Channel estimation method, device, equipment, storage medium and program product

Abstract

The invention provides a channel estimation method, device and equipment, a storage medium and a program product, which are applied to the technical field of wireless communication. The method comprises the following steps: collecting state information of a current communication environment; inputting the state information to the pilot frequency scheduling strategy network to obtain a pilot frequency activation mask, wherein the pilot frequency activation mask is used for indicating whether the user terminal sends the pilot frequency or not; constructing a sparse pilot frequency matrix based on the pilot frequency activation mask, and transmitting pilot frequency signals according to the sparse pilot frequency matrix to generate a pilot frequency observation matrix; the pilot frequency observation matrix and the pilot frequency activation mask serve as input of a channel estimator, missing channel information is complemented through back diffusion iteration, a channel estimation result is obtained, and the channel estimator is constructed based on a generation diffusion model; and calculating a normalized mean square error of the channel estimation result and a real channel, constructing a reward function in combination with the normalized mean square error and the pilot frequency usage amount, and updating a scheduling strategy of the pilot frequency scheduling strategy network based on the reward function.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and in particular to a channel estimation method, apparatus, device, storage medium, and program product. Background Technology

[0002] In fifth-generation and future sixth-generation mobile communication systems, the cellular-free massive MIMO architecture, with its distributed access point collaborative operation, significantly improves system capacity and coverage, making it widely applicable to scenarios such as disaster emergency communication and industrial IoT. Under this architecture, accurate acquisition of channel state information is a prerequisite for realizing core functions such as beamforming and interference management, while channel estimation relies on pilot resources.

[0003] In existing technologies, traditional linear estimation methods, compressed sensing and low-rank matrix recovery techniques, deep learning models and generative diffusion models, and reinforcement learning methods have all been used for channel estimation or pilot scheduling. However, traditional linear estimation suffers from a sharp performance drop when pilots are insufficient, non-orthogonal, or observations are missing; compressed sensing methods rely on strong prior assumptions and are difficult to adapt to complex channel and environmental changes; deep learning models have limited generalization and reliability; while generative diffusion models possess sparse recovery capabilities, pilot configurations are fixed and rigid; and reinforcement learning is not deeply integrated with channel estimation. Therefore, existing technologies cannot simultaneously achieve both estimation accuracy and dynamic resource optimization requirements under sparse pilot conditions. Summary of the Invention

[0004] This invention provides a channel estimation method, apparatus, device, storage medium, and program product to solve the problem that existing technologies cannot simultaneously meet the estimation accuracy and dynamic resource optimization requirements under sparse pilots in decellularized massive MIMO architectures. It can achieve accurate estimation of large-scale high-dimensional channel states with limited pilot resources, and dynamically optimize pilot transmission strategies to save communication resources while ensuring estimation performance.

[0005] This invention provides a channel estimation method, comprising: collecting state information of the current communication environment; inputting the state information into a pilot scheduling policy network to obtain a pilot activation mask, wherein the pilot activation mask is used to indicate whether a user terminal sends a pilot; constructing a sparse pilot matrix based on the pilot activation mask, and transmitting pilot signals according to the sparse pilot matrix to generate a pilot observation matrix; using the pilot observation matrix and the pilot activation mask as inputs to a channel estimator, and completing the missing channel information through back-diffusion iteration to obtain a channel estimation result, wherein the channel estimator is an estimator constructed based on a generative diffusion model; calculating the normalized mean square error between the channel estimation result and the actual channel, constructing a reward function by combining the normalized mean square error and pilot usage, and updating the scheduling policy of the pilot scheduling policy network based on the reward function.

