CN113746776B - Signal Receiving Method Based on Constellation Point Sorting and Dynamic Tree Search - Google Patents

Abstract

The invention belongs to the technical field of wireless communication, and in particular relates to a signal receiving method based on constellation point sorting and dynamic tree search. The key of the method of the present invention is to pre-sort the constellation points, so as to avoid the calculation complexity generated by calculating the Euclidean distance. And after the QRM-MLD search is completed, the dynamic tree search is performed again, and the discarded constellation points are searched again to improve the correctness of signal recovery. The beneficial effect of the present invention is that the calculation complexity can be effectively reduced under the condition of satisfying the system performance.

Description

Signal Receiving Method Based on Constellation Point Sorting and Dynamic Tree Search

technical field

The invention belongs to the technical field of wireless communication, and in particular relates to a signal receiving method based on constellation point sorting and dynamic tree search.

Background technique

In massive MIMO, as the number of antennas is greatly increased, the system capacity also increases linearly. In order to meet the needs of high-speed communication, MIMO systems make full use of space resources, multipath effects and frequency selectivity will cause serious interference between symbols, and the quality of signal detection algorithms will directly affect the performance of the system. Moreover, the computational complexity of signal detection increases exponentially with the number of antennas, so introducing a low-complexity and high-efficiency detection algorithm that can accurately recover transmitted signals is the key to MIMO technology.

At present, the Maximum likelihood Detection (MLD) receiver has been extensively studied, and it obtains better performance than the linear receiver. However, since it needs to detect all possible sending signals one by one, the complexity is too high to be acceptable in engineering. Aiming at the challenge of the extremely high complexity of the MLD algorithm, the maximum likelihood receiver based on QR decomposition (QRM-MLD) algorithm obtains the performance close to MLD with a complexity much lower than that of the MLD receiver, and is easier to implement in engineering. The QRM-MLD algorithm performs QR decomposition on the channel at the receiving end, and converts the channel matrix into a triangular matrix, which can be searched layer by layer during search and detection to ensure that useful signals are not affected by other interference signals. First, search layer by layer from the end signal, and then combine the candidate sets of the searched signals to search for the candidate set of the next signal. The traditional QRM-MLD algorithm combined with QR decomposition and M algorithm reduces the detection complexity of the traditional MLD algorithm, but to make the QRM-MLD algorithm gradually approach the traditional MLD, then the value of M must gradually increase, then with the number of layers With the improvement of the modulation method, the computational complexity of QRM-MLD will also increase greatly.

In order to make the performance of the QRM-MLD algorithm gradually approach the MLD algorithm, then M must approach C as much as possible, so there is a problem of how to compromise the system performance and computational complexity.

Contents of the invention

The present invention aims at the problem of how to compromise the above-mentioned computational complexity and system performance, and based on the traditional QRM-MLD algorithm, proposes a signal receiving method based on the combination of constellation point pre-sorting and dynamic tree search. The method uses a certain storage space, The calculation complexity can be effectively reduced, and the system performance close to the MLD algorithm can be obtained through the dynamic tree search algorithm.

The technical solution adopted by the present invention is as follows, the key of which is to pre-sort the constellation points to avoid the computational complexity generated by calculating the Euclidean distance. And after the QRM-MLD search is completed, the dynamic tree search is performed again, and the discarded constellation points are searched again to improve the correctness of signal recovery.

The specific scheme of constellation point pre-sorting is as follows, assuming that the number of transmitting and receiving antennas is 2, and the modulation method is 16QAM, where M is the number of candidate constellation point sets, M=8, and each constellation point is composed of a fixed real part and imaginary part. The signal sent by the terminal is represented as follows:

Among them, x ₁ and x ₂ represent two transmitted symbols respectively, and the signal received at the receiving end is represented as follows:

Among them, y ₁ and y ₂ represent two received symbols respectively. After performing QR decomposition on the channel matrix H at the receiving end, the signal matrix H is transformed into an upper triangular matrix, and the MLD metric is expressed as follows:

Where r ₁₁ represents the element in the first row and the first column in the R matrix, r ₁₂ represents the element in the first row and the second column in the R matrix, r ₂₂ represents the element in the second row and the second column in the R matrix, y _m1 , y _m2 respectively represent the received symbols after QR decomposition, then the detection of x ₂ will not be affected by other interference signals, and the pre-estimation of x ₂ is performed first, that is:

Where x _p2 represents the estimated point. The traditional QRM-MLD needs to calculate the Euclidean distance between the estimated point and each constellation point, which has high complexity. However, the algorithm only needs to calculate the real part of the estimated point and each constellation point. And the difference between the imaginary part can be sorted. The specific process is as follows to calculate the difference between the pre-estimated point and the real part of each constellation point:

diff_r _k ＝|rr _k |, k＝1,2,...,16 (5)

Where r _k represents the value of the real part of the kth constellation point, and diff_r _k represents the absolute value of the difference between the real part of the kth constellation point and the real part of the predicted point. Since the modulation method is 16QAM, that is, the signal has 16 samples, and each 4-bit binary number represents a sample point, so there are 4 values of the real part of the constellation point, that is, _{diff_rk} has 4 values, and they are sorted from small to large , sort the real part of the constellation points according to the serial number.

Similarly, calculate the difference between the estimated point and the imaginary part of each constellation point:

diff_i _k ＝|ii _k |, k＝1,2,...,16 (6)

Where i _k represents the value of the imaginary part of the kth constellation point, and diff_i _k represents the absolute value of the difference between the imaginary part of the kth constellation point and the imaginary part of the pre-estimated point. Similarly, the value of the imaginary part of the constellation point is 4 kinds, that is, _{diff_ik} has 4 values, which are sorted from small to large, and the imaginary part of the constellation point is sorted according to the serial number;

According to the sorting values of the real and imaginary parts of the constellation points, the coordinates of the constellation points are expressed: (a, b), where a represents the sorting value of the real part of the constellation point, and b represents the sorting value of the imaginary part of the constellation point.

Before the algorithm is performed, the sorted list is saved in advance, and the sorted value of the distance between each constellation point and the pre-estimated point is determined according to the coordinate sorted value of each constellation point in the list. The specific sorting operation is as follows: Constellation point coordinates: (1,1), corresponding constellation point sorting value: 1; constellation point coordinates: (1,2), corresponding constellation point sorting value: 2; constellation point coordinates: ( 2,1), the corresponding constellation point sorting value: 3; constellation point coordinates: (2,2), corresponding constellation point sorting value: 4; constellation point coordinates: (3,1), corresponding constellation point sorting value: 5; Constellation point coordinates: (1,3), corresponding to constellation point sorting value: 6; Constellation point coordinates: (3,2), corresponding to constellation point sorting value: 7; Constellation point coordinates: (2,3), The corresponding constellation point sorting value: 8; the constellation point coordinates: (3,3), the corresponding constellation point sorting value: 9; the constellation point coordinates: (1,4), the corresponding constellation point sorting value: 10; the constellation point coordinates : (4,1), corresponding to constellation point sorting value: 11; constellation point coordinates: (2,4), corresponding to constellation point sorting value: 12; constellation point coordinates: (4,2), corresponding to constellation point sorting Value: 13; Constellation point coordinates: (3,4), corresponding constellation point sorting value: 14; Constellation point coordinates: (4,3), corresponding constellation point sorting value: 15; Constellation point coordinates: (4,4 ), then the corresponding constellation point sorting value: 16, and finally get the sorting of all constellation points, and then select M candidate vector sets with the smallest distance.

The specific scheme of dynamic tree search is as follows, the configuration of the transmitter is as shown above, and the candidate constellation point set of x ₂ is obtained by using the constellation point pre-sorting scheme:

x ₂ _candidate＝{c ₁ ,c ₂ ,...,c _M } (7)

According to the set of candidate constellation points, the discarded non-candidate constellation points are 16-M:

x ₂ _discarded＝{c _M+1 ,c _M+2 ,...,c ₁₆ } (8)

The calculation of LLR using the known candidate constellation point set can be approximated as:

以及

如果利用非候选向量集计算的度量值小于候选向量集计算的最小度量值，则更新最小度量值，利用更新过的最小度量值重新计算LLR软信息。where σ ² represents the power of noise. When the number of candidate vector sets is small, the MLD solution is likely to be discarded in advance. Therefore, the non-candidate vector set is used to recalculate the metric value

as well as

If the metric value calculated by using the non-candidate vector set is smaller than the minimum metric value calculated by the candidate vector set, the minimum metric value is updated, and the LLR soft information is recalculated by using the updated minimum metric value.

The beneficial effect of the present invention is: the calculation complexity can be effectively reduced under the condition of satisfying the system performance.

