JP3537044B2 - Turbo decoding method and method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] The present invention relates to a turbo decoding system and method.

[0002]

2. Description of the Related Art In recent years, a turbo decoding method has been spotlighted as a decoding method approaching the limit of Shannon, and W-CDMA and CDMA-200, which are standards for next-generation mobile communication, have been highlighted.
0 is also used.

[0003] However, since the turbo decoder performs iterative decoding using an interleaver, an increase in circuit size and processing time poses a serious problem.

As a soft output decoding algorithm applied to an element decoder of turbo decoding, it is said that a method using a maximum a posteriori probability (hereinafter, referred to as MAP) is the best at present. However, since the device scale and the processing amount are significantly large, the implementation is simplified by selecting the maximum value (MAX) as a result of the likelihood calculation as to whether the path between the points is “1” or “0”. Max-log-MAP (Max Logarithmic Maximum
A Posteriori) is generally and widely used.

[0005]

However, the Max-lo
Although the amount of processing is reduced by g-MAP, especially in data transmission, an enormous memory area is required for decoding as the data length increases. In particular, a memory area for storing the likelihood information of the entire trellis is required. In recent years, memory devices have become smaller and have higher performance, and although the memory area inside the device has become larger, it is difficult to store state metrics for the decoding length. Therefore, a method called a sliding window as shown in FIG. 7 is widely used. In this method, only the likelihood information in the trellis for the window size is stored, and this window position is shifted until the decoding length is reached, so that a large saving of memory can be achieved. However, in order to reduce the characteristic degradation due to having only part of the likelihood information of the trellis, it is necessary to increase the window size to some extent, which results in an increase in processing time.

SUMMARY OF THE INVENTION An object of the present invention is to provide a turbo decoding method and a turbo decoding method capable of significantly reducing the window size with a slight increase in memory usage and consequently reducing the processing time. And

[0007]

According to the present invention, there is provided a first soft output decoder for performing forward processing and backward processing, and a second soft output decoder for performing the forward processing and backward processing. , Interleaving the reliability information likelihood output from the first soft-output decoder,
An interleaver for supplying to the soft output decoder of
And a deinterleaver for deinterleaving the reliability information likelihood outputted by the soft output decoder of the first soft output decoder and supplying the same to the first soft output decoder, wherein the first soft output decoder and the second In a turbo decoding system in which decoding is alternately repeated by a soft output decoder, the forward
Decoding that performs processing first and performs the backward processing later
Perform the backward processing first for each window
The decoding performed after the forward processing is alternately repeated, and at least one of the first soft-output decoder and the second soft-output decoder performs the forward processing for each window first in a certain iteration, and When the word processing is performed later, the state metric for each state at the end point of each window obtained by the forward processing in the repetition is stored, and in the next repetition, the backward processing is performed on each window first. Performing the forward processing later, and in the forward processing in the next iteration,
Means for using the stored state metric as the initial value of the state metric of each state at the start point of each window, and when performing the backward processing first for each window in a certain iteration and performing the forward processing later, The state metric for each state at the starting point of each window obtained by the backward processing in the repetition is stored, and in the next repetition, the forward processing is performed for each window first, and the backward processing is performed later. Means for using the stored state metric as an initial value of the state metric of each state at the end point of each window in the backward processing in the next iteration. .

In the above turbo decoding system, the forward processing is an operation for executing the Viterbi algorithm from the start point to the end point of the data, and the backward processing is the operation for executing the Viterbi algorithm from the end point to the start point of the data. May be an operation to be executed.

In the above turbo decoding method, at the time of the first iteration, a window size used in an operation performed after the forward processing or the backward processing performed on each window is calculated by an operation performed before the window size. May be provided with means for making the window size larger than that used in.

According to the present invention, a first soft output decoder that performs forward processing and backward processing, a second soft output decoder that performs the forward processing and the backward processing, and the first soft output decoder An interleaver that interleaves the reliability information likelihood output from the output decoder and supplies the reliability information likelihood output to the second soft output decoder, and deinterleaves the reliability information likelihood output from the second soft output decoder. A deinterleaver for supplying the first soft output decoder with
The provided, in the turbo decoding process performed by the turbo decoding device repeat the decoding alternately by the first soft-output decoder and the second soft-output decoder, the full for each window
Forward processing is performed first, and the backward processing is performed later.
Decoding and the backward processing for each window first
And the decoding performed after the forward processing is alternately repeated.
And when at least one of the first soft-output decoder and the second soft-output decoder performs the forward processing for each window in a certain iteration first and performs the backward processing later, The state metric for each state at the end point of each window obtained by the forward processing in the repetition is stored, and in the next repetition, the backward processing is performed for each window first, and the forward processing is performed later. Performing a step of using the stored state metric as an initial value of the state metric of each state at the start point of each window in the forward processing in the next iteration, and first performing the backward processing for each window in a certain iteration. Do the Foy When performing the post-processing, the state metric for each state at the start point of each window obtained by the backward processing in the repetition is stored, and in the next repetition, the forward processing is performed for each window first. Performing the backward processing later, and using the stored state metric as an initial value of the state metric of each state at the end point of each window in the backward processing in the next iteration. Is provided.

In the above turbo decoding method, the forward processing is an operation for executing the Viterbi algorithm from the start point to the end point of the data, and the backward processing is an operation for executing the Viterbi algorithm from the end point to the start point of the data. May be an operation to be executed.

In the above turbo decoding method, at the time of the first iteration, the window size used in the operation performed after the forward processing or the backward processing performed on each window is calculated by the operation performed before the window size. May have a step of making it larger than the window size used in.