[0006] This invention also provides a channel estimation device, comprising the following modules: an information acquisition module, a pilot scheduling decision module, a pilot signal interaction module, a diffusion channel estimation module, and a strategy optimization feedback module; the information acquisition module is used to acquire the status information of the current communication environment; the pilot scheduling decision module is used to input the status information into the pilot scheduling strategy network to obtain a pilot activation mask, the pilot activation mask being used to indicate whether the user terminal should send a pilot; the pilot signal interaction module is used to construct a sparse pilot matrix based on the pilot activation mask, and transmit pilot signals according to the sparse pilot matrix to generate a pilot observation matrix; the diffusion channel estimation module is used to use the pilot observation matrix and the pilot activation mask as inputs to a channel estimator, and complete the missing channel information through back diffusion iteration to obtain a channel estimation result, the channel estimator being an estimator constructed based on a generative diffusion model; the strategy optimization feedback module is used to calculate the normalized mean square error between the channel estimation result and the actual channel, construct a reward function by combining the normalized mean square error and the pilot usage, and update the scheduling strategy of the pilot scheduling strategy network based on the reward function.

[0007] The present invention also provides an electronic device, 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 any of the channel estimation methods described above.

[0008] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the channel estimation method as described above.

[0009] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the channel estimation method as described above.

[0010] The channel estimation method, apparatus, device, storage medium, and program product provided by this invention, by receiving the current communication environment status information through a pilot scheduling strategy network, generating a pilot activation mask, and constructing a sparse pilot matrix, can dynamically adjust the pilot transmission strategy based on the real-time environment status. This avoids the problem of fixed and rigid pilot configuration in existing technologies and achieves adaptive and optimized allocation of pilot resources. Furthermore, it employs a channel estimator based on a generative diffusion model, using the pilot observation matrix and pilot activation mask as input, and recovering the true channel information through back-diffusion iteration. Because the diffusion model possesses strong sparse completion capabilities... Therefore, even in extremely sparse pilot scenarios where the pilot length is much smaller than the number of users, it can accurately recover high-dimensional channel information, breaking through the performance limitations of traditional linear estimation and compressed sensing methods. By combining the normalized mean square error and pilot usage to construct a reward function and using it to update the scheduling strategy of the pilot scheduling strategy network, a closed-loop optimization mechanism is formed. Therefore, it can dynamically balance the channel estimation accuracy and pilot resource overhead, solving the pain point that existing technologies cannot take both into account. This significantly improves the communication efficiency and robustness of non-cellular massive MIMO architecture in complex scenarios such as disaster emergency communication and industrial IoT. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0012] Figure 1 This is one of the flowcharts illustrating the channel estimation method provided by the present invention; Figure 2 This is the second flowchart illustrating the channel estimation method provided by the present invention; Figure 3 This is a schematic diagram of the channel estimation device provided by the present invention; Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions 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.

[0014] It should be noted that in the embodiments of this application, the words "exemplary" or "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design scheme described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of the words "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0015] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0016] To facilitate a clear description of the technical solutions of the embodiments of this application, the terms "first" and "second" are used in the embodiments of this application to distinguish the same or similar items with essentially the same function and effect. Those skilled in the art can understand that the terms "first" and "second" are not intended to limit the quantity or execution order.

[0017] This application describes some exemplary embodiments for illustrative purposes. It should be understood that this application may be implemented in other ways not specifically shown in the accompanying drawings.

[0018] like Figure 1 As shown in the figure, this application provides a channel estimation method that can be applied to a channel estimation device. It is suitable for extreme communication environments such as limited pilot resources, rapidly changing channel states, and complex user distribution, achieving efficient, low-overhead, and high-precision wireless channel estimation. It is particularly suitable for scenarios such as disaster emergency communication, UAV relay networks, and industrial IoT. The channel estimation method may include steps S101-S105: S101, The channel estimation device collects the status information of the current communication environment.

[0019] The status information includes the spatial distribution of user terminals and access points, average channel gain, channel signal-to-noise ratio, historical pilot configuration, and historical channel estimation error.

[0020] Specifically, the channel estimation device can collect the required state information from the current wireless communication environment. This state information may include: the spatial locations of all user terminals and access points, the average channel gain, the signal-to-noise ratio (SNR) level of the current frame, historical pilot usage, and the error feedback from the previous channel estimation. These state data are then combined into a high-dimensional state vector.

[0021] S102. The channel estimation device inputs the state information into the pilot scheduling strategy network to obtain the pilot activation mask.

[0022] The pilot activation mask is used to indicate whether the user terminal should send a pilot signal.

