JP4599625B2 - Error correction decoder - Google Patents

Description

  The present invention relates to an error correction decoder, and more specifically, an RS-C serial concatenated code having a Reed-Solomon (RS) code as an outer code and a convolutional (C) code as an inner code. The present invention relates to an error correction decoder.

  In recent years, terrestrial digital television broadcasting has attracted attention. The standard of terrestrial digital television broadcasting adopted in Japan is an integrated services digital broadcasting for terrestrial (hereinafter also referred to as ISDB-T).

  ISDB-T employs many technologies for the purpose of improving frequency utilization efficiency and transmission efficiency and diversifying services. In broadcasting using ISDB-T, various services such as high-definition broadcasting, data broadcasting, and broadcasting for mobile terminals are being developed by utilizing such technical features.

  In recent years, it has been demanded to spread digital broadcast receivers in all fields where current analog broadcast receivers are used. In addition, the demand for receivers is diversified, and there is a strong demand for mounting receivers not only on fixed receivers but also on mobile objects such as automobiles and mobile phones.

  However, it is more difficult to receive on a mobile unit than on a fixed receiver, and this is still a problem.

  In mobile reception, the antenna height and size are limited in consideration of mobility, and in many cases, the transmission antenna and the reception antenna cannot be seen linearly (Non-Line Of Sight; NLOS). .

  The NLOS environment is a multipath propagation environment in which a transmission wave is diffused in a plurality of paths by reflection, diffraction, or the like and reaches a reception point. At the reception point, incoming waves from these multiple paths cause interference, and the received signal level fluctuates greatly.

  Further, in reception by a mobile unit, a frequency shift due to the Doppler effect occurs in the received wave, so that interference occurs between sub-channels of the OFDM symbol, and reception characteristics are greatly degraded.

  In such an unstable reception environment, it is difficult to always ensure good reception characteristics. In particular, in the error correction decoding method used in the conventional ISDB-T receiver, the probability of occurrence of a burst error caused by error correction at the time of error correction decoding in a low CNR (Carrier to Noise Power Ratio) environment tends to increase. .

  An excessive burst error exceeding the limit of the correction capability becomes a cause of deterioration of coding gain and leads to deterioration of reception quality. Therefore, in order to realize reception with less quality degradation, it is necessary to correct such a burst error and improve the coding gain.

  In ISDB-T, a Reed-Solomon (RS) -convolutional (C) series concatenated code is adopted as an error correction code.

  It is known that the RS-C serial concatenated code can improve the coding gain by devising the decoding method (Non-Patent Document 1). There are two specific methods: soft decision RS decoding and iterative decoding.

  The soft-decision RS decoding method is a technique for improving the decoding capability of an RS decoder by using likelihood information of a code sequence. In this method, if the quality of likelihood information is guaranteed, it is possible to correct a minute burst error.

  On the other hand, the iterative decoding method is a decoding method proposed for turbo codes. The turbo code adopts a recursive convolutional code, an interleave method, and an iterative decoding method, and has succeeded in achieving a gain close to the Shannon limit for the first time in history. As a result, it has been generally recognized that the iterative decoding method is effective in correcting a burst error that occurs during decoding, which is a problem of convolutional codes.

  Also in the ISDB-T receiver, it is considered that the coding gain can be improved by devising such a decoding method.

In overseas digital television broadcasting standards and digital video broadcasting (DVB), repeated decoding is performed by combining a soft decision convolutional decoder and a Reed-Solomon decoder having a soft input / output function, and a coding gain is obtained. A technique for improving the above (Non-patent Document 2) has already been proposed. Therefore, even in ISDB-T adopting a similar code structure, an improvement in coding gain can be expected by introducing an iterative decoding method.
O. Aitsab and R. Pyndiah: “Performance of Concatenated Reed-Solomon / Convolutional Codes with Iterative Decoding”, Proc. Of IEEE (1997) M. Lamarca, J. Sala-Alvarez and A. Martinez: “Iterative Decoding Algorithms for RS-Convolutional Concatenated Codes”, Proc. Of 3rd intl. Symposium on Turbo Codes and Related Topics (2003)

  However, the technique disclosed in Non-Patent Document 2 has a problem that each element decoder requires a soft input / output / soft decision function, and the decoder configuration becomes very complicated.

  The present invention has been made to solve the above problems, and an object thereof is to provide an error correction decoder capable of improving the coding gain with a simple configuration.

An error correction decoder according to the present invention is an error correction decoder that repeatedly executes error correction processing of received data, and includes a first decoder that performs convolutional decoding upon receiving received data, and a first decoder A first data rearrangement circuit that rearranges the data order array as a decoding result by a predetermined method, and a second that performs hard decision Reed-Solomon decoding based on the decoding result of the first decoder via the first data rearrangement circuit And the first data rearrangement circuit are paired, and the second data decoder receives the decoding result of the second decoder and rearranges the data order in the opposite manner to the first data rearrangement circuit. A second data rearrangement circuit that inputs to one decoder, and a delay element that delays received data. The second decoder includes a syndrome generation unit that generates a syndrome indicating the reliability of the decoded data output from the second decoder. The first decoder performs convolution decoding again on the received data input through the delay element according to the decoded data and syndrome output from the second decoder input through the second data rearrangement circuit. I do. The second decoder performs hard-decision Reed-Solomon decoding again based on the result of the second convolutional decoding performed by the first decoder via the first data rearrangement circuit.

  Preferably, the first decoder calculates a metric that is an index of the state transition probability based on the likelihood information included in the input received data, and calculates a maximum likelihood path that represents a likely state transition based on the metric. In the calculation and metrics calculation, if the data indicating the direction of state transition matches the decoded data output from the second decoder, a predetermined value based on the syndrome is added.

  Preferably, the first decoder performs convolutional decoding based on a soft decision Viterbi decoding algorithm.

Another error correction decoder according to the present invention is an error correction decoder that performs error correction processing of received data, and receives N-stage (N ≧ 2) errors that receive received data and output decoding results. Each of the N stages of error correction blocks includes a first decoder that receives input of received data and performs convolutional decoding, and a block decoder that performs block decoding based on a decoding result of the first decoder. 2 decoders. The i-stage (i ≧ 1) error correction block is provided between the first decoder and the second decoder, and the arrangement of the data order as the decoding result of the first decoder is performed in a predetermined manner. The first data rearrangement circuit to be rearranged and the first data rearrangement circuit are paired, and the data order arrangement is opposite to that of the first data rearrangement circuit in response to the decoding result of the second decoder. And a second data rearrangement circuit that is input to the first decoder of the (i + 1) -stage error correction block. The second decoder of the i-stage error correction block includes a syndrome generation unit that generates a syndrome indicating the reliability of the decoded data output from the second decoder. The first decoder included in the (i + 1) (N ≧ i + 1> 1) stage error correction block receives the i stage error correction input via the second data rearrangement circuit with respect to the input of the received data. Convolution decoding is performed according to the decoded data and syndrome output from the second decoder included in the block. Decoded data is output from the second decoder included in the final error correction block.

  Since the error correction decoder according to the present invention is configured to repeatedly perform convolutional decoding using the decoding result of the second decoder as a prior probability sequence in the first decoder, the coding gain can be improved. And the configuration using the syndrome as the prior probability sequence, the configuration of the second decoder can be easily configured, and the configuration of the error correction decoder as a whole can be easily configured. is there.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In the drawings, the same reference numerals indicate the same or corresponding parts.

  ISDB-T is established by the Association of Radio Industries and Businesses (ARIB), and its technical specifications are published as ARIB standards ("Transmission standard for terrestrial digital television broadcasting", Japan Association). Japan Radio Industry Association (2001)). ISDB-T employs many technologies for the purpose of improving frequency utilization efficiency and transmission efficiency and diversifying services.

FIG. 1 is a schematic functional block diagram of an ISDB-T transmitter 100.
A signal processing procedure in the transmitter 100 will be described with reference to FIG. The receiver performs signal processing opposite to that of the transmitter.

