JPH06177778A - Viterbi decoder for sub-set discrimination - Google Patents

Abstract

PURPOSE:To improve deterioration of error rate after decoding by constituting a branch metric calculation section as to obtain a branch metric with the use of the multidimensional probability distribution. CONSTITUTION:A branch metric calculation section 1 obtains the branch metric of a branch from a state Si at the time point (t) to a state Sj at the time point t+1 in a trellis chart. An ACS section 2 calculates the pass metric of a remaining pass of the state Si and the branch metric from the state Si to the state Sj, comparing the addition value on all the branches to the state Sj. The pass of the branch to give the minimum value is newly made the remaining pass in the state Sj. A pass metric memory 3 stores the remaining pass metric of each state. A pass memory 4 stores the remaining pass of each state.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a Viterbi decoder for subset determination used in a digital data communication system requiring error correction.

Usually, maximum likelihood decoding such as Viterbi decoding using a convolutional code is used in a bad communication channel with a high error rate, but the distribution of the reception sequence at this time is a normal distribution.

However, since the coded and modulated signal is processed before being input to the Viterbi decoder, the signal applied to the Viterbi decoder differs from the normal distribution. Therefore, the branch metric value cannot be obtained correctly, and the error rate after decoding deteriorates. Therefore, it is necessary to improve the deterioration of the error rate.

[0004]

2. Description of the Related Art FIG. 5 is an example of a configuration diagram of a main part of an encoder, FIG. 6 is an example of a configuration diagram of a main part of a Viterbi decoder, FIG. 7 is an example of an operation explanatory diagram of the encoder, and FIG. FIG. 9 is an explanatory diagram of the coding and modulation method, FIG. 9A is an explanatory diagram of the relationship between transmission signal points and a subset, FIG. 9B is an explanatory diagram of intra-subset determination, and FIG.
FIG. 3 is an explanatory diagram of a relationship between a code sequence and a reception sequence.

First, the reception sequence Y = y0 y1 y2 ... yk ...
Let y (n-1) be received (where n is an arbitrarily large positive integer and yk is the received block at time k). "Code theory"
According to Chapter 12 "Maximum Likelihood Decoding of Convolutional Codes" by Hideki Imai, published by The Institute of Electronics, Information and Communication Engineers, December 20, 1990, the likelihood function P (Y | X) is used for maximum likelihood decoding. It is sufficient to determine that the code sequence X = x0 x1 x2 ... xk ... x (n-1) that maximizes is transmitted (xk represents the code block at time point k).

Here, the code block at time point k is xk = [x0k,
x1k, ..., x (m−1) k], reception block yk =
[Y0k, y1k, ..., y (m−1) k] can be written. At this time, we can write P (Y | X) = Π P (yk | xk) ... (1), take the logarithm of this and multiply by -1, -Log P (Y | X) = -ΣLog P ( yk | xk) (2) Here, k = 0 to ∞ in both equations (1) and (2).

Therefore, maximum likelihood decoding can be performed by obtaining −Log P (yk | xk) for all code sequences for each reception sequence yk and forming this sum, and obtaining the code sequence with the minimum sum. Here, in a normal transmission path where thermal noise is dominant, the distribution of the received signal follows a normal distribution, so P (yk | xk) = (2πσ ² ) ^{-m / 2} · exp [−d (xk, yk) ^{2/2 σ 2]} written as (3). Where d (xk, yk) is the Euclidean distance between yk and xk.

Actually, when the minimum value of -Log P (Y | X) is obtained, the sum of the quantities proportional to -Log P (yk | xk) (such quantities are generally called branch metrics) is calculated. I'm fine, so use d (xk, yk) to find the minimum.

Now, the Viterbi decoder performs maximum likelihood decoding based on the above principle, and its operation will be specifically described with reference to FIGS. The encoder shown in FIG. 5 is configured using two flip-flops (FF ₁ and FF ₂ ) and EX-OR gates 71 to 73, and one output has two outputs (output 1, output 1
2) is sent out. FIG. 7 is a state transition diagram of FIG. 5. For example, “1/11” in FIG. 7 is an output when the input is 1.
Both 1 and output 2 are 1, indicating that the states of FF ₁ and FF ₂ transit from "0,0" to "0,1".

