Self-adaptive SCL decoding algorithm based on path metric value

Abstract

The invention belongs to the technical field of communication, and relates to an adaptive serial Cancellation List (SCL) decoding algorithm based on a path metric (PM), which mainly comprises the following steps of: 1, calculating a PM value of each path of a current bit; 2, the calculated PM values are arranged in an ascending order and recorded as PMl; 3, calculating a difference value between every two adjacent PM values, and recording the difference value as delta PMl; 4, comparing the delta PMl with an upper limit threshold alpha H and a lower limit threshold alpha L one by one, and determining a path to be reserved according to a result; and 5, sequentially carrying out operations from the step 1 to the step 4 on each bit position until all bits are decoded to obtain a metric value set of the (N-1) th bit, and selecting a path corresponding to the minimum metric value to output in the metric value set to obtain a decoding result. According to the invention, the number of lists can be flexibly increased and reduced, so that the SCL decoding complexity is adaptively adjusted, and higher throughput and lower processing delay are realized on the premise of almost no loss of performance.

Description

Technical Field

[0001] The invention belongs to the field of communication technology, and specifically refers to an adaptive SCL decoding algorithm based on path metric value. Background Art

[0002] Polar code was proposed by Professor Arikan of Turkey in 2009. It is a channel coding method that theoretically proves that the channel capacity can reach the Shannon limit. It has become a research hotspot at home and abroad due to its excellent error correction performance, low computational complexity and hardware complexity. The SCL decoding algorithm improves the error correction performance of SC decoding by retaining a candidate information list L in each information node of the SC decoding process. However, the SCL-based decoder needs to expand L paths to 2L paths at each information node, sort the extended paths, and then prune them to L paths. When L is large, the time complexity of the sorting operation for 2L paths is very high. When applied to the actual Polar code decoder, the delay overhead is too large and the throughput is low.

[0003] Bin et al. proposed an adaptive serial cancellation decoding list decoding method (Adaptive Successive Cancellation List, AD-SCL) (B. Li, H. Shen and D. Tse, "An Adaptive Successive Cancellation List Decoder for Polar Codes with Cyclic Redundancy Check," in IEEE Communications Letters, vol. 16, no. 12, pp. 2044-2047). AD-SCL uses L = 2 as the starting list size. If the decoding passes the CRC check, it is considered a successful decoding. Otherwise, the decoding list is expanded to twice the previous one and decoded again until the decoding is successful or the decoding list reaches the maximum list L. max Output the path with the minimum PM value. The AD-SCL decoder can reduce some complexity in scenarios with extremely demanding performance requirements (where the list L is very high), but for most practical scenarios, iteratively increasing the list will increase the average number of decoding attempts, thereby introducing higher latency. From this we observe that for most received frames, an SCL decoder with very small L can successfully decode the information bits, and few frames require a larger L for successful decoding.

[0004] In the actual Polar code decoder, the list size can be adaptively adjusted according to different scenarios and environments to achieve higher throughput and lower processing delay. In the SCL decoding algorithm, the PM value represents the probability of the path being correct. The smaller the PM value, the greater the probability of the path being correct and the more reliable the path; the larger the PM value, the smaller the probability of the path being correct and the relatively unreliable the path. Based on this, the difference in PM values ​​under different environments can be used for pruning to adaptively adjust the list size. Therefore, an adaptive SCL decoding algorithm based on path metrics has extremely high application value. Summary of the invention

[0005] In view of the above background, the present invention provides an adaptive SCL decoding algorithm based on path metric value, sorts the PM value of the current information bit, calculates the difference between each two adjacent PM values, and compares it with the upper limit threshold α of the PM difference. H and the lower threshold α L Comparison is made to determine the path to be retained, and then pruning is performed to generate the current path list. The present invention can flexibly increase and decrease the number of lists, thereby adaptively adjusting the complexity of SCL decoding, and greatly reducing the SCL decoding delay without almost losing performance.

[0006] To achieve the above object, the present invention provides the following technical solution, and the specific steps are as follows:

[0007] Step 1: Calculate the PM value of each path of the current bit according to formula (1), where refers to the PM value of the lth path of the ith bit, Refers to the hard decision result of the bit in the current path. Refers to the log-likelihood ratio soft information of the current bit.

