way for the other iterative decoding codes.
• the information entered in the CRC is, for example, a binary value, acceptable message or incorrectly decoded message.
• the method according to the invention implements in particular the following steps:
• the metric vector of the bits considered at the output of the demodulator is decoded
• the super decoder transmits the decoded word to the recipient
• the super-decoder modifies the iterative decoding parameters of the initial message received from the demodulator and executes at least one new step of iterative decoding with these new parameters.
• the modification of the decoding parameters can then be a modification of the speed of convergence, a modification of the computing power, a modification of the inputs of the decoder as detailed in the following.
• the method comprises a step where this intrinsic verifier is added and the aforementioned decoding steps are then performed.
• a CRC is added adapted to detect a false decoding and to restart a second iterative decoding, when the overall coding rate is decreased by the insertion of the CRC, the decoded code iteratively maintaining its coding rate.
• the added CRC is for example adapted to restart a second iterative decoding, when the overall coding rate is not changed by the insertion of the CRC, thanks to a perforation, that is to say the arbitrary deletion of certain coded bits, the more often uniformly distributed in the coded message) performed in the decoded code, and the decoder is informed by sending for the perforated bit a value corresponding to "no information a priori", most often the LLR (Log Likelihood Ratio) null).
• the method comprises, for example, an existing CRC adapted to restart a second iterative decoding, starting from the same initialization vector (ie the weighted values provided by the demodulator) provided by the demodulator, with a more expensive decoding process, but closer to optimal.
• an existing CRC adapted to restart a second iterative decoding, starting from the same initialization vector (ie the weighted values provided by the demodulator) provided by the demodulator, with a more expensive decoding process, but closer to optimal.
• the method uses an already existing CRC to decide to restart a second iterative decoding, starting from the same initialization vector provided by the, but with a slower convergence decoding method using coefficients weights of the extrinsic metric weaker than during the first decoding.
• This second decoding can also use more iterations than the first attempt.
• Conventional iterative decoding provided by the conventional decoder has several iterations.
• the "decoder 7" provides the "CRC comparator 8" with the decoded information.
• the CRC comparator then chooses to validate the result proposed by the decoder, or chooses to ask the "inverse gradient operator 6" a new starting point for a new decoding of the code word to be decoded. This new decoding will then take several iterations.
• the second decoding is executed starting from an initialization vector different from the initialization vector supplied by the demodulator.
• the corrective vector representing the difference between the initial vector and the vector on which the new decoding is attempted, is provided by a noise of low amplitude (low temperature) according to a simulated annealing method.
• the second decoding can be executed starting from an initialization vector different from the initialization vector supplied by the demodulator, when the corrective vector representing the difference between the initial vector and the vector on which is attempted the new decoding is provided by the anti-gradient of the last half iterations.
• the anti-gradient consists of a hollow matrix where only the N_correct non-zero coefficients correspond to the least reliable decoded values during the last half-iteration, the penultimate (or the two before the last) half-iteration of the turbo code.
• N_correct is typically% * d- free, where djree is the distance of the code, and the matrix coefficients are each of opposite sign to the decoded value during the penultimate iteration
• the number of non-zero values retained in this anti-gradient is typically of the order of three quarters of the free code distance without this value being an obligation.
• the absolute value of non-zero points anti-gradient is typically equal to half of the average metric. Their sign is the opposite of the decision corresponding to the decoding.
• the second decoding is a combination of the two preceding methods, when the corrective vector representing the difference between the initial vector and the vector on which the new decoding is attempted is zero on the points having a metric. decoded, and a random noise (eg Gaussian) on the points corresponding to a poorer convergence, characterized by a weak decoded metric.