  The content encoded by MPEG-2 is multiplexed by the MPEG2 system 102 in frame units defined as a transport stream (TS).

  In the transmitter 100, this is re-multiplexed into a frame whose transport unit is a transport stream packet (TSP) suitable for signal processing.

  The remultiplexed TS is encoded by the shortened RS (204, 188) code in the RS encoding unit 104 to generate a packet called a transmission TSP.

  After generation of the transmission TSP, in the layer division 106, the TS is divided into a maximum of three layers according to the number of layers to be used, and energy diffusion 108a to 108c, byte interleaving 110a to 110c, and convolutional coding 112a to 112c for each layer. Each process is executed. The puncturing codes are used in the punctures 114a to 114c. After bit interleaving 116a to 116c, symbol mapping is performed in QAM mapping 118a to 118c. Then, in the hierarchy synthesis 120, the hierarchy divided after each processing is synthesized. The synthesized data is subjected to time interleaving 122 and frequency interleaving 124, and a pilot signal and a TMCC signal are added based on the configuration of the OFDM frame to generate an OFDM symbol in the OFDM frame configuration 126. The generated OFDM symbol is frequency-multiplexed by IFFT128, and further, a guard interval is added by a guard interval adding unit 130 and transmitted.

Next, features of main technologies adopted in ISDB-T will be described.
First, the segment will be described.

  In ISDB-T, a transmission frequency band of 5.6 MHz is divided into units called 13 segments each having a frequency band of 430 kHz. Furthermore, it is possible to divide the segment into a maximum of three layers and give different transmission parameters to each layer. As a result, various service operations are possible. For example, high-definition television broadcasting (High Definition TeleVision; HDTV) using all 13 segments, three-channel standard television broadcasting (Standard Definition TeleVision; SDTV) divided into three segments of 4 segments, SDTV and data broadcasting Combination broadcasting, segment broadcasting for mobile terminals, and the like are possible.

  Next, frequency multiplexing is a technique for improving frequency utilization efficiency and transmission efficiency, and has been adopted in many communication systems.

  In ISDB-T, an Orthogonal Frequency Division Multiplex (OFDM) system is adopted for frequency multiplexing.

  In the OFDM scheme in ISDB-T, there are three transmission modes (Modes 1-3) according to the number of subcarriers. The transmission parameters and transmission rates for one segment are shown in the following table.

Mode 1 has a minimum symbol length and a maximum subcarrier interval.
On the other hand, mode 3 has the maximum symbol length and the minimum subcarrier spacing. For this reason, Mode 1 is resistant to frequency offset and Doppler frequency shift, and it can be said that Mode 3 is resistant to multipath delay spread.

  Which mode is to be used can be selected by each broadcaster considering the propagation environment in the service area.

  In general, mode (Mode) 3 is selected in consideration of transmission efficiency and resistance to multipath delay spread.

  Further, by using OFDM, it is possible to construct a broadcasting network by a single frequency network (SFN).

  For this reason, the relay station can use the same frequency as the master station, and can effectively use the frequency of the overcrowded UHF band.

  The transmission using the OFDM scheme has a feature that the symbol length is longer than that of transmission using a single carrier, and is resistant to inter-symbol interference due to symbol delay spread.

  However, since it is impossible to completely prevent interference, the tolerance against interference is enhanced by adding an invalid symbol section called a guard interval to the head of the effective symbol.

  Further, since the signal of the guard interval is set to be the same as the signal at the end of the effective symbol, it can be used to establish symbol synchronization.

  The OFDM signal is delayed by an effective symbol time, the peak of signal correlation is detected, and the synchronization timing is determined.

The ratio of the guard interval length to the symbol length is called the guard interval ratio.
The guard interval ratio is desirably set to be equal to or longer than the maximum delay time of the delayed wave, but if it is longer than necessary, the transmission efficiency is lowered. Therefore, it is necessary to appropriately determine the guard interval ratio according to the propagation environment.

  In ISDB-T, the guard interval ratio is selected from 1/4, 1/8, 1/16, and 1/32 according to the propagation environment.

  The method of once fetching the input sequence into the memory and outputting it in a different order from the time of input is called the interleaving method, and the mechanism for realizing this is called the interleaver (interchanger) (data rearrangement circuit).

  By randomizing the sequence positional relationship by interleaving, burst errors that occur in a narrow range of received sequences can be converted into random errors that occur in a wide range of received sequences. This makes it possible to ensure a high coding gain even with a code having a small error correction capability.

FIG. 2 is a diagram illustrating an interleaver.
Interleavers are broadly divided into block types and convolution types.

  FIG. 2A is a diagram for explaining a block type interleaver. FIG. 2B is a diagram illustrating a convolutional interleaver.

  A typical block-type method is to configure a memory as a matrix, write an input sequence in the row direction, and read from the column direction.

  On the other hand, in the convolution type, an input sequence is divided into a plurality of shift registers, and is multiplexed at the time of output.

  In ISDB-T, a convolutional interleaver with higher real-time processability and non-uniformity is used.

  The ISDB-T interleaver is used in four stages. In order, the four axes of packet, symbol, time, and frequency are interlaced. Thereby, it has high burst error tolerance by crossing symbols over a wide band for a long time.

  Currently, ISDB-T uses two modulation systems, namely, quadrature phase shift keying (QPSK) and quadrature amplitude modulation (QAM).

  When QPSK is used for mapping, transmission efficiency is not high, but stable reception is possible compared to QAM. On the other hand, when QAM is used, high transmission efficiency can be ensured, but resistance to propagation path errors is lower than that of QPSK.

  For this reason, QPSK is used in broadcasting where the reception environment for portable devices such as one-segment broadcasting is assumed to be unstable, and there is a relatively stable reception environment for stationary bodies such as HDTV broadcasts such as stationary TVs. QAM is used in the assumed broadcast.

Next, ISDB-T employs an error correction code to improve propagation characteristics.
Using a Reed-Solomon (RS) code, which is a symbol error correction code, and a convolution (C) code, which is a random error correction code, a high error correction capability is secured by performing double encoding. .

  The error correction code is a code used for the purpose of correcting an error in an information sequence and restoring the information sequence before the error occurred. Currently, error correction codes are roughly classified into two types: block codes and convolutional codes.

  A block code divides data into symbol units, and generates a codeword by adding a parity symbol generated based on the data symbol.

  At the time of encoding, the convolutional code refers not only to the current input data but also to the past input data, and correlates the past input data and the generated codeword. In general, an algebraic decoding method is used for decoding a block code, and a statistical decoding method is used for decoding a convolutional code.

  Error correction code decoding methods can be classified into hard decision decoding methods and soft decision decoding methods according to how information sequences are handled. A method of handling a digital sequence of only 0 or 1 as an input is called a hard decision decoding method, and a method of handling analog weights such as likelihood information of the sequence as decoding auxiliary information in addition to the digital sequence is called a soft decision decoding method.

  The hard decision decoding method has a simple decoder configuration. On the other hand, the soft decision decoding method is said to be able to improve the coding gain by about 2-3 dB over the hard decision decoding method although the decoder configuration is complicated.

The RS code, which is a block code, is a symbol error correction code whose code word is GF (2 ⁸ ).

  In ISDB-T, a shortened RS (204,188) code obtained by shortening an RS (255,329) code having a code length N = 255 symbols and a data length K = 239 symbols by 51 symbols is used.

An encoding procedure will be described. First, 51 symbols of null symbols are added to a 188 symbol data word to generate a 239 symbol data word. Next, a 16-symbol parity word is added to the 239-symbol data word to generate a 255-symbol codeword. Then, after generating the code word, the null symbol is removed to obtain a (204,188) code. At this time, the following equation is used as the primitive polynomial of GF (2 ⁸ ).

  In the error correction decoding of the RS code, (NK) / 2 errors randomly generated in the code word can be corrected. Therefore, in the case of ISDB-T, errors of up to 8 symbols can be corrected.

