CN107124249B - Method for reducing stack overflow probability of decoder of Fano algorithm - Google Patents

Abstract

The invention discloses a method for reducing stack overflow probability of a Fano algorithm decoder, which mainly solves the problem that the stack overflow probability of the Fano algorithm decoder is too high. The specific idea of the invention is to increase the observation distance to screen out the optimal node so as to reduce the probability of the decoder entering the wrong path; the second is to determine the reliability of a specific point to modify its Fano metric to reduce the probability of the decoder leaving the correct path. The method comprises the following specific steps: (1) initializing parameters of a decoder; (2) screening out an optimal node; (3) processing the decoding node; (4) judging the shutdown condition; (5) correcting the Fano measurement of a specific point; (6) tightening the threshold; (7) obtaining a backward peeking measurement; (8) reducing the threshold; (9) the observation direction is selected. The invention has the advantages of low complexity of system realization, high decoding speed and low stack overflow probability. The method obviously reduces the stack overflow probability of the Fano algorithm decoder, and improves the decoding speed and the decoding stability of the convolutional code Fano algorithm decoding when the channel is poor.

Description

A method to reduce the stack overflow probability of Fano algorithm decoder

technical field

The invention belongs to the field of communication technology, and further relates to a method for reducing the stack overflow probability of a Fano algorithm decoder in the field of channel coding. Decoding speed and decoding stability when the channel is poor.

Background technique

Convolutional codes, proposed by Elias et al. in 1955, are a very promising coding method and have been widely used. The convolutional code decoding method mainly adopts probability decoding. Probabilistic decoding is further divided into two categories: Viterbi decoding and sequence decoding.

Viterbi decoding is a maximum likelihood decoding algorithm and an optimal probability decoding method. When the constraint length of the code is small, it has the advantages of fast speed, constant computation and simple decoder. Since it was proposed, it has developed rapidly, both in theory and in practice, and is widely used in various data transmission systems, especially in satellite communications. However, the complexity of Viterbi decoding of convolutional codes increases exponentially, so it cannot be applied to too large codes, thus limiting the bit error rate output by Viterbi decoding to be too low, resulting in Viterbi decoding. The code method is limited in application.

Sequence decoding is a quasi-maximum likelihood decoding algorithm, and the decoding complexity has nothing to do with the constraint length of the code, so that the application of larger codes is possible, thereby obtaining a lower error rate. The calculation amount of sequence decoding varies with the size of the channel interference, so that the average calculation amount can be reduced and the decoding speed can be accelerated.

The Fano algorithm was proposed by Fano in 1963 and is "one of the two major algorithms" in sequence decoding. In a nutshell, the basic principle of decoding is to continuously move the observation point (fork node in the code tree graph) in the code tree graph, obtain the Fano metric of each branch by comparing the received code group and the code tree branch, and select the code tree. Move forward on the path with the smallest metric, through the total metric value of the path and the preset threshold value, detect back and forth, search forward and backward repeatedly, to eliminate the wrong path as soon as possible, and return to the correct path as soon as possible with the greatest correct probability, Thereby, the average calculation amount of the decoder is reduced. Another large algorithm for sequence decoding is the stack decoding algorithm. Compared with the stack decoding algorithm, Fano algorithm decoding needs to search more nodes, but the stack decoding algorithm requires a very large amount of memory. In addition, the multi-layer judgment and nested loop of Fano algorithm are very suitable for hardware implementation.

The disadvantage of Fano's algorithm is that the decoder of Fano's algorithm needs an input buffer to store the input received sequence, for the decoder to store the subsequent input sequence when searching for the separation point, and to provide the previously received received sequence. . But when the channel interference is very large, the decoder search time is very long, which may cause buffer overflow. Therefore, in the Fano algorithm, the decoding error probability is not a major problem, but buffer overflow is a major problem.

Mao Shuhua et al. proposed a threshold in the article "A New Convolutional Code Decoding Algorithm Based on Branch Metrics" (Journal of Yunnan University (Natural Science Edition), 2010, 32(SI): 376-378) Adjustable sequence decoding algorithm to overcome the problem of buffer overflow. The algorithm uses a new branch metric and adjusts the threshold level by sending a specific pilot sequence to estimate the noise level. The disadvantage of this method is that it inserts pilot frequency, which reduces the information rate.

