JP4025226B2 - Error correction transmission device - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to an error correction transmission apparatus, and more particularly to an error correction transmission apparatus that performs soft-decision Viterbi decoding by transmitting a convolutional code using a multilevel modulation scheme.
[0002]
[Prior art]
In digital communication, transmission data subjected to error correction coding is converted into transmission symbols, modulated and transmitted. The receiver decodes the error correction code to obtain information data. In various digital communications, convolutional codes that are advantageous for burst errors are used. The Viterbi algorithm is used as a method for decoding the convolutional code. The Viterbi algorithm is a decoding method that efficiently performs maximum likelihood decoding using an iterative structure of a convolutional code. Basically, the bit string that can be taken at each time point on the trellis diagram and the received bit string (or the modulation symbol corresponding to the bit string) are compared, and the path (maximum likelihood path) with the largest likelihood function is selected. The transmission signal is estimated without performing a brute force method with all of the code sequence patterns existing in (1).
[0003]
Decoding methods using the Viterbi algorithm include hard decision and soft decision. Hard decision decoding uses a Hamming distance as a likelihood function, and is a method of determining a path after determining a received signal as a binary value. On the other hand, soft decision decoding uses the Euclidean distance as a likelihood function, and is a method for determining a path by treating a received signal as an analog value. Since soft decision decoding makes a decision based on the analog value of the received signal, it has higher error correction capability than hard decision decoding. Refer to Non-Patent Document 1 for a decoding method of an error correction code using a conventional soft decision Viterbi algorithm.
[0004]
As a multi-level modulation system with high frequency utilization efficiency, there is a multi-level QAM system (QPSK, 16QAM, 64QAM, etc.) in which code identification points are arranged in a lattice pattern by a linear modulation system, and is used in many systems. In the linear modulation method, it is possible to independently use two channels of a carrier in-phase component (Ich) and a carrier quadrature component (Qch). Therefore, the modulation symbol can be represented by a complex number such as (I + jQ). Here, (I + jQ) is referred to as a modulation symbol (or simply a symbol), and when referring to each of I and Q, it is referred to as a modulation sub-symbol (or simply a sub-symbol).
[0005]
By the way, the soft decision Viterbi algorithm cannot be used unconditionally when the convolutional code is transmitted by the multi-level modulation method. The reason is that the transmission unit (subsymbol) of the multi-level modulation method may be 1 bit or multiple bits, but amplitude information at the time of demodulation can be obtained only in the transmission unit. This is because when the number of bits is 2 bits or more, it is impossible to know amplitude information in 1-bit units.
[0006]
A conventional soft decision Viterbi decoding method will be described with reference to the trellis diagram shown in FIG. Taking the coding rate R = 1/2 and the modulation scheme as QPSK as an example, there are two state transition paths for one step of the encoder. This is the case where the input bit value is 0 and 1. In the Viterbi algorithm, in each case, in order to efficiently compare the transmission sequence and the reception sequence, the transmission symbol corresponding to the branch joining each state at each time point is compared with the reception amplitude, and the Euclidean distance is compared. Is obtained and added to the path metric of the state one point before, and the branch with the smaller value is selected to determine the paths and path metrics for all states (S0 to S15). When attention is paid to one surviving path, the result is as shown in FIG. Here, M is the value of the path memory. Such processing is executed from time point 1 to time point N × R (N is the code length and R is the coding rate). When the convolutional code is terminated to the state 0 by the tail bit, since the last state is S0, the state transition is traced in reverse from the S0 at the time N × R in the direction of the time 0 according to the value of the path memory. When this is done (this is called traceback), a decoded bit string can be obtained.
[0007]
FIG. 15 shows a configuration of a conventional radio device that performs soft-decision Viterbi decoding by transmitting a convolutional code using a multi-level modulation method. Here, since a bit interleaver is arranged immediately after the convolutional encoder on the transmission side, generally, if the modulation method is not 1-bit / sub-symbol QPSK or the like on the reception side, the original arrangement is changed by the sub-symbol deinterleaver. I can't get it back.
[0008]
[Non-Patent Document 1]
Hideki Imai: "Code Theory" (The Institute of Electronics and Communication Engineers, published in 1997) pp.280-312.
[0009]
[Problems to be solved by the invention]
In the conventional error correction transmission system, the soft decision can be applied only when the sub-symbol of multilevel modulation has a 1-bit configuration (1 bit / sub-symbol), and there is a problem that the error correction capability cannot be sufficiently increased. . The reason is as follows.
[0010]
2 for the number of multi-level modulation symbols^2q, Sub-symbols I and Q are represented by q bits. Incidentally, q = 1 for 4 values, q = 2 for 16 values, q = 3 for 64 values, and q = 4 for 256 values. On the other hand, the coding rate of the convolutional code is expressed as R = k / n by the number of input sequences k and the number of output sequences n (the number of bits of the code block. Hereinafter, the code block is abbreviated as a code word). . If q is a divisor of n, one codeword can be expressed by an integer number of sub-symbols, so that the probability that a received symbol is a transmission symbol corresponding to each branch for each state transition in one step (conditional probability) (This is called a likelihood function.) However, if q is not a divisor of n, one codeword cannot be expressed by an integer number of subsymbols. Therefore, the conditional probability of each branch for each state transition of one step cannot be defined. Therefore, although hard decision decoding is possible, normal soft decision Viterbi decoding that evaluates the state transition for every k bits of encoder input is impossible.