[0013]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The processing time of the Max-log-MAP element decoder itself is dominated by the time required for forward processing and backward processing when sliding window processing is performed. Here, the forward processing
The operation of executing the Viterbi algorithm from the start point to the end point of data and the backward processing indicate the operation of executing the Viterbi algorithm from the end point of data to the start point (hereinafter, forward processing and backward processing are referred to as forward processing and backward processing, respectively). α operation, β operation).

The turbo decoding processing speed-up circuit according to the present invention can be applied to the case where an iterative decoding operation is performed using a sliding window or an element decoder having a configuration equivalent thereto. At the time of α calculation (or at the time of β calculation), the output result (state metric) at the end of each window size is held, and the held state metric is used in the β calculation (or α calculation ) at the next repetition calculation. Provided is a circuit configuration that realizes a significant reduction in decoding processing time compared to the conventional method with a slight increase in memory usage.

As shown in FIG. 7, the data length to be decoded is k
In order to obtain the reliability information likelihood for the time point Q1 in one process, the number of repetitions of the process is k / Q1. Except for the last block, the ACS processing of the α operation in one pass memory processing (ACS processing means that when there are a plurality of paths that transit to a certain state, each path corresponds to a likelihood value (state metric) in that state. (Addition) the likelihood value (metric) when passing through, comparing (Compare) the added values of the respective paths, and selecting (Select) the path with the highest likelihood) The number of time points is Q1, and the number of ACS processing
+ Q2). Therefore, one code block is defined as Max-
The number of times of ACS required for log-MAP decoding is estimated by the following value.

S · (k / Q1) · (Q1 + (Q1 + Q)
2)) = Sk (2 + Q2 / Q1) where S is the number of states at each time point.

In general, when having the entire trellis, Q1 =
When k (information length) and Q2 = k, the ACS count is 2Sk, which is the minimum value. In order to reduce the number of times of ACS, it is necessary to make Q1 larger than Q2. Generally, no weight is assigned to each state for the initial value at the time of the backward operation processing. Therefore, it is necessary to secure a certain size of Q2 in order to prevent decoding deterioration.

In the method according to the present invention, the initial value of the state metric at the time of the backward (or forward) arithmetic processing at a certain repetition is set as the state obtained at the time of the forward (or backward) arithmetic processing at the previous iteration. By using the metric, it is possible to reduce the size of Q2 with a slight increase in the amount of memory, or to set Q2 = 0, thereby reducing the ACS processing time and greatly reducing the entire iterative decoding processing time. And

Further, if the processing of the present invention is added without changing the magnitude of Q2, the decoding result can be expected to converge quickly. Therefore, by using some effective decoding stop criterion, the required number of repetitions can be reduced. It can be expected that the processing time will be shortened due to the decrease.

FIG. 8 shows the configuration of an embodiment of the receiving apparatus according to the present invention. 8, 1 is an antenna, 2 is a transmission / reception separation unit, 3 is a reception radio unit, 4 is a despreading unit, 5 is a demodulation unit,
Reference numeral 6 denotes a reception data memory unit, 7 denotes a reliability information likelihood memory unit, 8 denotes a turbo decoding unit, and 9 denotes an information extraction unit.

The operation of the receiving device will be described with reference to FIGS.

A spread signal transmitted from a base station or a mobile terminal is received by an antenna 1 and input to a reception radio section 3 via a transmission / reception separation section 2. In the receiving radio unit 3,
The received signal passes through a band-pass filter (BPF), and the received signal from which out-of-band components have been removed is sent to a local oscillator (not shown).
Is frequency-converted into an intermediate frequency band (IF) by the local oscillation signal. The received signal whose frequency has been changed to the IF band is BP
After passing through F, the signal is corrected to a signal of an appropriate level by an automatic gain control circuit (AGC), quasi-synchronous detection is performed, and the frequency is converted to a baseband signal. After passing through a low-pass filter (LPF), the received signal frequency-converted to the baseband is A / D converted and output as a digital signal.

The received digital signal output from receiving radio section 3 is despread in despreading section 4 and output as a narrow-band modulated signal. The signal output from the despreading unit 4 is demodulated by the demodulation unit 5, subjected to predetermined soft decision processing, and stored in the reception data memory unit 6. The turbo decoding unit 8 performs decoding while updating the reliability information likelihood based on the data in the received data memory unit 6 and the reliability information likelihood memory unit 7. After decoding, a hard decision process is performed, and information bits are extracted in the information source extraction unit 9.

The soft-decision reception data (inf, par) of the turbo code (encoding rate is 1/3) obtained from the demodulation unit 5
entity1, parity2) is the received data memory unit 6
And decoding is performed in the turbo decoding unit 8.

When the decoding process is started, first, initialization is performed prior to the ACS process of each decoder. Here, the initial setting means that soft decision data PIL-inf used in the soft output decoder 102 is generated by the interleaver 103 and stored in the reception data memory unit 6.

When the initialization is completed, the soft output decoder 10
1 reads the soft decision data (inf, parity1) and the reliability information likelihood from the reception data memory unit 6 and the reliability information likelihood memory unit 7, respectively, and starts decoding. Here, the reliability information likelihood is any reliability information likelihood given in advance (A priori) to each information bit. After the normalization processing, the reliability information likelihood Le1 obtained from the soft output decoder 101 is calculated by the interleaver 105.
Thus, the reliability information likelihood La1 used in the soft output decoder 102 is stored in the reliability memory unit 7.