FIG. 8 shows the state transition of FIG. 7 with the horizontal axis as the time axis. If the initial state S ₀₀ changes to S ₀₁ at the time point 1 by "1" input at the time point 0, the code is The output of the instrument (code sequence in the figure) should be "11" (Fig. 7
However, since the received sequence is "11", there is no error and the branch metric is 0. However, when "0" is input, S ₀₀ is intact, but code sequence should "00" (see 0/00 in FIG. 7), the received sequence is as described above "11" So the error is 2 and the branch metric is 2. These processes are performed in the branch metric calculation part 1 of FIG.

The branch metric at time 2 is S₀₀
Is 2, S₀₁Is 0, S_TenWhen is 1, S₁₁Is 1 when
It Therefore, the branch metric from time 0 to time 2
The sum of (called path metric) is S₀₀→ S₀₀→ S₀₀so
Is 4, S₀₀→ S₀₀→ S₀₁Then 2, S ₀₀→ S₀₁→ S_TenThen 1, S
₀₀→ S₀₁→ S₁₁Then it becomes 1.

[0012] In addition, the S ₀₀ of point 3, and S ₀₀ of point 2
The branch from S ₁₀ comes in, but the branch metric of this is 1 for S _{00 and} 1 for S ₁₀ , so S
The path metric of ₀₀ → S ₀₀ → S ₀₀ → S ₀₀ is 5, S ₀₀ → S
The path metric of ₀₁ → S ₁₀ → S ₀₀ is 2, and the latter is smaller than the former, so the latter is the surviving path.

Similarly, for each time point, the survivor paths that enter each state are obtained, and four survivor paths remain at the final time point (when there are two encoder outputs). A path is selected, and the information sequence 110010 corresponding to the selected path is output as the result of maximum likelihood decoding of the reception sequence.

In addition, these are referred to as ACS (add-compare-selec
t) Part. The path memory stores the surviving path in each state, and the path metric memory stores the path metric of the surviving path in each state.

However, in the case of a receiver that performs a subset determination in coded modulation, the received signals are multiplied before being input to the Viterbi decoder, so that the noise component of the input signal to the Viterbi decoder is not always normal. Not distributed.

For example, in the QPSK modulation shown in FIG. 9 (a), the transmission codes A ₁ , A ₂ , B ₁ , B ₂ are divided into two groups of A ₁ , A ₂ and B ₁ , B ₂ , and the reception code is divided into two groups. Consider a case where it is determined which one belongs to (which is called a Cebuset determination).

For this reason, when multiplied by a code as shown by Q × I in FIG. 9 (b), if it is +1, it is the set of A ₁ and A ₂ , and if it is -1, it is the set of B ₁ and B ₂ . We call them subset A and subset B, respectively. Then, as shown in FIG. 9 (b), Q ch and
If the sum of I ch is +2, it is judged to be the code A ₁ , if it is −2, it is judged to be the code A _2, and if the difference is calculated to be +2, it is the code.
If B ₁ and −2, four codes can be determined by determining the code as B ₂ .

Generally, in the case of subset determination as described above, it is assumed that the relationship between the code sequence and the reception sequence is as shown in FIG. In the figure, the code sequence XI is composed of I ch code blocks from time 0 to time (n−1), and the code block at time k is 0.
It is composed of m-bit I ch codes from to (m−1). The reception sequence is similar to the code sequence.

That is, the transmission code is I as shown by in the figure.
The ch and Q ch code sequences are transmitted, but the code block at each time point is composed of m-bit codes as shown in the figure. The same applies to the receiving side (YI, YQ) (in the figure, ´, ´
See).

The transmission code is divided into subsets A and B as shown in FIG. 9 (b). Whether it is subset A or subset B depends on xIk0 × xQk0, xIk1 × xQk1 ,.
・, XIk (m −1) × xQk (m −1) is “1” or −1
It is the same as saying that

Conventionally, a general Viterbi decoder is also used for decoding a received sequence used for subset determination, and the input of the decoder is yIk0 × yQk0, yIk1 × yQk1, ..., yIk (m−1) × yQk. (m
−1) as branch metrics xIk × xQk and yIk × yQ
Decoding was done using the Euclidean distance of k.

At this time, the block k xIk × xQk at the time point k is
[xIk0 × xQk0, xIk1 × xQk1, ... xIk (m −1) × xQk (m
−1)], yIk × yQk is [yIk0 × yQk0, yIk1 × yQk
1, ... yIk (m −1) × yQk (m −1)] can be written.

