User scheduling method and system

Abstract

The present invention provides a user scheduling method and system, comprising: obtaining channel state data between a base station and a user terminal to be scheduled; inputting the channel state data into a trained user selection model to obtain a user candidate set that satisfies maximum channel gain; wherein the trained user selection model is obtained by training a neural network using sample channel state data labeled with beam strength between the user terminal and the base station, and the trained user selection model includes a first neural network and a second neural network; and sorting spatially multiplexed users in the user candidate set to obtain a user scheduling result. The present invention introduces multi-task learning, fusing the tasks corresponding to the two neural networks into an overall user selection model, thereby reducing network parameters, lowering computational complexity, and improving user scheduling performance.

Description

Technical Field

[0001] The present invention relates to the technical field of wireless communications, and in particular to a user scheduling method and system. Background Art

[0002] 5G communication systems are built on the 4G technology architecture and are an evolution of existing communication technologies. They require significantly faster transmission rates, hundreds or even thousands of times faster than 4G. To achieve these higher transmission rates, millimeter wave communication technology has become a highly anticipated research area within 5G.

[0003] In millimeter-wave massive multiple-input, multiple-output (MIMO) systems, the complexity of fully digital precoding increases dramatically with the number of antennas, making it impractical in practical communication systems. Furthermore, configuring a radio frequency link for each antenna incurs significant hardware costs and power consumption. Due to the ultra-large bandwidth and multipath characteristics of millimeter-wave channels, they are frequency-selective, necessitating the use of orthogonal frequency division multiple access (OFDMA). Consequently, user scheduling is no longer simply selecting a subset of users to be served in the time-frequency domain; it also considers beam pair selection in the analog domain.

[0004] Existing user scheduling schemes mainly include the following: 1. A scheduling scheme based on spatial-time division multiple access (STDMA): This scheme first introduces an adjustment factor to achieve a trade-off between user fairness and throughput, and then designs a low-complexity link scheduling scheme based on this. 2. A two-stage user scheduling scheme that takes into account both inter-user channel correlation and channel energy: This scheme ranks users based on channel orthogonality, and then selects users and beams within a narrow range to maximize channel energy. 3. A joint optimization of simulated beam selection and user scheduling based on limited effective channel state information: The joint simulated beam selection and user scheduling (JBSUS) problem is formulated as a non-convex and combinatorial optimization problem, and a low-complexity joint optimization method based on a DC process and a greedy algorithm is proposed to solve this problem. 4. A Lyapunov drift optimization framework: This scheme develops a solution for joint user scheduling and beam selection, and obtains the optimal scheduling strategy in a closed form. However, current user scheduling solutions are highly complex and suffer from poor performance. Therefore, there is an urgent need for a user scheduling method and system to address these issues. Summary of the Invention

[0005] The present invention provides a user scheduling method and system to solve the technical problems of high complexity and poor performance of existing user scheduling solutions.

[0006] In a first aspect, the present invention provides a user scheduling method, comprising:

[0007] Acquire channel state data between the base station and the user terminal to be scheduled;

[0008] Inputting the channel state data into a trained user selection model to obtain a user candidate set that satisfies a maximum channel gain; wherein the trained user selection model is obtained by training a neural network using sample channel state data labeled with a beam strength label between a user terminal and a base station, and the trained user selection model includes a first neural network and a second neural network;

[0009] Sort the spatial multiplexing users in the user candidate set to obtain a user scheduling result.

[0010] In one embodiment, the trained user selection model is obtained by training through the following steps:

[0011] Constructing a first training sample set based on sample channel state data labeled with a beam strength label between the user terminal and the base station;

[0012] Inputting the first training sample set into a first neural network for training to obtain a virtual user channel and a pre-trained first neural network model, wherein the virtual user channel is a channel corresponding to a user group using the same beam;

[0013] Constructing a second training sample set according to the virtual user channel;

[0014] Inputting the second training sample set into a second neural network for training to obtain a pre-trained second neural network model;

[0015] A trained user selection model is obtained based on the pre-trained first neural network and the pre-trained second neural network.

