Symbol detection method, system, electronic device, and storage medium

Abstract

The application discloses a symbol detection method and system, electronic equipment and a storage medium, relates to the technical field of communication signal detection, and the method comprises the following steps: initializing a symbol estimation vector of a sending end based on an equivalent real-value MIMO system model; performing reverse diffusion on the symbol estimation vector of the current time step, wherein in the reverse diffusion process, prior guide information of the symbol estimation vector of the current time step is calculated based on a modulation constellation, an estimated sending symbol vector is determined, the estimated sending symbol vector is mapped through a real-value channel tensor to obtain a first reconstructed receiving signal vector, and likelihood guide information is calculated according to the difference between the first reconstructed receiving signal vector and a real-value receiving signal vector, so that the intermediate estimation result is corrected to obtain a symbol estimation vector of the next time step, and the reverse diffusion process of the current time step is completed; and the reverse diffusion process is repeated to obtain a final symbol estimation vector of the current trajectory.

Description

Technical Field

[0001] This application relates to the field of communication signal detection technology, specifically to symbol detection methods, systems, electronic devices, and storage media. Background Technology

[0002] With the continuous development of mobile communication technology, Multiple-Input Multiple-Output (MIMO) technology has become one of the key technologies in fifth-generation and subsequent mobile communication systems due to its ability to significantly improve system capacity and spectrum utilization efficiency. By configuring multiple antennas at both the transmitting and receiving ends, massive MIMO systems can transmit multiple data streams simultaneously on the same spectrum resources, effectively meeting the communication requirements of high speed and large capacity. In massive MIMO systems, the receiving end needs to estimate the discrete modulation symbols at the transmitting end based on the received signal, channel state information, and noise characteristics; this process is usually called symbol detection. The performance of symbol detection directly affects the system's bit error rate and overall communication quality, thus becoming one of the key technical issues in massive MIMO systems.

[0003] Existing symbol detection methods mainly include maximum likelihood detection, linear detection, and neural network-based detection methods. Maximum likelihood detection theoretically achieves optimal detection performance, but its complexity increases exponentially with the number of antennas and modulation order, making it difficult to directly apply to large-scale MIMO scenarios. Linear detection methods, such as zero-forcing (ZF) detection and minimum mean square error (MMSE) detection, while computationally efficient, still show a significant performance gap compared to maximum likelihood detection. Neural network-based detection methods, such as DnCNN, ComNet, and GNN, face two challenges: weak generalization ability for different SNRs and significant performance differences compared to traditional maximum likelihood detection methods. Currently, neural network-based detection methods are trained on datasets with specific SNRs and validated on test sets with similar SNRs; ​​when these models encounter data different from the training set, their detection performance drops significantly. Summary of the Invention

[0004] This application addresses the technical problem of high computational complexity in existing symbol detection methods while achieving optimal detection performance under conditions of large-scale antenna configuration and high-order modulation, and provides a symbol detection method, system, electronic device, and storage medium.

[0005] To achieve the above objectives, this application adopts the following technical solution: A first aspect of this application is a symbol detection method, comprising: Obtain the received signal at the receiving end, as well as the channel matrix between the transmitting and receiving ends; An equivalent real-valued MIMO system model is constructed based on the received signal and the channel matrix. Based on the equivalent real-valued MIMO system model, the real-valued received signal vector and the real-valued channel tensor are obtained. Initialize the symbol estimation vector at the transmitter based on the equivalent real-valued MIMO system model; The symbol estimation vector at the current time step is back-spread. During the back-spreading process, prior guidance information of the symbol estimation vector at the current time step is calculated based on the modulation constellation. The estimated transmitted symbol vector is determined based on the prior guidance information. The estimated transmitted symbol vector is used to update the symbol estimation vector at the current time step to obtain intermediate estimation results. The estimated transmitted symbol vector is mapped through a real-valued channel tensor to obtain the first reconstructed received signal vector. The likelihood guidance information is calculated based on the difference between the first reconstructed received signal vector and the real-valued received signal vector. The intermediate estimation results are corrected based on the likelihood guidance information to obtain the symbol estimation vector for the next time step, thus completing the reverse diffusion process for the current time step. The reverse diffusion process is repeated to obtain the final symbol estimation vector for the current trajectory.

[0006] In some implementations, after repeating the reverse diffusion process to obtain the final symbol estimation vector of the current trajectory, the following steps are further included: The symbol estimation vector at the transmitter is initialized multiple times based on the equivalent real-valued MIMO system model, and each initialized symbol estimation vector is back-divided to obtain the final symbol estimation vectors of multiple trajectories. After mapping the final symbol estimation vectors of multiple trajectories through real-valued channel tensors, several second reconstructed received signal vectors are obtained. The final symbol estimation vector corresponding to the second reconstructed received signal vector with the smallest difference from the real-valued received signal vector is selected as the optimal symbol estimation vector.