Description of drawings

When Fig. 1 is M=4, the emulation comparison schematic diagram of the inventive method and traditional method, (a) is the throughput emulation figure, (b) is the throughput emulation figure;

Fig. 2 is a schematic diagram of a simulation comparison between the method of the present invention and the traditional method when M=16, (a) is a throughput simulation diagram, and (b) is a throughput simulation diagram.

Detailed ways

The technical solution of the present invention has been described in detail in the part of the content of the invention, and the practicability of the present invention will be illustrated below in conjunction with the accompanying drawings and simulation examples.

In the present invention, the number of real number multiplications of each transmission vector is used as a parameter to measure the complexity, and the complexity difference between the traditional QRM-MLD algorithm and the QRM-MLD algorithm of constellation point pre-sorting is compared.

For the traditional QRM-MLD algorithm, according to formula (1-2), it can be known that the weight calculation formula of each layer can be expressed as:

需要4×(N_t-i)次实乘；计算复数平方需要2次实乘。因此每层权值计算共需要的实乘次数表示为：For any constellation point in the constellation diagram, the above formula must be calculated, then: calculating r _{i, i} x _i requires 2 real multiplications; calculating

4×(N _t -i) real multiplications are required; 2 real multiplications are required to calculate the square of a complex number. Therefore, the total number of real multiplications required for each layer of weight calculation is expressed as:

num_i=4×(N _t -i)+4 (12)

For each layer of _si × 2 ^bps nodes, real multiplication is required, and the final real multiplication times of the traditional QRM-MLD algorithm is expressed as:

Among them, s _i represents the number of surviving candidate constellation point sets calculated by the upper layer, 2 ^bps represents the number of all constellation points, and bps represents the base number of QAM quadrature amplitude modulation.

For the QRM-MLD algorithm for constellation point pre-sorting, according to formula (1-2), the formula for calculating the weight of each layer of constellation point pre-sorting can also be expressed as formula (1-15), but only need to calculate the central node:

需要4×(N_t-i)次实乘；不需计算复数平方。因此每层权值计算共需要的实乘次数表示为：To traverse all the constellation points and calculate the central node, the above equation needs to be solved: calculating r _{i, i} x _i requires 2 real multiplications; calculating

4×(N _t -i) real multiplications are required; no complex square calculation is required. Therefore, the total number of real multiplications required for each layer of weight calculation is expressed as:

num_i ₁ =4×(N _t -i)+2 (15)

After obtaining the central node, it is necessary to calculate the position of the horizontal and vertical coordinates of all constellation points from the central node, that is, to calculate |(xx _i ) _r |, |(xx _i ) _i |, and to express the coordinates of the constellation points. The real multiplication required for this process The number of times is:

num_i ₂ =2×bps (16)

Where bps represents the base number of QAM quadrature amplitude modulation, and the number of horizontal and vertical coordinates of all constellation points is determined by the QAM base number. For each layer, it is only necessary to calculate the real multiplication of the s _i surviving candidate constellation points, and the final real multiplication times of the QRM-MLD algorithm for constellation point pre-sorting is:

Comparing Equation (13) with Equation (17), we can see that the complexity of the constellation point pre-sorting algorithm is significantly lower than that of the traditional QRM-MLD algorithm. As the number of transmitting antennas gradually increases, the complexity of the constellation point pre-sorting algorithm decreases significantly, while the complexity of the traditional QRM-MLD algorithm increases exponentially. Therefore, under the large-scale number of antennas, the constellation point pre-sorting algorithm has obvious advantages.

The present invention simplifies the calculation of each layer's weight through the pre-stored sorting table, thereby achieving the purpose of reducing the calculation complexity. When the number of transmitting antennas increases gradually, the complexity of the algorithm decreases more obviously. From the simulation results shown in Figure 1, it can be seen that when M=4, the performance of the traditional QRM-MLD algorithm has a greater drop than the throughput of the MLD algorithm. This is because the dynamic tree search algorithm finds solutions that may be discarded, and through Constellation point pre-sorting reduces the complexity of the traditional QRM-MLD algorithm, so that the algorithm of the present invention has a good balance performance and complexity; as can be seen from Figure 2, when M=8 is set, the performance of the algorithm of the present invention and the traditional QRM-MLD algorithm The difference is not big, because with the gradual increase of M, the performance of the traditional QRM-MLD algorithm is gradually approaching that of the MLD algorithm. From the simulation results, the performance of the algorithm of the present invention is better than that of the traditional QRM-MLD algorithm.