SUMMARY OF THE INVENTION

The object of the present invention is to overcome the above-mentioned deficiencies of the prior art, and to provide a method for reducing the stack overflow probability of the Fano algorithm decoder. Only a small amount of calculation is added, and the performance loss of Fano algorithm can be significantly reduced in the channel. The average number and maximum number of searches when the interference is large, thereby improving the decoding speed and reducing the probability of stack overflow.

The specific ideas for realizing the purpose of the present invention are as follows: one is to increase the observation distance to filter out the optimal node, so as to reduce the probability that the decoder enters the wrong path; Probability of leaving the correct path.

The present invention realizes above-mentioned purpose concrete steps are as follows:

(1) Decoder parameter initialization:

Set the threshold value of the decoder to zero, set the Fano metric of the first node of the code tree to zero, and set the first node of the code tree to be the decoding node;

(2) Screen out the optimal node:

(2a) read the code group corresponding to the current decoding node from the sequence buffer;

(2b) according to the code group corresponding to the current decoding node, calculate the Fano metric of each branch starting from the current decoding node and the Fano metric of the node reached by each branch starting from the current decoding node;

(2c) According to the candidate point screening method, select one of the above branches as the optimal node;

(3) Process the decoding node:

(3a) Calculate the Fano metric of the optimal node;

(3b) Judging whether the Fano metric of the optimal node is greater than or equal to the threshold T, if so, move the current decoding node to the optimal node, otherwise, execute step (7);

(4) Judging the shutdown conditions:

Determine whether the current decoding node is the end point of the code tree, if so, stop, otherwise, continue to the next step;

(5) Correct the Fano metric of a specific point:

Correct the Fano metric of a specific point according to the point reliability judgment method, where the specific point is the point specified in the point reliability judgment method;

(6) Compress the threshold:

(6a) Judging whether the current decoding node is visited for the first time, and if so, perform a tightening threshold processing; otherwise, keep the current threshold unchanged;

(6b) perform step (2);

(7) Get the backward peek measure:

Taking the Fano metric of the previous node of the current decoding node as the backward peeping metric of the current decoding node, if the current decoding node is the initial node, then setting the backward peeping metric of the current node to be negative infinity;

(8) Reduce the threshold:

It is judged whether the backward peeping measure is less than the threshold, if so, the threshold is reduced, and then step (2) is performed, otherwise, the decoding node is rolled back to the previous node of the current decoding node;

(9) Select the observation direction:

According to the worst branch judgment method, it is judged whether the regression of the decoding node in the previous step is from the worst branch, and if so, execute step (7); Step (3).

Compared with the prior art, the present invention has the following advantages:

First, since the present invention adopts the method of increasing the observation distance and judging the reliability of a specific point, through a small amount of calculation, it overcomes the excessive average number of visits and the maximum number of visits, and the probability of stack overflow when the original Fano algorithm is in poor channel conditions. Excessive shortcomings make the present invention have the advantages of fast decoding speed, low stack overflow probability and high decoding stability.

Second, because the method of increasing the observation distance and judging the reliability of a specific point is adopted in the present invention, only a small amount of calculation is added, which overcomes the shortcoming of inserting a training sequence in the prior art to reduce the information rate, so that the present invention has the complexity of system implementation. The advantages of low and high information rate.

Description of drawings

Fig. 1 is the algorithm flow chart of the present invention;

Fig. 2 is the flow chart of the candidate point screening method of the present invention;

Fig. 3 is the flow chart of the point reliability judgment method of the present invention;

Fig. 4 is the simulation diagram of the average number of visits that the present invention decodes a coded information block to search for access state;

5 is a CCDF simulation diagram of the present invention when E _b /N ₀ is 0dB to decode a coded information block to search for access state times;

6 is a CCDF simulation diagram of the present invention when E _b /N ₀ is 0.5dB to decode a coded information block to search for access state times;

FIG. 7 is a simulation diagram of the received decoding bit error rate according to the present invention.

Detailed ways

Referring to FIG. 1 , the implementation method of the present invention will be further described below with reference to the embodiments.

Step 1, the decoder parameters are initialized.

The threshold value of the decoder is set to zero, the Fano metric of the first node of the code tree is set to zero, and the first node of the code tree is set as the decoding node.

