Low-complexity deep learning MIMO detection method based on partial MAP

Abstract

The invention discloses a low-complexity deep learning MIMO detection method based on partial MAP, and the method comprises the steps: obtaining a plurality of groups of training and testing samples, carrying out the layering and data preprocessing of the training and testing data samples through QR decomposition, obtaining K training sub-layers, and obtaining the input and output labels of each training sub-layer through calculation; training a DNN of a local LLR generated in a detection stage of each layer in the K training sub-layers based on a partial MAP method; layer-by-layer inspection is carried out from the Kth layer to the first layer, for the detection stage of each layer, the local LLR is calculated through the trained DNN, the LLR result of the current layer is updated in combination with prior information transmitted by the previous layer, and the LLR result is transmitted to the next layer to serve as prior information until the final LLR result is obtained; a final LLR result is obtained and input to an FEC decoder for error correction, and a final signal estimation result is obtained; partial MAP adopted by the method directly outputs log-likelihood ratio soft information, and better detection performance can be obtained by combining the log-likelihood ratio soft information with forward error correction codes.

Description

Technical Field

[0001] The present invention belongs to the technical field of signal detection, and in particular relates to a low-complexity deep learning MIMO detection method based on partial MAP. Background Art

[0002] Multiple-input multiple-output (MIMO) can realize diversity multiplexing by utilizing the spatial dimension of signals, and can multiply the reliability and transmission rate of signal transmission without increasing power and spectrum resources. In MIMO systems, signal detection technology recovers the original transmitted signal from the received signal transmitted through multiple paths, thereby ensuring communication performance. Therefore, multi-antenna signal detection is of great significance. The classic maximum likelihood detection method (ML) and maximum a posteriori detection method can achieve the best detection performance, but the computational complexity increases exponentially with the number of transmitting antennas, which seriously affects practical applications. Therefore, it is urgent to study low-complexity MIMO detection methods. Existing linear detectors, such as minimum mean square error (MMSE) detectors and zero forcing (ZF) detectors, can significantly reduce the computational complexity through weighted matrix multiplication, but the detection performance is poor. In order to better balance the computational complexity and detection performance, suboptimal methods such as approximate message passing and semidefinite relaxation (SDR) are proposed. In addition, detection methods based on successive interference cancellation (SIC) have also been widely studied. The performance of SIC is limited by error propagation, but the impact of error propagation can be effectively mitigated through soft estimation methods. The idea of ​​partial MAP can reduce the MIMO detection into several sub-detection problems, achieving a compromise between detection performance and computational complexity.

[0003] At present, deep learning has been widely used in MIMO signal detection. Existing deep learning signal detection can be divided into data-driven methods and model-driven methods. Data-driven detectors regard the signal detection process as a black box. Through the automatic feature extraction capability of deep neural networks (DNNs), end-to-end mapping from received signals to transmitted signals can be achieved in complex environments, but a large amount of training data and training time are required. In contrast, model-driven detection methods use domain knowledge and physical models to construct neural networks, usually adding trainable parameters to traditional iterative detection algorithms to achieve faster convergence speed. Recently, a hybrid strategy combining model-based systems with data-driven methods has been proposed, which uses domain knowledge to design network structures, so that learning can achieve better performance on relatively small data sets. However, existing deep learning-assisted signal detection methods are often based on expanding or improving traditional iterative detection algorithms, such as the classic data-driven MIMO detection network DetNet, the model-driven MIMO detection network OAMP-Net, etc. Its complexity is related to the number of iterations. Although the scale of training data and learning parameters can be controlled, it still faces the challenge of large computational complexity. Summary of the invention

[0004] In view of this, an object of the present invention is to provide a low-complexity deep learning MIMO detection method based on partial MAP to solve the above technical problems.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] A low-complexity deep learning MIMO detection method based on partial MAP, comprising:

[0007] Obtain multiple sets of training and test samples, and stratify and preprocess the training and test data samples through QR decomposition to obtain K training sub-layers, and calculate the input and output labels of each training sub-layer;

[0008] Based on the partial MAP method, a DNN is trained to generate local LLRs in each detection stage in K training sub-layers;

[0009] Perform layer-by-layer inspection from the Kth layer to the 1st layer. For each layer inspection stage, use the trained DNN to calculate the local LLR, combine the prior information transmitted by the previous layer to update the LLR result of the current layer, and pass it to the next layer as prior information until the final LLR result is obtained;

[0010] The final LLR result is obtained and input into the FEC decoder for error correction to obtain the final signal estimation result.