Next, the soft output decoder 102 compares the soft decision data (PIL-inf, parity2) and the reliability information likelihood La1 output from the soft output decoder 101 with the reception data memory 6 and the reliability information likelihood La1, respectively. Read from the memory 7 and start decoding. The resulting soft output is
Since it is used in the soft-output decoder 101, it is rearranged in the deinterleaver 104, and the reliability information likelihood La
2 is stored in the reliability information likelihood memory unit 7.

Thereafter, this decoding process is repeated a set number of times or the number of times of processing based on an appropriate decoding stop criterion, and the decoding is repeated using the reliability information likelihood finally obtained from the soft output decoding unit 102. A decoding result is obtained by performing the determination.

Next, the operation of the soft output decoder 101 will be described. Here, a feature of the present invention, and the part that actually affects the processing time most is the ACS processing time,
A sliding window operation currently widely used in the circuit configuration of the S processing unit will be described. After that, the configuration and operation of the embodiment using the method according to the present invention will be described.

Although the soft output decoder 1 and the soft output decoder 2 are distinguished in the description, they are actually realized by one decoder. Therefore, the soft output decoder 1 will be described.

Referring to FIG. 10, soft output decoders 101 and 102 according to the conventional example have two ACS units (ACS1).
Unit 202 and ACS2 unit 203) and internal memory 204
Is provided. The ACS 1 unit 202 performs the α operation and the β operation, and the ACS 2 unit 203 also performs the α operation and the β operation. AC
The S1 unit 202 has a γ operation unit 202-1 and
The unit 203 has a γ operation unit 203-1. γ operation unit 20
2-1 and 203-1 are received data and reliability information likelihood L
Calculate a branch metric from a.

Hereinafter, the data length k = 330 for decoding the procedure of the ACS processing circuit adopting the sliding window method according to the conventional example, the number of states S at each time point = 8, and the window size Q1 for performing the α operation and the β operation = A description will be given of an example of an operation process of a window size Q2 = 96 in which only the 32 and β operations are performed. FIGS. 11 and 12 show the ACS operation procedure and the ACS operation timing. In this example, ACS1 part and A
There are CS2 sections, which are configured to be able to operate in parallel.

(1) Start of α32-1 operation in ACS1 section Received data Ain1 at each time point from time 0 to time 31 from the received data memory section 6 is transferred from the reliability information likelihood memory section 7 at each time point from time 0 to time point 31. Reliability information likelihood Rin
1 is read, and the ACS1 unit 202 performs an α operation using the read received data Ain1 and the reliability information likelihood Rin1. At the same time, the read reception data Ain
1 and the reliability information likelihood Rin1 and the state metric Sin1 of each state at the time points 0 to 31 obtained from the α operation are stored in the internal memory 204.

(2) Start of β96-1 operation in the ACS1 unit Next, the received data Ain3 at each of the time points 95 to 64 and the received data Ain2 at each of the time points 63 to 32 are read from the received data memory unit 6 in this order. From the reliability information likelihood memory unit 7, the reliability information likelihood Rin3 at each of the time points 95 to 64 and the reliability information likelihood Rin2 at each of the time points 63 to 32 are read out in this order,
Using the read received data Ain3 and Ain2 and the reliability information likelihoods Rin3 and Rin2, the ACS1 unit 202 performs a β operation. At the same time, the read reception data Ain3, Ain2 and the reliability information likelihoods Rin3, Ri
n2 is stored in the internal memory 204. In the β operation,
An initial value is given to the likelihood of the state metric for all the states at the time point 95, and this initial value is all zero.

Next, the state metric Sout1 (= Sin1) for each state at the time points 31 to 0 stored in the internal memory 204, and the time points 31 to 0
Of received data Aout1 (= Ain1) at each time point
And the reliability information likelihood Ro at each of the time points 31 to 0
ut1 (= Rin1) is read, and the ACS1 unit 202 performs a β operation using the read state metric Sout1, received data Aout1, and reliability information likelihood Rout1. The reliability information likelihood obtained as a result of the β operation is written to the reliability information likelihood memory 7 via a subtractor, a multiplier, and an interleaver (or a deinterleaver).

(3) Start of α32-2 operation in the ACS2 section Two ACS sections (ASC1 section 202 and ASC2 section 20)
3) are operated in parallel, so that the ACS1 unit 202 executes the β operation using the state metric Sout1, the received data Aout1, and the reliability information likelihood Rout1, and at the same time, the ACS1 unit 202
Received data Aout2 (= Ain2) and reliability information likelihood Ro at each of the time points 32 to 63 stored in
ut2 (= Rin2), and reads AC using the received data Aout2 and the reliability information likelihood Rout2.
In step S2, the α operation is performed. The state metric Sin2 for each state at the time points 32 to 63 obtained from the α operation is stored in the internal memory 204.

(4) The α1-3 operation is started in the ACS1 section.

Next, at time 6 when the data is stored in the internal memory 204,
The received data Aout3 (= A
in3) and reliability information likelihood Rout3 (= Rin3)
And the ACS1 unit 202 performs an α operation using the read received data Aout3 and the reliability information likelihood Rout3. The state metric Si for each state at each of the time points 64 to 95 obtained from the α operation
n3 is stored in the internal memory 204. Next, in the same procedure as (2), the ASC1 unit executes the β96-3 operation.