[0007] In some implementations, prior guidance information for calculating the symbol estimation vector at the current time step based on the modulation constellation includes: The real and imaginary parts of the symbol estimation vector at the current time step are subtracted from the real and imaginary parts of each constellation point in the modulation constellation, and then rearranged to obtain the real distance tensor and the imaginary distance tensor. The real and imaginary distance tensors are squared element-wise and summed to obtain the first distance tensor. The first distance tensor is weighted according to the noise intensity of the current time step, and the softmax function is used to perform probability normalization on the dimension of the modulation constellation to obtain the weight of each constellation point. After multiplying the weight of each constellation point by the real and imaginary parts of the corresponding constellation point, sum them in the dimension of the modulated constellation to obtain the real part expectation vector and the imaginary part expectation vector; concatenate the real part expectation vector and the imaginary part expectation vector to obtain the expectation matrix; The difference between the expected matrix and the symbol estimation vector at the current time step is calculated, and the difference is scaled according to the noise intensity at the current time step to obtain the prior guidance information.

[0008] In some implementations, obtaining the first reconstructed received signal vector by mapping the estimated transmitted symbol vector through a real-valued channel tensor includes: Multiply the estimated transmitted symbol vector by the real-valued channel tensor to obtain the first reconstructed received signal vector; After mapping the final symbol estimation vectors of multiple trajectories through real-valued channel tensors, several second reconstructed received signal vectors are obtained, including: Multiplying the final symbol estimation vectors of multiple trajectories with the real-valued channel tensor yields several second reconstructed received signal vectors.

[0009] In some implementations, calculating likelihood guidance information based on the difference between the first reconstructed received signal vector and the real-valued received signal vector includes: Calculate the square of the Euclidean distance between the first reconstructed received signal vector and the real-valued received signal vector; Solve for the gradient of the square of the Euclidean distance with respect to the sign estimate vector at the current time step, and use the gradient as the likelihood guidance information.

[0010] In some implementations, selecting the second reconstructed received signal vector that has the smallest difference from the real-valued received signal vector includes: Calculate the Euclidean distance between several second reconstructed received signal vectors and real-valued received signal vectors; After sorting the Euclidean distances between the second reconstructed received signal vector and the real-valued received signal vector, the final symbol estimation vector corresponding to the second reconstructed received signal vector with the smallest Euclidean distance is selected as the optimal symbol estimation vector.

[0011] A second aspect of this application is a symbol detection system, comprising: The signal acquisition module is used to acquire the received signal at the receiving end, as well as the channel matrix between the transmitting end and the receiving end; The real-valued module is used to construct an equivalent real-valued MIMO system model based on the received signal and the channel matrix, and to obtain the real-valued received signal vector and the real-valued channel tensor based on the equivalent real-valued MIMO system model. The symbol estimation vector initialization module is used to initialize the symbol estimation vector of the transmitter based on the equivalent real-valued MIMO system model. The prior guidance module is used to back-diffuse the symbol estimation vector of the current time step. During the back-diffusing process, the prior guidance information of the symbol estimation vector of the current time step is calculated based on the modulation constellation. The estimated transmitted symbol vector is determined according to the prior guidance information. The estimated transmitted symbol vector is used to update the symbol estimation vector of the current time step to obtain the intermediate estimation result. The likelihood guidance module is used to map the estimated transmitted symbol vector through a real-valued channel tensor to obtain the first reconstructed received signal vector. It calculates the likelihood guidance information based on the difference between the first reconstructed received signal vector and the real-valued received signal vector, corrects the intermediate estimation results based on the likelihood guidance information, and obtains the symbol estimation vector for the next time step, thus completing the reverse diffusion process for the current time step. The reverse diffusion process is repeated to obtain the final symbol estimation vector for the current trajectory.

[0012] A third aspect of this application is a computer device comprising: a processor and a computer-readable storage medium; A processor, adapted to execute computer programs; A computer-readable storage medium containing a computer program that, when executed by a processor, implements a symbol detection method.

[0013] A fourth aspect of this application is a computer-readable storage medium storing a computer program adapted to be loaded and executed by a processor for a symbol detection method.

[0014] A fifth aspect of this application is a computer program product comprising a computer program that, when executed by a processor, implements a symbol detection method.