[0011] Furthermore, multiple sets of training and test samples are obtained, and the training and test data samples are stratified and preprocessed by QR decomposition to obtain K training sub-layers, including:

[0012] It is assumed that the communication system used for current detection has K transmitting antennas at the transmitting end and L receiving antennas at the receiving end;

[0013] transmitting and receiving a plurality of sets of training and testing samples according to the communication system;

[0014] The data received on the lth receiving antenna is expressed as:

[0015]

[0016] Among them, n l ~CN(0,N0) is the Gaussian white noise on the lth receiving antenna, h l，k is the channel coefficient between the kth transmitting antenna and the lth receiving antenna, s k is the data transmitted by the kth transmitting antenna, s k ∈S, S is a constellation point set, and the size of the constellation point set is M, that is, |S| = M, l∈[1, ..., L], k∈[1, ..., K];

[0017] The transmission process of the data y received on the receiving antenna is expressed in matrix form as:

[0018] y=Hs+n

[0019] Where H is the channel matrix, s is the data transmitted by the transmitting antenna, and n is the Gaussian white noise on the receiving antenna;

[0020] Decompose H according to QR decomposition:

[0021] H=QR

[0022] Get is the submatrix consisting of the first K columns of Q, and obtains The matrix representation of each training sublayer is obtained by submatrix consisting of the first K rows of R:

[0023]

[0024] According to the equation y l And equation z, the entire signal detection process is converted into K training sub-layers;

[0025] Among them, the equation corresponding to the kth training sublayer is:

[0026]

[0027] Among them, r k，i For the matrix The element in the kth column and the ith row in s i is the data transmitted by the i-th transmitting antenna, n k is the equivalent received Gaussian white noise of the kth sub-layer.

[0028] Furthermore, the input and output labels of each training sub-layer are calculated, including:

[0029] make Indicates that the signal s is detected in the kth layer i LLR soft information, where i∈{k,…,K}, b∈{1,…,B}, B=log2 M is the number of information bits carried by a transmission symbol;

[0030] make Indicates that the signal s is detected in the kth layer i Local LLR of

[0031] Calculated according to the following formula:

[0032]

[0033] Among them, s -i ={s1, ..., s i-1 ,s i+1 , ..., s K},v + =[r k，k , r k，k+1 , ..., r k，K ] T ,s k =[s k , ..., s K ];

[0034] Take an extended input set I containing 2((K-k+1)M+1) real-valued elements k，ex Perform training and testing of each layer of DNN:

[0035] I k，ex ={z k , r k，k ⊙S,…,r k，K ⊙S}

[0036] in,

[0037] according to and I k，ex Calculate the input and output labels for each training sublayer

[0038] Furthermore, based on the partial MAP method, a DNN is trained to generate local LLRs in each detection stage in the K training sub-layers, including:

[0039] When K=2, the LLR soft information of the relevant bits in s2 is obtained, and Q(s2) is calculated to be passed to the first layer as the prior information for its verification; the calculation formula is:

[0040]

[0041] Among them, c b，k Indicates k The bth bit of Represents c obtained from the previous layer b，k The prior probability of

[0042] In the test of level 1, z1 is processed and calculated:

[0043] z1=r 1，1 s1+r 1，2 s2+n1

[0044] Given z1 and c b，k After the prior probability of b，k LLR soft information; the calculation formula is:

[0045]

[0046] Among them, Θ b，k It is the prior information obtained from the previous layer;

[0047] Get the local LLR, expressed as:

[0048]

[0049] Among them, s -k ={s1, ..., s k-1 ,s k+1 , ..., s K};

[0050] According to Λ b，k and∑ b，k , the LLR update process of layer 1 detection is determined as:

[0051] Λ b，1 =Σ b，1

[0052] Λ b，k =Σ b，2 +Θ b，2 , b=0,…,B-1

[0053] because

[0054] The local LLR in layer 1 detection is expressed as a function of the following input:

[0055]

[0056] in represents taking the real part of a complex number, S N represents the N-th Cartesian product of S;

[0057] The output of the function is represented as:

[0058] O={∑ 0，1 , ..., ∑ B-1，1 ,∑ 0，2 , ..., ∑ B-1，2}

[0059] According to the input function and the output function, the sub-detection problem of the current layer is regarded as a regression problem with input I and output O, and the regression problem is solved based on the DNN to generate the DNN of the local LLR of the current layer detection stage.

[0060] Furthermore, we conduct layer-by-layer inspection from the Kth layer to the 1st layer. For each layer detection stage, we use the trained DNN to calculate the local LLR, and update the LLR result of the current layer in combination with the prior information transmitted by the previous layer, and transmit it to the next layer as prior information until the final LLR result is obtained, including:

[0061] The update process of the k-th layer detection is:

[0062]

[0063] in, It is the prior information, which comes from the detection result of the previous layer:

[0064]

[0065] After K layers of iteration and update, the final decoded LLR result is obtained

[0066] According to the LLR result, the hard decoding result is directly obtained:

[0067]

[0068] Wherein, sign(·) represents a sign function, and when x≥0, sign(x)=1, otherwise sign(x)=-1.

[0069] The beneficial effects of the present invention are:

[0070] The present invention applies a DNN-assisted data-model hybrid-driven deep learning method to a partial MAP algorithm, realizes low-complexity MIMO detection with the help of the hierarchical detection idea of ​​partial MAP, and proposes a low-complexity deep learning signal detection network PMAP-Net based on partial MAP; wherein, since the partial MAP algorithm can be reduced to multiple sub-detection stages, each layer of detection can be regarded as a regression problem, because the exponential computational complexity brought by traversal can be replaced by a DNN-assisted learning method; in addition, unlike most existing MIMO detection methods that learn hard bits, the partial MAP used in the present invention directly outputs log-likelihood ratio soft information, which can obtain better detection performance when combined with a forward error correction code, thereby taking into account both low complexity and high detection performance.

[0071] Other advantages, objectives and features of the present invention will be described in the following description and will be apparent to those skilled in the art to some extent, or they may be taught from the practice of the present invention. The purposes and other advantages of the present invention may be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings.

[0072] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0074] Figure 1 It is a method flow chart of a low-complexity deep learning MIMO detection method based on partial MAP in an embodiment of the present invention;

[0075] Figure 2 This is a structural diagram of a deep learning signal detection based on partial MAP in a low-complexity deep learning MIMO detection method based on partial MAP in an embodiment of the present invention;

[0076] Figure 3 It is a schematic diagram of the BER performance of PMAP-Net under a fixed channel in a low-complexity deep learning MIMO detection method based on partial MAP in an embodiment of the present invention;

[0077] Figure 4 This is a schematic diagram of the BER performance of PMAP-Net under a changing channel in a low-complexity deep learning MIMO detection method based on partial MAP in an embodiment of the present invention. DETAILED DESCRIPTION

[0078] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0079] like Figure 1 , Figure 2 , Figure 3 and Figure 4 As shown, the present invention proposes a low-complexity deep learning MIMO detection method based on partial MAP, comprising:

[0080] S101, obtaining multiple groups of training and test samples, and stratifying and preprocessing the training and test data samples through QR decomposition to obtain K training sub-layers, and calculating the input and output labels of each training sub-layer;

[0081] S102, based on the partial MAP method, training the DNN that generates local LLRs in each detection stage in the K training sub-layers;

[0082] S103, perform layer-by-layer inspection from the Kth layer to the 1st layer. For each layer inspection stage, use the trained DNN to calculate the local LLR, update the LLR result of the current layer in combination with the prior information transmitted by the previous layer, and transmit it to the next layer as prior information until the final LLR result is obtained;

[0083] S104, obtaining the final LLR result and inputting it into the FEC decoder for error correction to obtain the final signal estimation result;