The code tree is an expression of the state transition of the convolutional code. The nodes of the code tree correspond to the state of the convolutional code, and the state transitions generated by different inputs of the convolutional code correspond to each branch in the code tree. The depth of the code tree is determined by The coding length is determined, and the number of branches of each node is determined by the parameters of the convolutional code.

Step 2, filter out the optimal node.

First, the code group corresponding to the current decoding node is read from the sequence buffer.

Secondly, according to the code group corresponding to the current decoding node, the Fano metric of each branch starting from the current decoding node and the Fano metric of the node reached by each branch starting from the current decoding node are calculated respectively.

The Fano metric of each branch starting from the current decoding node is calculated according to the following formula:

where log ₂ represents the base 2 logarithm, R _i represents the i-th code group vector received, c _i represents the i-th code group vector sent, and P(R _i | _ci ) represents the ith code group vector at c _i The probability of occurrence of R _i under the condition, R _c represents the probability of occurrence of R _i , n ₀ represents the length of the convolutional code segment, R _c represents the code rate, and _MF (R _i |c _i ) represents the Fano metric of the ith branch .

The Fano metric of the node reached by each branch from the current decoding node is calculated according to the following formula:

M _i+1 =M _i +M _F (R _i |c _i )

Among them, M _i is the Fano metric of the i-th decoding node, M _i+1 is the Fano metric from the i+1-th decoding node, and M _F (R _i |c _i ) is the i-th decoding node and the Fano metric of the corresponding branch between the i+1th decoding node.

Finally, one of the above branches is selected as the optimal node according to the candidate point screening method.

The specific steps of the candidate point screening method in the present invention will be described below with reference to FIG. 2 .

The first step is to select the branch with the largest Fano metric as the best branch from all branches from the current decoding node;

The second step is to determine whether the number of optimal branches is greater than 1. If so, then multiple optimal branches are obtained. Otherwise, this branch is set as the optimal branch, and the optimal node is set as the optimal branch from the current decoding node along the optimal branch. Step 10 is performed on the next node reached in the branch direction;

The third step is to take any optimal branch, and set the node reached by the optimal branch as a temporary node;

The fourth step is to calculate the Fano metric of each branch starting from the temporary node, and select the branch with the largest Fano metric as the second best branch;

The fifth step is to determine whether the number of secondary optimal branches is equal to 1, if so, set the node reached by this secondary optimal branch as the secondary pre-selected node, otherwise, select a secondary optimal branch and set the The node reached by this second best branch is set as the second preselected node;

The sixth step is to judge whether all the best branches have been taken, if so, obtain multiple secondary pre-selected nodes, otherwise, perform the third step;

The seventh step is to calculate the Fano metric of multiple secondary preselected nodes;

The eighth step is to select the node with the largest Fano metric from all the secondary preselected nodes;

The ninth step is to judge whether the number of nodes with the largest Fano metric selected in the previous step is greater than 1. If so, set the optimal node to the previous node of any one of them, otherwise, set the optimal node to the previous node of the node. a node.

The tenth step is to end all steps and get the optimal node.

Step 3, processing the decoding node.

First, calculate the Fano metric of the optimal node.

Then, it is judged whether the Fano metric of the optimal node is greater than or equal to the threshold T, if so, move the current decoding node to the optimal node, otherwise, go to step 7.

Step 4, judging the shutdown condition.

It is judged whether the current decoding node is the end point of the code tree, if so, stop the operation, otherwise, continue to execute the next step.

Step 5: Correct the Fano metric of a specific point.

The Fano metric of a specific point is modified according to the point reliability judgment method, and the specific point is the point specified in the point reliability judgment method.

The specific steps of the midpoint reliability judgment method of the present invention will be described below with reference to FIG. 3 .

The first step is to set the number of backtracking points to a fixed value, and set the reliability threshold to a fixed value;

The second step is to judge whether the order of the current decoding node in the code tree is less than or equal to the number of backtracking points, if so, end all steps, otherwise, continue to the next step;

The third step is to calculate the difference obtained by subtracting the Fano metric of the backtracking end point from the Fano metric of the current decoding node, where the backtracking end point represents the node reached after returning to the backtracking point several nodes after the current decoding node;

The fourth step is to judge whether the difference obtained in the previous step is less than the reliability threshold value, if so, end all steps, otherwise, continue to execute the next step;

The fifth step is to judge whether the decoder reaches the current node for the first time, and if so, reset the Fano metric of the previous node of the backtracking end point to negative infinity, otherwise, end all steps.