[0011]
In the case of a convolutional code with a coding rate R = 1/2, if 1 bit / subsymbol QPSK (the same applies to π / 4 shift QPSK) or 2 bits / subsymbol 16QAM, one codeword is an integer. Since it can be expressed by the number of sub-symbols, normal soft decision Viterbi decoding is possible. However, when bit interleaving is performed, in 16QAM, information of one code word is separated into two sub-symbols and can be extracted only in a mixed form with information of other code words. Also, since 64QAM is 3 bits / subsymbol, it is impossible to express one codeword by one subsymbol. In the case of a convolutional code with a coding rate R = 2/3, QPSK and 64QAM can be used, but 16QAM cannot be used. In addition, when bit interleaving is performed, information of a single codeword cannot be extracted with 64QAM. As described above, in the conventional method, only a specific multilevel modulation method (hereinafter referred to as symbol size) can be used for a convolutional code of a specific coding rate, and the optimal multilevel is used. The transmission efficiency cannot be increased by selecting the modulation method.
[0012]
The present invention solves the above-mentioned conventional problem, and enables a soft decision to be made even if a convolutional code having an arbitrary coding rate is transmitted using a multilevel modulation scheme having an arbitrary symbol size, thereby reducing the error rate. The purpose is to lower the transmission efficiency.
[0013]
[Means for Solving the Problems]
In order to solve the above-described problem, according to the present invention, an error correction transmission apparatus includes a coding rate k / n convolutional coding unit that outputs an n-bit codeword for each k-bit information sequence input, Symbol encoding means for assembling symbols consisting of two q-bit sub-symbols from the bit sequence of the codeword output from the multi-code encoding means, sub-symbol interleaving means for interleaving in units of sub-symbols, Transmitting means for orthogonally modulating a carrier wave and transmitting; receiving means for synchronously detecting a received signal to obtain subsymbols and carrier amplitude information in subsymbol units; and subsymbol deinterleaving for deinterleaving subsymbols together with carrier amplitude information And a soft decision Viterbi decoding means for soft decision in units of sub-symbols.
[0014]
With this configuration, soft-decision Viterbi decoding can be performed even if a multilevel modulation scheme of an arbitrary symbol size is selected for a convolutional code of an arbitrary coding rate, so that the error rate is reduced. Transmission efficiency can be increased.
[0015]
Further, the symbol encoding means is provided with means for associating the least common multiple LCM (q, n) of q and n with the processing unit bit number p and corresponding the p-bit code word to the p-bit sub-symbol. The Viterbi decoding means is provided with an evaluation means for calculating a branch metric for each state transition of the convolutional coding means corresponding to code words for p bits. With this configuration, soft decision Viterbi decoding can be performed even when the number of bits of the codeword is not divisible by the number of bits of the subsymbol, so that the error rate can be reduced and the transmission efficiency can be increased.
[0016]
The evaluation means is provided with calculation means for calculating the Euclidean distance between the code word for p bits and the sub-symbol for p bits. With this configuration, soft-decision Viterbi decoding can be performed more precisely than when the Hamming distance is used, so that the error rate can be reduced and the transmission efficiency can be increased.
[0017]
The receiving means is provided with means for multiplying the normalized synchronous detection output by the carrier level at each time point and returning it to the original coordinate system, and the calculating means calculates the branch metric using the square of the Euclidean distance as the logarithmic likelihood. Means were provided. With this configuration, it is possible to accurately perform soft decision Viterbi decoding by removing the influence of thermal noise, thereby reducing the error rate and increasing the transmission efficiency.
[0018]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, embodiments of the present invention will be described in detail with reference to FIGS.
[0019]
(Embodiment)
In the embodiment of the present invention, the output of the convolutional encoder is converted into a multi-level modulation symbol, then interleaved in sub-symbol units, orthogonally modulated, transmitted, received, and received subsymbols obtained by synchronous detection A set of values (analog values) and reproduced carrier level information (collectively referred to as “sub-symbol amplitude information”) is deinterleaved in units of sub-symbols, and the number of bits of the code word and the number of sub-symbols The error correction transmission apparatus performs soft-decision Viterbi decoding using amplitude information of sub-symbols by state transition for each of a plurality of codewords, with the least common multiple of the number of bits as the number of processing unit bits.
[0020]
FIG. 1 is a functional block diagram of a transceiver of an error correction transmission apparatus according to an embodiment of the present invention. FIG. 1A is a configuration diagram of a transmitter. FIG.1 (b) is a block diagram of a receiver. In FIG. 1, a convolutional encoder 1 is a circuit that performs convolutional encoding of an information sequence input. The encoding unit 2 is a circuit that collectively converts a plurality of bits into one sub-symbol. The sub symbol interleaver 3 is means for interleaving in units of sub symbols. The synchronous detection processing unit 4 is a circuit that performs synchronous detection by extracting quasi-synchronous detection, A / D conversion, band-limited by an LPF configured by a digital filter, and extracting synchronous timing from sampling data. The sub-symbol deinterleaver 5 is means for rearranging data interleaved in units of sub-symbols in the original order. The soft decision Viterbi decoder 6 is means for performing Viterbi decoding by making a soft decision in units of a plurality of steps of state transition.