[0084] Among them, multiple groups of training and test samples are obtained, and the training and test data samples are stratified and preprocessed by QR decomposition to obtain K training sub-layers, including:

[0085] It is assumed that the communication system used for current detection has K transmitting antennas at the transmitting end and L receiving antennas at the receiving end;

[0086] transmitting and receiving a plurality of sets of training and testing samples according to the communication system;

[0087] The data received on the lth receiving antenna is expressed as:

[0088]

[0089] Among them, n l ~CN(0,N0) is the Gaussian white noise on the lth receiving antenna, h l，k is the channel coefficient between the kth transmitting antenna and the lth receiving antenna, s k is the data transmitted by the kth transmitting antenna, s k∈S, S is a constellation point set, and the size of the constellation point set is M, that is, |S| = M, l∈[1, ..., L], k∈[1, ..., K];

[0090] The transmission process of the data y received on the receiving antenna is expressed in matrix form as:

[0091] y=Hs+n

[0092] Where H is the channel matrix, s is the data transmitted by the transmitting antenna, and n is the Gaussian white noise on the receiving antenna;

[0093] Decompose H according to QR decomposition:

[0094] H=QR

[0095] Get is the submatrix consisting of the first K columns of Q, and obtains The matrix representation of each training sublayer is obtained by submatrix consisting of the first K rows of R:

[0096]

[0097] According to the equation y l And equation z, the entire signal detection process is converted into K training sub-layers;

[0098] Among them, the equation corresponding to the kth training sublayer is:

[0099]

[0100] Among them, r k，i For the matrix The element in the kth column and the lth row in s i is the data transmitted by the i-th transmitting antenna, n k is the equivalent received Gaussian white noise of the kth sublayer;

[0101] The input and output labels of each training sublayer are calculated, including:

[0102] make Indicates that the signal s is detected in the kth layer l LLR soft information, where l∈{k,…,K}, b∈{1,…,B}, B=log2 M is the number of information bits carried by a transmission symbol;

[0103] make Indicates that the signal s is detected in the kth layer l Local LLR of

[0104] Calculated according to the following formula:

[0105]

[0106] Among them, s -i ={s1, .., s i-1i ,s i+1 ,..,s K},v k =[r k，k , r k，k+1 , ..., r k，K ] T ,s k =[s k , ..., s K ].

[0107] Take an extended input set I containing 2((K-k+1)M+1) real-valued elements k，ex Perform training and testing of each layer of DNN:

[0108] I k，ex ={z k , r k，k ⊙S,…,r k，K ⊙S}

[0109] in,

[0110] according to and I k，ex , calculate the input and output labels of each training sublayer

[0111] Based on the partial MAP method, a DNN is trained to generate local LLRs in each detection stage in K training sub-layers, including:

[0112] When K=2, the LLR soft information of the relevant bits in s2 is obtained, and Q(s2) is calculated to be passed to the first layer as the prior information for its verification; the calculation formula is:

[0113]

[0114] Among them, c b，k Indicates k The bth bit of Represents c obtained from the previous layer b，k The prior probability of

[0115] In the test of level 1, z1 is processed and calculated:

[0116] z1=r 1，1 s1+r 1，2 s2+n1

[0117] Given z1 and c b，k After the prior probability ofb，k LLR soft information; the calculation formula is:

[0118]

[0119] Among them, Θ b，k It is the prior information obtained from the previous layer;

[0120] Get the local LLR, expressed as:

[0121]

[0122] Among them, s -k ={{s1, ..., s k-1 ,s k+1 , ..., s K};

[0123] According to Λ b，k and∑ b，k , the LLR update process of layer 1 detection is determined as:

[0124] Λ b，1 =Σ b，1

[0125] Λ b，k =Σ b，2 +Θ b，2 , b=0,…,B-1

[0126] because

[0127] The local LLR in layer 1 detection is expressed as a function of the following input:

[0128]

[0129] in represents taking the real part of a complex number, S N Represents the Nth Cartesian product of S.