Step 6, compress the threshold.

First, it is judged whether the current decoding node is visited for the first time, and if so, the tightening threshold processing is performed; otherwise, the current threshold is kept unchanged.

The specific steps of the tightening threshold processing are: cyclically execute T←T+Δ, until the Fano metric M of the current decoding node satisfies T≤M<T+Δ, where T represents the threshold value, and Δ represents the threshold increment , M represents the Fano metric of the current node, and the threshold increment is usually an integer, ranging from 2 to 8.

Next, go to step 2.

Step 7, get the backward peeking metric.

The Fano metric of the previous node of the current decoding node is used as the backward peeping metric of the current decoding node. If the current decoding node is the initial node, the backward peeping metric of the current node is set to negative infinity.

Step 8, reducing the threshold.

It is judged whether the backward peeping metric is smaller than the threshold, and if so, the threshold is reduced, and then step 2 is performed; otherwise, the decoding node is rolled back to the previous node of the current decoding node.

The specific steps for reducing the threshold are as follows: let T←T-Δ, where T represents the threshold value, Δ represents the threshold increment, and the threshold increment usually takes an integer and ranges from 2 to 8. .

Step 9, select the observation direction.

According to the worst branch judgment method, it is judged whether the regression of the decoding node in the previous step is from the worst branch, if so, go to step 7, otherwise, look forward to the optimal node of the remaining nodes, and then go to step 3 .

The concrete steps of the described worst branch judgment method are: judging whether the current branch is the last branch selected and advanced in all branches of the current decoding node, if so, the current branch is the worst branch, otherwise, the current branch is not the worst branch. bad branch.

The effect of the present invention will be further described below through the simulation experiment of the present invention.

1. Simulation conditions:

The simulation software of the present invention uses Matlab 8.5.0, the simulation channel is a Gaussian channel, the modulation mode is BPSK modulation, the convolutional code is (2,1,7) convolutional code, and its generator polynomial is as follows:

G(D)=[1+D ² +D ³ +D ⁵ +D ⁶ ,1+D ¹ +D ² +D ³ +D ⁶ ]

The length of the encoded information block at the sending end is set to 128 bits, and the decoding at the receiving end adopts hard-decision decoding. is the sum of the Fano metrics for these branches.

2. Simulation content:

In the simulation experiment of the present invention, the method of the present invention and the original Fano algorithm are used to decode the convolutional codes respectively, and compare in multiple aspects.

2.1. Comparison of the average number of visits:

Simulation results:

FIG. 4 is a graph of the average number of visits obtained by simulation, wherein the horizontal axis represents the signal-to-noise ratio E _b /N ₀ , and the unit is dB. The vertical axis is the average number of visits for decoding a coded information block and searching for the access state. In accompanying drawing 4, use the method that the present invention proposes to decipher a coding information block, and search the average access times of the access state with the curve of the triangular mark; represent the original Fano algorithm to decode a coded information block with the circle mark, search for the access state average number of visits.

Analysis of simulation results:

It can be seen from the simulation results of FIG. 4 that when the signal-to-noise ratio E _b /N ₀ of the present invention is close to 0dB, the average number of visits of the present invention is less than 2500, while the average number of visits of the original Fano algorithm is greater than 4000. When the channel conditions are good, the two are basically the same. This shows that, compared with the original Fano algorithm, the present invention can significantly reduce the average access times and improve the decoding speed when the channel conditions are poor.

2.2. CCDF comparison of search access state times:

Simulation results:

Accompanying drawing 5, accompanying drawing 6 are the CCDF (Complementary Cumulative Distribution Function) curve of the search access state times of decoding a coded information block obtained by simulation, the abscissa is the number of visits, the ordinate is the probability, on the curve The meaning of one point is that the probability that the number of search access times of decoding a coded information block is greater than the value of the abscissa is the value of the ordinate. In the figure, the curve marked with a triangle represents the CCDF curve using the method proposed by the present invention; the curve marked with a circle represents the CCDF curve of the original Fano algorithm. Fig. 5 is obtained under the condition that the signal-to-noise ratio E _b /N ₀ is 0 dB, and Fig. 6 is obtained under the condition that the signal-to-noise ratio E _b /N ₀ is 0.5 dB.