  For decoding the RS code, erasure error correction decoding can be used in addition to normal error correction decoding.

  The erasure error correction decoding is a technique of performing decoding by designating e ≦ NK symbols, which are expected to have lost data, as erasures in advance.

  In this case, assuming that the code length is Ne, an error of e + (Ne-K) / 2 can be corrected at the maximum.

  In the conventional ISDB-T receiver, 8-symbol error correction decoding with the simplest decoder configuration is performed, but generalization that uses erasure error correction decoding to improve decoding capability and minimize the error rate after decoding. There is also a technique such as Generalized Minimum Distance (GMD) decoding (GDForney, Jr: “Generalized Minimum Distance Decoding” (1966)).

  When performing GMD decoding, information such as the likelihood of a code sequence is required as decoding auxiliary information. From the viewpoint of handling information sequences, it can be said that the conventional method is a hard decision decoding method and the GMD decoding method is a soft decision decoding method.

  The ISDB-T convolutional (C) code is called a punctured-convolutional (PC) code.

  In the PC code, a code word generated by a C encoder having a constant coding rate is punctured by a mask pattern, thereby making the coding rate variable.

FIG. 3 is a diagram for explaining the outline of the convolutional encoder.
The encoding procedure will be described with reference to FIG.

C encoding is performed using a shift register.
Here, as an example, a C encoder having a constraint length K = 7 and a coding rate of ½ used in ISDB-T will be described.

In ISDB-T, a convolutional encoder generates a mother code with a coding rate of 1/2.
Furthermore, the coding rate is controlled by deleting some bits of the mother code according to the mask pattern shown in the following table and outputting in order of transmission signal sequences corresponding to each coding rate.

  In general, a Viterbi (V) decoder based on the Viterbi (V) algorithm is used for decoding the C code.

  In a conventional ISDB-T receiver, it is common to use soft decision Viterbi decoding that uses two sequences of a code sequence and its likelihood sequence as an input sequence.

  Details of Viterbi decoding by the Viterbi algorithm in ISDB-T will be described later.

  In addition, the Viterbi algorithm includes an algorithm with a function for efficiently generating a soft output, and a soft-output Viterbi algorithm (hereinafter also referred to as SOVA).

  In the SOVA decoder, various application methods such as soft decision decoding of concatenated codes, iterative decoding, synchronization and delay detection have been devised.

  In order to improve the coding gain, it is necessary to use a code having a large code length and minimum distance. However, decoding is rapidly complicated in proportion to these two elements.

  Therefore, a method has been proposed in which a code having an effect of extending the minimum distance is configured by combining two codes having a relatively small code length. Typical examples are a product code and a concatenated code.

Product code, q source (n _1, k ₁₎ linear codes C ₁ and q source (n _2, k ₂₎ linear code C ₂ in combination composed q source _{(n 1, k 1, n} 2, k ₂ ) Linear code.

FIG. 4 is a diagram for explaining the product code.
Referring to FIG. 4, when the information source is constituted by a two-dimensional array of k ₁ × k ₂ , each column is encoded by C ₁ and each row is encoded by C ₂ . In general, a product code can be applied only to a linear code. This is because if a non-linear code is used, the parity symbol C ₂ (C ₁ ) may differ depending on the coding order.

  Since ISDB-T uses an RS code of a linear code and a C code of a nonlinear code, it is not classified into a product code in terms of structure.

The concatenated code is a code configured as shown in FIG. 5 by combining a q ^k (k ≧ 2) element code and a q element code.

FIG. 6 is a schematic block diagram illustrating encoding and decoding.
Referring to FIG. 6, outer encoder 200 performs outer encoding, inner encoder 205 performs inner encoding, and outputs the result to a communication path. Then, the inner coded data is decoded by the inner decoder 210, and the outer coded data is decoded by the outer decoder 215. That is, decoding is performed in reverse order with respect to the encoding order.

In the concatenated code, each element code is called an inner code and an outer code in order from the one closest to the communication path.
ISDB-T employs an RS-C serial concatenated code that uses an RS code as an outer code and a C code as an inner code.

  Although not shown here, an interleaver is arranged between the encoders for the purpose of randomizing the sequence.

  RS-C serial concatenated codes are used as codes that can realize characteristics closest to the Shannon limit before turbo codes, low-density parity check (LDPC) codes, etc. are put into practical use. It was.

  Further, it is known that RS-C concatenated codes can improve the coding gain by devising the decoding method.

  This includes a technique for performing soft-decision RS decoding and an iterative decoding method in which RS decoding output is used for re-decoding by a V decoder.

  For the purpose of applying these in the overseas digital broadcasting standard DVB, a soft output is generated by using a SOVA decoder for C decoding and this is used as decoding auxiliary information in the RS decoder, or an RS decoder A technique has been proposed in which a soft output function is added to the SVA decoder to generate a soft output to the SOVA decoder and perform iterative decoding.

Here, the configuration of a conventional ISDB-T receiver will be described.
FIG. 7 is a diagram for explaining the configuration of a conventional ISDB-T receiver.

  Referring to FIG. 7, in a conventional ISDB-T receiver, decoding by a soft decision Viterbi (V) -hard decision Reed-Solomon (RS) serial concatenated decoder having a simple decoder configuration is used.

A decoding procedure in a conventional receiver will be described.
First, the received signal (I, Q) is demapped by QAM demapping 300 to obtain a data symbol that is a demodulated bit stream and its likelihood information.

  This is depunctured by the depuncture 305 using the puncturing pattern table described above, and inputs X and Y to the soft decision Viterbi (V) decoder 310 are obtained.

  Then, the output of soft decision Viterbi (V) decoder 310 is deinterleaved by byte deinterleaver 315 and input to hard decision Reed-Solomon (RS) decoder 320. Finally, RS decoding is performed by the hard decision RS decoder 320 to obtain TSP as decoded data.

  Next, likelihood information generated at the time of demapping, decoding by a soft decision Viterbi (V) decoder, and decoding by a hard decision Reed-Solomon (RS) decoder will be described.

First, the generation of likelihood information is described.
In general, Euclidean distance and log likelihood ratio (LLR) are used for the likelihood information.

  The Euclidean distance represents a geometric distance from a nearby symbol when the received symbol is arranged on the geometric plane. The log likelihood ratio represents the probability that a received symbol is a transmitted symbol by a log ratio. In the digital circuit, the likelihood information is appropriately quantized and used.

An outline of likelihood information generation by LLR will be described.
When the transmission bit is d and the reception bit is r, the posterior probabilities of d = 0 and d = 1 are expressed as P (d = 0 | r) and P (d = 1 | r), respectively, and the LLR of r is It is defined by the following formula.

The generation of posterior probabilities is performed simultaneously with decoding, such as when performing statistical decoding.
Generally, it is generated as decoding auxiliary information at the time of decoding by a maximum a posteriori probability (MAP) algorithm, a maximum logarithmic MAP algorithm, SOVA, or the like.

  When the LLR is a positive value, the likelihood of d = 1 is high, and when the LLR is a negative value, the likelihood of d = 0 is high.

In addition, the closer the LLR value is to 0, the lower the likelihood of the symbol.
When a log likelihood ratio is used in a digital circuit, a sign is used to represent a positive / negative concept.

  1 bit is assigned to the code word, and several bits are assigned to the absolute value of the LLR. In this case, the larger the absolute value, the higher the likelihood of the codeword, and the closer the absolute value is to 0, the lower the likelihood of the codeword.

  Since the concept of posterior probability cannot be introduced at the time of demapping, the likelihood information that can be generated at the time of demapping by the ISDB-T receiver is the likelihood information based on the Euclidean distance.

  In a conventional ISDB-T receiver, several bits are assigned to such distance information and used as likelihood information.

Next, decoding by the Viterbi decoder will be described.
Decoding of the convolution (C) code is performed using the Viterbi algorithm as described above. In the Viterbi algorithm, the state transition probability of the internal state of the encoder is sequentially calculated based on the received word, and a plausible state transition path is obtained.