[0008]

[0009] Step 2: Arrange the calculated 2L PM values ​​in ascending order, and the arranged sequence is recorded as PM l ∈{PM1,PM2,…,PM L ,PM L+1 ,…,PM 2L}.

[0010] Step 3: Calculate the difference between each two adjacent PM values, recorded as ΔPM l ∈{ΔPM1,ΔPM2,…,ΔPM L ,ΔPM L+1 ,…,ΔPM 2L-1}.

[0011] Step 4: ΔPM l One by one and the upper threshold αH and the lower threshold α L Make a comparison and increase or decrease the number of lists based on the results to determine which paths to keep.

[0012] First, ΔPM1 and the upper threshold α H For comparison, update the path list set List and the number of path lists L num , then in L num When ΔPM is less than L l With the upper threshold α H For comparison, in L num ≥L, ΔPM l With the lower threshold α L Compare and update the path list set List and the number of path lists L each time num , and judge L num With L max The size of L num =L max , terminate the path selection process, and set the current List and L max As the output result of path selection, otherwise continue to compare the next ΔPM l Until the final List and L are selected num Output.

[0013] Step 5: Perform steps 1 to 4 for each bit position, and then perform path splitting and expansion until all bits are decoded to obtain the metric value set of the N-1th bit. The path corresponding to the minimum metric value is selected from the 2L paths and output to obtain the decoding result. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to make the purpose, technical solution and beneficial effects of the present invention clearer, the present invention provides the following drawings for illustration:

[0015] Figure 1 This is a flow chart of an adaptive SCL decoding algorithm based on path metric value described in the present invention.

[0016] Figure 2 This is a schematic diagram of a case where the number of lists is reduced according to the present invention.

[0017] Figure 3 This is a schematic diagram of the case where the number of lists described in the present invention increases. DETAILED DESCRIPTION

[0018] The present invention is further described below with reference to the accompanying drawings. The following is a specific implementation step of an adaptive SCL decoding algorithm based on path metrics. Assume that the code length N of the Polar code is 256, the code rate R is 1 / 2, and the maximum number of lists L ismax 16 is used to illustrate a specific embodiment of the present invention.

[0019] like Figure 1 As shown, the present invention provides an adaptive SCL decoding algorithm based on path metric value, which mainly includes the following steps:

[0020] Step 1: Calculate the PM value of each path of the current information bit according to formula (2), where refers to the PM value of the lth path of the ith bit, Refers to the hard decision result of the bit in the current path. Refers to the log-likelihood ratio soft information of the current bit. If the list size of the previous information bit is L, then the number of optional paths for the current information bit is 2L, i.e., l∈{1,2,…,L,L+1,…2L}.

[0021]

[0022] When the code length of the Polar code is 256 and the code rate is 1 / 2, the number of information bits contained is 128. The pruning method is for information bits, because if the i-th bit is a frozen bit, the SCL algorithm only retains That is, if the i-th bit is a frozen bit and its list number is L, then the list number of the i+1-th bit is still L, and no path splitting and expansion will be performed, so there is no need for pruning. So here we only discuss the information bit. Assuming that the number of lists for the m-1th (m>1)th information bit is 8, that is, L=8, then after path splitting and expansion, the number of candidate lists at the m-th information bit is 16, that is, the number of candidate paths at this time is L′=2L=16, l∈{1,2,…,8,9,…16}, and the calculated PM value is also 16.

[0023] Step 2: Arrange the 16 calculated PM values ​​in ascending order. The arranged sequence is recorded as PM l ∈{PM1,PM2,…,PM8,PM9,…,PM 16}. The smaller the PM value is, the greater the probability of the corresponding path being correct, and the more reliable the path is; the larger the PM value is, the smaller the probability of the corresponding path being correct, and the path is relatively unreliable. For example, the value of PM1 is smaller than that of PM2, the correct probability of l1 is greater than that of l2, and l1 is relatively more reliable.

[0024] Step 3: Calculate the difference between each two adjacent PM values, recorded as ΔPM l ∈{ΔPM1,ΔPM2,…,ΔPM8,ΔPM9,…,ΔPM 15}. Due to ΔPM l It refers to the difference between the paths, so it is only 15.