[0015] Compared with the prior art, this application has the following beneficial effects: This application provides a symbol detection method that transforms the detection process into an iterative update linearly related to the number of transmit antennas by performing back-diffusion on the symbol estimation vector at the current time step. This fundamentally avoids the problem that while maximum likelihood detection achieves optimal performance, its complexity increases exponentially with the number of antennas and modulation order. In this back-diffusion process, prior guidance information for the symbol estimation vector at the current time step is first calculated based on the modulation constellation. The discreteness of the modulation constellation is used to transform the hard constraint that the symbol must fall on a legal point into a differentiable gradient direction, so that the symbol estimation vector is guided by the modulation constraint at each step. Then, the estimated transmitted symbol vector is mapped through the real-valued channel tensor to obtain the first reconstructed received signal vector. The likelihood guidance information is calculated based on the difference between the first reconstructed received signal vector and the real-valued received signal vector, and the physical observation constraint is also transformed into a gradient direction, so that the symbol estimation vector is simultaneously guided by data consistency. Subsequently, the estimated transmitted symbol vector is determined based on prior guidance information. This estimated transmitted symbol vector is used to update the symbol estimation vector at the current time step, obtaining an intermediate estimation result. The intermediate estimation result is then corrected based on likelihood guidance information to obtain the symbol estimation vector for the next time step. This ensures that the posterior distribution sampling process is explicitly decomposed into independently computable prior guidance and likelihood guidance terms at each step. This guarantees that the estimation result conforms to the modulation constellation constraints and is consistent with the received signal, thus approximating optimal detection performance without introducing exponential complexity. Simultaneously, an equivalent real-valued MIMO system model is constructed based on the received signal and channel matrix, transforming all operations into real-domain operations. This enables the method to achieve optimal detection performance under large-scale antenna configurations and high-order modulation conditions.

[0016] Furthermore, this application initializes the symbol estimation vector at the transmitter multiple times based on an equivalent real-valued MIMO system model, and performs back-diffusion on each initialized symbol estimation vector to obtain the final symbol estimation vectors for multiple trajectories. This expands the detection process from single sampling to multiple independent samplings, which is equivalent to exploring the posterior distribution multiple times independently, increasing the probability of covering the optimal solution. Based on this, the final symbol estimation vector corresponding to the second reconstructed received signal vector with the smallest difference from the real-valued received signal vector is selected as the optimal symbol estimation vector. This retains the sampling advantage of the diffusion model for complex posterior distributions while effectively suppressing random fluctuations from single sampling through optimal output, significantly improving the reliability and robustness of the final detection results. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart of the symbol detection method provided in the embodiments of this application; Figure 2 This is a structural diagram of the symbol detection system provided in an embodiment of this application; Figure 3 A structural diagram of a large-scale MIMO system based on diffusion posterior sampling provided in an embodiment of this application; Figure 4 A structural diagram of a computer device provided for an embodiment of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] The following is a description of the technical terms used in this application: Multiple-Input Multiple-Output (MIMO) technology; Zero-Forcing (ZF) detection; Minimum Mean Square Error (MMSE) detection; Denoising Convolutional Neural Network (DnCNN); Communication Neural Network (ComNet); Graph Neural Network (GNN) Signal-to-noise ratio (SNR); Symbol Error Rate (SER); Variance-Preserving Stochastic Differential Equation (VP-SDE); Quadrature Phase Shift Keying (QPSK); 16-Quadrature Amplitude Modulation (16QAM). Vertical-Bell Laboratories Layered Space-Time (V-BLAST) Diffusion Posterior Sampling (DPS).

[0021] like Figure 1 As shown, this application provides a symbol detection method based on diffusion posterior sampling, applicable to large-scale MIMO systems, comprising the following steps: S1, obtain the received signal at the receiving end, and the channel matrix between the transmitting end and the receiving end; In this embodiment, the receiver is equipped with multiple receiving antennas for receiving wireless analog signals from the transmitter. The receiver performs down-conversion, sampling, and analog-to-digital conversion on the received wireless analog signals through a radio frequency front-end to obtain a received signal in digital baseband form. The received signal can be represented as a complex-valued received signal vector. ; Complex-valued received signal vector The dimension is ,in, It is the number of received signal samples. This refers to the number of antennas at the receiving end; Simultaneously, the receiving end acquires the channel matrix between the transmitting and receiving ends, which is represented as a complex-valued channel matrix. Complex-valued channel matrix The channel matrix can be estimated from the pilot signal transmitted by the transmitter, or provided by the channel state information acquisition module in the communication system. This embodiment does not limit the channel matrix. The specific method of obtaining it; Complex-valued channel matrix Dimensions ,in, It refers to the number of transmitting antennas; To construct an equivalent MIMO system model, the complex-valued transmitted signal vector is used. This indicates that complex-valued Gaussian noise is represented by... express.