[0016] In one embodiment, obtaining a trained user selection model based on the pre-trained first neural network and the pre-trained second neural network includes:

[0017] Obtaining a pretrained user selection model based on the pretrained first neural network and the pretrained second neural network;

[0018] The pre-trained user selection model is used to predict candidate user terminals; the beam intensity labels corresponding to the predicted candidate user terminals are compared with the beam intensity labels of the candidate user terminals in the actual training set, and backpropagation is performed based on the comparison results to fine-tune the pre-trained user selection model to obtain a trained user selection model.

[0019] In one embodiment, sorting the spatially multiplexed users in the user candidate set to obtain a user scheduling result includes:

[0020] S1, arranging the spatially multiplexed users in the user candidate set in descending order according to the channel gains of the user terminals in the user candidate set, and constructing an index set;

[0021] S2, constructing a user index set according to the maximum intensity beam index of each spatially multiplexed user in the index set;

[0022] S3, determining a frequency resource allocation prediction value of a spatially multiplexed user of a maximum intensity beam according to the user index set;

[0023] S4, arranging the classification probability value of each resource block in the frequency resource allocation prediction value in descending order;

[0024] S5, repeat S2 to S4 until all user terminals are allocated corresponding frequency resources, and then determine the user scheduling result.

[0025] In one embodiment, after sorting the users in the user candidate set and obtaining the user scheduling result, the method further includes:

[0026] Perform inverse fast Fourier transform on the precoding codebook and obtain simulated precoding based on the user scheduling result;

[0027] Acquire digital precoding according to the analog precoding and the user scheduling result;

[0028] According to the analog precoding and the digital precoding, user scheduling is performed on uplink and downlink channels.

[0029] In one embodiment, performing user scheduling on uplink and downlink channels according to the analog precoding and the digital precoding includes:

[0030] Acquire a total system rate according to the analog precoding and the digital precoding;

[0031] Performing user scheduling on uplink and downlink channels according to a maximum signal-to-noise ratio criterion and the total system rate;

[0032] The system rate sum calculation formula includes:

[0033]

[0034]

[0035] Among them, R k,n R represents the achievable rate of the nth resource block for the kth user.sum Indicates the total system rate, N RB Indicates the number of resource blocks, N RF Indicates the number of RF links, H k,n Indicates dimension N r ×N t mmWave massive MIMO channel, s k,n Indicates dimension N s ×1 transmitted signal vector, Indicates dimension N RF ×N s The digital precoder, Indicates dimension N t ×N RF The analog precoder, N t Indicates the number of transmitting antennas, N s represents the number of data stream communications between the base station and the user terminal, σ 2 represents variance, RF represents the analog domain, BB represents the digital domain, and j represents the jth data stream.

[0036] In a second aspect, the present invention provides a user scheduling system, comprising:

[0037] A channel state data acquisition module is used to acquire channel state data between the base station and the user terminal to be scheduled;

[0038] a user candidate set acquisition module, configured to input the channel state data into a trained user selection model to obtain a user candidate set that satisfies a maximum channel gain; wherein the trained user selection model is obtained by training a neural network using sample channel state data labeled with a beam strength label between a user terminal and a base station, and the trained user selection model includes a first neural network and a second neural network;

[0039] The user scheduling module is used to sort the spatial multiplexing users in the user candidate set and obtain user scheduling results.