[0021]
FIG. 2 is a flowchart for explaining the error correction transmission apparatus according to the embodiment of the present invention in comparison with the conventional example. FIG. 3 is an example of a circuit diagram of the convolutional encoder. FIG. 4 is a trellis diagram showing all state transitions. FIG. 5 is a trellis diagram for explaining a state transition branch metric calculation. FIG. 6 is a diagram illustrating state transition for each of a plurality of codewords focusing on one surviving path. Here, M is the value of the path memory.
[0022]
FIG. 7 is a flowchart showing a demodulation procedure of the error correction transmission system in the embodiment of the present invention. FIG. 8 is a flowchart showing a decoding procedure. FIG. 9 is a flowchart showing a traceback procedure. FIG. 10 is an explanatory diagram of the interleave function. FIG. 11 is an explanatory diagram of a burst format. FIG. 12 is a graph of static characteristics of error correction, a graph of error correction characteristics under flat fading, and a graph of error correction characteristics under multipath fading.
[0023]
The operation of the error correction transmission system in the embodiment of the present invention configured as described above will be described. First, an outline of the operation of the transceiver will be described with reference to FIG. The information sequence input SD [i] is input to the convolutional encoder 1. The output of the convolutional encoder 1 is gray-coded every q bits by the encoding unit 2 and mapped to symbols (I [m], Q [m]) of multilevel modulation. The subsymbol interleaver 3 performs interleaving (I [n], Q [n]) in units of subsymbols. The subsequent operation is the same as that of a normal quadrature modulation transmitter.
[0024]
In the receiver, the received signal is quasi-synchronously detected (I '(t), Q' (t)), and this is sampled and A / D converted at a frequency several times the baud rate frequency (for example, 8 times) and digital low-pass. The sampling data (I ′ (k), Q ′ (k)) output through the filter (LPF) is sent to the synchronous detection processing unit 4. The synchronous detection processing unit 4 detects a synchronous symbol from the transmitted sampling data, extracts a symbol timing, and regenerates a carrier wave using a pilot symbol defined in a burst format (cx (n)) Then, the absolute value (W (n)) is calculated, and synchronous detection (I [n], Q [n]) is performed by the synchronous detection unit using the regenerated carrier wave. The sub-symbol information (I [n], W [n], Q [n], W [n]) obtained here is deinterleaved by the sub-symbol deinterleaver 5 (I [m], W [m] , Q [m], W [m]). Soft decision Viterbi decoder 6 performs soft decision in units of sub-symbols.
[0025]
The error correction transmission apparatus according to the embodiment of the present invention is generalized as follows. An n-bit code word is output for each k-bit information sequence input by a convolutional encoding means with a coding rate k / n. A sub-symbol sequence is generated by dividing the bit sequence of the codeword output from the convolutional coding means into q bits. Interleave in units of subsymbols. Two interleaved sub-symbols are taken out as Ich and Qch, and the carrier wave is orthogonally modulated and transmitted. On the receiving side, the received signal is synchronously detected to obtain sub-symbols (Ich, Qch) and sub-symbol unit carrier amplitude information (W). The sub-symbol is deinterleaved together with the carrier amplitude information. Let the least common multiple LCM (q, n) of q and n be the number of processing unit bits p, and code words for p bits correspond to sub-symbols for p bits. For each state transition of the convolutional encoding means corresponding to the p-bit codeword, a value obtained by multiplying the Euclidean distance between the p-bit codeword and the p-bit sub-symbol by the carrier amplitude and squared. A branch metric is calculated to make a soft decision.
[0026]
When performing soft decision decoding, when the modulation method is 16QAM (2 bits / subsymbol), codes such as coding rate R = k / 3 (k = 1 to 2), k / 5 (k = 1 to 4) are used. When the modulation method is 64QAM (3 bits / subsymbol), codes such as coding rate R = 1/2 and k / 4 (k = 1 to 3) have no state transition for each step (time point). It becomes possible. 3 steps (LMC (2,3) = 6 bits) when R = k / 3 in 16QAM, 2 steps (LMC (2,5) = 10 bits) when R = k / 5 in 16QAM, 64QAM 3 steps (LMC (3,2) = 6 bits) when R = 1/2, and every 3 steps (where LMC (4,3) = 12 bits) when R = k / 4 with 64QAM Transition is required. In this way, the method of determining the path by calculating the branch metric and path metric based on the state transition for each of the plurality of steps is referred to as “modified Viterbi decoding” in order to distinguish it from normal Viterbi decoding. .
[0027]
With reference to FIG. 2, the present invention will be described in comparison with a conventional example in the case of transmitting a code having a generator polynomial degree of 5, a coding rate R = 1/2, and an information point of 100 bits by 64QAM. The processing up to the point where the information sequence input (100 bits) is convolutionally encoded (210 bits) is the same as the conventional example. Conventionally, bit interleaving is followed by conversion to multilevel modulation symbols (35 symbols). On the other hand, in the present invention, after conversion to multi-level modulation symbols (70 subsymbols), interleaving (70 subsymbols) is performed in units of subsymbols. After that, sub-symbols are mapped to the burst signal (70 sub-symbols), orthogonally modulated and transmitted. On the receiving side of the conventional example, data detection is performed by synchronous detection (35 symbols), gray code is decoded (210 bits), deinterleaved in bit units (210 bits), and then hard decision Viterbi decoding (100 bits) )I do. On the other hand, in the present invention, data subsymbols are extracted from the quasi-synchronous detection output (70 subsymbols), deinterleaved in units of subsymbols (70 subsymbols), and then soft decision Viterbi decoding (100 bits) is performed.