[0022] S2, construct an equivalent real-valued MIMO system model based on the received signal and the channel matrix, and obtain the real-valued received signal vector and the real-valued channel tensor based on the equivalent real-valued MIMO system model; In this embodiment, in order to facilitate subsequent symbol detection processing based on diffusion a posteriori sampling, the complex-valued received signal vector and the corresponding complex-valued channel matrix obtained in step S1 are expanded into real-valued forms. Specifically, the complex-valued received signal vector The real and imaginary parts are extracted and concatenated to construct an equivalent real-valued received signal vector. Its dimensions are For complex-valued transmitted signal vectors Complex Gaussian noise Perform the same operation to construct a real-valued transmitted signal vector. and real Gaussian noise Simultaneously, the complex-valued channel matrix... The real and imaginary parts are combined to create an equivalent real-valued channel tensor; Real-value transmitted signal vector Real-value received signal vector Real-valued channel tensor Real-value Gaussian noise They are represented as follows: , , ,

[0023] in, For complex-valued transmitted signal vectors The real part, For complex-valued transmitted signal vectors The imaginary part, For complex-valued received signal vector The real part, For complex-valued received signal vector The imaginary part, For complex-valued channel matrix The real part, For complex-valued channel matrix The imaginary part, For complex Gaussian noise The real part, Complex Gaussian noise The imaginary part.

[0024] Therefore, the equivalent real-valued MIMO system model is expressed as: .

[0025] S3, Initialize the symbol estimation vector of the transmitter based on the equivalent real-valued MIMO system model; Specifically, the number of trajectories to be run by the program is set. The number of sampling steps to be performed for each trajectory First, based on the real-valued channel tensor obtained in step S2... The number of received signal samples contained in the dimension and the number of antennas at the transmitting end The dimension of the initialized transmitted signal vector is determined to be... Then, within the normalized range of the modulation constellation, the transmitted signal vector is randomly initialized based on a uniform distribution to obtain a noise symbol sequence. This serves as the starting point for the iterative back diffusion of the symbol estimation vector.

[0026] S4, perform back-diffusion on the symbol estimation vector of the current time step. During the back-diffusion process, calculate the prior guidance information of the symbol estimation vector of the current time step based on the modulation constellation. Determine the estimated transmitted symbol vector based on the prior guidance information. The estimated transmitted symbol vector is used to update the symbol estimation vector of the current time step to obtain the intermediate estimation result. Specifically, it needs to be based on posterior guiding terms. Construct a VP-SDE to perform backdiffusion on the symbol estimation vector, which is then used to update the forward diffusion stochastic differential equation VP-SDE of the symbol estimation vector at time step. The expression is as follows:

[0027] The forward diffusion stochastic differential equation is used to model noise perturbations in the real-valued transmitted signal vector at the transmitter during symbol detection, thus serving symbol detection in large-scale MIMO systems; wherein, Forward diffusion steps, Real-value transmitted signal vector With diffusion time change; For noise scheduling functions, This represents standard Brownian motion in forward diffusion; the drift term in the forward diffusion process is... The strength of the diffusion term is ; Real-value transmitted signal vector The distribution is defined as: when hour, ;when hour, , To adjust the upper limit of the normalized values ​​of constellations, To adjust the upper limit of the normalized values ​​of constellations, Indicates in and The uniform distribution between them; by updating the symbol estimation vector using the reverse diffusion SDE formula corresponding to forward diffusion SDE, the original distribution of the symbol estimation vector can be obtained. Medium sampling, the formula is as follows:

[0028] in, For the reverse diffusion time, For standard Brownian motion with reverse diffusion, the a priori guiding term in the above formula is... Replace with posterior guiding terms It can achieve sampling from the posterior distribution. H For real-valued channel tensors, it serves as a condition for the posterior guiding term; As one example, the diffusion model and its parameter configuration are initialized, the linear scheduling factor corresponding to each time step is saved, and the mean coefficient and standard deviation coefficient of the reverse diffusion process are calculated, as follows: S401, Based on the parameter configuration of the diffusion model, generate and save the linear scheduling factor and correlation coefficient; The equivalent discrete form of the forward diffusion stochastic differential equation is:

[0029] in, Indicates the original transmitted symbol sequence; Indicates the first The transmitted symbol sequence containing noise after the next forward diffusion; It is a preset linear scheduling factor, and ; It is by Obtained by multiplication; Indicates the number of forward diffusion steps. ; As forward diffusion proceeds, Starting from 0.0001, it increases to 0.02; Continuously decreasing, while due to As forward diffusion proceeds, Approaching 0, It tends to 1; when executed The resulting symbol sequence after the first forward diffusion Completely polluted by noise; the parameters , , , All need to be saved for later use. Perform index calls; S402, based on the linear scheduling factor and correlation coefficient Calculate the mean coefficient and standard deviation coefficient of the reverse diffusion process in order to convert the noise symbol sequence... To recover the true sequence of transmitted symbols, it is necessary to use the reverse diffusion process corresponding to the forward diffusion process to reverse-engineer the symbol data. Furthermore, in order to solve the problem using a computer, the discrete form of the reverse diffusion process needs to be defined, as shown below:

[0030] , is the standard deviation coefficient of the reverse diffusion process calculated based on the linear scheduling factor; Calculate the mean coefficient and standard deviation coefficient for each time step to prepare for subsequent back diffusion.