[0040] In one embodiment, the system further comprises:

[0041] A first training sample set construction module is used to construct a first training sample set based on sample channel state data marked with a beam strength label between the user terminal and the base station;

[0042] a virtual user channel acquisition module, configured to input the first training sample set into a first neural network for training, and acquire a virtual user channel and a pre-trained first neural network model, wherein the virtual user channel is a channel corresponding to a user group using the same beam;

[0043] A second training sample set construction module, configured to construct a second training sample set according to the virtual user channel;

[0044] A second neural network model pre-training module, configured to input the second training sample set into a second neural network for training, and obtain a pre-trained second neural network model;

[0045] The user selection model training module is used to obtain a trained user selection model based on the pre-trained first neural network and the pre-trained second neural network.

[0046] In a third aspect, the present invention provides an electronic device comprising a memory and a memory storing a computer program, wherein the processor implements the steps of the user scheduling method described in the first aspect when executing the program.

[0047] In a fourth aspect, the present invention provides a processor-readable storage medium, wherein the processor-readable storage medium stores a computer program, and the computer program is used to enable the processor to execute the steps of the user scheduling method described in the first aspect.

[0048] The user scheduling method and system provided by the present invention train a first neural network and a second neural network using sample channel state data labeled with a beam strength label between a user terminal and a base station terminal to obtain a trained user selection model. By introducing multiple tasks, the tasks corresponding to the two neural networks are integrated into an overall user selection model, thereby reducing network parameters and computational complexity. The channel state data between the base station terminal and the user terminal to be scheduled is input into the trained user selection model to obtain a user candidate set that meets the maximum channel gain, and the spatially multiplexed users in the user candidate set are sorted to obtain a user scheduling result, thereby improving the performance of user scheduling. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0050] Figure 1 A flow chart of the user scheduling method provided by the present invention;

[0051] Figure 2 A schematic diagram of the structure of the user scheduling system provided by the present invention;

[0052] Figure 3 This is a structural diagram of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0053] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0054] In the millimeter wave MU-MIMO system, the base station is configured with N t transmitting antennas and passing N RF RF links communicate with K MU-MIMO spatial multiplexing users, where K ≤ N RF In order to serve as many users as possible, K can be set to a maximum value N RF In addition, the MU-MIMO spatial multiplexing users served are from N all Selected from users, each user is configured with N r In order to achieve N antennas between the base station and the user end s For data stream communication, the number of transmitting antennas and RF chains should satisfy N s ≤N RF ≤N t In addition, the millimeter wave MU-MIMO system uses resource blocks for OFDMA transmission. The entire frequency band contains N RB Resource blocks (RBs), each RB occupies N f adjacent subcarriers and N o consecutive OFDM symbols.

[0055] Figure 1 A flow chart of the user scheduling method provided by the present invention is shown as follows: Figure 1 As shown, the present invention provides a user scheduling method, comprising:

[0056] Step 101: Acquire channel state data between a base station and a user terminal to be scheduled.

[0057] In the present invention, the channel state data between the base station and the user terminal to be scheduled is obtained, wherein the channel state data is the channel attribute of the communication link, which describes the attenuation factor of the signal on each transmission path, including signal scattering, environmental attenuation and distance attenuation.

[0058] Step 102: Input the channel state data into a trained user selection model to obtain a user candidate set that satisfies the maximum channel gain; wherein the trained user selection model is obtained by training a neural network using sample channel state data labeled with a beam strength label between the user terminal and the base station, and the trained user selection model includes a first neural network and a second neural network.

[0059] In the present invention, the user selection model is formed by fusing a first neural network and a second neural network, wherein the first neural network is an RB network, which represents a resource block network; and the second neural network is an RF network, which represents a radio frequency link network.

[0060] Furthermore, the beam strength label between the user terminal and the base station is the label of the user selecting the strongest beam. The sample channel state data marked with the label of the user selecting the strongest beam is input into the fused neural network for training to obtain a trained user selection model.

[0061] Furthermore, the channel state data is input into the trained user selection model to obtain a user candidate set that satisfies the maximum channel gain. The user that satisfies the maximum channel gain must meet the maximum SNR signal-to-noise ratio condition, which is:

[0062]

[0063] Among them, n k represents the beam index of the i-th candidate user, σ 2 represents the variance, represents the channel corresponding to the i-th candidate user beam.