[0023] Optionally, the pilot scheduling strategy network inputs the state information to obtain the pilot activation mask, including: extracting decision features from the state information; inputting the decision features into the pilot scheduling strategy network, and outputting a binary sequence with a length consistent with the number of user terminals as the pilot activation mask, wherein the pilot activation mask follows the constraint of the maximum number of user terminals that are active in sending pilots in the current time slot.

[0024] Specifically, the channel estimation device can extract and filter features from the collected high-dimensional state information, extract topological features such as relative distance from the spatial distribution of user terminals and access points, extract statistical features such as gain amplitude range from the average channel gain, extract environmental features such as signal quality level and interference intensity from the channel signal-to-noise ratio, and retain the activation ratio of historical pilot configuration, the distribution characteristics of activated users, and the magnitude and trend characteristics of historical channel estimation errors. After removing redundant noise data, it forms dimension-optimized decision features.

[0025] Then, the decision feature is standardized and normalized, converted into a data format suitable for the input requirements of the pilot scheduling strategy network, and input into the trained pilot scheduling strategy network. Through the network's fully connected layers and activation function operations, the pilot transmission priority of each user terminal is quantitatively evaluated. Based on the evaluation results, an initial binary sequence is generated, where user terminals with a priority higher than a preset threshold are marked with "1", and the rest are marked with "0". Subsequently, the initial binary sequence undergoes compliance verification. If the number of activated user terminals exceeds the maximum number constraint K set for the current time slot, the excess "1" markers are removed in descending order of priority until the number of activated terminals meets the constraint requirements, ultimately forming a pilot activation mask with a length consistent with the number of user terminals.

[0026] It should be noted that the pilot scheduling policy network is provided with decisions by a reinforcement learning controller. This network is modeled as a Markov Decision Process (MDP), with the goal of learning a policy. In state Selecting the pilot activation mode maximizes long-term benefits. The pilot activation mask can be used to indicate which user terminal nodes should send pilot signals in this frame.

[0027] For example, the pilot activation mask can be a length of binary sequence, ,in This indicates that the i-th terminal transmits a pilot signal in the current time slot. This indicates that the i-th terminal does not send a pilot signal in the current time slot.

[0028] It should be noted that the pilot activation mask not only considers the goal of maximizing estimation accuracy, but also introduces resource control capabilities through policy learning, which can achieve a balance between pilot minimization and channel reconstruction maximization.

[0029] S103. The channel estimation device constructs a sparse pilot matrix based on the pilot activation mask, and transmits pilot signals according to the sparse pilot matrix to generate a pilot observation matrix.

[0030] Specifically, the channel estimation device parses the binary indication information corresponding to each user terminal in the pilot activation mask, marks the user terminal corresponding to "1" as pilot transmission active state, and the user terminal corresponding to "0" as pilot silent state. Based on the marking result, a sparse pilot matrix with dimensions matching the total number of user terminals and the length of the pilot sequence is constructed. Only the positions corresponding to the active state user terminals in the matrix are filled with a preset pilot sequence, and the positions corresponding to the silent state user terminals are filled with zero values ​​to achieve sparse transmission of pilot signals.

[0031] Subsequently, the channel estimation device sends pilot transmission instructions to each user terminal through the distributed access point of the non-cellular massive MIMO system. The instructions instruct the active user terminals to transmit pilot signals on the specified time-frequency resources according to the pilot sequence allocated in the sparse pilot matrix, while the silent user terminals do not transmit pilot signals.

[0032] The distributed access point receives pilot signals sent by each active user terminal. Taking into account the actual situation of signal attenuation and noise interference during channel transmission, it performs synchronization, filtering and amplitude calibration on the received signals. The processed signals are then recombined according to the user terminal order and access point number to generate a pilot observation matrix containing channel observation information, which provides input data support for the subsequent channel estimation module.

[0033] S104. The channel estimation device uses the pilot observation matrix and pilot activation mask as inputs to the channel estimator, and completes the missing channel information through back-diffusion iteration to obtain the channel estimation result.

[0034] The channel estimator is an estimator built based on a generative diffusion model.