[0130] The output of the function is represented as:

[0131] O={∑ 0，1 , ..., ∑ B-1，1 ,∑ 0，2 , ..., ∑ B-1，2}

[0132] According to the input function and the output function, the sub-detection problem of the current layer is regarded as a regression problem with input I and output O, and the regression problem is solved based on the DNN to generate the DNN of the local LLR of the current layer detection stage;

[0133] From the Kth layer to the 1st layer, each layer is tested layer by layer. For each layer detection stage, the trained DNN is used to calculate the local LLR, and the LLR result of the current layer is updated in combination with the prior information transmitted by the previous layer, and then transmitted to the next layer as prior information until the final LLR result is obtained, including:

[0134] The update process of the k-th layer detection is:

[0135]

[0136] in, It is the prior information, which comes from the detection result of the previous layer:

[0137]

[0138] After K layers of iteration and update, the final decoded LLR result is obtained

[0139] According to the LLR result, the hard decoding result is directly obtained:

[0140]

[0141] Wherein, sign(·) represents the sign function, when x≥0, sign(x)=1, otherwise sign(x)=-1;

[0142] The working principle of the above technical solution is as follows: Step S1, stratification and data preprocessing by QR decomposition, including: preparing to randomly generate N T Group training data samples where s (j) , H (j) ,y (j) are the data transmitted on the jth sample transmitting antenna, the channel matrix, and the data received on the receiving antenna; perform QR decomposition on the channel matrix of all training data and convert the data into K training sublayers; calculate the input and output labels of each training sublayer according to formula (13) and formula (15)

[0143] Step S2, training the DNN network that generates local LLR in each layer of detection, including: each training sub-layer can be trained independently; the DNN for the k-th layer detection is a fully connected structure with 2 ((K-k+1)M+1) neurons as input and (K-k+1)B neurons as output, where M is the size of the constellation point set, and B=log2M is the number of bits carried in a symbol. Each DNN network has n hidden layers and uses the ReLU activation function; a batch processing scheme is used, and each batch uses the minimum mean square error (MSE) of the training data as the loss function, and the parameters are optimized by gradient descent training using the ADAM optimizer;

[0144] Step S3: After the network is trained, a layer-by-layer detection and update process is performed from the Kth layer to the 1st layer, including: for each layer detection, the local LLR is calculated using the trained deep learning network, and the LLR result of the layer is updated using the local LLR and the prior information transmitted from the previous layer;

[0145] Step S4: input the final LLR soft information into the FEC decoder for error correction to obtain the final signal detection result;

[0146] Specifically, consider that the communication system transmitter is equipped with K transmitting antennas and the receiving end is equipped with L receiving antennas; let the data transmitted by the kth transmitting antenna be s k ∈S, where S is the set of constellation points; let h l，k is the channel coefficient between the kth transmitting antenna and the lth receiving antenna, so the data received at the lth receiving antenna can be expressed as formula (1):

[0147]

[0148] where n l ~CN(0, N0) is the Gaussian white noise on the lth antenna; this transmission process can be expressed in matrix form as formula (2):

[0149] y=Hs+n

[0150] In the detection, H is first subjected to QR decomposition, which is expressed as formula (3):

[0151] H=QR

[0152] make is a submatrix consisting of the first K columns of Q, is the submatrix composed of the first K rows of R, then we get formula (4):

[0153]

[0154] Therefore, the detection process can be transformed into K equations, where the kth equation is expressed as formula (5):

[0155]

[0156] Taking K = 2 as an example, in SIC detection, the maximum likelihood strategy is first used to detect s2 from z2, and the next layer removes s2 as interference before detecting s1. Without considering the complexity of QR decomposition, the complexity of the SIC detection algorithm increases linearly with the number of transmitting antennas K. However, the SIC algorithm is affected by error propagation. To solve this problem, consider using a partial MAP method in detection, expressed as formula (6):

[0157]