[0025] Step 4: ΔPM l One by one and the upper threshold α H and the lower threshold α L Make a comparison and increase or decrease the number of lists based on the results to determine which paths to keep.

[0026] To select the L with the smallest PM value from L′=2L=16 paths max In this process, the number of selected paths is L num Taking L=8 as the reference, given L max =16 is the upper limit, where L num ≤16.L num The size may be different from L. num When >8, it means that the number of selected lists is larger than the previous bit. num =8 indicates that the number of lists remains unchanged compared to the previous bit. num When <8, it means that the number of selected lists is reduced compared with the previous bit. Upper threshold α H and the lower threshold α L is an empirical threshold obtained through a large amount of simulation data. If the PM difference is greater than the upper threshold α H , indicating that the two paths are far apart, and a bit decision error is likely to have occurred. In this case, the previous path can be retained and the subsequent path can be discarded. If the PM difference is less than the lower limit threshold α L , indicating that there is little difference between the two paths, and it is very likely that both paths are correct. In this case, both paths can be retained.

[0027] First, start from ΔPM1 and compare ΔPM1 with the upper threshold α H For comparison, if ΔPM1≤α H , then l1 and l2 are stored in the currently selected and retained path list set List, that is, List = {l1, l2}, L num =2, where L num Indicates the number of paths currently selected for retention. Conversely, if ΔPM1>α H , update List = {l1}, L num =1 as the output result of path selection, that is, l1 is retained and l2 and the subsequent paths are discarded.

[0028] Judge L num With size 8.

[0029] When L num When ΔPM is less than 8, l With the upper threshold α H For comparison, if ΔPM l ≤αH , then store the lth path and the l+1th path in List, that is, List = {l1,l2,…,l l ,l l+1}, L num =l+1. On the contrary, if ΔPM l >α H , update List = {l1,l2,…,l l}, L num = l as the output result of path selection, that is, the lth path is retained and the l+1th and subsequent paths are discarded. This is the case where the number of lists is reduced. num With L max The size of L num =L max , terminate the path selection process, and set the current List and L max As the output result of path selection.

[0030] For example, when compared to ΔPM6, it is found that ΔPM6≤α H , then store the 6th and 7th paths in List, that is, List = {l1,l2,…,l6,l7}, L num =7, and then continue to compare ΔPM7 with the upper threshold α H Conversely, if ΔPM6>α H , update List = {l1,l2,…,l6}, L num = 6 as the output result of path selection, that is, the sixth path is retained and the seventh and subsequent paths are discarded. This is the case where the number of lists is reduced, such as Figure 2 shown.

[0031] When L num ≥8, set ΔPM l With the lower threshold α L For comparison, if ΔPM l >α L , update List = {l1,l2,…,l l}, L num = l as the output result of path selection, that is, keep the lth path and discard the l+1th and subsequent paths. On the contrary, if ΔPM l ≤α L , then store both the lth path and the l+1th path in List, that is, List = {l1,l2,…,l l ,l l+1}, L num = l + 1, this is the case where the number of lists increases. num With L max The size of Lnum =L max , terminate the path selection process, and set the current List and L max As the output result of path selection.

[0032] For example, when compared to ΔPM 10 , it was found that ΔPM 10 >α L , update List = {l1,l2,…,l 10}, L num =10 as the output result of path selection, that is, keep the 10th path and discard the 11th and subsequent paths. On the contrary, if ΔPM 10 ≤α L , then the 10th path and the 11th path are stored in List, that is, List = {l1,l2,…,l 10 ,l 11}, L num =11, this is the case where the number of lists increases, such as Figure 3 Then continue to increase ΔPM 11 With the lower threshold α L Make comparison.