[0035] Optionally, the pilot observation matrix and pilot activation mask are used as inputs to the channel estimator. Missing channel information is filled in through backdiffusion iteration to obtain the channel estimation result. This includes: converting the complex channel data format of the pilot observation matrix and pilot activation mask into a real tensor format using a feature coding network to generate a conditional feature vector adapted to the channel estimator; the diffusion model automatically fills in missing information at non-pilot locations using its own characteristics; initializing Gaussian distributed samples, initiating backdiffusion iteration, and predicting the noise component of each noisy sample based on the conditional feature vector using a parameterized neural network; and using a predictor-corrector sampling method to gradually reconstruct the channel matrix based on the noise component.

[0036] Specifically, the channel estimation device can fuse the complex domain channel observation data contained in the pilot observation matrix with the binary indication information corresponding to the pilot activation mask, and then input it into the feature coding network. Through complex separation, dimension mapping and normalization operations within the network, the real and imaginary parts of the complex channel data are separated into independent dimensions and mapped to a preset real space, which is then transformed into a real tensor format that meets the processing requirements of the generation and diffusion model. At the same time, the indication features of the pilot position are preserved, and a conditional feature vector that combines channel observation information and position constraint information is generated.

[0037] Then, an initial noisy sample with the same dimension as the channel matrix is ​​randomly generated using a standard Gaussian distribution as the initial distribution. The back-diffusion iteration process is started. In each iteration, the current noisy sample, the corresponding iteration time step information, and the conditional feature vector are input into the parameterized neural network. The noise distribution law of the channel data is learned through the multi-layer convolution and fully connected operations of the network, and the noise component superimposed in the current noisy sample is accurately predicted.

[0038] The channel matrix is ​​then restored using a predictor-corrector sampling method. The predictor selects an appropriate iteration step size based on the solver of the stochastic differential equation (SDE) and updates the noisy samples according to the noise prediction results to approximate the true channel distribution. The corrector fine-tunes and optimizes the predicted samples through a Markov chain Monte Carlo process to improve the fit between the samples and the target channel distribution, gradually reduce the noise interference in the samples, and restore the effective characteristics of the channel.

[0039] After each round of back-diffusion iteration, pilot consistency constraints are applied based on the pilot positions indicated by the pilot activation mask. The values ​​of the corresponding pilot positions in the estimated channel are replaced with the measured data in the pilot observation matrix through position index matching to ensure the observation consistency of pilot positions. The missing channel information at non-pilot positions is automatically filled in by the generative diffusion model based on the learned global channel distribution characteristics and the correlation of adjacent channel features.

[0040] The above noise prediction, sample update, and constraint application process is continuously executed according to the preset total number of back diffusion iterations until the iteration terminates. The final real tensor format data is then converted back into a complex domain channel matrix, and a complete and accurate channel estimation result is output.

[0041] Optionally, before using the pilot observation matrix and pilot activation mask as input to the channel estimator and completing the missing channel information through back-diffusion iteration to obtain the channel estimation result, the method further includes: using the real complete channel matrix as a sample, constructing a diffusion trajectory by gradually adding noise, so that the channel estimator learns the global distribution characteristics and condition generation capability of the channel.

[0042] Specifically, the real complete channel matrix collected in a non-cellular large-scale multiple-input multiple-output scenario is used as the training sample set. The complex domain channel data corresponding to each sample is converted into a real tensor format and then input into the forward diffusion module. According to the preset noise scheduling table, a noise component conforming to a Gaussian distribution is superimposed on the current sample in each diffusion step. The variance of the noise gradually increases with the number of diffusion steps. Through multi-step iteration, a diffusion trajectory is constructed that gradually transitions from the real channel data distribution to an isotropic Gaussian distribution, so that the sample gradually loses the original channel characteristics and approaches a random noise distribution during the diffusion process.

[0043] Simultaneously, the pilot activation mask and pilot observation matrix corresponding to the training samples are input into the feature encoding network to generate conditional feature vectors that match the training samples. In each diffusion step, the current noisy sample, the corresponding iteration time step information, and the conditional feature vector are input into the parameterized neural network to predict the noise component superimposed in the current sample as the training objective. The network parameters are optimized by minimizing the mean square error between the predicted noise and the actual superimposed noise.