(5) The ACS2 unit uses the received data Aout3 and the reliability information likelihood Rout3 to start the β96-2 operation. The ACS1 unit 202 performs the α operation, and at the same time, from the reception data memory unit 6 to the time points 127 to 96. The received data Ain4 at the time point is stored in the reliability information likelihood memory unit 7 from the reliability information likelihood R at each of the time points 127 to 96.
in4 using the received data Ain4 and the reliability information likelihood Rin4,
Perform the operation. At the same time, the read reception data Ain4 and the reliability information likelihood Rin4 are stored in the internal memory 2.
04. In the β operation, an initial value is given to the likelihood of the state metric for all the states at the time point 127, and this initial value is all zero.

Next, from the internal memory 204 to the time points 95 to 6
4 at each time point Aout3 (= Ain
3) and the reliability information likelihood Rout3 (= Rin3) at each of the time points 95 to 64 are read, and the ACS2 unit 203 executes the β operation using the read received data Aout3 and the reliability information likelihood Rout3.

Next, the state metric Sout2 (= Sin2) of each state at each of the time points 63 to 32 stored in the internal memory 204, the received data Aout2 (= Ain2) at each of the time points 63 to 32, and the time point The reliability information likelihood Rou at each time point of 63 to 32
t2 (= Rin2) is read, and the ACS2 unit 203 executes a β operation using the read state metric Sout2, received data Aout2, and reliability information likelihood Rout2. The reliability information likelihood obtained as a result of the β operation is
The data is written to the reliability information likelihood memory 7 via a subtractor, a multiplier and an interleaver (or a deinterleaver).

By repeating the above operation,
Decoding results can be sequentially extracted.

Next, the turbo decoding system according to the embodiment of the present invention will be described. The turbo decoding method according to the present embodiment is an improvement of the sliding window method, and is different from the methods shown in FIGS.

Referring to FIG. 6, the soft output decoders 101 and 102 according to the embodiment of the present invention include two ACS units (ACS1 unit 202 and ACS2 unit 203), a first internal memory 205 and a second internal memory. 206 is provided.
The ACS 1 unit 202 performs an α operation and a β operation,
The unit 203 also performs α calculation and β calculation. ACS1 section 202
Has a γ operation unit 202-1 and the ACS2 unit 203
It has a calculation unit 203-1. γ operation units 202-1 and 20
3-1 calculates a branch metric from the received data and the reliability information likelihood La. The first internal memory 205
It is similar to the internal memory 204 of the conventional example.

Hereinafter, in the present embodiment, the data length k = 330 to be decoded, the number of states S at each time point S = 8, the window size Q1 = 32 for performing α and β operations, and the window size Q2 = for performing only β operations = Set to 0.

First, the number of repetitions referred to in the present invention is defined. One repetition in the present invention means that the soft-output decoder 101 and the soft-output decoder 102 each perform the decoding operation once. Therefore, the relationship between the decoding operation of the soft output decoder 101 and the soft output decoder 102 and the number of repetitions is as shown in FIG.

According to the present invention, when the number of repetitions is an odd number, as shown in FIG. 1, the α operation is performed first for each sliding window, and the β operation is performed later, but when the number of repetitions is even, As shown in FIG. 2, the β operation is performed first and the α operation is performed later for each sliding window. If the number of repetitions is odd,
As shown in FIG. 1, the sliding window is slid from time 0 to time 329. When the number of repetitions is an even number, as shown in FIG.
The sliding window is slid from 29 to time 0.

When the number of repetitions is odd,
(1) The state metric for each state at the starting point of the window used when the ASC1 unit 202 starts the α operation for a window is the ASC2 unit 2
03 is a state metric for each state at the end point of the immediately preceding window obtained by performing an α operation on the immediately preceding window, and (2) AS
The state metric for each state at the starting point of the window used when the C2 unit 202 starts the α operation on a window is obtained by the ASC1 unit 202 performing the α operation on the window one window before the window. (3) ASC1 unit 20 is a state metric for each state at the end point of the previous window.
The state metric for each state at the end point of the window used when starting the β operation for a certain window is the immediately preceding repetition (the number of repetitions is an even number) in the ASC1 section 202 or the ASC2 section 20.
3 is a state metric for each state at the starting point of the window after the window obtained by performing the β operation on the window after the window, and (4) AS
The state metric for each state at the end point in the window used when the C2 unit 203 starts the β operation for a certain window is the last repetition (the number of repetitions is even), the ASC1 unit 202 or the AS
This is a state metric for each state at the start point of the next window obtained by the C2 unit 203 performing β operation on the window immediately after the window.

When the number of repetitions is odd, A
The state metric for each state at the end point of the window obtained by performing the α operation on the window of the SC1 unit 202 is stored in the second internal memory 206,
The state metric for each state at the end point of the window obtained by performing the α operation on the window of the ASC2 unit 202 is also stored in the second internal memory 206.

When the number of repetitions is an even number,
(1) The state metric for each state at the end point of the ASC1 unit 202 used when starting the β operation for a certain window is ASC2 unit 2
03 is a state metric for each state at the starting point of the next window obtained by performing the β operation on the next window after the window, and (2) AS
The state metric for each state at the end point of the window used when the C2 unit 202 starts the β operation for a window is obtained by the ASC1 unit 202 performing the β operation on the window one window after the window. (3) ASC1 unit 20 is a state metric for each state at the start point of the next window.
The state metric for each state at the start point in the window used when the α operation is started for a certain window is the last repetition (the number of repetitions is an odd number), and the ASC1 unit 202 or the ASC2 unit 20
3 is a state metric for each state at the end point of the immediately preceding window obtained by performing the α operation on the immediately preceding window, and (4) AS
The state metric for each state at the starting point in the window used when the C2 unit 203 starts the α operation for a certain window is the last repetition (the number of repetitions is an odd number) in the ASC1 unit 202 or AS
The state metrics at the end point of the immediately preceding window obtained by the C2 unit 203 performing the α operation on the window immediately before the window.
In addition, the starting point, the ending point, and the relation before and after the preamble are those when viewed without inverting the time axis.