Analysis of simulation results:

It can be seen from Fig. 5 that when the signal-to-noise ratio E _b /N ₀ is 0dB, the probability that the number of visits of the original Fano algorithm exceeds 10 ⁵ is about 1/250, while the probability of the present invention is less than 1/5000. It can be seen from Fig. 6 that when the signal-to-noise ratio E _b /N ₀ is 0.5dB and the probability of 1/1000, the number of visits of the present invention is in the order of 10 ⁴ , while the original Fano algorithm is in the order of 10 ⁵ . about an order of magnitude smaller. Since the CCDF is directly related to the decoding stack overflow probability, it can be seen from FIG. 5 and FIG. 6 that the present invention can greatly reduce the stack overflow probability during Fano decoding.

2.3. Comparison of decoding bit error rate:

Simulation results:

7 is a decoding bit error rate curve obtained by simulation, the abscissa is the signal-to-noise ratio E _b /N ₀ , the unit is dB, and the ordinate is the bit error rate. In accompanying drawing 7, the curve of the triangle mark represents the decoding bit error rate curve using the method proposed by the present invention; the curve marked with the circle represents the decoding bit error rate curve of the original Fano algorithm; the dotted line represents the theoretical bit error rate of uncoded BPSK curve.

It can be seen from the simulation results of Fig. 7 that, compared with the original Fano algorithm, under the same bit error rate, when the signal-to-noise ratio is low, the present invention loses about 0.3dB; and when the signal-to-noise ratio is high, the two are wrong. The bitrate curves overlap. It can be seen that, with a small bit error rate loss, the present invention significantly improves the decoding speed and reduces the probability of stack overflow to a great extent. The increased decoding speed enables the application of longer constrained convolutional codes, resulting in better performance.

The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and changes according to the concept of the present invention without creative efforts. Therefore, any technical solutions that can be obtained by those skilled in the art through logical analysis, reasoning or limited experiments on the basis of the prior art according to the concept of the present invention shall fall within the protection scope determined by the claims.

Claims (5)

1. A method for reducing the stack overflow probability of a Fano algorithm decoder, comprising the steps of: (1) Decoder parameter initialization: Set the threshold value of the decoder to zero, set the Fano metric of the first node of the code tree to zero, and set the first node of the code tree to be the decoding node; (2) Screen out the optimal node: (2a) read the code group corresponding to the current decoding node from the sequence buffer; (2b) According to the code group corresponding to the current decoding node, calculate the Fano metric of each branch starting from the current decoding node and the Fano metric of the node reached by each branch starting from the current decoding node; The Fano metric of each branch from the code node is calculated according to the following formula:

where log ₂ represents the base 2 logarithm, R _i represents the i-th code group vector received, c _i represents the i-th code group vector sent, and P(R _i | _ci ) represents the ith code group vector at c _i The probability of occurrence of R _i under the condition, P(R _i ) represents the probability of occurrence of Ri, n ₀ represents the length of the convolutional code segment, R _c represents the code rate, and M _F (R _i |c _i ) represents the _ith decoding Fano metric of the corresponding branch between the code node and the i+1th decoding node; (2c) according to the candidate point screening method, select one as the optimal node from the above-mentioned branches; the specific steps of the candidate point screening method are: The first step is to select the branch with the largest Fano metric as the best branch from all branches from the current decoding node; The second step is to determine whether the number of optimal branches is greater than 1. If so, then multiple optimal branches are obtained. Otherwise, this branch is set as the optimal branch, and the optimal node is set as the optimal branch from the current decoding node along the optimal branch. Step 10 is performed on the next node reached in the branch direction; The third step is to take any optimal branch, and set the node reached by the optimal branch as a temporary node; The fourth step is to calculate the Fano metric of each branch starting from the temporary node, and select the branch with the largest Fano metric as the second best branch; The fifth step is to determine whether the number of secondary optimal branches is equal to 1. If so, set the node reached by this secondary optimal branch as the secondary pre-selected node, otherwise, select a secondary optimal branch and set the The node reached by this second best branch is set as the second preselected node; The sixth step is to judge whether all the best branches have been taken, if so, obtain multiple secondary pre-selected nodes, otherwise, perform the third step; The seventh step is to calculate the Fano metric of multiple secondary preselected nodes; The eighth step is to select the node with the largest Fano metric from all the secondary preselected nodes; The ninth step is to judge whether the number of nodes with the largest Fano metric selected in the previous step is greater than 1. If so, set the optimal node to the previous node of any one of them, otherwise, set the optimal node to the previous node of the node. a node; The tenth step, end all steps and get the optimal node; (3) Process the decoding node: (3a) Calculate the Fano metric of the optimal node; (3b) Judging whether the Fano metric of the optimal node is greater than or equal to the threshold T, if so, move the current decoding node to the optimal node, otherwise, execute step (7); (4) Judging the shutdown conditions: Determine whether the current decoding node is the end point of the code tree, if so, stop, otherwise, continue to the next step; (5) Correct the Fano metric of a specific point: The Fano metric of a specific point is modified according to the point reliability judgment method, wherein the specific point is the point specified in the point reliability judgment method; the specific steps of the point reliability judgment method are: The first step is to set the number of backtracking points to a fixed value, and set the reliability threshold to a fixed value; The second step is to judge whether the order of the current decoding node in the code tree is less than or equal to the number of backtracking points, if so, end all steps, otherwise, continue to the next step; The third step is to calculate the difference obtained by subtracting the Fano metric of the backtracking end point from the Fano metric of the current decoding node, where the backtracking end point represents the node reached after returning to the backtracking point several nodes after the current decoding node; The fourth step is to judge whether the difference obtained in the previous step is less than the reliability threshold value, if so, end all steps, otherwise, continue to execute the next step; The fifth step is to judge whether the decoder reaches the current node for the first time, and if so, reset the Fano metric of the previous node of the backtracking end point to negative infinity, otherwise, end all steps; (6) Compress the threshold: (6a) Judging whether the current decoding node is visited for the first time, and if so, perform a tightening threshold processing; otherwise, keep the current threshold unchanged; (6b) perform step (2); (7) Get the backward peek measure: Taking the Fano metric of the previous node of the current decoding node as the backward peeping metric of the current decoding node, if the current decoding node is the initial node, then setting the backward peeping metric of the current node to be negative infinity; (8) Reduce the threshold: It is judged whether the backward peeping measure is less than the threshold, if so, the threshold is reduced, and then step (2) is performed, otherwise, the decoding node is rolled back to the previous node of the current decoding node; (9) Select the observation direction: According to the worst branch judgment method, it is judged whether the regression of the decoding node in the previous step is from the worst branch, and if so, execute step (7); step (3); The concrete steps of the described worst branch judgment method are: judging whether the current branch is the last branch selected and advanced in all branches of the current decoding node, if so, the current branch is the worst branch, otherwise, the current branch is not the worst branch. bad branch. 2. a kind of method that reduces Fano algorithm decoder stack overflow probability according to claim 1, is characterized in that, the code tree described in step (1) is a kind of expression mode of convolutional code state transition, code The nodes of the tree correspond to the state of the convolutional code, and the state transitions generated by different inputs of the convolutional code correspond to each branch in the code tree. The depth of the code tree is determined by the encoding length, and the number of branches of each node is determined by the parameters of the convolutional code. 3. a kind of method that reduces Fano algorithm decoder stack overflow probability according to claim 1, is characterized in that, the Fano of the node that each branch that departs from current decoding node described in step (2b) arrives The metric is calculated according to the following formula: M _i+1 =M _i +M _F (R _i |c _i ) Among them, M _i is the Fano metric of the i-th decoding node, M _i+1 is the Fano metric from the i+1-th decoding node, and M _F (R _i |c _i ) is the i-th decoding node and Fano metric of the corresponding branch between the i+1th decoding node. 4. a kind of method that reduces Fano algorithm decoder stack overflow probability according to claim 1, is characterized in that, the concrete steps of the tightening threshold processing described in the step (6a) are: cyclically execute T←T+Δ , until the Fano metric M of the current decoding node satisfies T≤M<T+Δ, where T represents the threshold value, Δ represents the threshold increment, M represents the Fano metric of the current node, and the threshold increment is usually an integer in the range of 2 to 8. 5. a kind of method that reduces Fano algorithm decoder stack overflow probability according to claim 1, is characterized in that, described in step (8), the concrete step that threshold is reduced is: make T←T-Δ , where T represents the threshold value, Δ represents the threshold increment, and the threshold increment usually takes an integer, ranging from 2 to 8.

Images