[0064] Step 103: sort the spatial multiplexing users in the user candidate set to obtain a user scheduling result.

[0065] In the present invention, spatially multiplexed users in a user candidate set are sorted and matched to the results of users selecting the strongest beam, so as to perform user scheduling according to the corresponding beam selected by the user.

[0066] The user scheduling method provided by the present invention trains a first neural network and a second neural network using sample channel state data labeled with a beam strength label between a user terminal and a base station terminal to obtain a trained user selection model. By introducing multiple tasks, the tasks corresponding to the two neural networks are integrated into an overall user selection model, thereby reducing network parameters and computational complexity. The channel state data between the base station terminal and the user terminal to be scheduled is input into the trained user selection model to obtain a user candidate set that meets the maximum channel gain, and the spatially multiplexed users in the user candidate set are sorted to obtain a user scheduling result, thereby improving the performance of user scheduling.

[0067] Based on the above embodiment, the trained user selection model is obtained by training through the following steps:

[0068] Constructing a first training sample set based on sample channel state data labeled with a beam strength label between the user terminal and the base station;

[0069] Inputting the first training sample set into a first neural network for training to obtain a virtual user channel and a pre-trained first neural network model, wherein the virtual user channel is a channel corresponding to a user group using the same beam;

[0070] Constructing a second training sample set according to the virtual user channel;

[0071] Inputting the second training sample set into a second neural network for training to obtain a pre-trained second neural network model;

[0072] A trained user selection model is obtained based on the pre-trained first neural network and the pre-trained second neural network.

[0073] In the present invention, the first neural network is the RB network and the second neural network is the RF network. Assuming that the perfect channel state information of the user terminal is known, under this assumption, the input data x of the RB network can be obtained by real-numbering the channel state information. RB .

[0074] Specifically, each user terminal searches for the codeword with the highest correlation with its strongest beam and feeds its index back to the base station. The base station then establishes a K-dimensional all ×N c Matrix To represent the beam selection of the user terminal, if Then the maximum intensity beam of the i-th user terminal is f j ,like This means that the user terminal has not selected the maximum strength beam. In addition, all users using the same beam form an OFDMA group.

[0075] It should be noted that in order to ensure that the sum rate of frequency multiplexing users using the same beam is maximized, the maximum SNR criterion should be adopted for frequency domain scheduling. c On the nth resource block in an OFDMA group, user k must meet the maximum SNR signal-to-noise ratio condition:

[0076]

[0077] Furthermore, n c represents the OFDMA group number, k represents the user, σ2 represents the variance, H k represents the channel of user k.

[0078] Furthermore, based on the sample channel state data labeled with the beam strength label between the user terminal and the base station, a first training sample set is constructed and used to input the RB network and train the RB network to obtain a pre-trained RB neural network model. As the label data required for training the RB network. After completing the frequency resource allocation, each OFDMA group is regarded as a virtual user, and its channel is merged from the channels of the group members. The channel reflects the spatial characteristics of the users on each resource. c The channel of a virtual user can be expressed as:

[0079]

[0080] in, Indicates the nth c channels for virtual users, Indicates the nth c The channels of the first link in the resource blocks, Indicates the nth c The second link channel in the resource blocks, Indicates the nth c The Nth resource block RB Channel of the link.

[0081] Furthermore, a second training sample set is constructed based on the virtual user channel, that is, the virtual user channel is processed into real numbers to obtain the input data of the second neural network. The input data x RF Input into the RF neural network for training to obtain a pre-trained RF neural network model.