A plausible state transition path is called a maximum likelihood path, and a decoded output is obtained by following this path.
The procedure for following the maximum likelihood path is called traceback.

  The traceback length in decoding by the Viterbi algorithm is an important value that determines the likelihood of the output sequence.

  Further, since the memory for storing the path value is required as the traceback length increases, it can be said that there is a trade-off between the decoding performance in the Viterbi algorithm and the hardware scale.

In the computer simulation in this example, the traceback length is 192 bits.
In ISDB-T, two inputs of X and Y are required for the Viterbi (V) decoder corresponding to the output of the C encoder in FIG. 3 described above.

  Therefore, the code is represented as C, the likelihood information is represented as L, X is configured as (CxLx), and Y is configured as (CyLy).

The Viterbi (V) decoder calculates path metrics based on these two inputs.
The path metrics are values that serve as criteria for determining the maximum likelihood path in the Viterbi algorithm.

In the Viterbi algorithm, path metrics Mk (Q _i ) for the internal state Q _i of each register are calculated, and a state transition diagram called a trellis tree is constructed.

FIG. 8 is a diagram for explaining a trellis tree.
Referring to FIG. 8, l represents the constrained code constraint length, and k represents the Viterbi (V) decoder traceback length (number of state transitions).

In soft decision Viterbi (V) decoding, when the number of state transitions is k (1 ≦ k ≦ K), the path metric M _k (Q _i ) of the internal state Q _i is calculated as follows.

FIG. 9 is a diagram for explaining a trellis tree when the state transition paths from the states Q _i0 and Q _i1, which are the number of state transitions k−1, merge at the state Q _i at the state transition count k.

FIG. 10 is a diagram for explaining a calculation procedure of the path metrics M _k (Q _i ) in the state Q _i .

Referring to FIG. 10, here, Gx and Gy indicate generator polynomials for codewords Cx and Cy, respectively, and it is assumed that a codeword can be generated according to the rule C = G × Q. A path from Q _i0 and Q _i1 is called a branch.

Referring to FIG. 10, first, the path metrics M _k-1 (Q _i0 ) and M _k-1 (Q _i1 ) of the states Q _i0 and Q _i1 when the number of state transitions is k−1 are _represented by m ₀ and m _1. (Steps S0 and S10).

Then, Q _i0 × Gx is calculated according to the rule C = G × Q and compared with Cxk (step S1).

In step S1, Q _i0 × Gx is calculated according to the rule C = G × Q and compared with Cxk. If the result is equal, the process proceeds to step S2, and Lxk is added to m ₀ .

On the other hand, if Q _i0 × Gx is calculated according to the rule C = G × Q and compared with Cxk in step S1, if not equal, the process proceeds to step S3, and Lxk is subtracted from m ₀ .

Next, Q _i0 × Gy is calculated according to the rule C = G × Q, and is compared with Cyk (step S4).

In step S4, Q _i0 × Gy is calculated according to the rule C = G × Q and compared with Cyk. If the result is equal, the process proceeds to step S5, and Lyk is added to m ₀ .

On the other hand, if Q _i0 × Gy is calculated according to the rule C = G × Q and compared with Cyk in step S4, if not equal, the process proceeds to step S6 and Lyk is subtracted from m ₀ .

Similarly, next, Q _i1 × Gx is calculated according to the rule of C = G × Q and compared with Cxk (step S11).

In step S11, when Q _i1 × Gx is calculated according to the rule C = G × Q and compared with Cxk, if the result is equal, the process proceeds to step S12, and Lxk is added to m ₁ .

On the other hand, if Q _i1 × Gx is calculated according to the rule C = G × Q and compared with Cxk in step S11, if not equal, the process proceeds to step S13, and Lxk is subtracted from m ₁ .

Next, Q _i1 × Gy is calculated according to the rule C = G × Q and compared with Cyk (step S14).

In step S14, Q _i1 × Gy is calculated according to the rule C = G × Q and compared with Cyk. If the result is equal, the process proceeds to step S15, and Lyk is added to m ₁ .

On the other hand, if Q _i1 × Gy is calculated according to the rule C = G × Q and compared with Cyk in step S14, if not equal, the process proceeds to step S16, and Lyk is subtracted from m ₁ .

Then, m ₀ and m ₁ obtained based on the above result are compared (step S7).
In step S7, the larger one of the path metrics m ₀ and m ₁ is stored in the memory as the path metric of the surviving path. If they are equal, any of them may be stored in the memory. In this example, m ₀ is stored if they are equal.

Specifically, if m ₁ > m ₀ , m ₁ is set as the path metric M _k (Q _i ) (step S8). On the other hand, if m ₁ ≦ m ₀ , m ₀ is set as the path metric M _k (Q _i ) (step S9).

  At this time, the branch in which the metrics are stored in the path metrics becomes a surviving path.

  The path metrics calculation is repeated until the number of state transitions reaches the traceback length K, and the surviving paths are sequentially stored in the memory. The state with the largest path metric is the end of the state transition, and the maximum likelihood path can be determined by tracing back the surviving path until the end of the state transition.

Next, decoding by the Reed-Solomon (RS) decoder will be described.
The decoding of the RS code is performed by obtaining an error position polynomial σ (X) and an error evaluation polynomial ω (X) based on the syndrome polynomial S (X).

  The error position of the code word can be obtained from the error position polynomial σ (X), and the error size of the code word can be obtained from the error evaluation polynomial ω (X).

  The RS code in ISDB-T is a shortened code obtained by shortening the RS code of code length N = 255 symbols and data length K = 239 symbols by 51 symbols as described above. Therefore, the number of correctable errors τ is considered as τ = (N−K) / 2 = 8.

First, a syndrome polynomial is obtained from the received word.
Assuming that the received code polynomial is R (X), the coefficient of the kth-order term of the syndrome polynomial is obtained by substituting the root of Equation (3.2) into R (X):

  Next, an error position polynomial is obtained from the syndrome polynomial. The error locator polynomial can be obtained by solving the following equation using the so-called Peterson method, Berlekamp-Massey method, Euclid method, or the like.

However, t ≦ τ. When the error position of the received word is β ₀ , β ₁ ,..., Β _t−1 , these reciprocals are roots of σ (X). That is, the error position of the received word can be determined by obtaining the root of σ (X).

  When the syndrome polynomial and the error position polynomial are obtained, an error evaluation polynomial can be generated from the product of these.

  As described above, it is difficult to generate a high-quality likelihood sequence in a low CNR environment.

  However, the quality of the likelihood sequence is an important factor when performing soft decision decoding, and if the input likelihood sequence is low quality, the probability that a burst error will occur in the decoded sequence increases.

  Therefore, in order to ensure a higher coding gain, a technique for preventing such burst errors during decoding is required.

  In general, as a countermeasure against burst errors, the interleave method and the iterative decoding method are effective methods as described above. In 1993, the turbo code proposed by Berrou et al. Constructed a pseudo-long constraint by constructing a parallel concatenated code using two recursive convolutional (RSC) codes with a short constraint length and an interleaver. It is possible to obtain a long codeword.

  Another element that characterizes the turbo code is an iterative decoding method used as the decoding method.

  The iterative decoding method is a method of optimizing the entire sequence by using the decoding result obtained by one decoder in the other decoder. By applying this, it is possible to survive the erroneous survival when constructing the trellis tree. The probability of selecting a path can be reduced, and the probability of occurrence of a burst error during decoding can be reduced.

  The turbo code is a code that reaches a gain close to the Shannon limit for the first time in a practical range by using such an interleaver and iterative decoding method.

FIG. 11 is a diagram illustrating the configuration of a turbo encoder and a turbo decoder.
Referring to FIG. 11, input information input to turbo encoder 500 is encoded by first encoder 510 and output to modulator 525. Further, the input information is encoded by the second encoder 520 after being rearranged in the order of the data via the interleaver 505 and output to the modulator 525.