[0044] During multiple training rounds, the model learns the diffusion trajectory patterns under different channel scenarios and gradually masters the global distribution characteristics of channel data, including the correlation of channel features under different user topologies and channel gains. At the same time, combined with the constraints of conditional feature vectors, it learns the conditional generation capability to complete missing channel data based on pilot observation information. After training, the network parameters are fixed, enabling the channel estimator to quickly reconstruct the complete channel based on sparse pilot observations and pilot masks during the deployment phase.

[0045] S105. The channel estimation device calculates the normalized mean square error between the channel estimation result and the real channel, constructs a reward function by combining the normalized mean square error and the pilot usage, and updates the scheduling strategy of the pilot scheduling strategy network based on the reward function.

[0046] The reward function is constructed by weighting the normalized mean square error and the amount of pilot usage with a negative correlation.

[0047] Specifically, the channel estimation device can compare the channel estimation result output by the channel estimator with the actual channel matrix element by element. By calculating the sum of squares of the differences between the two and dividing it by the sum of squares of the elements in the actual channel matrix, the normalized mean square error (NMSE) of the quantization estimation accuracy is obtained. This index directly reflects the degree of deviation in the channel estimation; the smaller the value, the higher the estimation accuracy. At the same time, the number of "1"s in the pilot activation mask is counted to determine the number of user terminals that are actively transmitting pilots in the current time slot. This is used as a quantitative indicator of pilot usage to measure the degree of consumption of pilot resources.

[0048] Then, based on the preset accuracy importance coefficient and resource sensitivity coefficient ( , (All numbers are positive) Construct a reward function By using a weighted negative correlation method, the estimation accuracy is bound to the pilot overhead. The larger the NMSE and the more pilots are used, the smaller the reward value, and vice versa, thus achieving a positive incentive for the "high accuracy, low overhead" scheduling strategy.

[0049] Subsequently, the channel estimation device constructs a "state-action-reward" triple from the current communication time slot's state information, the output pilot activation mask, and the calculated reward value. The data is stored in the experience replay buffer. When the amount of data in the buffer reaches a preset threshold, triplet samples are randomly selected in batches and input into the pilot scheduling policy network. The gradient of the reward value with respect to the network parameters is calculated using the policy gradient method. The network weights are iteratively optimized through backpropagation. At the same time, the target network is used to stabilize the training process and avoid training oscillations.

[0050] During training, the parameters of the generated diffusion model can be frozen periodically, focusing only on updating the pilot scheduling policy network. This ensures that each round of parameter adjustment optimizes the long-term cumulative reward, enabling the policy network to gradually learn the optimal pilot scheduling policy that adapts to different communication environments, thus achieving a dynamic balance between estimation accuracy and pilot resource consumption.

[0051] like Figure 2 The diagram shows the complete channel estimation workflow. Some user terminals transmit uplink pilot signals, while others do not. Multiple access points (base stations) receive these pilot signals and transmit the mixed signal, superimposed with interference and noise, to the Central Processing Unit (CPU). The CPU then inputs the mixed signal into the "Generative Diffusion Channel Estimation" module. This module, based on the Generative Diffusion Model (GDM), uses the received sparse pilot signals to iteratively fill in the missing channel information through back-diffusion, outputting the channel estimation result, where "Y" represents the received signal and "H" represents the estimated channel matrix. Simultaneously, the "Reinforcement Learning Pilot Allocation Optimization" module, based on the channel estimation results, learns the rationality of the current pilot allocation and generates a better downlink pilot scheduling strategy, which is then fed back to the user terminals. Subsequent time slots will dynamically select which terminals transmit pilots based on this optimization strategy, achieving the goal of obtaining higher channel estimation accuracy with fewer pilot resources.