• a random noise eg Gaussian
• the second decoding is executed starting from an initialization vector different from the initialization vector supplied by the demodulator, when the corrective vector, representing the difference between the initial vector and the vector on which it is attempted.
• the new decoding is provided by a low amplitude noise corresponding to a low temperature in a simulated annealing method.
• the second decoding noise is injected as described above, on a reduced number of points at the input of the decoder.
• This reduced number of points is of the order of the free distance of the code, and the points correspond to the least reliable values coming from the penultimate or the second last iteration.
• the second decoding is for example of multi hypothesis type and consists in forcing a decoded bit to a given value, corresponding for example to the inverse of the value found by the decoding the first super iteration; the corrective vector then corresponds to a hard decision, and forces the metric (LLR) of the considered bit to the maximum value corresponding to an extremely safe bit.
• the second super iteration is repeated for different bit positions. The position of the inverted bit is taken a priori (as for a simulated annealing) or a posteriori thanks to the decoding of the first super-iteration (as for a gradient reverse).
• the new successive input vectors therefore consist of the vector initially received, but in which a value has been forced to bit 0 (thus to the minimum LLR) or bit 1, ie to the maximum metric.
• bit 0 thus to the minimum LLR
• bit 1 ie to the maximum metric.
• the position of the inverted bit (with respect to the decoded value) during the different successive super-iterations is random or follows the order of increasing reliability, ie starting with the least reliable of the decoded bits.
• the number of super iterations can be greater than two and the starting point of the new super iteration consisting of the initial point at the output of the demodulator, and a correction vector of size (number of non-zero coordinates) or weight (Euclidean norm ) variable depending on the iteration (typically increasing) or fixed.
• the number of super iterations is for example greater than 2 and the starting point of the new super iteration consists, iteratively so, of the entry point of the previous superiteration, and of a correction vector of fixed size or weight or no.
• the number of super-iterations is limited by a fixed maximum. Super iterations stop as soon as the CRC found matches.
• Figures 2 and 3 express the gain from a theoretical point of view; gain in BER (bit error rate) for a given signal-to-noise ratio.
• the first component of the terminal of the union is therefore substantially:
• the initial product code is shortened by one line.
• the new encoder adds 10 bits of CRC (for example). It should not add stuffing bits.
• the product code is 992 bits out of which no stuffing bit.
• the new encoder thus perforates 10 coded bits at specific positions. These positions are not the subject of the process.
• a practical example is given on a small code, consisting of 96 bits of data, 4 CRC bits, and protected by a parity-extended BCH (H) on the rows and columns and a shortcut of a row and a column.
• the code rate without CRC is 96/221.
• the code with CRC is therefore perforated with 4 bits.
• the input matrix is therefore the following, considering that the initial information is uniformly zero.
• the redundancy is in italics, the CRC is in bold, and the 4 punched positions are marked ?? , and correspond to a null metric.
• the output of the decoder is more likely than the initial message: it is verified that the sums of the metric products times decision (Log Likelihood Ratio) for the decoding corresponding to the initial message (all zero) is lower than that of the decoded message.
• the decoding then converges to the value of the initial message.
• the CRC is then obviously checked.

Landscapes

• Physics & Mathematics (AREA)
• Probability & Statistics with Applications (AREA)
• Engineering & Computer Science (AREA)
• Theoretical Computer Science (AREA)
• Error Detection And Correction (AREA)

Applications Claiming Priority (2)

Publications (1)

Family

ID=36128608

Family Applications (1)

Country Status (4)

Families Citing this family (12)

Family Cites Families (12)

• 2005
  • 2005-09-09 FR FR0509220A patent/FR2890806B1/fr not_active Expired - Fee Related
• 2006
  • 2006-09-11 WO PCT/EP2006/066227 patent/WO2007028834A2/fr active Application Filing
  • 2006-09-11 EP EP06793408A patent/EP1932241A2/de not_active Withdrawn
  • 2006-09-11 US US12/066,119 patent/US8332717B2/en not_active Expired - Fee Related

Non-Patent Citations (1)

Also Published As

Similar Documents

Legal Events