  The received information input to the turbo decoder 600 is decoded by a SISO (Soft Input Soft Out) first decoder 610 and further decoded by a SISO second decoder 620 via an interleaver 615. The output of the SISO second decoder 620 is fed back to the SISO first decoder 610 via the deinterleaver 625. By repeating this process, highly reliable information can be decoded.

Here, the iterative decoding method will be described.
When double encoding is performed using codes having no statistical correlation with each other, an error sequence generated in a received sequence is also a sequence having no correlation between codewords.

  For this reason, there is a possibility that the result obtained by one decoding is determined to be inappropriate at the time of the other decoding. By using this property and using one decoding result as decoding auxiliary information at the time of the other decoding, a decoding result with a lower error rate can be obtained.

  Further, the error rate can be gradually lowered by repeating the exchange of decoding auxiliary information between the decoders in a structure like the turbo code of FIG. Such a method is called an iterative decoding method.

  The iterative decoding method is a decoding method used in error correction codes that have attracted the most attention in recent years, such as turbo codes and LDPC codes.

  When iterative decoding is introduced, in order to exchange decoding auxiliary information with each other, the element decoder is required to have a soft input / soft output function.

The turbo code uses a technique of calculating and outputting the posterior probability by SOVA as decoding auxiliary information. The path metrics in the turbo code are calculated by adding the term L _Pk representing the prior probability to the equation (3.4) and the following equation.

In SOVA, when calculating the path metrics, storing a history of merged branch metrics difference value delta _k and survivor path. After decoding, a soft output is generated based on these.

  In the turbo code, the decoding process is repeated with the soft output of the SOVA as a prior probability, and the path metrics are optimized by updating the decoding process.

Hereinafter, the iterative decoding method disclosed in Non-Patent Document 2 will be described.
There are several problems in applying the iterative decoding method to RS-C serial concatenated codes.

  A decoding auxiliary information generation function, an interface between decoders for efficiently exchanging decoding auxiliary information, and the like.

  Since the above-described function does not exist in the conventional decoder described with reference to FIG. 7, it is necessary to add these functions.

  Non-Patent Document 2 proposes a method of applying an iterative decoding method in DVB of overseas standards.

  When the iterative decoding method is applied to the ISDB-T receiver by the method shown in Non-Patent Document 2, an interface for performing soft input / output to each element decoder and a function for performing soft decision by the RS decoder are required. It becomes.

  FIG. 12 is a diagram illustrating a soft decision Viterbi (SOVA) decoder-soft decision Reed-Solomon (RS) serial concatenated decoder.

  Referring to FIG. 12, in the SOVA decoder-soft-decision RS serial concatenated decoder, SOVA decoder 700 receives the received sequence (X, Y) and outputs hard-coded decoded data and soft output. Decoding auxiliary information is generated. Then, the soft decision RS decoder 720 receives the output result of the SOVA decoder 700 via the deinterleavers 705 and 710 and performs soft decision RS decoding.

  Soft decision RS decoder 720 generates decoded data and decoding auxiliary information and feeds back the generated decoding result to SOVA decoder 700 via interleavers 725 and 730 to perform iterative decoding.

FIG. 13 is a diagram for explaining the configuration of the SOVA decoder 700.
Referring to FIG. 13, the input to SOVA decoder 700 requires two sequences, a reception sequence (X, Y) and a prior probability sequence (Priori). Feedback from the RS decoder 720 is used for the prior probability sequence (Priori). Specifically, the prior probability sequence (Priori) includes a code sequence (RS-decoded) and a likelihood sequence (LLR _RS ) input via interleavers 725 and 730 that are outputs of the RS decoder 720. .

The above formula (4.1) is used for calculation of path metrics in SOVA. Here, assuming that the feedback from the soft-decision RS serial concatenated decoder is composed of the code word C _RS and its likelihood information L _RS , equation (4.1) is replaced as the following equation.

  Specifically, the SOVA decoder 700 stores a metric difference value, a path history, and the like at the time of storing path metrics and uses them for later generation of soft output.

  Since SOVA is an algorithm that statistically selects the path with the highest state transition probability, it can be said that the metric difference value indicates the LLR of the path. However, since the likelihood of the path before the time when the difference value is stored is indicated, it is necessary to refer to the path history at the same time. In SOVA, a posterior probability sequence is generated with reference to these values and used as a soft output.

FIG. 14 is a diagram illustrating a soft decision Reed-Solomon (RS) decoder 720.
Referring to FIG. 14, soft decision RS decoder 720 uses a hard output code sequence (C-decoded) and soft output likelihood sequence (LLRc) input via deinterleavers 705 and 710. Perform soft decision decoding.

Since decoding of the RS code is performed algebraically, GMD decoding or the like is used.
In soft-decision decoding of an RS code, decoding is performed by combining error correction and erasure error correction, and a codeword that is estimated to have the least error is determined as a transmission codeword by associating a codeword with a likelihood sequence. .

At this time, it is possible to obtain information such as the number of symbols determined to be erasure errors and whether the minimum distance condition is satisfied. In the method of Non-Patent Document 2, a code sequence (RS-decoded) that is a soft output and a likelihood sequence (LLR _RS ) that is a soft output are configured based on these pieces of information and fed back to the SOVA decoder 700. It is shown.

  Next, iterative decoding method using soft decision Viterbi decoder-hard decision Reed-Solomon (RS) serial concatenated decoder according to the embodiment of the present invention will be described.

  It is shown by the turbo code or the like that the iterative decoding method is effective for correcting burst errors during decoding. In the ISDB-T receiver, the iterative decoding method can be applied because of the structure of the code, but as described above, the iterative decoding method is not introduced in the conventional receiver.

  For this reason, it is considered that the coding gain can be improved in the ISDB-T receiver by applying the iterative decoding method.

  When the iterative decoding method is applied to the ISDB-T receiver, a decoding auxiliary information generation function, an interface between decoders for efficiently exchanging decoding auxiliary information, and the like are required.

  As described above, the SOVA decoder-soft decision RS serial concatenated decoder shown in Non-Patent Document 2 proposes a method for satisfying these in the overseas standard DVB by the SOVA decoder and the soft input / output / soft decision RS decoder. Has been. In this method, the posterior probability generated by the SOVA decoder is passed to the RS decoder, and soft decision decoding is also performed by the RS decoder. Further, a soft output is generated based on information such as the number of erasure corrections at the time of RS decoding, and iterative decoding is performed by feeding back to the SOVA decoder as decoding auxiliary information.

The problem with this technique is that the decoder configuration is complicated as described above.
This is because the amount of storage and the amount of calculation for soft output are necessary, and in particular, soft decision RS decoding has a large amount of calculation because the maximum likelihood sequence is estimated by repeating algebraic decoding.

  In this embodiment, therefore, an iterative decoding method using a soft-decision Viterbi decoder-hard-decision Reed-Solomon (RS) serial concatenated decoder that uses decoding success / failure information based on the RS code syndrome as decoding auxiliary information is proposed.

  In conventional hard-decision RS codes, decoding fails if the received sequence contains an amount of errors that exceeds the decoding capability of the codeword. For this reason, a very high likelihood is guaranteed in a sequence that has been successfully decoded, and conversely, a sequence that has failed in decoding has a very large number of errors and thus has a low likelihood. Therefore, it is considered that this decoding success / failure information can be used as high-quality decoding auxiliary information by feeding back to the soft decision Viterbi (V) decoder.

  FIG. 15 is a schematic diagram illustrating a configuration of a soft decision Viterbi decoder-hard decision Reed-Solomon (RS) serial concatenated decoder according to an embodiment of the present invention.

  Referring to FIG. 15, soft decision Viterbi decoder-hard decision Reed-Solomon (RS) serial concatenated decoder 1 includes delay element 12, soft decision Viterbi decoder 5, deinterleaver 15, and hard decision Reed-Solomon. It includes a decoder 10 and interleavers 20a and 20b.