[0031] In this embodiment, the posterior guiding term Decomposed into prior guiding terms used to characterize symbolic prior constraints and the likelihood guiding term used to characterize data consistency. The guiding term participates in the diffusion sampling process in the form of gradient or equivalent direction update. By introducing the prior guiding term and the likelihood guiding term respectively, the diffusion sampling process is jointly guided, satisfying the following relationship:

[0032] The above-mentioned guiding items The prior guidance information used to characterize the symbolic prior constraints is represented by the following formula:

[0033] in, This indicates that the sequence of transmitted symbols corresponds to the real-valued transmitted signal vector. The expectation matrix is ​​used to calculate the prior guidance information, and then... The first reverse diffusion process update includes the following steps: S410, firstly, the... After the first forward spread, the transmitted symbol sequence contains noise. , as the symbol estimation vector at the current time step The symbol estimation vector at the current time step The real and imaginary parts are separated, and the difference between the real and imaginary parts of each modulation constellation point is calculated separately; in this embodiment, the modulation order is . M , The dimension is ,by Taking the real part as an example, the dimension is Reshaping yields a dimension of The matrix, and then respectively with The dimension is obtained by subtracting the real parts of each modulation constellation point. The matrix; this matrix represents The real part of each symbol and The distance between each modulation constellation point; It refers to the number of transmitting antennas.

[0034] S420, rearrange the distance matrices of the real and imaginary parts obtained in S410 into dimensions respectively. Given the real and imaginary distance tensors, we can sum the sum of the element-wise squares of each tensor to obtain the first distance tensor. The equivalent mathematical formula for each element in the element-wise square is as follows:

[0035] in, These are the coordinates of a point in a specific constellation within the modulation constellation. ; Computational Dimensions First distance tensor and After calculating the ratio, the last dimension is normalized using the softmax function to obtain the weights of each constellation point, specifically using the following formula:

[0036] S430, the current symbol estimation vector under given conditions. and noise intensity Next, calculate the random variable element by element. The expected value; specifically, rearranging the probability normalization matrix formed by the weights of the above constellation points into dimensions is The matrix is ​​then multiplied by the real and imaginary parts of each constellation point, and then rearranged into a matrix with dimensions of . The tensor is summed along the last dimension and concatenated with the matrix to obtain the expectation matrix of the current symbol estimation vector, with dimension 1. The corresponding formula is as follows:

[0037] in, X is the set of modulation constellation points.

[0038] S440, using the expectation matrix and the symbol estimation vector at the current time step, construct the prior guiding term, satisfying the following relationship:

[0039] Based on the above relationships, determine the estimated transmitted symbol vector. and prior guiding items The relationship between them is as follows:

[0040] Estimating the transmitted symbol vector It is based on the first After the first forward spread, the transmitted symbol sequence contains noise. Time step , Related guidance items The obtained estimate of the original transmitted symbol sequence, where This is the symbol estimation vector for the current time step. The correlation coefficient, The value is the expectation matrix. The value of .

[0041] S450, based on the estimated symbol vector, transmits the symbol vector. The symbol estimation vector at the current time step The standard deviation coefficient and correlation coefficient of the reverse diffusion process Perform the first update at the current time step to obtain the intermediate estimate. The specific formula is as follows:

[0042] The mean coefficient of the reverse diffusion process is and ; S5, the estimated transmitted symbol vector is mapped through the real-valued channel tensor to obtain the first reconstructed received signal vector. The likelihood guidance information is calculated based on the difference between the first reconstructed received signal vector and the real-valued received signal vector. The intermediate estimation results are corrected based on the likelihood guidance information to obtain the symbol estimation vector for the next time step, thus completing the reverse diffusion process of the current time step. The reverse diffusion process is repeated to obtain the final symbol estimation vector of the current trajectory.

[0043] In this embodiment, the likelihood-guided term is used to evaluate the intermediate estimation results. The corrections include the following steps: S501, transmit the estimate from S460 to the symbol vector. Treating it as the actual data from the sending end, calculate its real-valued channel tensor. After mapping, the first reconstructed received signal vector is obtained. The first reconstructed received signal vector and the real-valued received signal vector are then compared. The square of the L2 norm is used to measure data consistency, specifically using the following formula:

[0044] S502, Solve for the sign estimate vector of the square of the L2 norm in S501 with respect to the current time step. The gradient is used as likelihood guidance information and as a direction representing data consistency, as shown in the following equation:

[0045] The gradient satisfies:

[0046] S503, Based on the likelihood-guided information, the intermediate estimation results are... Make corrections to obtain The symbolic estimation vector of the step :

[0047] Repeat S4 and S5 T This yields the final estimated symbol sequence for the transmitter corresponding to a trajectory. As the final symbol estimation vector of the current trajectory, the real-valued received signal vector is calculated and stored. and The Euclidean distance.