[0082] in, represents the user channel matrix of the first antenna after real number processing, represents the user channel matrix of the first antenna after real number processing, Indicates the Nth t The user channel matrix after real number processing of the root antenna, Indicates the Nth t The user channel matrix after real number processing of root antennas, Nt represents the number of antennas, n c Indicates the maximum intensity beam index of the virtual user, l indicates the number of virtual users, and satisfies l≤N c All virtual users constitute the spatial multiplexing user candidate set

[0083] Furthermore, in order to reduce the computational complexity of the selected spatial multiplexing users, it is necessary to select the user with the maximum channel gain from the candidate set Ω, that is, to meet the maximum SNR condition:

[0084]

[0085] Among them, n k represents the beam index of the i-th candidate user. Let As label data for training RF network, Indicates the nth k The virtual user corresponding to each beam is selected as one of the spatial multiplexing users, and the remaining spatial multiplexing users are found according to the maximum system rate criterion.

[0086] Furthermore, the pre-trained RB neural network model and the pre-trained RF neural network model are fused into one network model, that is, a trained user selection model is obtained.

[0087] This paper implements an underlying sharing mechanism by sharing network parameters. This mechanism allows for feature sharing between different tasks, thereby mutually promoting learning outcomes and improving system performance. Simultaneously, the two neural networks are fused into a single user-selected model, reducing the number of network parameters and computational complexity.

[0088] On the basis of the above embodiment, the method of obtaining a trained user selection model based on the pre-trained first neural network and the pre-trained second neural network includes:

[0089] Obtaining a pretrained user selection model based on the pretrained first neural network and the pretrained second neural network;

[0090] The pre-trained user selection model is used to predict candidate user terminals; the beam intensity labels corresponding to the predicted candidate user terminals are compared with the beam intensity labels of the candidate user terminals in the actual training set, and backpropagation is performed based on the comparison results to fine-tune the pre-trained user selection model to obtain a trained user selection model.

[0091] In the present invention, the RB neural network is fused with the RF neural network to obtain a pre-trained user selection model. The candidate user terminal is predicted by the pre-trained user selection model, and the beam intensity label corresponding to the predicted candidate user terminal is compared with the beam intensity label of the candidate user terminal in the actual training set, and the prediction accuracy is calculated, wherein the prediction accuracy can be obtained by dividing the number of correctly predicted labels by the total number of labels in the training set. According to the prediction accuracy result, the neural network parameters are reversely adjusted and updated through the back propagation algorithm and the RMSProp (Root Mean Square Prop) optimization algorithm to obtain the loss function value of the user selection model until it is determined that the loss function value meets the convergence condition, thereby obtaining a trained user selection model.

[0092] In one embodiment, since the dataset used for neural network training is large, directly using the entire dataset for iterative training will severely reduce learning efficiency. Therefore, a mini-batch mechanism can be used for training, and dropout can be used to further improve learning efficiency while preventing overfitting. The multi-task deep neural network training process is as follows: constructing the model structure and initializing the model parameters, with the initial learning rate set to 0.001, the dropout rate set to 0.5, the number of iterations set to 100, and the batch size set to 200. Each set of training data is divided into a training set and a validation set, and both the training set and validation set are input into the constructed network model for training. After training each set of training data, the classification category of the predicted data is determined, and then the classification category of the predicted data is compared with the true labels of the training set and the accuracy is calculated. The prediction accuracy is calculated by dividing the number of correctly predicted labels by the total number of labels in the training set. After completing a mini-batch iteration, the network parameters are adjusted and updated inversely using the backpropagation algorithm and the RMSProp optimization algorithm. The user selection model is repeatedly trained according to the above method until the training loss and prediction loss converge. When the training loss and test loss reach a minimum and converge, the user selection model training is completed. The current user selection model is then saved. Finally, the saved user selection model is tested using the validation dataset.

[0093] Furthermore, the saved user selection model is tested with a validation dataset. The user scheduling result can be obtained by processing the predicted data. Then, a hybrid precoding is designed, and the system and rate performance as well as the computational complexity performance are statistically analyzed and compared with the traditional user scheduling algorithm.