  The soft decision Viterbi decoder 5 receives the received sequence (X, Y) and generates decoded data TSP that is a hard output. The hard decision RS decoder 10 receives the output result of the soft decision Viterbi decoder 5 via the deinterleaver 15 and performs hard decision RS decoding. The interleaver 20 rearranges the input data order array according to a predetermined method and outputs the result. Further, the deinterleaver 15 is paired with the interleaver 20, and outputs the data order that is input after the data order is rearranged to the original data order by a method opposite to the predetermined method.

  The hard decision RS decoder 10 generates decoded data and auxiliary decoding information, and feeds back the generated decoding result to the soft decision Viterbi decoder 5 via the interleavers 20a and 20b. Thereby, iterative decoding is executed.

  Specifically, in soft decision Viterbi decoder 5, again based on the reception sequence (X, Y) delayed by delay element 12 and the decoded data and decoding auxiliary information generated by hard decision RS decoder 10, soft decision Viterbi decoder 5 again. Performs determination Viterbi decoding. The hard decision Reed-Solomon decoder 10 receives the decoding result input via the deinterleaver 15, performs hard decision RS decoding, and outputs the decoded data obtained thereby.

  As is clear from the configuration of FIG. 12, in the configuration according to the present embodiment, the likelihood information from the convolution (C) decoder, which is a soft-decision Viterbi decoder, to the block decoder, which is a hard-decision RS decoder. Is not necessary.

  Therefore, the soft output function of the convolution (C) decoder, which is a soft decision Viterbi decoder, the soft input function of the block decoder, which is an RS decoder, and the likelihood information generation function in each decoder can be omitted or simplified. Yes, the decoder can be configured simply.

FIG. 16 is a schematic diagram illustrating the configuration of the soft decision Viterbi decoder 5.
Referring to FIG. 16, soft decision Viterbi (V) decoder 5 receives the same two sequences of reception sequence (X, Y) and prior probability sequence (RS-decoded, S) as those used in SOVA. Use a decoder that

  Here, as described above, the input of the hard decision RS decoder 10 does not require the input of decoding auxiliary information, so the soft decision Viterbi (V) decoder 5 is realized with a simple configuration without the need for providing a soft output function. Is possible.

FIG. 17 is a schematic diagram illustrating the configuration of the hard decision Reed-Solomon (RS) decoder 10.
Referring to FIG. 17, hard decision RS decoder 10 includes a hard decision RS decoding unit 12 and a syndrome generator 14.

  The hard decision RS decoding unit 12 receives the input data TSP generated via the deinterleaver 15 generated by the soft decision Viterbi decoder 5 and outputs a hard decision output (RS-decoded) decoding result.

  The syndrome generator 14 calculates and outputs decoding auxiliary information (syndrome) S based on the hard decision output (RS-decoded). The RS code detects errors by calculating the syndrome at the time of decoding.

  The syndrome polynomial is expressed by equation (3.7) based on the coefficient of equation (3.6). Here, assuming that an error represented by the error polynomial e (X) has occurred in the received code polynomial R (X), R (X) can be expressed by the following equation using the transmission code polynomial T (X). .

  As described above, when some error occurs in the code word, since e (X) ≠ 0, s (x) ≠ 0.

  On the other hand, when there is no error in the code word, since e (X) = 0, S (x) = 0. That is, it can be assumed that there is no error in the codeword when the syndrome polynomial is 0, and that there is an error in the codeword when it is not 0. However, since an error has occurred but the syndrome polynomial may be 0, it cannot always be said that there is no error. In this case, since an error cannot be detected due to the structure of the code, an error is missed.

  Since the structure of the RS code itself does not change even after decoding, the presence or absence of an error can be detected by recalculating the syndrome after decoding, and the success or failure of decoding can be confirmed.

  In the present embodiment, such a syndrome polynomial is calculated even after decoding, and the success or failure of decoding is confirmed by checking whether there is an error in the decoded packet. At this time, binary data of 1 is assigned to the decoding result and 0 is assigned to the decoding failure, and is fed back to the soft decision Viterbi (V) decoder 5 as decoding auxiliary information (syndrome) S.

  As is clear from the configuration of FIG. 12, in the configuration according to the present embodiment, the hard-decision RS decoder 10 can omit or simplify the soft input function and the likelihood information generation function. It is possible to simply configure the decoder.

  Hard-decision RS decoder 10 according to the present embodiment does not generate likelihood information of prior probability in the actual sense, and uses syndrome S as an alternative value of prior probability.

  That is, the path metrics are calculated by substituting Equation (4.1) as follows:

However, usually, several bits are assigned to the likelihood information of the received symbol in the system, and the value of the decoding auxiliary information (syndrome) S is smaller than the value of L _Rk . If the value of the decoding auxiliary information is too small, an appropriate effect cannot be obtained. Therefore, an appropriate weight W is assigned to S as shown in the following equation.

  FIG. 18 is a diagram illustrating a path metric calculation procedure according to the embodiment of the present invention.

  Referring to FIG. 18, the path metrics calculation procedure according to the embodiment of the present invention is obtained by adding a flow relating to decoding auxiliary information to path metrics in addition to the calculation procedure of FIG. Specifically, Step S20 to Step S25 are further added, and the other points are the same, so the same points will not be repeated.

In step S4, Q _i0 × Gy is calculated according to the rule of C = G × Q and compared with Cyk. If they are equal, the process proceeds to step S5, and Lyk is added to m ₀ .

On the other hand, if Q _i0 × Gy is calculated according to the rule C = G × Q and compared with Cyk in step S4, if not equal, the process proceeds to step S6 and Lyk is subtracted from m ₀ .

Next, MSB (Q _i ) is compared with C _RS k (step S20). Here, MSB (Q _i ) indicates the most significant bit (input bit in the convolutional code) of the internal state Q _i of the shift register. C _RS indicates the RS codeword fed back.

That is, in step S20, the RS code bit fed back is compared with the MSB (Q _i ) corresponding to the input bit of the convolutional code corresponding thereto. Specifically, the data 0 or 1 indicating the direction of the state transition of the internal state Q _i in the trellis tree configured inside the soft decision Viterbi decoder 5 is compared with the RS code bit fed back.

  If the received word and the RS code word match in step S20, the soft decision auxiliary data (SkW) of the matched sequence is added to the path metric in the previous state (step S21). Here, the decoding assistance data corresponds to syndrome S × weight W.

  On the other hand, if the received word and the RS codeword do not match in step S20, the decoding auxiliary data (SkW) is subtracted from the previous path metrics (step S22).

Similarly, in step S14, C = G × Q results compared to Cyk calculate the Q _i1 × Gy in rule, if they are equal, the process proceeds to step S15, adds Lyk to m _1.

On the other hand, if Q _i1 × Gy is calculated according to the rule C = G × Q and compared with Cyk in step S14, if not equal, the process proceeds to step S16, and Lyk is subtracted from m ₁ .

Next, MSB (Q _i ) is compared with C _RS k (step S23). Here, MSB (Q _i ) indicates the most significant bit (input side bit) of the internal state Q _i of the shift register. C _RS indicates the RS codeword fed back.

That is, in step S23, the RS code bit fed back is compared with the MSB (Q _i ) corresponding to the input bit of the convolutional code corresponding thereto. Specifically, the data 0 or 1 indicating the direction of the state transition of the internal state Q _i in the trellis tree configured inside the soft decision Viterbi decoder 5 is compared with the RS code bit fed back.

  In step S23, when the received word and the RS code word match, the decoded auxiliary data (SkW) of the matched sequence is added to the path metrics in the previous state (step S24). Here, the decoding assistance data corresponds to syndrome S × weight W.

  On the other hand, if the received word and the RS code word do not match in step S23, the auxiliary decoding data (SkW) is subtracted from the previous path metrics (step S25).