When the number of repetitions is an even number, A
The state metric for each state at the start point of the window obtained by performing the β operation on the window of the SC1 unit 202 is stored in the second internal memory 206,
The state metric for each state at the start point of the window obtained by performing the β operation on the window of the ASC2 unit 202 is also stored in the second internal memory 206. In addition, the starting point, the ending point, and the relation before and after the preamble are those when viewed without inverting the time axis.

When the number of repetitions is an odd number, similarly to the conventional example, when performing the β operation on a certain window, each state at each time point of the window obtained by the α operation on the window is obtained. State metrics are used. On the other hand, when the number of repetitions is even, unlike the conventional example, when performing the α operation on a certain window, the state metric for each state at each time point of the window obtained by the β operation on the window is obtained. Is used.

Next, the turbo decoding method according to the embodiment of the present invention will be described more specifically.

FIG. 1 is a conceptual diagram for explaining the operation when the number of repetitions of the soft output decoders 101 and 102 employing the turbo decoding method according to the embodiment of the present invention is an odd number, and FIG. FIG. 7 is a timing chart for explaining an operation when the number of repetitions of the soft output decoders 101 and 102 employing the turbo decoding method according to the embodiment of the present invention is an odd number.

Referring to FIG. 1 and FIG. 3, the reception data A
in1 is obtained from the reliability information likelihood memory unit 7 at time points 0 to 31.
Read the reliability information likelihood Rin1 at each time point of
The read reception data Ain1 and the reliability information likelihood Ri
The α operation is performed by the ACS 1 unit 202 using n1. At the same time, the read reception data Ain1 and the reliability information likelihood Rin1 and the time points 0 to 31 obtained from the α operation
Is stored in the first internal memory 205 for each state at each time point. Also,
The state metric αs (1) for each state at the time point 31 obtained from the α operation is stored in the second memory 206 (indicated by S301 in FIG. 1).

Next, in the immediately preceding repetition (the number of repetitions is an even number), the ACS 2
6 (indicated by S415 in FIG. 2), the state metric βs for each state at time 31
After reading (1) (indicated by S302 in FIG. 1),
The received data Aout1 (= Ain1) and the reliability information severity Rout1 (= Rin1) at each of the time points 31 to 0 are read from the internal memory 205, and the read βs
(1) Received data Aout1 and reliability information likelihood Ro
The β calculation is executed by the ACS1 unit 202 using ut1.
The reliability information at each of the time points 31 to 0 obtained as a result of the β operation is written to the reliability information likelihood memory 7 via a subtractor, a multiplier, and an interleaver (or a deinterleaver). In this example, in S302, the second memory 2
The state metric βs (1) read from 06 is A
Although the CS2 unit 203 has written the data, the ACS1 unit 202 depends on the data length and the window size.
May have been written.

Since the ACS 1 unit 202 and the ACS 2 unit 203 operate in parallel, the ACS 1 unit 202 performs the β operation at the time points 31 to 0, and at the same time, the ACS 2 unit 203 performs the α operation at the time points 32 to 63. ACS1 section 202
Reads out αs (1) written in the second internal memory 206 when the α operation is performed for the time points 0 to 31 (S303 in FIG. 1), and then reads out each of the time points 32 to 63 from the reception data memory unit 6. Received data Ain2 at the time
Is read out from the reliability information likelihood memory unit 7 at the time points 32 to 63, and the read reception data Ain2 and the reliability information likelihood Rin2 are read out.
Is performed by the ACS 2 unit 203 using At the same time, the read data Ain2 and the reliability information likelihood Rin2 and the time points 32 to 63 obtained from the α operation
Is stored in the first internal memory 205 for each state at each point in time. Also,
The state metric αs (2) for each state at the time point 63 obtained from the α operation is stored in the second memory 206 (indicated by S304 in FIG. 1).

Next, the ACS 2 unit 203 determines that the time points 31 to 63
.Alpha.s (2) written in the second internal memory 206 when the .alpha. Operation was performed on .alpha.
Indicated at 06. ), From the reception data memory unit 6 to the time point 6
The reception data Ain3 at each time point of 4 to 95 is read from the reliability information likelihood memory unit 7 to read the reliability information likelihood Rin3 at each time point of 64 to 95, and the read reception data Ain3 and reliability information likelihood Rin3 are read. Is performed by the ACS 1 unit 202 using. At the same time, the read reception data Ain3 and the reliability information likelihood Rin
3 and the state metric Sin3 for each state at the time points 64 to 95 obtained from the α operation are stored in the first internal memory 205. Further, the state metric αs (3) for each state at the time point 95 obtained from the α operation is stored in the second memory 206 (indicated by S307 in FIG. 1).

