Internet of Things large-scale multi-user detection method based on precise structure learning

Abstract

The invention provides an Internet of Things large-scale multi-user detection method based on accurate structure learning. The method comprises the following steps: acquiring a base station receiving signal and an equivalent channel matrix, and initializing a sparse proportion, a noise variance and a maximum iteration number; executing signal estimation under the constraint of a user frame sparse model, wherein the user frame sparse model requires that the active user sets of all time slots are the same; the method comprises the following steps of: (a) executing accurate structure learning based on a memory linear estimation result, and updating a user active probability through full-time-slot Bayesian fusion; (b) generating a detection output in the non-linear estimation; (c) when iteration continues, optimizing the intermediate variable by adopting dynamic damping control; (d) performing multi-time-slot joint optimization on the sparse proportion and updating noise parameters; and (e) judging an iteration state based on the reconstruction error: if a termination condition is met, outputting a detection result, otherwise, returning to the step (a).

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technologies, and in particular to a multi-user detection method based on Accurate Structure Learning-Memory Approximate Message Passing (ASL-MAMP) in a scenario where a channel matrix is ​​pre-equalized in uplink Grant-Free Non-Orthogonal Multiple Access (GF-NOMA). Background Art

[0002] In the fifth generation of mobile communication systems (5G), massive machine type communication (mMTC) has attracted widespread attention. It features large user scale, low latency, low energy consumption, and sporadic communication. However, traditional orthogonal multiple access (OMA) schemes, limited by the limited number of orthogonal resources, are no longer able to meet the requirements of large-scale communication scenarios in mMTC. Furthermore, authorization-based communication in mMTC scenarios incurs significant signaling overhead. To address this, unlicensed non-orthogonal multiple access (GF-NOMA) schemes have been proposed. This scheme avoids complex signaling scheduling and allows active users to directly send their data to the base station, reducing access latency. Because GF-NOMA allows users to share a limited number of the same resources, it can free up more radio resources for payload transmission. However, the base station lacks any information about active users, making the research of multi-user detection (Joint Activity and Data Detection (JADD)) algorithms challenging.

[0003] Most existing multi-user detection schemes perform active user detection (AUD) by leveraging prior information about the sparsity of active users. However, this prior information is unknown in real-world applications. Furthermore, when considering user activity models with multiple observation vectors, a frame-sparse model is generally adopted, where potential users maintain the same active or inactive state throughout the observation frame. To exploit the temporal correlation of active users, detection accuracy can be improved by combining detection results from multiple time slots. However, simply treating the active user signals in each time slot as independent signals and detecting them separately fails to effectively exploit temporal correlation. Summary of the Invention

[0004] To address the shortcomings of existing multi-user detection algorithms, which struggle to balance inter-slot correlation utilization and computational efficiency, this paper provides an IoT multi-user detection method and system based on slot-correlated sparse learning. By establishing a sparse model constraint for user frames, this method innovatively leverages the temporal correlation of active users as a detection advantage. Combined with cross-slot Bayesian fusion and dynamic damping control, this method significantly improves detection accuracy and stability in GF-NOMA (Grant-Free Non-Orthogonal Multiple Access) scenarios.

[0005] In designing a user frame sparse model, this paper proposes for the first time the principle of time slot consistency for active users within an observation frame. This method requires that any user maintains the same active state throughout the frame duration, ensuring that the set of active users is completely consistent across different time slots. This constraint effectively captures the continuity of device behavior in IoT scenarios, laying the foundation for subsequent processing.

[0006] Based on this model, a cross-slot Bayesian probability fusion mechanism represents a key breakthrough. Unlike traditional independent slot detection, this mechanism combines the intermediate variables of all slots for probability fusion, fully exploiting the inter-slot correlation of data symbols and more accurately describing the true sparse structure of random access signals. This all-slot joint design significantly improves detection accuracy in low signal-to-noise ratio scenarios.

[0007] To address the convergence problem of high-dimensional signals, the dynamic damping mechanism inherent in the MAMP algorithm is used to optimize the iterative process in the following ways:

[0008] Fuse the historical nonlinear estimation output with the current linear estimation input;

[0009] Suppress oscillations by a normalization factor that decreases with the number of iterations;

[0010] Optimize intermediate variables to improve the stability of subsequent iterations.

[0011] This regulation mechanism based on history-current coupling ensures the convergence efficiency of the algorithm, reduces the number of iterations, and greatly improves the iteration efficiency.