[0094] Based on the above embodiment, sorting the spatially multiplexed users in the user candidate set to obtain the user scheduling result includes:

[0095] S1, arranging the spatially multiplexed users in the user candidate set in descending order according to the channel gains of the user terminals in the user candidate set, and constructing an index set;

[0096] S2, constructing a user index set according to the maximum intensity beam index of each spatially multiplexed user in the index set;

[0097] S3, determining a frequency resource allocation prediction value of a spatially multiplexed user of a maximum intensity beam according to the user index set;

[0098] S4, arranging the classification probability value of each resource block in the frequency resource allocation prediction value in descending order;

[0099] S5, repeat S2 to S4 until all user terminals are allocated corresponding frequency resources, and then determine the user scheduling result.

[0100] In the present invention, the predicted output values of the spatial multiplexing user selection task are arranged in descending order by a neural network, and the predicted output values can be the channel gains of the user terminals in the user candidate set. RF The virtual users corresponding to the codewords will be used as spatial multiplexing users, and their indexes will be constructed as sets

[0101] Furthermore, according to the maximum intensity beam index of each spatial multiplexing user in the index set, the maximum intensity beam is determined to be All users and build a user index collection It can be understood that the beam with the maximum intensity is the selected optimal beam.

[0102] Furthermore, based on the user index set Determine the maximum intensity beam as Frequency resource allocation prediction value of spatial multiplexing users

[0103] Furthermore, the classification probability values of the spatial multiplexing users in the index set in each resource block are sorted in descending order, that is, The average of each column is calculated, and users with large probability values will use the corresponding frequency resources.

[0104] Furthermore, the user index set is continuously constructed to determine the frequency resource allocation prediction value of the spatial multiplexing user with the maximum intensity beam, and the classification probability value of each resource block in the frequency resource allocation prediction value is arranged in descending order until the corresponding frequency resources are finally allocated to all user terminals, thereby obtaining the final user scheduling result.

[0105] Based on the above embodiment, after sorting the users in the user candidate set and obtaining the user scheduling result, the method further includes:

[0106] Perform inverse fast Fourier transform on the precoding codebook and obtain simulated precoding based on the user scheduling result;

[0107] Acquire digital precoding according to the analog precoding and the user scheduling result;

[0108] According to the analog precoding and the digital precoding, user scheduling is performed on uplink and downlink channels.

[0109] In the present invention, the predetermined RF precoding codebook can be expressed as:

[0110]

[0111] Among them, N c The kth beamforming vector f k is through its time domain form A k A is obtained by inverse fast Fourier transform. k It can be expressed as:

[0112]

[0113] Where N represents the number of antennas, φ θ Indicates the azimuth quantization accuracy, φ θ It can be set to 15. λ represents the wavelength and d represents the distance between antenna units. Based on the prediction results of the multi-task neural network spatial multiplexing user selection task, in the predetermined RF precoding codebook Parallel selection N RF The codewords form an analog precoding matrix, which can be expressed as:

[0114]

[0115] in, represents the l1th beamforming vector, Indicates the beamforming vectors.

[0116] Furthermore, digital precoding is derived based on the user scheduling results and analog precoding. Digital precoding can be designed using traditional linear precoding. First, the analog equivalent channel of spatially multiplexed users using different beams and multiplexing the same resource block is calculated. The analog equivalent channel of the kth spatially multiplexed user in the nth resource block can be expressed as:

[0117]

[0118] The ZF precoding algorithm is used to design the digital precoding F corresponding to each spatial multiplexing user. BB , which can be expressed as:

[0119]

[0120] in, represents the simulated equivalent channel matrix of all users that reuse the nth resource block. If W is expressed as Then the digital precoding of the kth spatial multiplexing user on the resource block can be expressed as F BB =W k,n .