Since the ACS 1 unit 202 and the ACS 2 unit 203 operate in parallel, the ACS 1 unit 202
At the same time as performing the α operation, the ACS 2 unit 203 performs the β operation for the time points 63 to 32. In the immediately preceding repetition (the number of repetitions is an even number), the ACS 1 unit 202 writes the state metric βs (2) for each state at the time 63 written in the second internal memory 206 (indicated by S412 in FIG. 2). Read (S30 in FIG. 1)
Indicated by 5. ), Time points 63 to 3 from the internal memory 205
2 Received data Aout2 (= Ain2) at each time point
And the reliability information severity Rout2 (= Rin2) is read out, and the ACS2 unit 203 performs β calculation using the read βs (2), the received data Aout2, and the reliability information likelihood Rout2. Time points 63 to 3 obtained as a result of β operation
The reliability information at each point of time 2 is written to the reliability information likelihood memory 7 via a subtractor, a multiplier, and an interleaver (or a deinterleaver). In this example, S3
The state metric βs (2) read from the second memory 206 at 05 is written by the ACS1 unit 202, but may be written by the ACS2 unit 203 depending on the data length and window size. There is also.

Hereinafter, the ACS unit 202 and the AC
The S2 unit 203 repeats the parallel operation.

FIG. 2 is a conceptual diagram for explaining the operation of the soft-output decoders 101 and 102 employing the turbo decoding method according to the embodiment of the present invention when the number of repetitions is an even number. FIG. 4 is a timing chart for explaining an operation when the number of repetitions of the soft output decoders 101 and 102 employing the turbo decoding method according to the embodiment of the present invention is an even number.

Referring to FIGS. 2 and 4, received data Ain11 at each time point from time 329 to time 320 from received data memory section 6 is stored in reliability information likelihood memory section 7 at reliability time at each time point 329 to 320. Information likelihood Rin
11 'is read out, and the ACS1 unit 20 is read using the read reception data Ain11 and the reliability information likelihood Rin11'.
In step 2, a β operation is performed. At the same time, the readout data Ain11 and the reliability information likelihood Rin11 ′ and the state metric Sin11 ′ for each state at the time points 329 to 320 obtained from the β operation are stored in the first internal memory 205. Further, the state metric βs (n−1) of each state at the time point 320 obtained from the β operation is stored in the second memory 206 (indicated by S401 in FIG. 2).

Next, in the immediately preceding repetition (the number of repetitions is an odd number), the ACS 2
6 (indicated by S315 in FIG. 1), the state metric α for each state at time 319.
s (n-1) is read out (indicated by S402 in FIG. 2).
From the internal memory 205, read out the received data Aout11 (= Ain11) and the reliability information severity Rout11 ′ (= Rin11 ′) at each of the time points 329 to 320, read αs (n−1), and the received data Ao.
ut11 and reliability information likelihood Rout11 ′
The CS unit 202 performs an α operation. The reliability information at each of the time points 329 to 320 obtained as a result of the α operation is written to the reliability information likelihood memory 7 via a subtractor, a multiplier, and an interleaver (or a deinterleaver).
In this example, the state metric αs (n−1) read from the second memory 206 in S402 is ACS
The data is written by the second unit 203, but may be written by the ACS1 unit 202 depending on the data length and the window size.

Since the ACS 1 unit 202 and the ACS 2 unit 203 operate in parallel, the ACS 1 unit 202
At the same time as performing the α operation on the
No. 3 performs the β calculation for the time points 319 to 288. ACS
Βs written in the second internal memory 206 when the first unit 202 performed the β operation for the time points 329 to 320
After reading (n-1) (indicated by S403 in FIG. 2), the received data Ain10 at each of the time points 319 to 288 from the received data memory unit 6 is transferred from the reliability information likelihood memory unit 7 to the time points 319 to 319. In step 288, the reliability information likelihood Rin10 ′ at each time point is read, and the ACS 2 unit 203 performs a β operation using the read received data Ain10 and the reliability information likelihood Rin10 ′. At the same time, the read reception data Ain10 and the reliability information likelihood Rin10 ′ and the time points 319 to 2 obtained from the β operation
The state metric Sin 10 ′ for each state at each point of time 88 is stored in the first internal memory 205. The state metric βs (n−2) for each state at time 288 obtained from the β calculation is
(Indicated by S404 in FIG. 2).

Next, the ACS 2 unit 203 determines whether
When the β operation is performed on 88, the second internal memory 20
After reading out βs (n−2) written in No. 6 (indicated by S406 in FIG. 2), the received data Ai at each of the time points 287 to 256 is read from the received data memory unit 6.
n9 from the reliability information likelihood memory unit 7 at time points 287 to 2
The reliability information likelihood Rin 9 ′ at each time point 56 is read out, and the ACS unit 202 performs a β operation using the read received data Ain 9 and the reliability information likelihood Rin 9 ′. At the same time, the read reception data Ain9 and the reliability information likelihood Rin9 ′ and the state metric Sin9 ′ for each state at the time points 287 to 256 obtained from the β operation are stored in the first internal memory 205.
To be stored. Further, the state metric βs (n−
3) is stored in the second memory 206 (S407 in FIG. 2).
Indicated by ).

Since the ACS 1 unit 202 and the ACS 2 unit 203 operate in parallel, the ACS 1 unit 202
At the same time as performing the β operation on
No. 3 performs an α operation for time points 319 to 288. In the last repetition (the number of repetitions is even), ACS1 part 2
02 is written to the second internal memory 206 (S in FIG. 1).
Shown at 312. ), The state metric αs (n−2) for each state at the time point 287 is read (indicated by S405 in FIG. 2), and the received data Aou at the time points 319 to 288 from the internal memory 205 is read.
t10 (= Ain10) and reliability information severity Rout1
0 ′ (= Rin10 ′), and the read αs
(N-2), the ACS 2 unit 203 executes the α operation using the received data Aout10 and the reliability information likelihood Rout10 ′. The reliability information at each of the time points 319 to 288 obtained as a result of the α operation is written to the reliability information likelihood memory 7 via a subtractor, a multiplier, and an interleaver (or a deinterleaver). In this example, S405
Although the state metric αs (n−2) read from the second memory 206 is written by the ACS1 unit 202, it may be written by the ACS2 unit 203 depending on the data length and window size. In some cases.