[0048] In some embodiments of this application, the above method further includes the following steps: S6, based on the equivalent real-valued MIMO system model, the symbol estimation vector of the transmitter is initialized multiple times, and each initialized symbol estimation vector is back-divided to obtain the final symbol estimation vector of the transmitter given by all trajectories and the corresponding Euclidean distance matrix, with dimensions respectively. and The process of selecting the final symbol estimation vector that has the smallest Euclidean distance to the real-valued received signal vector as the optimal symbol estimation vector includes: S6.1, after mapping the final symbol estimation vectors of the multiple trajectories through the real-valued channel tensor, a second reconstructed received signal vector is obtained. The Euclidean distance matrix between the second reconstructed received signal vector and the real-valued received signal vector is calculated, and the Euclidean distance matrix is ​​sorted in ascending order in the second dimension. S6.2, Traverse all The final symbol estimation vector corresponding to the trajectory with the smallest Euclidean distance is selected as the optimal symbol estimation vector.

[0049] To verify the feasibility of the symbol detection method for large-scale MIMO systems based on diffusion posterior sampling described in this embodiment, a large-scale MIMO communication system model is constructed in a simulation environment to verify the method. The system structure is as follows: Figure 3 As shown; In this embodiment, a structure containing One transmitting antenna and A large-scale MIMO system with multiple transmit and receive antennas is presented, where the number of transmit and receive antennas can be configured according to simulation requirements. In the specific simulation, system a uses 8 transmit antennas and 64 receive antennas, while system b uses 32 transmit antennas and 64 receive antennas to verify the performance of the symbol detection method under different system scales.

[0050] The transmitting end generates an independent and identically distributed random bit sequence through a bit generation module, and modulates the random bit sequence using digital modulation through a modulation module to obtain the baseband symbol sequence to be transmitted. This baseband symbol sequence is then converted into a radio frequency (RF) signal by the transmitting antenna array and transmitted through a wireless channel. The receiving antenna array receives the RF signal and converts it into a baseband signal. The channel state information is obtained by the channel estimation module, and the signal preprocessing is performed by the baseband processing module. Finally, the final symbol estimation vector is output by the spread a posteriori sampling symbol detection module, which is the recovered original transmitted symbol sequence. In this embodiment, system a uses QPSK modulation, and system b uses 16QAM modulation. The modulated symbols undergo average energy normalization to ensure fairness under different modulation methods.

[0051] The wireless channel model adopts an independent and identically distributed Rayleigh fading channel model, with a real-valued channel tensor. All elements of the denominator follow a complex Gaussian distribution with zero mean and unit variance. The real-valued Gaussian noise is modeled using additive white Gaussian noise, with a mean of zero and a variance determined by the signal-to-noise ratio parameter.

[0052] To evaluate the performance of the symbol detection method under different channel conditions, the signal-to-noise ratio (SNR) was set to -1 dB to 4 dB for system a and to 11 dB to 16 dB for system b during the simulation. At the receiving end, for the received signal, the symbol detection process based on diffusion posterior sampling is executed according to steps S1 to S6 as described above to obtain the corresponding transmitted symbol estimation result, i.e. the optimal symbol estimation vector. Table 1. Examples of symbol detection results for system a under different signal-to-noise ratio conditions.

[0053] Table 2. Examples of symbol detection results for System b under different signal-to-noise ratio conditions.

[0054] Through the above simulation process, corresponding transmitted symbol estimation results can be obtained under different system scales, different modulation methods, and different signal-to-noise ratios. The simulation results are shown in Tables 1 and 2. System a can maintain the SER of the transmitted symbol estimation results at a low level in the low signal-to-noise ratio scenario and system b can maintain the SER at a low level in the high signal-to-noise ratio range. Therefore, the symbol detection method based on diffusion posterior sampling in this embodiment can operate stably under different MIMO system configurations and output symbol estimation results that meet the data consistency constraints, thereby verifying the feasibility of the method of this invention.

[0055] The following comparative experiments using a 16x32 MIMO system further validate the symbol detection method described in this invention: Step 1: Set the modulation scheme to 16-QAM, the signal-to-noise ratio range to 11dB to 16dB, select Rayleigh fading channel as the channel type, and obtain the transmitted signal, received signal, and channel matrix. Step two: The received signal from step one is detected using MMSE, V-BLAST, and the symbol detection algorithm based on diffusion posterior sampling described in this invention, respectively. The comparison results of the symbol detection error rates of each method are shown in Table 3. Table 3 Comparison of symbol error rates for different methods

[0056] As can be seen from Table 3, for each signal-to-noise ratio range, the symbol error rate (DPS-SER) based on diffusion posterior sampling provided in this embodiment is lower than the symbol error rate of other methods.