[0012] In terms of parameter optimization, multi-slot joint innovations are embodied in a dual design: Expectation maximization achieves sparse multi-slot averaging, effectively enhancing parameter robustness; while synchronously updating noise parameters prevents the algorithm from falling into local extrema. This, combined with user-side pre-equalization technology—which normalizes the equivalent channel matrix through pre-equalization—significantly reduces base station implementation complexity. When the iterations meet the termination criteria, the reconstructed signal mean and variance are output as the final detection result.

[0013] Tests show that this solution achieves a significant improvement in detection performance while keeping computational complexity manageable (the maximum number of iterations is significantly smaller than the signal dimension). The pre-equalization design simplifies the channel matrix to the identity matrix, enabling efficient execution of the detection algorithm even on resource-constrained IoT devices.

[0014] Through an adaptive initialization mechanism, the sparsity ratio and noise variance are correlated and calculated based on the Gaussian distribution function and the received signal energy, respectively. This design supports a wide range of parameter variations and ensures stable algorithm convergence in complex environments. The resulting closed-loop technology provides a high-precision, low-complexity system-level solution for large-scale IoT access.

[0015] The technical solution specifically adopted by the present invention to solve the technical problem is:

[0016] A large-scale multi-user detection method for the Internet of Things based on precise structure learning:

[0017] Obtain the base station received signal and equivalent channel matrix, initialize the sparse ratio, noise variance and maximum number of iterations;

[0018] performing signal estimation under the constraints of a user frame sparse model requiring the same set of active users for all time slots;

[0019] Perform the following operations in order:

[0020] (a) Perform accurate structure learning based on the memory linear estimation results and update the user activity probability through full-slot Bayesian fusion;

[0021] (b) generating detection outputs in nonlinear estimation;

[0022] (c) As the iterations continue, the intermediate variables are optimized using dynamic damping control;

[0023] (d) perform multi-slot joint optimization on the sparsity ratio and update the noise parameters;

[0024] (e) Determine the iteration status based on the reconstruction error: if the termination condition is met, output the detection result; otherwise, return to step (a).

[0025] Furthermore, the user frame sparse model satisfies: for any two time slots t1 and t2, user k is active in time slot t1 if and only if it is active in time slot t2.

[0026] Furthermore, the intermediate variables of each time slot, including the signal estimation value and its variance, are independently calculated through memory linear estimation; the intermediate variables of all time slots are combined and the user activity probability is updated through Bayesian probability fusion.

[0027] Furthermore, the dynamic damping control is achieved by:

[0028] Perform weighted fusion of historical nonlinear estimation output and current linear estimation input;

[0029] A normalization factor that decreases with the number of iterations is used to control the fusion weight;

[0030] The output signal after damping is used as the input for subsequent iterations.

[0031] Furthermore, the multi-slot joint optimization obtains a global sparsity ratio through average calculation.

[0032] Furthermore, the noise parameter update is achieved through expectation maximization.

[0033] Furthermore, it also includes performing user-side pre-equalization processing on the equivalent channel matrix, so that the equivalent channel matrix satisfies the identity matrix condition, that is, all diagonal elements in the matrix are 1.

[0034] Furthermore, the initialization sparse ratio and noise variance satisfy: the sparse ratio is calculated by associating a Gaussian distribution function with a user activity rate; and the noise variance is determined based on the received signal energy and a preset signal-to-noise ratio threshold.

[0035] And, a large-scale multi-user detection system for the Internet of Things based on precise structure learning, including:

[0036] A signal acquisition module is configured to obtain a base station received signal and an equivalent channel matrix; an initialization module is configured to initialize a sparsity ratio, a noise variance, and a maximum number of iterations; a signal estimation engine is configured to perform signal estimation under the constraints of a user frame sparse model, wherein the user frame sparse model requires that the set of active users in all time slots is the same; and a joint processor is configured to sequentially perform the following operations:

[0037] (a) Perform accurate structure learning based on the memory linear estimation results and update the user activity probability through full-slot Bayesian fusion;

[0038] (b) generating detection outputs in nonlinear estimation;

[0039] (c) As the iterations continue, the intermediate variables are optimized using dynamic damping control;

[0040] (d) perform multi-slot joint optimization on the sparsity ratio and update the noise parameters;

[0041] (e) Determine the iteration status based on the reconstruction error: if the termination condition is met, output the detection result; otherwise, return to step (a); an output controller is configured to output the detection result when the reconstruction error reaches the convergence criterion or the maximum number of iterations is reached.