Hereinafter, similarly, the ACS section 202 and the AC
The S2 unit 203 repeats the parallel operation.

In the conventional sliding window method, since the likelihood of each state at the time of starting the β calculation is unknown, the calculation must be started with the likelihood of all states set to 0 at the start of the β calculation. Therefore, unless the value of Q2 is set to be relatively large, the characteristics will deteriorate due to the convergence of the paths. However, as in the present invention, the value stored in the second internal memory 206 is used as the state metric when starting the likelihood calculation at the time of β calculation in the odd number operation and at the time of the α operation in the even number time. Even if the value is set to a small value (or 0 as in this example), deterioration of the characteristics is small and the processing time can be significantly shortened.

However, in the first iteration as described above, in order to reduce the characteristic deterioration caused by the state metrics starting with all 0 in both the β operation and the α operation, the operation is performed irregularly as follows. It is also conceivable to use a method for preventing the characteristic deterioration.

Only during the first iteration, the decoding operation is performed with Q2 at the time of the β operation being a sufficiently large value, and after the next iteration, the decoding operation based on the present invention is performed with the value of Q2 set to 0. By doing so, the processing time of the ACS processing required for the first iteration is increased, but Q2 becomes 0 from the next iteration, so that the decoding processing time as a whole is shortened and the characteristic deterioration is small. .

Note that the turbo decoding system according to the present invention can also be realized by a computer reading and executing a program for causing a computer to function as the turbo decoding system from a recording medium.

[0072]

As described above, according to the present invention,
The following effects are obtained.

The first effect is that the processing time of the ACS processing can be drastically reduced by using a small memory area as compared with the conventional sliding window algorithm, and the processing time required for turbo decoding is greatly reduced. That can be reduced.

The reason is that the use of the state metric obtained in the previous repetition operation as the initial value of the β operation or α operation makes it possible to greatly reduce Q2 with little characteristic deterioration. For example, Q1 = 32, Q2
Processing with Q = 64 window size and Q1 = 3
2. Comparing the processing time of the ACS processing with the window size of Q2 = 0, the latter can reduce the processing time to half. This means that the entire iterative decoding processing time can be reduced to half that of the conventional method.

The second effect is that, in the above embodiment, the second internal memory provided inside the soft output decoder may be provided outside the soft output decoder LSI. That is, the circuit scale of the LSI does not increase.

The reason is that the writing of the state metric to the second internal memory and the reading of the state metric from the second internal memory can be performed in parallel with the ACS processing.

[Brief description of the drawings]

FIG. 1 is a first conceptual diagram illustrating an operation of a turbo decoding scheme according to an embodiment of the present invention.

FIG. 2 is a second conceptual diagram for explaining the operation of the turbo decoding system according to the embodiment of the present invention.

FIG. 3 is a first timing chart for explaining an operation of the turbo decoding system according to the embodiment of the present invention.

FIG. 4 is a second timing chart for explaining the operation of the turbo decoding method according to the embodiment of the present invention.

FIG. 5 is a diagram illustrating the number of repetitions defined in the turbo decoding scheme according to the embodiment of the present invention.

FIG. 6 is a block diagram illustrating a configuration of a soft-output decoder according to an embodiment of the present invention.

FIG. 7 is a diagram for explaining a sliding window method as a kind of the turbo decoding method.

FIG. 8 is a block diagram illustrating a configuration of a receiving device according to an embodiment of the present invention.

FIG. 9 is a diagram illustrating a configuration of a turbo decoding scheme.

FIG. 10 is a block diagram showing a configuration of a conventional soft output decoder.

FIG. 11 is a timing chart for explaining an operation of a sliding window system according to a conventional example.

FIG. 12 is a conceptual diagram illustrating an operation of a sliding window system according to a conventional example.

[Explanation of symbols]

1 antenna 2 Transmission / reception separation unit 3 Reception radio section 4 Despreading part 5 Demodulation unit 6 Received data memory section 7 Reliability information likelihood memory unit 8 Turbo decoding unit 9 Information source extraction unit 101 Soft output decoding unit 102 Soft output decoding unit 103 interleaver 104 deinterleaver 105 interleaver 202 ACS1 section 203 ACS2 section 205 first internal memory 206 second internal memory

Continuation of the front page (56) References JP-A-2000-278144 (JP, A) WO 00/052833 (WO, A1) Kazuo Obuchi, Tetsuya Yano, Kazuaki Kakuda, Hideo Ichiba, Kazuo Kawabata, Ryuji Nakamura, RCS99-40: Circuit size reduction method of Sub-log-MAP in Turbo code, IEICE Technical Report [Wireless Communication System], Japan, Vol. 99, No. 142, p. 1-6 Viterbi, A .; J. , An initiative justification and a simplicity of implemen- tation of the MAP decoder for con- volution al cod, IEEE Journal of the United States. 16, No. 2, p. 260-264 (58) Fields investigated (Int. Cl. ⁷ , DB name) H03M 13/00-13/53

Claims (9)