[0028]
Next, with reference to FIG. 4 and FIG. 5, in the processing block of soft decision Viterbi decoding in FIG. An example will be described in which a code having a constraint length of 5 and a number of states of 16) and a coding rate R = 1/2 is transmitted by 64QAM. As described above, when the modulation sub-symbol has a 3-bit configuration and the coding rate R = 1/2, the state transition needs to be performed every three steps in order to execute the soft decision. The state transition every three steps is as shown in FIG. 4 because there are eight paths from the state at time t-3 to the state at time t in each state. The “encoder state” shown on the left side in FIG. 4 is the state (shift register state) displayed on the left side as the MSB regardless of the time point. In addition, “input of encoder” shown on the right side is a bit string input to the encoder when joining the state, and the left side is displayed as MSB (bits input earlier in time). .
[0029]
Among these, FIG. 5 is an explanatory diagram focusing on the state S1 at time t. The state before 3 time points in the path where the 3 bits input to the encoder between a certain time point t-3 and t-1 are “xxx” and join to “yxxx” at time point t is “000y”, “ There are eight types of 001y, “010y”, “011y”, “100y”, “101y”, “110y”, and “111y”. Here, the left side is represented as the upper side (MSB) in accordance with the notation of the shift register (shifted to the left) in the circuit diagram of the encoder of FIG. Here, the upper 3 bits in which there are 8 patterns are portions that are shifted and disappear when 3 bits are input to the encoder. The branch number i is associated with a decimal representation of the upper 3 bits of the state of the shift register three times before. Therefore, focusing on the state S1, the state before the three time points flows into the state S1 when the lower 1 bit “y” is “0” and the encoder input “xxx” is “001”. States S0 (i = 0), S2 (i = 1), S4 (i = 2), S6 (i = 3), S8 (i = 4), S10 (i = 5), S12 (i = 6), There are 8 types of S14 (i = 7).
[0030]
In FIG. 5, symbols λ (S, t) and λ (S, t−3) represent path metrics of state S at time t and time t−1, respectively. Taking the above-described encoding method and modulation method as an example, state transition is performed every three steps. That is, when the time number t is a multiple of 3, the received first modulation sub-symbol z (2n), its regenerated carrier level c (2n), and the received second modulation are received from the sub-symbol deinterleaver in FIG. Subsymbol z (2n + 1) and its recovered carrier level c (2n + 1) are given. In each state S, the immediately preceding state number is obtained for each branch number i using the table F0 (i, S), and the first modulation sub-symbol corresponding to the branch i is obtained using the table F1 (i, S). The sub symbol is obtained from the table F2 (i, S). The following minimum value is obtained using these values and the reception information acquired at this time.
λ (S, t) = min {(| F1 (i, S) −z (2n) | × | cx (2n) |)²
+ (| F2 (i, S) −z (2n + 1) | × | cx (2n + 1) |)²+ Λ (F0 (i, S), t-3)} (i = 0 to 7)
Therefore, the minimum value is stored as the path metric of the state S, and the branch number i giving the minimum value is stored in the path memory M (S, n). Here, the variable n has a relationship of the time point number t and 3n = t.
[0031]
The tables F0 (i, S), F1 (i, S), and F2 (i, S) used here are given by the following method. First, the following function is defined where i is a branch number (0 to 7), S is a state number (0 to 15), and b is a bit number (0 to 6).
d (i, S, b) = mod (int ((16i + S) / 2b), 2)
Here, int (x) is an integer part of x, and mod (x, 2) is a remainder obtained by dividing x by 2. d (i, S, b) is a function for calculating a bit string input to the encoder up to this point when the state number is S and the branch number is i. The LSB (D1 ) Is bit 0 (b = 0), and the MSB (D4) of the shift register of the encoder three times before is set to bit 6 (b = 6).
[0032]
When the function d (i, S, b) is used, the state number F0 (i, S) three times before is given by the following equation.
F0 (B, S) = d (i, S, 6) × 8 + d (i, S, 5) × 4 + d (i, S, 4) × 2 + d (i, S, 3)
Next, the first modulation sub-symbol value F1 (B, S) is given by the following equation.
F1 (i, S) = (1-2 × y0) × {(1-2 × y1) × (3-2 × y2) +4}
However,
y0 = mod (Σg1 (v) d (i, S, v + 2), 2) (total is v = 0 ~ 4)
y1 = mod (Σg2 (v) d (i, S, v + 2), 2) (total is v = 0-4)
y2 = mod (Σg1 (v) d (i, S, v + 1), 2) (total is v = 0-4)
F1 (i, S) is a 3-bit gray code (± 1, ± 3, ± 5, ± 7) of y0, y1, y2.