[0042] And, a computer device includes a memory, a processor and a computer program stored in the memory, and the processor implements the above method when executing the computer program.

[0043] A non-transitory computer-readable storage medium stores a computer program, which implements the method described above when executed by a processor.

[0044] Compared with the prior art, the present invention and its preferred embodiments have at least the following beneficial effects:

[0045] 1. Time slot consistency constraints significantly improve detection accuracy

[0046] The user frame sparse model enforces the same set of active users across all time slots, leveraging the temporal continuity of IoT devices. This constraint extends traditional independent time slot detection to frame-level joint detection, effectively reducing false alarms and improving detection robustness in low signal-to-noise ratio scenarios.

[0047] 2. Full-time Bayesian fusion breaks through the limitations of information isolation

[0048] The innovative all-time-slot joint probability fusion mechanism integrates intermediate variables across all time slots for Bayesian inference. This mechanism addresses the problem of signal interference between time slots and significantly enhances the reliability of active user identification, making it particularly suitable for IoT scenarios with bursty communications.

[0049] 3. Dynamic damping control ensures high-dimensional convergence stability

[0050] By weightedly fusing the historical nonlinear estimation output with the current linear estimation input, combined with a normalization factor that decreases with iteration, this method successfully suppresses divergent oscillations in large-scale matrix operations. This convergence control based on historical-current coupling overcomes the iterative efficiency bottleneck of traditional approximate message passing algorithms and provides a technical foundation for real-time detection.

[0051] 4. Parameter optimization and double enhancement

[0052] Multi-slot averaging of sparse ratios and joint updating of noise parameters form complementary optimizations: the former improves parameter robustness, while the latter prevents the algorithm from falling into local extrema. The expectation maximization module ensures global stability of parameter updates, enabling the algorithm to maintain stable performance in dynamic environments.

[0053] 5. Pre-equalization simplifies implementation complexity

[0054] User-side pre-equalization simplifies the equivalent channel matrix to the identity matrix, significantly reducing the computational burden of base station detection. This design, combined with the core algorithm, creates a hardware-software synergy, making the solution suitable for resource-constrained IoT devices.

[0055] 6. Adaptive initialization broadens application scenarios

[0056] The parameter initialization mechanism based on Gaussian distribution and signal energy supports network deployment of different scales. This wide range of adaptability ensures the algorithm's plug-and-play capability in diverse IoT environments.

[0057] These innovations form a tightly linked technical chain: the time-slot consistency model provides structural support for Bayesian fusion, dynamic damping control ensures efficient high-dimensional signal reconstruction, multi-slot optimization enhances parameter stability, and pre-equalization design lowers the implementation barrier. This closed-loop design enables a systematic improvement in detection accuracy while maintaining computational efficiency, providing a reliable technical foundation for large-scale IoT access. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0059] Figure 1 Schematic diagram of a user frame sparse model in uplink weighted non-orthogonal multiple access;

[0060] Figure 2 This is a flowchart of a method implementation according to an embodiment of the present invention;

[0061] Figure 3 This is a flowchart of precise structure learning in an embodiment of the present invention;

[0062] Figure 4 FIG. 5 is a diagram comparing the symbol error rate (SER) performance of different algorithms in an embodiment of the present invention. DETAILED DESCRIPTION

[0063] In order to make the features and advantages of the present invention more clearly understood, the following embodiments are given for detailed description:

[0064] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used in this specification have the same meanings as those commonly understood by those skilled in the art to which this application belongs.

[0065] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0066] like Figure 2 、 Figure 3 As shown, an embodiment of the present invention proposes a large-scale multi-user detection method for the Internet of Things based on accurate structure learning, which adopts the Accurate Structure Learning-Memory Approximate Message Passing (ASL-MAMP) method and is applied to the scenario where the channel matrix is ​​pre-equalized in the uplink Grant-Free Non-Orthogonal Multiple Access (GF-NOMA).