[0158] Where Q(s2) = cf(z2|s2), c>0 is the normalization coefficient; consider Q(s2) as the prior information of s2, then the above formula can be regarded as a partial MAP problem; different from the standard MAP method, the partial MAP method only needs the prior information of part of the signal to be available, but not the prior information of all the signals; in the SIC detection process of the entire MIMO system, k The relevant detection is called the k-th layer detection of the partial MAP; therefore, the entire detection process is carried out in descending order, that is, from the K-th layer detection to the l-th layer detection in sequence; the result of each layer detection is passed to the next layer as part of the prior information of the current detection;

[0159] It can be seen that this partial MAP-based SIC detection process can divide MIMO detection into multiple sub-stages, thereby reducing the computational complexity; compared with the traditional SIC method, interference cancellation is performed by soft message passing, and each layer of detection can update the prior information of some signals, thereby alleviating the problem of error propagation; in addition, since each layer of detection process can be regarded as a regression problem, it can be solved by a low-complexity deep learning method;

[0160] Taking K = 2 as an example, we first perform the second layer test to obtain the soft information of the log-likelihood ratios (LLRs) of the relevant bits in s2; based on this layer of test, we can calculate Q(s2) and pass it to the first layer as the prior information for its test; use c b，k Indicates k The b-th bit of , assuming that they are independent of each other, then we have formula (7):

[0161]

[0162] in Represents c obtained from the previous layer b，k Prior probability; in the test of the first level, z1 is processed and calculated to obtain formula (8):

[0163] z1=r 1，1 s1+r 1，2 s2+n1

[0164] So, given z1 and c b，k After the prior probability of c b，k The LLR soft information is obtained by formula (9):

[0165]

[0166] where Θb，k is the prior information obtained from the previous layer, and the first item is the local LLR, which can be expressed as formula (10):

[0167]

[0168] where s -k ={s1, ..., s k-11 ,s k+1 , ..., s K}; Therefore, the LLR update process of layer 1 detection can be summarized as formula (11):

[0169] Λ b，1 =∑ b，1

[0170] Λ b，k =∑ b，2 +Θ b，2 , b=0,…,B-1

[0171] because The local LLR in layer 1 detection can be expressed as a function of the following input (12):

[0172]

[0173] in represents taking the real part of a complex number, S N Represents the Nth Cartesian product of S.

[0174] The output of the function is O = {∑ 0，1 , ..., ∑ B-1，1 ,∑ 0，2 , ..., ∑ B-1，2 Therefore, the sub-detection problem of this layer can be regarded as a regression problem with input I and output O, and thus solved with the help of DNN;

[0175] When K>2, we can summarize the general detection process. In the detection of the kth layer, we can get k Get s k ,…,s K LLR soft information; let Indicates that the signal s is detected in the kth layer i LLR soft information, where i∈{k,…,K}, b∈{1,…,B}, B=log2 M is the number of information bits carried by a transmission symbol; Indicates that the signal s is detected in the kth layer i The local LLR of can be calculated by the following formula (13):

[0176]

[0177] Among them, v k =[r k，k , r k，k+1 , ..., r k，K ] T ,s k =[s k , ..., s K ].

[0178] This calculation process requires traversing all M K constellation points, as the number of antennas increases, an unbearable amount of calculation will occur; this method regards the calculation process of local LLR as a regression problem and realizes low-complexity detection by training the DNN network; similarly, the general input set I of the k-th layer detection k By 2M K-k+1 real-valued variables, expressed as formula (14):

[0179]

[0180] In order to further reduce the number of input variables for DNN training and thus reduce the training complexity, we use an extended input set I containing 2((K-k+1)M+1) real-valued elements. k，ex The training and testing of each layer of DNN is performed as formula (15):

[0181] I k，ex ={z k , r k，k ⊙S,…,r k，K ⊙S}

[0182] in, At this time, the size of the DNN input set grows linearly with K and M, which greatly reduces the training cost;

[0183] In the detection of the kth layer, there are (K-k+1)B local LLRs, namely Obtained by a low-complexity DNN deep neural network;

[0184] After calculating the local LLR of the k-th layer detection, the LLR of each signal of the k-th layer can be updated in combination with the prior information obtained from the k+1 layer; the update process of the k-th layer detection can be expressed as formula (16):

[0185]

[0186] The prior information The detection result from the previous layer is expressed as (17):

[0187]