Then, m ₀ and m ₁ obtained based on the above result are compared (step S7).
In step S7, the larger _one of m ₀ and m ₁ is stored in the memory as the surviving path metrics. If they are equal, any of them may be stored in the memory. In this example, m ₀ is stored if they are equal.

Specifically, if m ₁ > m ₀ , m ₁ is set as the path metric M _k (Q _i ) (step S8). On the other hand, if m ₁ ≦ m ₀ , m ₀ is set as the path metric M _k (Q _i ) (step S9).

  At this time, the branch in which the metrics are stored in the path metrics becomes a surviving path.

  The path metrics calculation is repeated until the number of state transitions reaches the traceback length K, and the surviving paths are sequentially stored in the memory. The state with the largest path metric is the end of the state transition, and the maximum likelihood path can be determined by tracing back the surviving path until the end of the state transition.

  The configuration according to the present embodiment includes a soft output function of a convolution (C) decoder that is a soft decision Viterbi decoder, a soft input function of a block decoder that is an RS decoder, and generation of likelihood information in each decoder. The function can be omitted or simplified, and the configuration of each decoder can be designed easily.

  Also, a deinterleaver may be provided corresponding to one input sequence between the soft decision Viterbi decoder 5 and the hard decision Reed-Solomon decoder 10, and the number of deinterleavers can be reduced.

  Since the configuration is simple, there is a possibility that the present embodiment can increase the number of repetitions in hardware having the same area.

  Next, a comparison between a case where iterative decoding is performed by the soft decision Viterbi decoder-hard decision Reed-Solomon (RS) serial concatenated decoder according to the embodiment of the present invention and a case where iterative decoding is not performed will be described.

First, the simulation environment will be described.
In general, there are many sources of noise added to a radio signal inside a receiver.

  At the receiver, the sum of signals generated from these many noise sources is received as noise. Such noise approximately follows a Gaussian distribution by the central limit theorem when the number of samples is large. Further, it can be approximated that the frequency characteristics of the probability distribution are uniform. For this reason, an additive white Gaussian noise (AWGN) model is generally used as the system internal noise model.

  Assume that a mobile body such as an automobile receives a broadcast. Since the antenna height is about 1-2 m, it can be generally regarded as an NLOS environment.

  In this case, it is known that an incoming wave becomes a multipath propagation environment in which a plurality of points in the vicinity of the mobile station reach the reception point, and the distribution of fading fluctuation is a Rayleigh distribution.

  Also, if there is no path length difference in the arriving wave and the delay time difference can be ignored, flat fading with uniform frequency characteristics results, the path length difference of the arriving wave is large, and the delay time difference cannot be ignored. Then, it becomes frequency selective fading with a distorted frequency characteristic.

  Since ISDB-T uses multicarrier symbols multiplexed by OFDM for transmission, the delay time difference cannot be ignored. For this reason, a multipath Rayleigh distribution model is used as the fading model. In particular, when receiving in an automobile, it is expected that there are many paths, so a 6-wave Rayleigh distributed fading noise model is used.

  Next, consider the case where reception is performed outdoors by a mobile object. In reception by a mobile object, a frequency shift due to the Doppler effect occurs. It is assumed that the i-th elementary wave arrives from the direction of the angle θi with respect to the traveling direction of the moving body. At this time, if the velocity of the moving body is v, the speed of light is c, and the carrier frequency is fc, this incoming wave undergoes a Doppler frequency shift given by the following equation.

  Next, the simulation specifications are shown in the following table.

  The propagation environment is set on the assumption that the receiver is an automobile traveling at about 60 km / h. The system internal noise model is an AWGN model. The fading noise model is a six-wave Rayleigh distribution noise model because radio waves arrive from other directions during traveling. The maximum Doppler frequency is 40 Hz considering that the carrier frequency is about 470-770 MHz.

  Assuming general HDTV, the ISDB-T transmission parameters are set to transmission mode mode (Mode) 3, guard interval ratio 1/8, time interleave length 2, modulation scheme 64QAM, and coding rate 3/4. For simplicity, full segment transmission without partial reception is used. The sample was 100 OFDM frames. When the mode is Mode 3 and the guard interval ratio is 1/8, the frame length is about 231.3 ms, so the sample length is about 23.13 s.

  For the soft decision Viterbi (V) decoder, the traceback length is 192 bits, and the input of the received sequence is quantized into “code 1 bit + likelihood information 3 bits”. One is used.

  As an evaluation index, a bit error rate (BER) and a packet error rate (PER) are used. The packet here indicates 204 bytes of transmission TSP.

In ISDB-T, BER after convolution (C) decoding is used as a receiver evaluation criterion, and it is required to satisfy BER ≦ 2 × 10 ⁻⁴ . This is because if BER ≦ 2 × 10 ⁻⁴ is satisfied after convolution (C) decoding, it is considered that the error becomes pseudo error-free after RS decoding.

Since the BER after RS decoding is not normally discussed, there is no clear evaluation criterion, but it is generally said that there is no practical problem if BER ≦ 1 × 10 ⁻¹¹ is satisfied. Evaluation criteria.

Similarly, there is no clear evaluation standard for PER.
However, in general, if PER ≦ 1 × 10 ⁻⁹ is satisfied, it is assumed that the environment is such that the content can be received almost without deterioration.

Further, if PER ≦ 1 × 10 ⁻⁶ is satisfied, it is said that the content can be understood even if deterioration is observed.

In this embodiment, it is considered appropriate to use these values as evaluation criteria. However, in the computer simulation, the calculation time is enormous, so the BER is about 1 × 10 ⁻⁶ and the PER is 1 ×. It can only measure up to about 10 ^-4 . Therefore, in this embodiment, the evaluation is performed with a CNR that achieves BER ≦ 1 × 10 ⁻⁶ and PER ≦ 1 × 10 ⁻⁴ .

  The comparative evaluation target of the present embodiment is decoding by a soft-decision Viterbi decoder-hard-decision Reed-Solomon (RS) serial concatenated decoder when the iteration is performed and when it is not performed.

  In this embodiment, when the number of iterations is set to 0, it can be said that it is equivalent to the conventional method due to the structure of the decoder.

  In each graph, the words, iteration and weight in the legend indicate the number of repeated decoding and the size of the assigned weight W in the present embodiment, respectively.

  Next, the effect of improving the coding gain according to the present embodiment will be verified under the influence of characteristic fading noise under the influence of fading noise.

  Assuming a 6-wave Rayleigh distributed fading noise channel as a propagation path, CNR vs. BER characteristics and CNR vs. PER characteristics were measured.

  FIG. 19 is a diagram illustrating CNR vs. BER characteristics after a hard decision RS decoder under the influence of fading noise.

Referring to FIG. 19, BER ≦ 1 × 10 ⁻⁶ is satisfied at CNR = 25.8 dB in the conventional method, and when iterative decoding is performed once according to this embodiment, CNR = 24.6 dB. At this time, in the 6-wave Rayleigh distributed fading noise channel, after RS decoding, an improvement in coding gain of about 1.2 dB can be expected as the CNR by the proposed method.

  FIG. 20 is a diagram illustrating CNR vs. PER characteristics after a hard decision RS decoder under the influence of fading noise.

Referring to FIG. 20, PER ≦ 1 × 10 ⁻⁴ is satisfied at CNR = 25.6 dB in the conventional method, and when iterative decoding is performed once according to the present embodiment, CNR = 24.4 dB. At this time, in the 6-wave Rayleigh distributed fading noise channel, after RS decoding, an improvement in coding gain of about 1.2 dB can be expected as the CNR by the proposed method.

  In this example, although iterative decoding has been described once, it is possible to further improve the coding gain by performing iterative decoding more than once.

  By using the proposed method as described above, it is expected that the ISDB-T receiver can improve the coding gain and realize more stable reception. Further, in this embodiment, the study was performed mainly on the premise that the ISDB-T receiver is mounted on the automobile, but the present invention can be applied to other fields because of the simplicity of the decoder configuration.