[0067] The method comprises the following steps: step S1, obtaining the received signals of multiple users at the base station, equivalent channel matrix information, damping length and the singular value diagonal matrix of the equivalent channel matrix, and optimizing initialization parameters, and setting the maximum number of iterations; step S2, using memory linear estimation (MLE) under the user frame sparse model to update the intermediate variables in the iteration process; step S3, using accurate structure learning (ASL) to update the sparsity ratio of the reconstructed signal according to the result of MLE, and then updating the user activity probability according to the updated sparsity ratio; step S4, using non-linear estimation (NLE) to update the mean and variance of the reconstructed signal according to the updated result of ASL on the user activity probability; step S5, using expectation maximization (ASL) to update the mean and variance of the reconstructed signal. The EM module performs a multi-slot averaging of the sparsity ratio of the reconstructed signal and further updates the equivalent Gaussian white noise based on the updated mean and variance of the reconstructed signal. Step S6 determines whether the iteration termination condition is met. If so, the reconstructed signal is output as the detection result. If not, steps S2-S5 are repeated until the iteration termination condition is met or the maximum number of iterations is reached. This invention effectively exploits the correlation of active users in different time slots under the user frame sparse model and uses the shared sparsity ratio learned through precise structure learning to detect active users, significantly improving the multi-user detection performance of the method.

[0068] like Figure 1 As shown, in the user frame sparse model, the IoT large-scale multi-user detection method based on precise structure learning of the present invention can effectively utilize the sparse ratio shared by active users between different time slots, and use the sum-product formula to enable the calculation process of the sparse ratio to fully utilize the information transmitted by the check nodes, which can greatly improve the detection performance and reconstruction performance of the algorithm.

[0069] As a preferred solution of this embodiment, step S1 specifically includes the following steps:

[0070] Step S11: The base station obtains the received signal It can be expressed as:

[0071] (1)

[0072] in, It is in The length of the time slot is The received symbol vector, is the number of orthogonal subcarriers; Indicates in The equivalent channel matrix under time slots is , is the total number of potential users, given by the channel gain matrix and spreading matrix It is assumed that after the active user equipment monitors the periodic broadcast beacon of the base station, it performs synchronization, power control and channel estimation, and introduces pre-equalization processing at the active user end, that is, ; Indicates in The original transmitted symbol modulated in time slots, the total number of time slots is ; Indicates in The mean value is 0 in the time slots and the variance is Additive white Gaussian noise.

[0073] Step S12: Initialize the sparse ratio parameters of the EM part and noise variance :Since the EM part may shrink to a local extreme point, an appropriate initialization formula is required. In actual use, when the initialization parameters are set to change in a large range, the EM part can also operate normally. Even so, when the initial value deviates too much from the actual value, the detection performance of the EM part will deteriorate. Use formula (2)-formula (3) to initialize and :

[0074] (2)

[0075] (3)

[0076] in, and are the cumulative distribution function and probability distribution function of the standard Gaussian distribution, express Non-zero probability, set by experience .

[0077] Step S13: Initialize parameters and set the maximum number of iterations: The singular value diagonal matrix of the equivalent channel matrix obtained by the base station is expressed as , ; Set weight parameters Set to 1; set auxiliary parameters ; Set the maximum number of iterations and ;set up , , update the orthogonal auxiliary parameters according to formula (4)-formula (7) , ,and .

[0078] (4)

[0079] (5)

[0080] (6)

[0081] (7)

[0082] Among them, the lower right corner Indicates the Iterations.

[0083] As a preferred solution of this embodiment, step S2 uses memory linear estimation to update intermediate variables in the iterative process under a model where user frames are sparse, and specifically includes the following steps:

[0084] Model design for sparse user frames: In a scenario where user frames are sparse, potentially active users will remain active in consecutive time slots within the observation frame, while inactive users will remain inactive. That is:

[0085] (8)

[0086] in, where represents the set of nonzero elements in the vector.

[0087] Step S22: Use the MAMP method to update the reconstructed signal: perform memory linear estimation (MLE) on the reconstructed signal and update the reconstructed signal using formula (9)-formula (19) The mean of the product with the equivalent channel matrix and variance :

[0088] (9)

[0089] (10)

[0090] (11)

[0091] (12)

[0092] (13)

[0093] (14)

[0094] (15)

[0095] (16)

[0096] (17)

[0097] (18)

[0098] (19)

[0099] Among them, the upper right corner Indicates the iterations, represents the normalization parameter, represents the orthogonalization parameter, .