[0033]
The second modulation sub-symbol value F2 (B, S) is given by
F2 (i, S) = (1-2 × y3) × {(1-2 × y4) × (3-2 × y5) +4}
However,
y3 = mod (Σg2 (v) d (i, S, v + 1), 2) (total is v = 0-4)
y4 = mod (Σg1 (v) d (i, S, v), 2) (total is v = 0-4)
y5 = mod (Σg2 (v) d (i, S, v), 2) (total is v = 0-4)
F2 (i, S) is a 3-bit gray code (± 1, ± 3, ± 5, ± 7) of y3, y4, y5.
y0 and y1 are codewords sent between 3 and 2 time points, y2 and y3 are codewords sent between 2 and 1 time points, y4 and y5 are from 1 time point to the current time point Codewords sent during G1 (v) and g2 (v) are the coefficients of the convolutional code generator polynomial.
g1 (4) = 1, g1 (3) = 1, g1 (2) = 0, g1 (1) = 0, g1 (0) = 1
g2 (4) = 1, g2 (3) = 0, g2 (2) = 1, g2 (1) = 1, g2 (0) = 1
It is.
[0034]
The above is the basic algorithm of modified Viterbi decoding. Next, a decoding operation using this algorithm will be described with reference to the flowcharts of FIGS. 7, 8, and 9, taking the burst format of FIG. 11 as an example. Multi-level modulation synchronous detection will be described by taking as an example the case where a signal is transmitted in a general burst format as shown in FIG. Normally, in a burst signal of a multilevel modulation system, data symbols are arranged following a series of synchronization symbols after a ramp time of transmission rise, and pilot symbols are inserted therein at regular intervals. Similar to the synchronization symbol, the pilot symbol is a symbol having a known value between transmission and reception. In the present embodiment, as an example of 64QAM, pilot symbol values are uniformly set to obtain a good S / N.
I + jQ = 7 + 7j
It is said.
[0035]
In the demodulation processing procedure of FIG. 7 (the operation of the synchronous detection processing unit 4 of FIG. 1B), sampling data (I ′ (k), Q ′ (k)) output from the A / D conversion unit / LPF unit. Among them, a “symbol synchronization acquisition process” block detects a synchronization symbol, extracts a symbol timing, and a “received symbol extraction” block extracts a pilot symbol and a data symbol following the synchronization symbol. The received symbol value rx (t) obtained here is a complex number. The variable t represents the absolute symbol number after the synchronization symbol, the symbol number of the first pilot symbol is 0, and the symbol number of the last pilot symbol is K + p0-1. K is the number of data symbols, and p0 is the number of pilot symbols.
[0036]
When rx (t) is extracted, carrier recovery is performed sequentially using the following equation, assuming that the data symbol number is k, the pilot symbol number is ≡, and the relative number of the data symbol between the two pilot symbols is h.
cx (k) = [{rx (b1) -rx (b0)} / b2 × h + rx (b0)] / [7 + 7j]
cx (k) is a complex number representing a regenerated carrier wave, and the kth data symbol is linearly interpolated with pilot symbols positioned before and after it. b0 is a pilot symbol absolute symbol number positioned in front, b1 is a pilot symbol absolute symbol number positioned later, and 7 + 7j is a value of a pilot symbol defined in advance. In linear modulation, a modulated wave is obtained by multiplying a carrier wave by a modulation component (I + jQ). Therefore, a data symbol value rx (b0 + h) obtained by quasi-synchronous detection is divided by a reproduced carrier wave cx (k). Thus, synchronous detection can be performed. cx (k) and rx (b0 + h) are complex numbers, and the amplitude and the phase are corrected simultaneously by dividing by cx (k). The real part of the complex number obtained here is the component of Ich, and the imaginary part is the component of Qch. As a result, Ich is stored as an even-numbered subsymbol and Qch is stored as an odd-numbered subsymbol according to the following equation.
y (2k) = Re [rx (b0 + h) / cx (k)] ... Ich
y (2k + 1) = Im [rx (b0 + h) / cx (k)] ... Qch
[0037]
Subsequently, a sub-symbol / deinterleaver deinterleaves a pair of sub-symbols and the absolute value of a reproduced carrier wave accompanying the sub-symbol using an interleave function ilv (k) function. Here, the variable k is a subsymbol number. In normal demodulation, the complex number rx (b0 + h) / cx (k) (= y (2k) + j · y (2k + 1)) obtained here is the signal at the closest symbol point in the signal space. It is determined that it has been transmitted, and a corresponding bit string is output. Here, in order to perform a soft decision, synchronous detection output (FIG. 1 (FIG. In b), it corresponds to I [n], Q [n].
[0038]
In the decoding processing procedure of FIG. 8, modified Viterbi decoding by soft decision is performed based on the received sub-symbol value z (•) and the absolute value w (•) of the reproduced carrier wave obtained in the demodulation processing procedure. Let PM (S) be the path metric value in state S, and M (S, n) be the path memory value in state S at time 3n. As initialization processing, 0 is stored in the path memory M (S, n) and path metric PM (0), and an appropriately large numerical value (2 15 here) is stored in PM (S) other than the state 0. This is a measure for preventing a branch that does not exist in the initial state transition from being selected. Here, the state number S is 0 to 15, and the state transition number n is 0 to K-1.
[0039]
The reception sub-symbols corresponding to the time points t = 3n, 3n + 1, and 3n + 2 and the absolute values of the regenerated carrier waves are given by the following array variables already stored in the demodulation processing procedure.