(57) [Claims] A first soft-output decoder that performs forward processing and backward processing; a second soft-output decoder that performs forward processing and the backward processing; and the first soft-output decoder. An interleaver for interleaving the output reliability information likelihood and supplying the output to the second soft output decoder; and deinterleaving the reliability information likelihood output from the second soft output decoder to the first soft output decoder. And a deinterleaver for supplying the soft output decoder to the soft output decoder. In a turbo decoding method in which decoding is performed alternately by the first soft output decoder and the second soft output decoder, the forward interleaving is performed for each window. Perform the processing first
Decryption after word processing and
Perform the backward processing first and the forward processing later
The decoding to be performed is alternately repeated, and at least one of the first soft-output decoder and the second soft-output decoder performs the forward processing for each window first in a certain iteration and performs the backward processing later. When performing, the state metric for each state at the end point of each window obtained by the forward processing in the repetition is stored, and in the next repetition, the backward processing is performed for each window first and the forward processing is performed. Means for using the stored state metric as an initial value of the state metric of each state at the starting point of each window in the forward processing in the next iteration; and performing the backward processing for each window in a certain iteration. Line first When the forward processing is performed later, the state metric for each state at the start point of each window obtained by the backward processing in the repetition is stored, and in the next repetition, the forward processing is performed for each window first. Means for performing the backward processing at a later time, and using the stored state metric as an initial value of the state metric of each state at the end point of each window in the backward processing at the next iteration. The turbo decoding method characterized by the above-mentioned. 2. The turbo decoding method according to claim 1, wherein the forward processing is an operation of executing a Viterbi algorithm from a start point to an end point of data, and the backward processing is an operation of executing the Viterbi algorithm on the data. A turbo decoding method characterized in that the calculation is performed from the end point to the start point of (1). 3. The turbo decoding method according to claim 1, wherein at the time of the first iteration, a window size used in an operation performed later in the forward processing or the backward processing performed on each window is set to one of them. And a means for making the window size larger than a window size used in an operation performed before the turbo decoding. 4. A first soft-output decoder for performing forward processing and backward processing, a second soft-output decoder for performing forward processing and backward processing, and a first soft-output decoder. An interleaver for interleaving the output reliability information likelihood and supplying the output to the second soft output decoder; and deinterleaving the reliability information likelihood output from the second soft output decoder to the first soft output decoder. And a deinterleaver for supplying to the soft output decoder of the above. A turbo decoding method performed by a turbo decoding device that repeats decoding by the first soft output decoder and the second soft output decoder alternately, The forward processing is performed first for each window, and the
Decryption after word processing and
Perform the backward processing first and the forward processing later
The decoding to be performed is alternately repeated, and at least one of the first soft-output decoder and the second soft-output decoder performs the forward processing for each window first in a certain iteration and performs the backward processing later. When performing, the state metric for each state at the end point of each window obtained by the forward processing in the repetition is stored, and in the next repetition, the backward processing is performed for each window first and the forward processing is performed. Using the stored state metric as an initial value of the state metric of each state at the starting point of each window in the forward processing in the next iteration. Word processing Is performed first and the forward processing is performed later, a state metric for each state at the start point of each window obtained by the backward processing in the repetition is stored, and in the next repetition, the Performing forward processing first, performing the backward processing later, and using the stored state metric as an initial value of the state metric of each state at the end point of each window in the backward processing in the next iteration. Performing a turbo decoding method. 5. The turbo decoding method according to claim 4, wherein the forward processing is an operation of executing a Viterbi algorithm from a start point to an end point of the data, and the backward processing is an operation of executing the Viterbi algorithm on the data. A turbo decoding method, characterized in that the calculation is performed from the end point to the start point. 6. The turbo decoding method according to claim 4, wherein at the time of the first iteration, a window size used in an operation performed after the forward processing or the backward processing performed on each window is set to one of them. A step of making the window size larger than a window size used in an operation performed before the decoding. 7. A first soft-output decoder that performs forward processing and backward processing, a second soft-output decoder that performs forward processing and backward processing, and the first soft-output decoder An interleaver for interleaving the output reliability information likelihood and supplying the output to the second soft output decoder; and deinterleaving the reliability information likelihood output from the second soft output decoder to the first soft output decoder. And a deinterleaver that supplies the soft output decoder to a soft output decoder, wherein a turbo decoding system that repeats decoding of the first soft output decoder and the second soft output decoder alternately is provided for each window. Perform the forward processing first and
Decryption after word processing and
Perform the backward processing first and the forward processing later
The decoding to be performed is alternately repeated, and at least one of the first soft-output decoder and the second soft-output decoder performs the forward processing for each window first in a certain iteration and performs the backward processing later. When performing, the state metric for each state at the end point of each window obtained by the forward processing in the repetition is stored, and in the next repetition, the backward processing is performed for each window first and the forward processing is performed. Means for using the stored state metric as an initial value of the state metric of each state at the starting point of each window in the forward processing in the next iteration; and performing the backward processing for each window in a certain iteration. Line first When the forward processing is performed later, the state metric for each state at the start point of each window obtained by the backward processing in the repetition is stored, and in the next repetition, the forward processing is performed for each window first. Means for performing the backward processing at a later time, and using the stored state metric as an initial value of the state metric of each state at the end point of each window in the backward processing at the next iteration. A program that causes a computer to function as a decoding method. 8. The program according to claim 7, wherein the forward processing is an operation for executing the Viterbi algorithm from a start point to an end point of the data, and the backward processing is an operation for executing the Viterbi algorithm at the end point of the data. A program, which is an operation to be executed from to a starting point. 9. The program according to claim 7, wherein in the turbo method, at the time of the first iteration, a window size used in an operation performed later in the forward processing or the backward processing performed for each window is set. A program comprising means for increasing a window size to be used in an operation performed before one of them.