[0100] As a preferred solution of this embodiment, step S3 updates the sparsity ratio of the reconstructed signal using precise structure learning based on the result of the memory linear estimation, and then updates the user activity probability based on the updated sparsity ratio, which specifically includes the following steps:

[0101] Step S31: Use the sparse ratio shared by each time slot to detect active users on the updated reconstructed signal: Considering the user static active scenario, if it is assumed that Although the assumption of independence simplifies the calculation of the posterior probability and the derivation of NLE, it cannot accurately describe the true sparse structure of the reconstructed signal. There is temporal correlation between user time slots, and different time slots share the same sparsity ratio. This embodiment introduces this correlation when calculating the sparsity ratio of the reconstructed signal, which improves the accuracy of user activity detection. Therefore, the process of active user detection can be expressed as formula (20)-formula (23):

[0102] (20)

[0103] (twenty one)

[0104] (twenty two)

[0105] (twenty three)

[0106] in, represents the modulation constellation set, represents the modulation order, express The conjugate transpose of Indicates time slot , No. User in The MLE result at the iteration is: express In the time slot The variance of express In the The sparsity ratio of the iteration, Indicates time slot , No. The probability of a user being active in The result at the iteration.

[0107] As a preferred solution of this embodiment, step S4 updates the mean and variance of the reconstructed signal using nonlinear estimation based on the updated result of the user activity probability through precise structure learning, and specifically includes the following steps:

[0108] Step S41, update the posterior probability of the reconstructed signal according to the result of active user detection: in step S3, the active user set is updated by updating the sparse ratio shared by different time slots. Next, nonlinear estimation NLE is performed on the reconstructed signal to update the posterior probability of the reconstructed signal, obtain the mean and variance of the reconstructed signal, and calculate the intermediate variable Damping operation is performed to ensure the convergence of the ASL-MAMP algorithm:

[0109] (twenty four)

[0110] (25)

[0111] (26)

[0112] in represents the damping vector, represents a separable Lipschitz continuous estimator, represents the mean of the reconstructed signal, represents the variance of the reconstructed signal.

[0113] As a preferred solution of this embodiment, step S5 expects the maximization module to perform a multi-slot averaging operation on the sparsity ratio of the reconstructed signal, and further updates the equivalent Gaussian white noise based on the updated results of the mean and variance of the reconstructed signal, specifically including the following steps:

[0114] Step S51: Use EM module to learn sparse ratio and noise variance : Residual and residuals Represent two alternating reconstruction parameters 、 The residuals between and the reconstruction parameters The retained residuals in the two iterations are expressed using formula (27)-formula (28):

[0115] (27)

[0116] (28)

[0117] Update the sparse ratio and noise variance After that, update the number of iterations , and start to judge whether the number of iterations meets the iteration termination condition. If not, continue to repeat steps S2-S5 until the maximum number of iterations is met. After the iteration is terminated, the variance of the final reconstructed signal is As the output signal of the detection algorithm.

[0118] The ASL-MAMP algorithm proposed in the embodiment of the present invention is based on the signal reconstruction module of the MAMP algorithm and the active user detection module based on ASL. Under the model of sparse active user frames, the algorithm considers the ASL strategy with a more complex prior distribution by establishing a graphical model, thereby making full use of the sparse ratio shared between different time slots. Figure 4 As shown in Figure 3, the simulation results show that the proposed ASL-MAMP algorithm has significantly better symbol error rate performance than the MAMP algorithm.

[0119] Compared with the prior art, the above solution of the embodiment of the present invention has the following beneficial effects:

[0120] 1. Under the model of sparse user frames, preliminary signal reconstruction is performed based on the MAMP algorithm. Compared with other algorithms, the MAMP algorithm is an approximate message passing algorithm with higher detection performance and efficiency, laying the foundation for the subsequent introduction of sparse ratios shared between time slots and multi-user detection using sparse ratios;

[0121] 2. Under different signal-to-noise ratio conditions, by establishing an image model, the ASL strategy's prior distribution is significantly closer to the true distribution of the signal. Utilizing the sum-product formula, the calculation process of the sparsity ratio can fully utilize the information transmitted by the check nodes, thereby significantly improving the algorithm's detection performance.

[0122] 3. Assuming that the active user equipment monitors the base station's periodic broadcast beacon, performs synchronization, power control and channel estimation, and introduces pre-equalization processing on the active user side, as long as the base station receives the signal, the total number of time slots The larger it is, the stronger the temporal correlation introduced from the check nodes is, and the better the detection performance of the algorithm is.

[0123] Based on the same inventive concept, the present invention also provides a computer device, which includes: one or more processors and a memory for storing one or more computer programs; the program includes program instructions, and the processor is used to execute the program instructions stored in the memory. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is used to implement one or more instructions, specifically for loading and executing one or more instructions in a computer storage medium to implement the above method.