[0057] In one embodiment of this application, such as Figure 2 As shown, a symbol detection system is provided, including: The signal acquisition module is used to acquire the received signal at the receiving end, as well as the channel matrix between the transmitting end and the receiving end; The real-valued module is used to construct an equivalent real-valued MIMO system model based on the received signal and the channel matrix, and to obtain the real-valued received signal vector and the real-valued channel tensor. The symbol estimation vector initialization module is used to initialize the symbol estimation vector of the transmitter based on the equivalent real-valued MIMO system model. The prior guidance module is used to back-diffuse the symbol estimation vector at the current time step. During the back-diffusing process, prior guidance information of the symbol estimation vector at the current time step is calculated based on the modulation constellation. The estimated transmission symbol vector is determined according to the prior guidance information. The estimated transmission symbol vector is used to update the symbol estimation vector at the current time step to obtain intermediate estimation results. The likelihood guidance module is used to map the estimated transmitted symbol vector through the real-valued channel tensor to obtain a first reconstructed received signal vector, calculate likelihood guidance information based on the difference between the first reconstructed received signal vector and the real-valued received signal vector, correct the intermediate estimation result based on the likelihood guidance information, obtain the symbol estimation vector for the next time step, and complete the reverse diffusion process of the current time step; repeat the reverse diffusion process to obtain the final symbol estimation vector of the current trajectory.

[0058] For specific limitations regarding the symbol detection system, please refer to the limitations of the symbol detection method above; the corresponding technical effects can be obtained equivalently and will not be repeated here. Each module in the above symbol detection system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0059] Figure 4 An internal structural diagram of a computer device is shown in one embodiment. This computer device may specifically be a terminal or a server. Figure 4 As shown, the computer device includes a processor, memory, network interface, display, camera, and input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a symbol detection method. The display screen can be an LCD screen or an e-ink display screen. The input device can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0060] As will be understood by those skilled in the art, computer equipment is Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the computer device to which the present invention is applied. Specific computing devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0061] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described above.

[0062] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0063] In summary, the symbol detection method, system, computer device, and storage medium provided in this application introduce a symbol detection framework based on diffused posterior sampling. This framework directly models and samples the posterior distribution of transmitted symbols during the symbol detection process, avoiding the exponential computational complexity caused by combinatorial search in traditional maximum likelihood detection. Therefore, it is applicable to MIMO systems with large-scale antenna configurations. This embodiment decomposes the symbol posterior probability into a prior guiding term characterizing modulation constellation constraints and a likelihood guiding term characterizing the consistency of the received signal. These guiding terms are jointly introduced during the diffused sampling process, which can gradually constrain the symbol estimation results while ensuring data consistency, thus improving the stability of the symbol detection process. Simultaneously, the symbol estimation vector is updated level by level at each diffused noise level. Through a reverse diffusion process, continuous recovery from a high-noise state to a low-noise state is achieved, avoiding the error accumulation problem caused by a single estimation. This is beneficial for obtaining reliable symbol detection results under complex channel conditions. Furthermore, this embodiment employs multi-track diffusion sampling and combines it with preset criteria to select the optimal symbol estimation result, enabling result filtering between different sampling trajectories and further improving the robustness of the symbol detection results. Moreover, this method does not rely on offline training processes for specific signal-to-noise ratio conditions. Compared with symbol detection methods based on neural networks, it has better adaptability to changes in signal-to-noise ratio or system scale. It is applicable to large-scale MIMO systems with different antenna sizes, modulation methods, and signal-to-noise ratio conditions, exhibiting good versatility and facilitating deployment in practical communication systems.

[0064] The various embodiments in this specification are described in a progressive manner. For directly identical or similar parts of the embodiments, refer to each other. Each embodiment focuses on its differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0065] The above-described embodiments are merely preferred embodiments of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various improvements and substitutions without departing from the principles of the present invention, and these improvements and substitutions should also be considered within the scope of protection of the present invention. Therefore, the scope of protection of this invention should be determined by the scope of the claims.

Claims

1. A symbol detection method, characterized in that, include: Obtain the received signal at the receiving end, as well as the channel matrix between the transmitting and receiving ends; An equivalent real-valued MIMO system model is constructed based on the received signal and the channel matrix, and a real-valued received signal vector and a real-valued channel tensor are obtained based on the equivalent real-valued MIMO system model. The symbol estimation vector of the transmitting end is initialized based on the equivalent real-valued MIMO system model; The symbol estimation vector at the current time step is back-diffused. During the back-diffusion process, prior guidance information of the symbol estimation vector at the current time step is calculated based on the modulation constellation. The estimated transmission symbol vector is determined according to the prior guidance information. The estimated transmission symbol vector is used to update the symbol estimation vector at the current time step to obtain intermediate estimation results. The estimated transmitted symbol vector is mapped through the real-valued channel tensor to obtain the first reconstructed received signal vector. The likelihood guidance information is calculated based on the difference between the first reconstructed received signal vector and the real-valued received signal vector. The intermediate estimation result is corrected based on the likelihood guidance information to obtain the symbol estimation vector for the next time step, thus completing the reverse diffusion process of the current time step. Repeat the reverse diffusion process to obtain the final symbol estimation vector of the current trajectory.