Receive sub-symbol: z (2n), z (2n + 1)
Absolute value of reproduced carrier wave: w (2n), w (2n + 1)
Therefore, for each state transition number n, the branch metric bm and the path metric pm are calculated by the following equations while the branch number i is changed from 0 to 7 for the state S from 0 to 15.
bm = {| F1 (i, S) -z (2n) | × w (2n)}²+ {| F2 (i, S) -z (2n + 1) | × w (2n)}²・ ・ ・ ・ ・ ・ Formula 1
pm = bm + PM (F0 (i, S))
The branch number i having the smallest path metric in the state S is obtained by successive comparison, and the path metric value at this time is temporarily stored in the array variable TPM (S) and the path memory M (S, n) is updated. . When this process is completed for all the states S, the path metric PM (S) is updated to the value of TPM (S). This process is sequentially executed from the value n of 0 to K-1.
[0040]
In the traceback process of FIG. 9, the path memory M (S, n) obtained by the decoding process procedure is used to reversely trace state transitions (this is called traceback), thereby estimating the encoder. An input sequence can be obtained. In the present embodiment, the convolutional code is terminated to state 0 by the tail bit. Therefore, it is assumed that the traceback is performed starting from the state 0 of the path memory and the state transition number K-1. By substituting the initial value M (0, K-1) for the array variable TB (n), where n is K-1 and S is 0, and successively calculating by the following equation while decrementing n: Decoding is complete.
S = F0 (TB (n + 1), S)
TB (n) = M (S, n)
RD (3n) = mod (int (TB (n) / 4), 2)
RD (3n + 1) = mod (int (TB (n) / 2), 2)
RD (3n + 2) = mod (TB (n), 2)
RD (·) is a decoded bit string obtained by soft decision modified Viterbi decoding. Further, F0 (i, S) is a function (table) indicating the state three points before the time corresponding to the branch number i of the state S described above. The above is the description of the decoding procedure by the soft decision modified Viterbi decoding with reference to FIG. 7, FIG. 8, and FIG.
[0041]
The reason why the branch metric calculation method in the decoding procedure described above is not simply the square of the Euclidean distance as shown in Equation 1 but is multiplied by the square of the absolute value of the carrier wave will be described. The normal symbol point of the transmitted data symbol is represented by sx + j · sy, the quasi-synchronous detection output of the data symbol is represented by rx + j · ry, and the recovered carrier at that time is represented by a complex number such as cx + j · cy Then, the Euclidean distances Di and Dq for each component between the normal symbol point and the data symbol point after synchronous detection are
Di ＝ Re {(sx + j ・ sy) − (rx + j ・ ry) / (cx + j ・ cy)}
Dq ＝ Im {(sx + j ・ sy) − (rx + j ・ ry) / (cx + j ・ cy)}
Given in.
[0042]
On the other hand, the thermal noise added to the data symbol during quasi-synchronous detection is represented by nx + j · ny. In the case of thermal noise, it is known that the in-phase component nx and the quadrature component ny follow a normal distribution N (0, σ) independent of each other. (The average value of thermal noise power is 2σ2.) When the line amplitude fluctuation is A (times) and the phase fluctuation is θ (rad), the quasi-synchronous detection output rx + j · ry is given by .
rx + j ・ ry ＝ (sx + j ・ sy) ・ A ・ exp (j ・ θ) + nx + j ・ ny
Therefore, if the recovered carrier cx + j · cy is ideal,
cx + j ・ cy ＝ A ・ exp (j ・ θ)
So,
(sx + j ・ sy) − (rx + j ・ ry) / (cx + j ・ cy)
= (Sx + j ・ sy) − (sx + j ・ sy) − (nx + j ・ ny) / (cx + j ・ cy)
= − (Nx + j ・ ny) / (cx + j ・ cy)
The relationship is established. Therefore, the aforementioned Euclidean distance is
Di ＝ Re {(nx + j ・ ny) / (cx + j ・ cy)}
Dq ＝ Im {(nx + j ・ ny) / (cx + j ・ cy)}
Can be rewritten.
[0043]
Next, when the Euclidean distance is multiplied by the absolute value (amplitude) of the carrier wave, the following equation is obtained.
Di '= Di · | cx + j · cy | = Re {(nx + j · ny) exp (−j · θ)}
Dq '= Dq · | cx + j · cy | = Im {(nx + j · ny) exp (−j · θ)}
As described above, since the in-phase component and the quadrature component of thermal noise are uncorrelated, the distribution of Di ′ and Dq ′ also follows the normal distribution N (0, σ). Denoting d as representative of Di ′ and Dq ′, the logarithm of the probability density function p (d) (hereinafter abbreviated as log likelihood) is
−d²/ (2σ²) −1/2 ・ ln (2π) −ln (σ)
It becomes. From the above consideration, when calculating a branch metric, not only the square value of the Euclidean distance Di or Dq of the sub-symbol value assumed for each branch and the sub-symbol value output for synchronous detection is calculated, It can be seen that by multiplying the square of the absolute value | cx + j · cy |, it is possible to directly obtain the log likelihood for thermal noise.