[0124] It should be further explained that, based on the same inventive concept, the present invention also provides a computer storage medium having a computer program stored thereon, which, when executed by a processor, performs the above-described method. The storage medium may be any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more conductors, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0125] It should be noted that, unless otherwise defined, the technical or scientific terms used in the present invention should have the usual meanings understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0126] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other manner. Any person skilled in the art may utilize the above-disclosed technical content to modify or modify the present invention into equivalent embodiments. However, any simple modifications, equivalent variations, and modifications to the above embodiments that do not depart from the technical content of the present invention and are based on the technical essence of the present invention remain within the scope of protection of the present invention.

[0127] The present invention is not limited to the above-mentioned best embodiment. Anyone can derive various other forms of a large-scale multi-user detection method for the Internet of Things based on precise structure learning based on the inspiration of the present invention. All equivalent changes and modifications made within the scope of the patent application of the present invention should fall within the scope of the present invention.

Claims

1. A large-scale multi-user detection method for the Internet of Things based on precise structure learning, characterized by: Obtain the base station received signal and equivalent channel matrix, initialize the sparse ratio, noise variance and maximum number of iterations; performing signal estimation under the constraints of a user frame sparse model requiring the same set of active users for all time slots; Do the following in order: (a) Perform accurate structure learning based on the memory linear estimation results and update the user activity probability through full-slot Bayesian fusion; (b) generating detection outputs in nonlinear estimation; (c) As the iterations continue, dynamic damping control is used to optimize the intermediate variables; (d) perform multi-slot joint optimization on the sparsity ratio and update the noise parameters; (e) Determine the iteration status based on the reconstruction error: if the termination condition is met, output the detection result; otherwise, return to step (a).

2. The method for large-scale multi-user detection in the Internet of Things based on precise structure learning according to claim 1, characterized in that: The user frame sparse model satisfies: for any two time slots t1 and t2, user k is active in time slot t1 if and only if it is active in time slot t2.

3. The method for large-scale multi-user detection in the Internet of Things based on precise structure learning according to claim 1, characterized in that: The intermediate variables of each time slot are calculated independently through memory linear estimation; the intermediate variables of all time slots are combined and the user activity probability is updated through Bayesian probability fusion.

4. The method for large-scale multi-user detection in the Internet of Things based on precise structure learning according to claim 1, characterized in that: The dynamic damping control is achieved by: Perform weighted fusion of historical nonlinear estimation output and current linear estimation input; A normalization factor that decreases with the number of iterations is used to control the fusion weight; The output signal after damping is used as the input for subsequent iterations.

5. The method for large-scale multi-user detection in the Internet of Things based on precise structure learning according to claim 1, characterized in that: The multi-slot joint optimization obtains the global sparsity ratio through average calculation.

6. The method for large-scale multi-user detection in the Internet of Things based on precise structure learning according to claim 1, characterized in that: The noise parameter update is achieved by expectation maximization.

7. The method for large-scale multi-user detection in the Internet of Things based on precise structure learning according to claim 1, characterized in that: The method also includes performing user-side pre-equalization processing on the equivalent channel matrix so that the equivalent channel matrix satisfies a unit matrix condition.

8. The method for large-scale multi-user detection in the Internet of Things based on precise structure learning according to claim 1, characterized in that: The initialization sparse ratio and noise variance satisfy: the sparse ratio is calculated by associating a Gaussian distribution function with a user activity rate; and the noise variance is determined based on the received signal energy and a preset signal-to-noise ratio threshold.

9. A large-scale multi-user detection system for the Internet of Things based on precise structure learning, characterized in that: include: A signal acquisition module configured to obtain a base station receiving signal and an equivalent channel matrix; An initialization module configured to initialize the sparsity ratio, noise variance, and maximum number of iterations; a signal estimation engine configured to perform signal estimation under the constraints of a user frame sparsity model, wherein the user frame sparsity model requires the same set of active users for all time slots; and a joint processor configured to sequentially perform the following operations: (a) Perform accurate structure learning based on the memory linear estimation results and update the user activity probability through full-slot Bayesian fusion; (b) generating detection outputs in nonlinear estimation; (c) As the iterations continue, dynamic damping control is used to optimize the intermediate variables; (d) perform multi-slot joint optimization on the sparsity ratio and update the noise parameters; (e) Determine the iteration status based on the reconstruction error: if the termination condition is met, output the detection result; otherwise, return to step (a); Output controller, configured to output the detection result when the reconstruction error reaches the convergence criterion or the maximum number of iterations is reached.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program implements the method according to any one of claims 1 to 8 when executed by a processor.

Images