[0188] After K layers of iteration and update, the final decoded LLR result is obtained. The hard decoding result (1 8) can be directly obtained based on LLR:

[0189]

[0190] Wherein, sign(·) represents the sign function, when x≥0, sign(x)=1, otherwise sign(x)=-1;

[0191] With the help of forward error correction code (FEC), the decision of signal detection can be better assisted: the hard bit information shown in (18) or the LLR soft decoding information Input into the FEC decoder, the final decision result after error correction can be obtained; most of the existing deep learning-based MIMO detectors are involved in learning hard bit decisions, so they can only directly support hard FEC decoders; and the partial MAP-based deep learning signal detection network PMAP-Net provided by the present invention decodes by updating LLR soft information, so it can be directly combined with the soft FEC decoder without special processing; because soft decoding can provide 1.5-2dB error correction performance gain compared with hard decoding, PMAP-Net can obtain significant performance gain with the assistance of FEC;

[0192] The beneficial effects of the above technical solution are as follows: the low-complexity MIMO detector based on partial MAP proposed in the present invention adopts a data-model hybrid driven deep learning detection architecture of DNN-assisted reasoning, decoupling the large-scale MIMO detection process into multiple low-dimensional sub-detection problems, wherein the sub-detection process of each layer can be regarded as a regression problem, thereby using DNN learning to replace the traversal calculation that grows exponentially with the number of antennas; by adopting an extended input set whose dimension grows linearly with the number of antennas, the training cost and computational complexity of the DNN are greatly reduced; in addition, since the partial MAP algorithm based on the present invention transmits LLR soft information between layers and continuously updates prior information, it can alleviate the error propagation problem of the SIC decoding method; at the same time, the DNN network directly learns LLR soft information instead of hard bits, so the detection accuracy can be greatly improved through FEC soft decoding assistance;

[0193] In the simulation, we tested the performance of the PMAP-Net proposed in the present invention by taking 4 input antennas as an example in fixed channels and variable channels. Figure 3 and Figure 4As shown; it can be seen that the performance of PMAP-Net not only exceeds that of traditional MMSE and SIC detectors, but also has obvious performance gains compared with DetNet and OMAP-Net which are also based on deep learning; under changing channels, although the performance advantage of the uncoded PMAP-Net under high SNR is no longer as obvious as that of fixed channels, the algorithm proposed in the present invention still performs well under low SNR; the uncoded PMAP-Net can quickly reduce the BER level to between to, which allows the soft FEC decoder to play its full role, so that the coding-assisted PMAP-Net shows a good performance advantage.

[0194] Finally, it should be noted that the above preferred embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail through the above preferred embodiments, those skilled in the art should understand that various changes can be made in form and details without departing from the scope defined by the claims of the present invention.

Claims

1. A low-complexity deep learning MIMO detection method based on partial MAP, characterized in that: include: Obtain multiple sets of training and test samples, and stratify and preprocess the training and test data samples through QR decomposition to obtain K training sub-layers, and calculate the input and output labels of each training sub-layer; Based on the partial MAP method, a DNN is trained to generate local LLRs in each detection stage in K training sub-layers; Perform layer-by-layer inspection from the Kth layer to the 1st layer. For each layer inspection stage, use the trained DNN to calculate the local LLR, combine the prior information transmitted by the previous layer to update the LLR result of the current layer, and pass it to the next layer as prior information until the final LLR result is obtained; The final LLR result is obtained and input into the FEC decoder for error correction to obtain the final signal estimation result.