However, since the above-mentioned distribution of yIk × yQk is not a normal distribution by multiplication, the error is obtained when the branch metric is obtained by the Euclidean distance d (xIk × xQk, yIk × yQk) obtained from the equation (3). Occurs.

[0024]

As described in detail above, since the distribution of the input signal of the Viterbi decoder is different from the normal distribution used when calculating the branch metric value,
There is a problem that the metric value cannot be calculated correctly and the error rate after decoding deteriorates.

An object of the present invention is to improve the deterioration of error rate after decoding.

[0026]

FIG. 1 is a block diagram showing the principle of the first invention, and FIG. 2 is a block diagram showing the principle of the second invention. In the figure, the branch metric calculating portion 1 to obtain the branch branch metric to the state S _j from the state S _i at the time t + 1 of point in time t in the trellis diagram, 2 states S _i path metric and the state S _i of the survivor path From state S _j
ACS part which adds the branch metric to the state S _j , compares this added value for all the branches to the state S _j , and makes the path of the branch giving the minimum value a new surviving path in the state S _j. The path metric memory for storing the path metric of the surviving path of 4 is a path memory for storing the surviving path of each state.

In the first aspect of the present invention, the branch metric calculation part is configured to obtain a branch metric by using a predetermined multidimensional probability distribution. The second invention is
A table storing the multidimensional probability distribution is provided, and the branch metric calculation part refers to the table to obtain the branch metric.

[0028]

FIG. 3 is a diagram for explaining the operation of FIG. First, according to the first aspect of the present invention, code blocks constituting a transmission signal belonging to a certain subset are xIk = (xIk0, xIk1), xQk = (xQk0, x
Qk1), the received block is yIk = (yIk0, yIk1), yQk = (y
Consider the case of Qk0, yQk1). This is m = 2 (encoder sends output 0 and output 1) for simplicity,
k

Here, xIk0, xQk0, xIk1, xQk1 are "1".
, Or "-1", and the relationship between the transmission code and the reception code at this time is shown in FIG. As shown in the figure, at the time point k from the encoder for I ch, output 0 is xIk0 and output 1 is xI.
Then, k1 is transmitted, and at the time point k, the encoder for Q ch transmits xQk0 as output 0 and xQk1 as output 1. Gaussian noise is superimposed on these outputs in the transmission line, and yIk0, yI
k1 and yQk0 and yQk1 are received on Q ch, respectively.

Next, when the transmission codes of I ch and Q ch belong to a certain subset, the reception codes of I ch and Q ch are (yIk0, yQk0
), (yIk1, yQk1), the transmit code is (xIk0,
The received code when (xQk0) or (−xIk0, −xQk0) is (y
Ik0, yQk0) and the transmit code is (xIk1, xQk1) or
It can be said that the received code at (−xIk1, −xQk1) is (yIk1, yQk1). When | (xk yk), P the conditional probability P (yk | xk) = | A 2 exp [-1 / 2σ 2 {(yIk0 -xIk0) 2 + (yQk0 -xQk0) 2}] + A 2 exp [−1 / 2σ ² {(yIk0 ＋ xIk0) ² ＋ (yQk0 ＋ xQk0) ² }] ｜ × ｜ A ² exp [−1/2 σ ² {(yIk1 −xIk1) ² ＋ (yQk1 −xQk1) ² }] ＋ A ² exp [−1 / 2σ ² {(yIk1 ＋ xIk1) ² ＋ (yQk1 ＋ xQk1) ² }] ｜ ・ ・ ・ (4) where A = (2π) ^-1/2 × σ ^-1 , and σ is the variance Is.

Here, in the first half of equation (4), where the transmission code is (xIk0, xQk0) or (-xIk0,
-XQk0) is the probability that the received signal will be (yIk0, yQk0). In the latter half of | ... |, the transmission code is (xIk0
1, xQk1) or (−xIk1, −xQk1) is the received signal
Probability of becoming (yIk1, yQk1), | ・ ・ ・ ｜ × ｜ ・ ・
-The "x" in | is for finding the probability that both occur at the same time.