  FIG. 21 is a diagram illustrating a configuration of a soft decision Viterbi decoder-hard decision Reed-Solomon (RS) serial concatenated decoder according to a modification of the embodiment of the present invention.

  Referring to FIG. 21, the soft decision Viterbi decoder-hard decision Reed-Solomon (RS) serial concatenated decoder according to the modification of the embodiment of the present invention has a configuration in which a plurality of (multiple stages) decoders are stacked. Is different. Specifically, a plurality of stages of soft decision Viterbi (V) decoders 5-1 to 5-n and hard decision Reed-Solomon (RS) decoders 10-1 to 10-n are provided. In the configuration of the next stage, information is input to the soft decision Viterbi (V) decoders 5-2,.

  The decoding result in the soft decision Viterbi decoder 5 is output to the hard decision Reed-Solomon (RS) decoder 10 via the deinterleaver 15. The decoding result of the hard decision Reed-Solomon (RS) decoder 10 is fed back to the next-stage soft decision Viterbi (V) decoder 5 through the interleavers 20a and 20b.

  For example, the decoding result of the soft decision Viterbi decoder 5-1 is output to the hard decision Reed-Solomon (RS) decoder 10-1 via the deinterleaver 15-1. The decoding result of the hard decision Reed-Solomon (RS) decoder 10-1 is fed back to the next-stage soft decision Viterbi (V) decoder 5-2 through the interleavers 20a-1 and 20b-1.

Then, decoded data is output from the hard decision Reed-Solomon decoder 10-n at the final stage.
Since the processing method is the same as that described in FIG. 12, detailed description thereof will not be repeated.

  For example, considering the case of a two-stage configuration, as compared to the configuration described in FIG. 12, a two-stage soft decision Viterbi (V) decoder 5-1 to 5-2 as shown in FIG. And the hard-decision Reed-Solomon (RS) decoders 10-1 to 10-2 are provided, the processing for the next input sequence is performed before the processing for the first input sequence is completed. Since processing can be performed by the decoder 5-1 and the hard decision Reed-Solomon (RS) decoder 10-1, error correction processing can be executed at high speed.

  In this example, ISDB-T has been described as an example, but the present invention is not limited to this, and can be applied to data broadcasting of other standards.

  The embodiment disclosed this time should be considered as illustrative in all points and not restrictive. The scope of the present invention is defined by the terms of the claims, rather than the description above, and is intended to include any modifications within the scope and meaning equivalent to the terms of the claims.

1 is a schematic functional block diagram of an ISDB-T transmitter 100. FIG. It is a figure explaining an interleaver. It is a figure explaining the outline of a convolutional encoder. It is a figure explaining a product code. It is a diagram illustrating a constructed concatenated code by combining a q ^{k (k} ≧ 2) source code and q-ary symbols. It is a schematic block diagram explaining encoding and decoding. It is a figure explaining the structure of the conventional ISDB-T receiver. It is a figure explaining a trellis tree. It is a figure explaining the trellis tree when the state transition path from the states Q _i0 and Q _i1 which are the state transition times k−1 in the state transition number k is merged in the state Q _i . It is a diagram illustrating a calculation procedure of the state Q _i path metrics in M _k (Q _i). It is a figure explaining the structure of a turbo encoder and a turbo decoder. FIG. 5 is a diagram illustrating a soft decision Viterbi (SOVA) decoder-soft decision Reed-Solomon (RS) serial concatenated decoder. FIG. 3 is a diagram for describing a configuration of a SOVA decoder 700. FIG. 6 illustrates a soft decision Reed-Solomon (RS) decoder 720. It is the schematic explaining the structure of the soft decision Viterbi decoder-hard decision Reed-Solomon (RS) serial connection decoder according to embodiment of this invention. 3 is a schematic diagram illustrating a configuration of a soft decision Viterbi decoder 5. FIG. 2 is a schematic diagram illustrating a configuration of a hard decision Reed-Solomon (RS) decoder 10. FIG. It is a figure explaining the calculation procedure of path metrics according to an embodiment of the invention. It is a figure explaining the CNR versus BER characteristic after the hard decision RS decoder under the influence of fading noise. It is a figure explaining the CNR versus PER characteristic after the hard decision RS decoder under the influence of fading noise. It is a figure explaining the structure of the soft decision Viterbi decoder-hard decision Reed-Solomon (RS) serial connection decoder according to the modification of embodiment of this invention.

Explanation of symbols

  5,310,700 Soft decision Viterbi decoder, 10,320 Hard decision Reed-Solomon decoder, 12,740 delay element, 15,705,710 deinterleaver, 20,725,730 interleaver, 100 ISDB-T transmission , 102 MPEG2 system, 104 RS encoder, 106 layer division, 108a-108c energy spreading, 110a-110c byte interleaving, 112a-112c convolutional coding, 114a-114c puncture, 116a-116c bit interleaving, 118a-118c QAM mapping 120 layer synthesis, 122 time interleaving, 124 frequency interleaving, 126 OFDM frame structure, 128 IFFT, 130 guard interval adding unit, 200 outer encoder 205 in encoder 210 within decoder 215 outer decoder, 300 QAM demapping, 305 depuncturing, 315 bytes deinterleaving, 500 turbo encoder 600 turbo decoder 720 soft-decision RS decoder.

Claims (4)

An error correction decoder that repeatedly executes error correction processing of received data,
A first decoder that receives the received data and performs convolutional decoding;
A first data rearrangement circuit for rearranging an array of data order which is a decoding result of the first decoder by a predetermined method;
A second decoder that performs hard decision Reed-Solomon decoding based on a decoding result of the first decoder via the first data rearrangement circuit;
The first data rearrangement circuit is paired and receives the decoding result of the second decoder, and rearranges the data order in the opposite manner to the first data rearrangement circuit. A second data rearrangement circuit that inputs to the decoder of
A delay element for delaying the received data,
The second decoder includes a syndrome generation unit that generates a syndrome indicating reliability of decoded data output from the second decoder;
The first decoder follows the decoded data and syndrome output from the second decoder input via the second data rearrangement circuit with respect to the received data input via the delay element. Perform convolution decoding again,
The error correction decoder, wherein the second decoder performs hard-decision Reed-Solomon decoding again based on the result of the second convolutional decoding performed by the first decoder via the first data rearrangement circuit. The first decoder calculates a metric that is an indicator of a state transition probability based on likelihood information included in input received data, and calculates a maximum likelihood path that represents a likely state transition based on the metric. And
2. The error correction according to claim 1, wherein, in the calculation of the metrics, a predetermined value based on a syndrome is added when data indicating a direction of state transition coincides with decoded data output from the second decoder. Decoder. The error correction decoder according to claim 1 or 2 , wherein the first decoder performs convolutional decoding based on a soft decision Viterbi decoding algorithm. An error correction decoder that performs error correction processing of received data,
An N-stage (N ≧ 2) error correction block that receives the received data and outputs a decoding result;
Each of the N stages of error correction blocks is:
A first decoder that receives the input of received data and performs convolutional decoding;
A second decoder that performs block decoding based on a decoding result of the first decoder;
The error correction block of i stages (i ≧ 1) is
A first data rearrangement circuit that is provided between the first decoder and the second decoder, and rearranges an array of data order that is a decoding result of the first decoder in a predetermined manner;
A pair of the first data rearrangement circuit is configured, and upon receiving the decoding result of the second decoder, the order of data is rearranged in the opposite manner to the first data rearrangement circuit, and the (i + 1) And a second data recombination circuit that inputs to the first decoder of the stage error correction block;
The second decoder of the i-stage error correction block includes a syndrome generation unit that generates a syndrome indicating reliability of decoded data output from the second decoder;
The first decoder included in the (i + 1) (N ≧ i + 1> 1) stage error correction block has an i-stage input to the received data through the second data recombination circuit. Mistake
Performing convolutional decoding according to the decoded data and syndrome output from the second decoder included in the correction block,
An error correction decoder in which decoded data is output from a second decoder included in an error correction block at the final stage .

Images