[0044]
Note that the constant term −1 / 2 · ln (2π) −ln (σ) and the coefficient 1 / (2σ²) Is common to the time, state, and branch, and can be omitted because it is canceled in the comparison operation. Further, in the comparison operation, the one with the larger likelihood is selected. At this time, the negative number of the log likelihood − (− d²) ＝ d²When using, the smaller value should be selected. On the other hand, when evaluating only by the Euclidean distance, when there is fading fluctuation or the like, the amplification degree differs at each time point, so the standard deviation of the thermal noise added to the synchronous detection output is different. If the comparison is made only with the branch metric, it is canceled out. However, since the Viterbi algorithm performs the comparison between the branches with the value added to the path metric value at the previous time point, the correct likelihood function is not given.
[0045]
The interleaving method will be described with reference to FIG. Interleaving is a method for converting a burst error into a random error. In general interleaving, assuming that the number of data is N (= m × n), the data is divided into n rows and arranged in m rows and n columns, and the interval between adjacent data is determined by reading this data column by column. Expanding. At this time, if m = n, an interval from the next adjacent data is also ensured at the same time, but if one of m and n is extremely small, the burst error cannot be converted into a random error. In the example of the burst signal format of FIG. 11, the number of data symbols is 48, that is, the number of subsymbols is 96. Therefore, in order to make this a matrix as described above, 6 × 16 or 8 × 12 is generally used. Is. As a method of interleaving this as close to a square matrix as possible, in this embodiment, an incomplete matrix as shown in Fig. 10 (10 × 10 matrix with 4 missing 10th row) is created. Taking a way to say skip. Here, when the number of data is N, m and n are the smallest integers greater than or equal to the square root value, and L is m × n−N. That is, since N is 96, the minimum integer is equal to or greater than the square root of 9.797, so that m = n = 10 and L = 4.
[0046]
This interleaving is given by the function ilv (k) in FIG. Here, k is the data reading order number, and ilv (k) is the corresponding data writing order number. When the matrix is a complete part (k = 0 to 59), the interleave function is given by
ilv (k) = mod (k, m) × n + int (k / m)
On the other hand, the interleaving function of the portion (k = 60 to 95) where the latter half of the last row is missing is given by the following equation.
ilv (k) = mod (k−m (n−L), m−1) × n + int ((k−m (n−L)) / (m−1)) + n−L
Here, mod (x, y) represents a remainder obtained by dividing x by y, and int (x / y) represents an integer part of a value obtained by dividing x by y.
[0047]
The effect of soft decision is shown in the graph of FIG. In these graphs, when there is no fading fluctuation (static characteristics), when there is fading fluctuation (flat fading), and when there is fading fluctuation in the multipath of the preceding wave and one delayed wave (multipath fading), It is the result of calculating the bit error rate before correction, after hard decision correction, and after soft decision correction by simulation. It can be seen that the bit error rate is lower in the soft decision than in the hard decision.
[0048]
As described above, in the embodiment of the present invention, the error correction transmission method is such that the output of the convolutional encoder is converted into multi-level modulation symbols, and then interleaved in units of sub-symbols, orthogonally modulated and transmitted. The received data is deinterleaved with the amplitude data in sub-symbol units, and the state transition between multiple time points corresponding to multiple code words is made with the least common multiple of the number of bits of the code word and the number of sub-symbols as the unit of processing In summary, since soft decision Viterbi decoding (modified Viterbi decoding is performed) using the amplitude information of the sub-symbol is made, soft decision can be made even if any multi-level modulation scheme is selected for any code. Therefore, it is possible to increase the transmission efficiency by reducing the error rate.
[0049]
However, the modified Viterbi decoding has a problem that the calculation amount is larger than that of the Viterbi decoding. If the coding efficiency is k / n, the sub-symbol size of multilevel modulation is q bits / sub-symbol, and the least common multiple LCM (n, q) of n and q is represented by a × n, a is 1 in the modified Viterbi decoding. Gives the number of times of state transition. The metric calculation amount (number of times) in one state transition of modified Viterbi decoding is Bm × S, where Bm is the number of branches and S is the number of states. Since the branch number Bm is a transition between time points a, 2^{k ・ a}It becomes a piece. On the other hand, the metric calculation amount between time points a in normal Viterbi decoding is a × B × S, but the number of branches at this time is 2^kIt is. From the above, the metric calculation amount per unit time of modified Viterbi decoding is 2 for Viterbi decoding.^{k (a-1)}It turns out that it becomes / a times. Note that a = 1 represents normal Viterbi decoding.
[0050]
Incidentally, when k = 1, the calculation amount does not increase when a = 2 when a = 2, but is 1.333 times when a = 3 (value of this embodiment), twice when a = 4, and 3.2 times when a = 5. Increases rapidly. Further, if k = 2, it is doubled when a = 2, and when k = 2, it is four times when a = 2. Therefore, when k = 1, a is 4 or less, and when k = 2, if a is 2, soft decision decoding can be performed with a calculation amount less than twice that of normal Viterbi decoding. If it exceeds, the amount of calculation will increase rapidly. In introducing modified Viterbi decoding, it is desirable to select a modulation method (value of q), a coding method (values of k and n), and an arithmetic element (DSP) in consideration of such characteristics. .