2. According to claim 1, a low-complexity deep learning MIMO detection method based on partial MAP is characterized in that: Get multiple sets of training and test samples, and stratify and preprocess the training and test data samples through QR decomposition to obtain K training sub-layers, including: It is assumed that the communication system used for current detection has K transmitting antennas at the transmitting end and L receiving antennas at the receiving end; transmitting and receiving a plurality of sets of training and testing samples according to the communication system; The data received on the lth receiving antenna is expressed as: Among them, n l ~CN(0,N0) is the Gaussian white noise on the lth receiving antenna, h l,k is the channel coefficient between the kth transmitting antenna and the lth receiving antenna, s k is the data transmitted by the kth transmitting antenna, s k ∈S, S is a constellation point set, the size of the constellation point set is M, |S|=M, l∈[1,...,L], k∈[1,...,K]; The transmission process of the data y received on the receiving antenna is expressed in matrix form as: y=Hs+n Where H is the channel matrix, s is the data transmitted by the transmitting antenna, and n is the Gaussian white noise on the receiving antenna; Decompose H according to QR decomposition: H=QR Get is the submatrix consisting of the first K columns of Q, and obtains The matrix representation of each training sublayer is obtained by submatrix consisting of the first K rows of R: According to equation yl and equation z, the entire signal detection process is converted into K training sub-layers; Among them, the equation corresponding to the kth training sublayer is: Among them, r k,i For the matrix The element in the kth column and the lth row in s i is the data transmitted by the i-th transmitting antenna, n k is the equivalent received Gaussian white noise of the kth sub-layer.

3. The low-complexity deep learning MIMO detection method based on partial MAP according to claim 2, characterized in that: The input and output labels of each training sublayer are calculated, including: make Indicates that the signal s is detected in the kth layer i LLR soft information, where i∈{k,…,K}, b∈{1,…,B}, B=log2M is the number of information bits carried by a transmission symbol; make Indicates that the signal s is detected in the kth layer i Local LLR of Calculated according to the following formula: where s -i ={s k ,..., s i-1 , s i+1 ,..., s K}, v k =[r k，k , r k，k+1 ,..., r k，K T , s k =[s k ,..., s K ;​ Take an extended input set I containing 2((K-k+1)M+1) real-valued elements k，ex Perform training and testing of each layer of DNN: I k，ex ={z k ,r k,k ⊙S,…,r k,K ⊙S} in, represents a complex domain; according to and I k，ex Calculate the input and output labels for each training sublayer 4. The low-complexity deep learning MIMO detection method based on partial MAP according to claim 1, characterized in that: Based on the partial MAP method, a DNN is trained to generate local LLRs in each detection stage in K training sub-layers, including: When K=2, the LLR soft information of the relevant bits in s2 is obtained, and Q(s2) is calculated to be passed to the first layer as the prior information for its verification; the calculation formula is: Among them, c b,k Indicates k The bth bit of Represents c obtained from the previous layer b,k The prior probability of In the test of level 1, z1 is processed and calculated: z1=r 1,1 s1+r 1,2 s2+n1 Given z1 and c b,k After the prior probability of b,k LLR soft information; the calculation formula is: Among them, Θ b,k It is the prior information obtained from the previous layer; Get the local LLR, expressed as: Among them, s -k ={s1, ..., s k-1 ,s k+1 , ..., s K }; According to Λ b,k and∑ b,k , the LLR update process of layer 1 detection is determined as: L b，1 =S b,1 L b,k =S b,2 +Θ b,2 ,b=0,…,B-1 because The local LLR in the l-th layer detection is expressed as a function of the following input: in represents taking the real part of a complex number, S N represents the N-th Cartesian product of S; The output of the function is represented as: O＝{∑ 0，1 ，...，∑ B-1，1 ，∑ 0，2 ，...,∑ B-1,2} According to the input function and the output function, the sub-detection problem of the current layer is regarded as a regression problem with input I and output O, and the regression problem is solved based on the DNN to generate the DNN of the local LLR of the current layer detection stage.

5. The low-complexity deep learning MIMO detection method based on partial MAP according to claim 1, characterized in that: From the Kth layer to the 1st layer, each layer is tested layer by layer. For each layer detection stage, the trained DNN is used to calculate the local LLR, and the LLR result of the current layer is updated in combination with the prior information transmitted by the previous layer, and then transmitted to the next layer as prior information until the final LLR result is obtained, including: The update process of the k-th layer detection is: in, It is the prior information, which comes from the detection result of the previous layer: After K layers of iteration and update, the final decoded LLR result is obtained According to the LLR result, the hard decoding result is directly obtained: Wherein, sign(·) represents a sign function, and when x≥0, sign(x)=1, otherwise sign(x)=-1.

Images