[0052] In this embodiment, since the pilot scheduling strategy network receives the current communication environment status information, generates a pilot activation mask, and constructs a sparse pilot matrix, it can dynamically adjust the pilot transmission strategy based on the real-time environment status, avoiding the problem of fixed and rigid pilot configuration in the prior art, and realizing adaptive optimization allocation of pilot resources. A channel estimator based on a generative diffusion model is used, taking the pilot observation matrix and pilot activation mask as input, and recovering the true channel information through back-diffusion iteration. Because the diffusion model has strong sparse completion capability, even in extremely sparse pilot scenarios where the pilot length is much smaller than the number of users, it can accurately recover high-dimensional channel information, overcoming the performance limitations of traditional linear estimation, compressed sensing, and other methods. Since a reward function is constructed by combining the normalized mean square error and pilot usage, and used to update the scheduling strategy of the pilot scheduling strategy network, forming a closed-loop optimization mechanism, it can dynamically balance channel estimation accuracy and pilot resource overhead, solving the pain point that the prior art cannot balance both, and significantly improving the communication efficiency and robustness of non-cellular large-scale multi-input multi-output architectures in complex scenarios such as disaster emergency communication and industrial IoT.

[0053] The foregoing mainly describes the solutions provided by the embodiments of this application from a methodological perspective. To achieve the above functions, it includes corresponding hardware structures and / or software modules for executing each function. Those skilled in the art should readily recognize that, in conjunction with the units and algorithm steps of the various examples described in the embodiments disclosed herein, the embodiments of this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0054] The channel estimation method provided in this application can be executed by a channel estimation device or a control module for channel estimation within that device. This application uses the example of a channel estimation device executing the channel estimation method to illustrate the channel estimation device provided in this application.

[0055] It should be noted that the embodiments of this application can divide the channel estimation device into functional modules according to the above method examples. For example, each function can be divided into its own functional modules, or two or more functions can be integrated into one processing module. The integrated modules can be implemented in hardware or as software functional modules. Optionally, the module division in the embodiments of this application is illustrative and only represents one logical functional division; other division methods may be used in actual implementation.

[0056] like Figure 3 As shown in the figure, this application provides a channel estimation device 300. The channel estimation device 300 includes: an information acquisition module 301, a pilot scheduling decision module 302, a pilot signal interaction module 303, a spread channel estimation module 304, and a strategy optimization feedback module 305; The information acquisition module 301 is used to collect the status information of the current communication environment; The pilot scheduling decision module 302 is used to input the status information into the pilot scheduling strategy network to obtain a pilot activation mask, and the pilot activation mask is used to indicate whether the user terminal sends a pilot. The pilot signal interaction module 303 is used to construct a sparse pilot matrix based on the pilot activation mask, and transmit pilot signals according to the sparse pilot matrix to generate a pilot observation matrix. The diffusion channel estimation module 304 is used to take the pilot observation matrix and pilot activation mask as input to the channel estimator, and obtain the channel estimation result by iteratively filling in the missing channel information through back diffusion. The channel estimator is an estimator built based on the generative diffusion model. The strategy optimization feedback module 305 is used to calculate the normalized mean square error between the channel estimation result and the actual channel, construct a reward function by combining the normalized mean square error and the pilot usage, and update the scheduling strategy of the pilot scheduling strategy network based on the reward function.

[0057] Optionally, the status information includes the spatial distribution of user terminals and access points, average channel gain, channel signal-to-noise ratio, historical pilot configuration, and historical channel estimation error.

[0058] Optionally, the pilot scheduling strategy network inputs the state information to obtain the pilot activation mask, including: extracting decision features from the state information; inputting the decision features into the pilot scheduling strategy network, and outputting a binary sequence with a length consistent with the number of user terminals as the pilot activation mask, wherein the pilot activation mask follows the constraint of the maximum number of user terminals that are active in sending pilots in the current time slot.

[0059] Optionally, the step of using the pilot observation matrix and pilot activation mask as input to the channel estimator, and completing the missing channel information through backdiffusion iteration to obtain the channel estimation result, includes: converting the complex channel data format of the pilot observation matrix and pilot activation mask into a real tensor format through a feature coding network to generate a conditional feature vector adapted to the channel estimator; initializing Gaussian distributed samples, initiating backdiffusion iteration, and predicting the noise component of each noisy sample based on the conditional feature vector through a parameterized neural network; using a predictor-corrector sampling method to gradually reconstruct the channel matrix according to the noise component; completing a preset number of iterations, and outputting the complete channel estimation result.