Therefore, from the equation (4), −Log P (yk | xk) = − 4 Log A −Log | exp [−1 / 2σ ² {(yIk0 −xIk0) ² + (yQk0 −xQk0) ² }] + exp [−1 / 2σ ² {(yIk0 ＋ xIk0) ² ＋ (yQk0 ＋ xQk0) ² }] ｜ − Log ｜ exp [−1/2 σ ² {(yIk1 −xIk1) ² ＋ (yQk1 −xQk1) ² }] ＋ exp [−1 / 2σ ² {(yIk1 ＋ xIk1) ² ＋ (yQk1 ＋ xQk1) ² }] ｜ ・ ・ ・ (5) Maximum likelihood decoding calculates −Log P (yk ｜ xk) for all received sequences and Since the most probable sequence is determined by taking the sum of the above, the constant term (-4Log A) may be ignored in the actual decoding. This is used to calculate the branch metric, and then the same processing as general Viterbi decoding is performed.

That is, the Viterbi decoding at the time of subset determination is performed without performing the I × Q multiplication process. Note that the same can be considered even if the number of subsets changes. This makes it possible to improve the deterioration of the error rate after decoding.

[0034]

FIG. 4 is a block diagram of the first and second embodiments of the present invention. In the figure, the received sequences of I ch and Q ch are not subjected to multiplication processing, and are added to the branch metric calculation part 1 in the Viterbi decoder 5 for subset determination as they are. The branch metric calculation part calculates the branch metric while using the multidimensional probability distribution stored in the table 11 and sends it to the ACS part 2.

The ACS portion compares the added values obtained by adding the surviving path metrics to all the obtained branch metrics, selects the surviving path / branch pair having the minimum added value, and selects the path metric. Is output via the path memory 4 as a decoding result of maximum likelihood decoding of the information sequence corresponding to the path with the minimum
Re-encode with 64 to subset A or subset B
The output corresponding to is sent to the selector 63.

The path metric memory 3 stores the selected surviving path / branch pair as the latest path metric. On the other hand, in order to perform the in-subset determination, the sum and difference of I ch and Q ch are taken by the addition part 61 and the subtraction part 62 and the addition output and the subtraction output are sent to the selector 63. Since the selector selects the addition output or the subtraction output corresponding to the output of the re-encoding part and adds it to the decoder 65, the decoder decodes the selector output and sends it as the decoded output of Q ch.

[0037]

As described in detail above, according to the present invention, it is possible to improve the deterioration of the error rate after decoding.

[Brief description of drawings]

FIG. 1 is a principle configuration diagram of a first present invention.

FIG. 2 is a principle configuration diagram of a second present invention.

FIG. 3 is an operation explanatory diagram of FIG. 1. Two

FIG. 4 is a configuration diagram of first and second embodiments of the present invention.

FIG. 5 is an example of a main part configuration diagram of an encoder.

FIG. 6 is an example of a main-part configuration diagram of a Viterbi decoder.

FIG. 7 is an example of an operation explanatory diagram of an encoder.

FIG. 8 is an example of a Viterbi decoding method explanatory diagram.

9A and 9B are explanatory diagrams of a coding modulation method, FIG. 9A is an explanatory diagram of a relationship between transmission signal points and a subset, and FIG.

[Fig. 10] Fig. 10 is an explanatory diagram of a relationship between a code sequence and a reception sequence.

[Explanation of symbols]

1 Branch metric calculation part 2 ACS part 3 Path metric memory 4 Path memory 11 table

Claims (2)

[Claims] 1. State S _i at time t in the trellis diagram
From (i = 1,2, ..., 2 ^{(n -1)} , n: constraint length), time t + 1
Branch metric calculation part for finding the branch metric of the branch to the state S _j (j = 1,2, ..., 2 ^(n-1) ) of
(1) and by adding the branch metric from the path metrics and state S _i of the surviving path of the state S _i to state S _j (A
dd), compare this added value for all branches to state S _j (Compare), and make the path of the branch giving the minimum value a new surviving path at state S _j (Seiect) ACS part (2) In the Viterbi decoder for subset determination having a path metric memory (3) storing the path metric of the surviving path of each state and a path memory (4) storing the surviving path of each state, the branch metric calculation part is , A Viterbi decoder for subset determination, which is configured to obtain a branch metric by using a predetermined multidimensional probability distribution. 2. The Viterbi decoding for subset determination according to claim 1, wherein a table (11) storing the multidimensional probability distribution is provided, and the branch metric calculation part refers to the table to obtain a branch metric. vessel.