Claims (1)

1. The signal receiving method based on constellation point sorting and dynamic tree search, the number of transmitting and receiving antennas in the communication system is 2, the modulation method is 16QAM, each constellation point is composed of a fixed real part and imaginary part, and the signal sent by the transmitter is expressed as In the following form:

Where x ₁ and x ₂ respectively represent two transmission symbols, characterized in that the receiving method includes: The signal received by the receiver is:

Among them, y ₁ and y ₂ represent two received symbols respectively. The channel matrix H is decomposed by QR at the receiving end, and the signal matrix H is transformed into an upper triangular matrix. The MLD metric expression is obtained as:

Where r ₁₁ represents the element in the first row and the first column in the R matrix, r ₁₂ represents the element in the first row and the second column in the R matrix, r ₂₂ represents the element in the second row and the second column in the R matrix, y _m1 , y _m2 respectively represent the received symbols after QR decomposition, first pre-estimate x ₂ :

Where x _p2 represents the pre-estimated point, calculate the difference between the pre-estimated point and the real part of each constellation point: diff_r _k ＝|rr _k |, k＝1,2,...,16 Where r _k represents the value of the real part of the kth constellation point, diff_r _k represents the absolute value of the difference between the real part of the kth constellation point and the real part of the pre-estimated point, the 16QAM modulation method corresponds to 16 samples, and each 4-bit binary The number represents a sample point, so the value of the real part of the constellation point is 4 kinds, that is, _{diff_rk} has 4 values, which are sorted from small to large, and the real part of the constellation point is sorted according to the serial number; Compute the difference between the predicted point and the imaginary part of each constellation point: diff_i _k ＝|ii _k |, k＝1,2,...,16 Where i _k represents the value of the imaginary part of the kth constellation point, and diff_i _k represents the absolute value of the difference between the imaginary part of the kth constellation point and the imaginary part of the pre-estimated point. Similarly, the value of the imaginary part of the constellation point is 4 types, That is, diff_i _k has 4 values, sort them from small to large, and sort the imaginary part of the constellation points according to the serial number; According to the sorting values of the real and imaginary parts of the constellation points, the coordinates of the constellation points are represented: (a, b), where a represents the sorting value of the real part of the constellation point, b represents the sorting value of the imaginary part of the constellation point, and the sorting rules are: constellation point coordinates : (1,1), corresponding to constellation point sorting value: 1; constellation point coordinates: (1,2), corresponding to constellation point sorting value: 2; constellation point coordinates: (2,1), corresponding to constellation point sorting Value: 3; Constellation point coordinates: (2,2), corresponding constellation point sorting value: 4; Constellation point coordinates: (3,1), corresponding constellation point sorting value: 5; Constellation point coordinates: (1,3 ), then the corresponding constellation point sorting value: 6; constellation point coordinates: (3,2), then the corresponding constellation point sorting value: 7; constellation point coordinates: (2,3), then the corresponding constellation point sorting value: 8; Point coordinates: (3,3), corresponding to constellation point sorting value: 9; constellation point coordinates: (1,4), corresponding to constellation point sorting value: 10; constellation point coordinates: (4,1), corresponding to constellation points Point sorting value: 11; constellation point coordinates: (2,4), corresponding constellation point sorting value: 12; constellation point coordinates: (4,2), corresponding constellation point sorting value: 13; constellation point coordinates: (3 ,4), then the corresponding constellation point sorting value: 14; constellation point coordinates: (4,3), then the corresponding constellation point sorting value: 15; constellation point coordinates: (4,4), then the corresponding constellation point sorting value: 16 , and finally get the sorting of all constellation points, and then select M candidate vectors with the smallest distance; After the candidate constellation point set is obtained according to the pre-sorting of the constellation points, the signal is recovered based on the dynamic tree search, specifically: The candidate constellation point set of _x2 obtained by pre-sorting the constellation points is: x ₂ _candidate＝{c ₁ ,c ₂ ,...,c _M } According to the candidate constellation point set, the discarded non-candidate constellation points are 16-M: x ₂ _discarded＝{c _M+1 ,c _M+2 ,...,c ₁₆ } The LLR is calculated using the known candidate constellation point set as:

where ^σ2 represents the power of the noise, B represents the candidate vector set of the MLD algorithm,

with

respectively represent the vector sets whose l-bit values of the i-th symbol are 1 and 0, and use the non-candidate vector set to recalculate the metric value

as well as

If the metric value calculated by using the non-candidate vector set is smaller than the minimum metric value calculated by the candidate vector set, the minimum metric value is updated, and the LLR soft information is recalculated by using the updated minimum metric value.

Images