[0060] Optionally, before using the pilot observation matrix and pilot activation mask as input to the channel estimator and completing the missing channel information through back-diffusion iteration to obtain the channel estimation result, the method further includes: using the real complete channel matrix as a sample, constructing a diffusion trajectory by gradually adding noise, so that the channel estimator learns the global distribution characteristics of the channel and the ability to generate conditions.

[0061] Optionally, the reward function is constructed by weighting the normalized mean square error and the pilot usage in a negative correlation.

[0062] In this embodiment, since the pilot scheduling strategy network receives the current communication environment status information, generates a pilot activation mask, and constructs a sparse pilot matrix, it can dynamically adjust the pilot transmission strategy based on the real-time environment status, avoiding the problem of fixed and rigid pilot configuration in the prior art, and realizing adaptive optimization allocation of pilot resources. A channel estimator based on a generative diffusion model is used, taking the pilot observation matrix and pilot activation mask as input, and recovering the true channel information through back-diffusion iteration. Because the diffusion model has strong sparse completion capability, even in extremely sparse pilot scenarios where the pilot length is much smaller than the number of users, it can accurately recover high-dimensional channel information, overcoming the performance limitations of traditional linear estimation, compressed sensing, and other methods. Since a reward function is constructed by combining the normalized mean square error and pilot usage, and used to update the scheduling strategy of the pilot scheduling strategy network, forming a closed-loop optimization mechanism, it can dynamically balance channel estimation accuracy and pilot resource overhead, solving the pain point that the prior art cannot balance both, and significantly improving the communication efficiency and robustness of non-cellular large-scale multi-input multi-output architectures in complex scenarios such as disaster emergency communication and industrial IoT.

[0063] Figure 4 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 4 As shown, the electronic device may include: a processor 410, a communications interface 420, a memory 430, and a communications bus 440, wherein the processor 410, the communications interface 420, and the memory 430 communicate with each other through the communications bus 440. The processor 410 can call logic instructions in the memory 430 to execute a channel estimation method, which includes: collecting state information of the current communication environment; inputting the state information into the pilot scheduling policy network to obtain a pilot activation mask, the pilot activation mask being used to indicate whether the user terminal sends a pilot; constructing a sparse pilot matrix based on the pilot activation mask, and transmitting pilot signals according to the sparse pilot matrix to generate a pilot observation matrix; using the pilot observation matrix and the pilot activation mask as inputs to a channel estimator, and completing the missing channel information through back-diffusion iteration to obtain a channel estimation result, the channel estimator being an estimator constructed based on a generative diffusion model; calculating the normalized mean square error between the channel estimation result and the real channel, constructing a reward function by combining the normalized mean square error and the pilot usage, and updating the scheduling policy of the pilot scheduling policy network based on the reward function.

[0064] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0065] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the channel estimation method provided by the above methods. The method includes: collecting state information of the current communication environment; inputting the state information into a pilot scheduling policy network to obtain a pilot activation mask, the pilot activation mask being used to indicate whether a user terminal should send a pilot; constructing a sparse pilot matrix based on the pilot activation mask, and transmitting pilot signals according to the sparse pilot matrix to generate a pilot observation matrix; using the pilot observation matrix and the pilot activation mask as inputs to a channel estimator, and completing the missing channel information through back-diffusion iteration to obtain a channel estimation result, the channel estimator being an estimator constructed based on a generative diffusion model; calculating the normalized mean square error between the channel estimation result and the real channel, constructing a reward function by combining the normalized mean square error and the pilot usage, and updating the scheduling policy of the pilot scheduling policy network based on the reward function.

[0066] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the channel estimation method provided by the above methods. The method includes: collecting state information of the current communication environment; inputting the state information into a pilot scheduling policy network to obtain a pilot activation mask, the pilot activation mask being used to indicate whether a user terminal should send a pilot; constructing a sparse pilot matrix based on the pilot activation mask, and transmitting pilot signals according to the sparse pilot matrix to generate a pilot observation matrix; using the pilot observation matrix and the pilot activation mask as inputs to a channel estimator, and completing the missing channel information through back-diffusion iteration to obtain a channel estimation result, the channel estimator being an estimator constructed based on a generative diffusion model; calculating the normalized mean square error between the channel estimation result and the actual channel, constructing a reward function by combining the normalized mean square error and pilot usage, and updating the scheduling policy of the pilot scheduling policy network based on the reward function.