[0051]
【The invention's effect】
As is apparent from the above description, according to the present invention, the error correction transmission apparatus comprises a coding rate k / n convolutional coding means for outputting an n-bit code word for each k-bit information sequence input, Symbol encoding means for assembling symbols consisting of two q-bit sub-symbols from the bit sequence of the codeword output from the multi-code encoding means, sub-symbol interleaving means for interleaving in units of sub-symbols, and interleaved sub-symbols Transmitting means for orthogonally modulating a carrier wave and transmitting; receiving means for synchronously detecting a received signal to obtain subsymbols and carrier amplitude information in subsymbol units; and subsymbol deinterleaving for deinterleaving subsymbols together with carrier amplitude information And soft decision Viterbi decoding means for soft decision in units of sub-symbols. So by selecting an optimal combination of symbols of the coding rate and the multi-level modulation of convolutional codes can soft-decision Viterbi decoding, there is an advantage that it is possible to reduce the transmission error rate to the minimum.
[Brief description of the drawings]
FIG. 1 is a functional block diagram of a transceiver of an error correction transmission apparatus according to an embodiment of the present invention;
FIG. 2 is a flowchart for explaining an error correction transmission apparatus according to an embodiment of the present invention in comparison with a conventional example;
FIG. 3 is a diagram showing a configuration of an encoder in the error correction transmission apparatus according to the embodiment of the present invention;
FIG. 4 is a diagram showing all the state transitions of the error correction transmission apparatus according to the embodiment of the present invention;
FIG. 5 is a diagram illustrating a state of calculating a branch metric of state transition of the error correction transmission apparatus according to the embodiment of the present invention;
FIG. 6 is a diagram showing a result of calculating a state transition in the error correction transmission apparatus according to the embodiment of the present invention;
FIG. 7 is a flowchart showing a demodulation procedure in the error correction transmission apparatus according to the embodiment of the present invention;
FIG. 8 is a flowchart showing a decoding procedure in the error correction transmission apparatus according to the embodiment of the present invention;
FIG. 9 is a flowchart showing a traceback procedure in the error correction transmission apparatus according to the embodiment of the present invention;
FIG. 10 is an explanatory diagram of an interleave function used in the error correction transmission apparatus according to the embodiment of the present invention;
FIG. 11 is an explanatory diagram of a burst format in the error correction transmission apparatus according to the embodiment of the present invention;
FIG. 12 is a graph of error correction characteristics of the error correction transmission apparatus according to the embodiment of the present invention;
FIG. 13 is a diagram for explaining a conventional convolutional code soft decision method;
FIG. 14 is a diagram for explaining a soft decision result of a conventional convolutional code;
FIG. 15 is a diagram illustrating a configuration of a radio using a conventional convolutional code.
[Explanation of symbols]
1 Convolutional encoder
2 Encoding unit
3 Sub-symbol interleaver
4 Synchronous detection processing section
5 Sub symbol deinterleaver
6 Soft decision Viterbi decoder

Claims (5)

A convolutional coding means of coding rate k / n that outputs an n-bit codeword for each input of k-bit information series, and a q-bit sub-code from the bit sequence of the codeword output from the convolutional coding means Symbol encoding means for assembling symbols consisting of two symbols, sub-symbol interleaving means for interleaving the sub-symbol as a unit, transmitting means for orthogonally modulating a carrier wave with the interleaved sub-symbol and transmitting, and synchronous detection of the received signal Receiving means for obtaining sub-symbols and carrier amplitude information in units of sub-symbols, sub-symbol de-interleaving means for de-interleaving the sub-symbols together with the carrier amplitude information, and soft-decision Viterbi decoding means for soft-decision in units of sub-symbols The symbol encoding means processes the least common multiple LCM ( q , n ) of q and n. A unit for associating a plurality of code words for p bits with a plurality of sub-symbols for p bits as the number of logical unit bits p is provided, and the soft decision Viterbi decoding unit corresponds to a plurality of code words for p bits. An error correction transmission apparatus comprising: an evaluation unit that calculates a branch metric for each state transition of the convolutional encoding unit. 2. The error correction according to claim 1 , wherein said evaluation means includes calculation means for calculating a square of Euclidean distance between a code word for p bits and a sub-symbol for p bits as a branch metric. Transmission equipment. 3. The error correction transmission apparatus according to claim 2 , wherein said calculating means is provided with means for calculating a square of a product of Euclidean distance and carrier wave amplitude as a branch metric . Each time k bits of the input information sequence are input to the convolutional encoding means having the coding rate k / n, an n-bit code word is output, and the least common multiple of the number of sub-symbol bits q and the number of code word bits n LCM ( q , n ) Is a processing unit bit number p, a plurality of code words for p bits are associated with a plurality of sub-symbols for p bits, and a symbol consisting of two sub-symbols of q bits is assembled from the bit sequence of the code word, A sequence of symbols is interleaved in units of the sub-symbols, a carrier wave is orthogonally modulated with the interleaved sub-symbols and transmitted, and a received signal is synchronously detected to obtain sub-symbols and sub-symbol unit carrier amplitude information A sub-symbol is deinterleaved together with the carrier wave amplitude information, and for each state transition of the convolutional encoding means corresponding to a plurality of code words for p bits, a plurality of code words for p bits and a plurality of code bits for p bits. Error correction transmission characterized by soft decision by calculating the square of Euclidean distance between sub-symbols as branch metric Law. 6. The error correction transmission method according to claim 5, wherein the square of the product of the Euclidean distance and the carrier amplitude is calculated as a branch metric .

Images