2. The symbol detection method according to claim 1, characterized in that, After repeating the reverse diffusion process to obtain the final symbol estimation vector of the current trajectory, the method further includes the following steps: The symbol estimation vector of the transmitter is initialized multiple times based on the equivalent real-valued MIMO system model, and each initialized symbol estimation vector is back-diffused to obtain the final symbol estimation vector of multiple trajectories. After mapping the final symbol estimation vectors of the multiple trajectories through the real-valued channel tensor, several second reconstructed received signal vectors are obtained. The final symbol estimation vector corresponding to the second reconstructed received signal vector with the smallest difference from the real-valued received signal vector is selected as the optimal symbol estimation vector.

3. The symbol detection method according to claim 1, characterized in that, The prior guidance information for calculating the symbol estimation vector at the current time step based on the modulation constellation includes: The real and imaginary parts of the symbol estimation vector at the current time step are subtracted from the real and imaginary parts of each constellation point in the modulation constellation, and then rearranged to obtain the real distance tensor and the imaginary distance tensor. The real and imaginary distance tensors are squared element-wise and summed to obtain the first distance tensor. The first distance tensor is weighted according to the noise intensity of the current time step, and the softmax function is used to perform probability normalization on the dimension of the modulation constellation to obtain the weight of each constellation point. After multiplying the weights of each constellation point by the real and imaginary parts of the corresponding constellation point, the results are summed along the dimension of the modulated constellation to obtain the real part expectation vector and the imaginary part expectation vector; the real part expectation vector and the imaginary part expectation vector are then concatenated to obtain the expectation matrix. The difference between the expected matrix and the symbol estimation vector at the current time step is calculated, and the difference is scaled according to the noise intensity at the current time step to obtain prior guidance information.

4. The symbol detection method according to claim 2, characterized in that, The step of obtaining the first reconstructed received signal vector by mapping the estimated transmitted symbol vector through the real-valued channel tensor includes: Multiply the estimated transmitted symbol vector by the real-valued channel tensor to obtain the first reconstructed received signal vector; The step of obtaining several second reconstructed received signal vectors by mapping the final symbol estimation vectors of the multiple trajectories through the real-valued channel tensor includes: The final symbol estimation vectors of the multiple trajectories are multiplied by the real-valued channel tensor to obtain several second reconstructed received signal vectors.

5. The symbol detection method according to claim 1, characterized in that, The step of calculating the likelihood guidance information based on the difference between the first reconstructed received signal vector and the real-valued received signal vector includes: Calculate the square of the Euclidean distance between the first reconstructed received signal vector and the real-valued received signal vector; The gradient of the square of the Euclidean distance with respect to the sign estimate vector at the current time step is calculated, and the gradient is used as the likelihood guidance information.

6. The symbol detection method according to claim 2, characterized in that, The second reconstructed received signal vector that has the smallest difference from the real-valued received signal vector includes: Calculate the Euclidean distance between the plurality of second reconstructed received signal vectors and the real-valued received signal vector; After sorting the Euclidean distances between the second reconstructed received signal vector and the real-valued received signal vector, the final symbol estimation vector corresponding to the second reconstructed received signal vector with the smallest Euclidean distance is selected as the optimal symbol estimation vector.

7. A symbol detection system, characterized in that, The symbol detection method according to any one of claims 1 to 6 includes: The signal acquisition module is used to acquire the received signal at the receiving end, as well as the channel matrix between the transmitting end and the receiving end; The real-valued module is used to construct an equivalent real-valued MIMO system model based on the received signal and the channel matrix, and to obtain a real-valued received signal vector and a real-valued channel tensor based on the equivalent real-valued MIMO system model. The symbol estimation vector initialization module is used to initialize the symbol estimation vector of the transmitter based on the equivalent real-valued MIMO system model. The prior guidance module is used to back-diffuse the symbol estimation vector at the current time step. During the back-diffusing process, prior guidance information of the symbol estimation vector at the current time step is calculated based on the modulation constellation. The estimated transmission symbol vector is determined according to the prior guidance information. The estimated transmission symbol vector is used to update the symbol estimation vector at the current time step to obtain intermediate estimation results. The likelihood guidance module is used to map the estimated transmitted symbol vector through the real-valued channel tensor to obtain a first reconstructed received signal vector, calculate likelihood guidance information based on the difference between the first reconstructed received signal vector and the real-valued received signal vector, correct the intermediate estimation result based on the likelihood guidance information, obtain the symbol estimation vector for the next time step, and complete the reverse diffusion process of the current time step; repeat the reverse diffusion process to obtain the final symbol estimation vector of the current trajectory.

8. A computer device, characterized in that, include: Processor and computer-readable storage media; A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program that, when executed by the processor, implements the symbol detection method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded by a processor and executed as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the symbol detection method as described in any one of claims 1 to 6.

Images