[0067] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0068] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A channel estimation method, characterized in that, include: Collect current communication environment status information; The pilot scheduling strategy network inputs the state information to obtain the pilot activation mask, which is used to indicate whether the user terminal should send a pilot. A sparse pilot matrix is ​​constructed based on the pilot activation mask, and pilot signals are transmitted according to the sparse pilot matrix to generate a pilot observation matrix. The pilot observation matrix and pilot activation mask are used as inputs to the channel estimator. Missing channel information is filled in through back-diffusion iteration to obtain the channel estimation result. The channel estimator is an estimator built based on the generative diffusion model. The normalized mean square error between the channel estimation result and the actual channel is calculated. A reward function is constructed by combining the normalized mean square error and the pilot usage. The scheduling strategy of the pilot scheduling strategy network is updated based on the reward function.

2. The channel estimation method according to claim 1, characterized in that, The status information includes the spatial distribution of user terminals and access points, average channel gain, channel signal-to-noise ratio, historical pilot configuration, and historical channel estimation error.

3. The channel estimation method according to claim 1, characterized in that, The pilot scheduling policy network inputs the state information to obtain the pilot activation mask, including: Extract decision features from the state information; The decision features are input into the pilot scheduling policy network, and the output binary sequence with the same length as the number of user terminals is used as the pilot activation mask. The pilot activation mask follows the constraint of the maximum number of user terminals that can activate and send pilots in the current time slot.

4. The channel estimation method according to claim 1, characterized in that, The process of using the pilot observation matrix and pilot activation mask as input to the channel estimator, and then iteratively completing the missing channel information through back-diffusion to obtain the channel estimation result includes: The complex channel data format of the pilot observation matrix and the pilot activation mask is converted into a real tensor format through a feature coding network to generate a conditional feature vector that fits the channel estimator. Initialize Gaussian distributed samples, initiate back diffusion iteration, and predict the noise component of each noisy sample based on the conditional feature vector using a parameterized neural network. The predictor-corrector sampling method is used to gradually reconstruct the channel matrix based on the noise components; After completing the preset number of iterations, output the complete channel estimation results.

5. The channel estimation method according to claim 4, characterized in that, Before using the pilot observation matrix and pilot activation mask as input to the channel estimator, and iteratively filling in the missing channel information through back-diffusion to obtain the channel estimation result, the method further includes: Using a real, complete channel matrix as a sample, a diffusion trajectory is constructed by gradually adding noise, enabling the channel estimator to learn the global distribution characteristics of the channel and its condition generation capability.

6. The channel estimation method according to claim 1, characterized in that, The reward function is constructed by weighting the normalized mean square error and the amount of pilot usage with a negative correlation.

7. A channel estimation device, characterized in that, include: The system includes an information acquisition module, a pilot scheduling decision module, a pilot signal interaction module, a spread channel estimation module, and a strategy optimization feedback module. The information acquisition module is used to collect status information of the current communication environment; The pilot scheduling decision module is used to input the status information into the pilot scheduling strategy network to obtain a pilot activation mask, and the pilot activation mask is used to indicate whether the user terminal should send a pilot. The pilot signal interaction module is used to construct a sparse pilot matrix based on the pilot activation mask, and transmit pilot signals according to the sparse pilot matrix to generate a pilot observation matrix. The diffusion channel estimation module is used to take the pilot observation matrix and pilot activation mask as input to the channel estimator, and obtain the channel estimation result by iteratively filling in the missing channel information through back diffusion. The channel estimator is an estimator built based on the generative diffusion model. The strategy optimization feedback module is used to calculate the normalized mean square error between the channel estimation result and the actual channel, construct a reward function by combining the normalized mean square error and the pilot usage, and update the scheduling strategy of the pilot scheduling strategy network based on the reward function.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the channel estimation method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the channel estimation method as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the channel estimation method as described in any one of claims 1 to 6.

Images