[0033] In particular, if, as described here, the mth information bit selects the L path with the smallest PM value from L′=2L=16 paths num = 10 paths. After path splitting and expansion, the number of candidate lists at the m+1th information bit is 20, that is, the number of candidate paths at this time is L′=20, l∈{1,2,…,10,11,…20}, and the calculated PM value is also 20. The path selection at this time is to select the L with the smallest PM value from the 20 paths. num Since we have limited the maximum number of lists L max =16, so in L num ≥10, ΔPM l With the lower threshold α L When making comparisons, pay attention to the number of lists selected, L num Cannot exceed L max , even if the path satisfies the condition ΔPM l ≤α L For example, when comparing to ΔPM 16 , it was found that ΔPM 16 ≤α L , and only the 16th path is stored in List, that is, List = {l1,l2,…,l 15 ,l 16}, L num = 16, discarding the 17th and subsequent paths, which is due to the maximum number of lists Lmax restrictions.

[0034] Step 5: Perform steps 1 to 4 for each bit position, and then perform path splitting and expansion until all bits are decoded to obtain the metric value set of the N-1th bit. The path corresponding to the minimum metric value is selected from the 2L paths and output to obtain the decoding result.

Claims

1. An adaptive SCL decoding algorithm based on path metrics, It is characterized in that The specific steps are as follows: Step 1: Calculate the PM value of each path of the current bit according to formula (1), where refers to the PM value of the lth path of the ith bit, Refers to the hard decision result of the bit in the current path. Refers to the log-likelihood ratio soft information of the current bit; Step 2: Arrange the calculated 2L PM values ​​in ascending order, and the arranged sequence is recorded as PM l ∈{PM 1 ,PM 2 ,…,PM L ,PM L+1 ,…,PM 2L }; Step 3: Calculate the difference between each two adjacent PM values, recorded as ΔPM l ∈{ΔPM 1 ,ΔPM 2 ,…,ΔPM L ,ΔPM L+1 ,…,ΔPM 2L-1 }; Step 4: ΔPM l One by one and the upper threshold α H and the lower threshold α L Make a comparison, increase or decrease the number of lists according to the result, and determine the path to be retained; Step 5: Perform steps 1 to 4 for each bit position, and then perform path splitting and expansion until all bits are decoded to obtain the metric value set of the N-1th bit. The path corresponding to the minimum metric value is selected from the 2L paths and output to obtain the decoding result.

2. The adaptive SCL decoding algorithm based on path metric according to claim 1, Features: In step 3, it is proposed to use the difference between each two adjacent PM values ​​as the basis for pruning, and the difference sequence is recorded as ΔPM l ∈{ΔPM 1 ,ΔPM 2 ,…,ΔPM L ,ΔPM L+1 ,…,ΔPM 2L-1 }, due to ΔPM l refers to the difference between the paths, so it is only 2L-1.

3. The adaptive SCL decoding algorithm based on path metric according to claim 1, Features: In step 4, an upper threshold α is proposed H and the lower threshold α L With ΔPM l For comparison, the upper threshold α H and the lower threshold α L is an empirical threshold obtained through a large amount of simulation data. If ΔPM l Greater than the upper threshold α H , indicating that the two paths are far apart, and it is likely that a current bit decision error has occurred. In this case, the previous path can be retained and the subsequent path can be discarded; if ΔPM l Less than the lower threshold α L , indicating that there is little difference between the two paths, and it is very likely that both paths are correct. In this case, both paths can be retained.

4. The adaptive SCL decoding algorithm based on path metric according to claim 1, Features: In step 4, after ΔPM l With the upper threshold α H and the lower threshold α L By comparison, the number of lists retained can be based on L, and the final pruning can flexibly increase or decrease the number of lists; in step 4, first ΔPM 1 With the upper threshold α H For comparison, update the path list set List and the number of path lists L num , then in L num When ΔPM is less than L l With the upper threshold α H For comparison, in L num ≥L, ΔPM l With the lower threshold α L Compare and update the path list set List and the number of path lists L each time num , and judge L num With L max If L num =L max , terminate the path selection process, and set the current List and L max As the output result of path selection, otherwise continue to compare the next ΔPM l Until the final List and L are selected num Output.

5. The adaptive SCL decoding algorithm based on path metric according to claim 1, Features: In step 5, the operations of steps 1 to 4 are performed on each bit position in sequence, and then the path is split and expanded until all bits are decoded to obtain the metric value set of the N-1th bit. The path corresponding to the minimum metric value is selected from the 2L paths and output to obtain the decoding result.

Images