[0121] Furthermore, based on the acquired analog precoding parameters and digital precoding parameters, mixed precoding is performed on multiple scheduling signals in the user scheduling result to perform user scheduling on uplink and downlink channels.

[0122] Based on the above embodiment, performing user scheduling on uplink and downlink channels according to the analog precoding and the digital precoding includes:

[0123] Acquire a total system rate according to the analog precoding and the digital precoding;

[0124] According to the maximum signal-to-noise ratio criterion and the total system rate, user scheduling is performed on the uplink and downlink channels.

[0125] In the present invention, when the base station communicates with the user, the data to be sent needs to pass through the digital precoder F in sequence. BB and analog precoder F RF Processing. Analog precoder It is realized by analog phase shifter, which can only realize phase change, satisfying the condition In the present invention, F RF The column element f i From the codebook Selected, where N c Indicates the number of code words. In order to meet the limit of base station transmission power, the analog precoder and digital precoder must meet The received signal Y of the kth multiplexed user in the nth resource block k,n The expression is:

[0126]

[0127] Among them, s k,n Indicates dimension N s ×1 transmitted signal vector, and satisfying Indicates dimension N RF ×N sThe digital precoder, F RF Indicates dimension N t ×N RF The analog precoder, H k,n Indicates dimension N r ×N t mmWave massive MIMO channel, n k,n It means the mean is zero and the variance is σ 2 Independent and identically distributed additive complex Gaussian noise. According to the above expression of the user received signal, the achievable rate can be expressed as:

[0128]

[0129] The total system rate expression is:

[0130]

[0131] Among them, R k,n R represents the achievable rate of the nth resource block for the kth user. sum Indicates the total system rate, N RB Indicates the number of resource blocks, N RF Indicates the number of RF links, H k,n Indicates dimension N r ×N t mmWave massive MIMO channel, s k,n Indicates dimension N s ×1 transmitted signal vector, Indicates dimension N RF ×N s The digital precoder, Indicates dimension N t ×N RF The analog precoder, N t Indicates the number of transmitting antennas, N s represents the number of data stream communications between the base station and the user terminal, σ 2 represents variance, RF represents the analog domain, BB represents the digital domain, and j represents the jth data stream.

[0132] Furthermore, based on the above principle, assuming that the uplink and downlink channels are reciprocal, that is, h 上行 =h T 下行 , (·) T Represents the transpose of the matrix. Frequency domain user scheduling is performed on the uplink and downlink channels based on the maximum SNR signal-to-noise ratio criterion, thereby ensuring that the sum rate of frequency-multiplexed users using the same beam is maximized.

[0133] It should be noted that the present invention provides a user scheduling and hybrid precoding algorithm based on multi-task deep learning. In order to solve the problems of sum rate performance loss and high training time complexity in existing user scheduling schemes, the two classification problems of frequency resource allocation and spatial multiplexing user selection are solved using the same neural network model by introducing multi-task learning. The model adopts a network structure in which the bottom layer shares the upper layer branches, and the bottom layer sharing mechanism is realized by sharing network parameters. This sharing mechanism not only enables feature sharing between different tasks to achieve mutual promotion of learning effects between different tasks and achieve the purpose of improving system performance, but also effectively reduces the scale of neural network parameters.

[0134] Figure 2 A schematic diagram of the structure of the user scheduling system provided by the present invention, such as Figure 2 As shown, the present invention provides a user scheduling system, including a channel state data acquisition module 201, a user candidate set acquisition module 202 and a user scheduling module 203, wherein the channel state data acquisition module 201 is used to obtain channel state data between a base station end and a user terminal to be scheduled; the user candidate set acquisition module 202 is used to input the channel state data into a trained user selection model to obtain a user candidate set that satisfies the maximum channel gain; wherein the trained user selection model is obtained by training a neural network using sample channel state data marked with a beam strength label between the user terminal and the base station end, and the trained user selection model includes a first neural network and a second neural network; the user scheduling module 203 is used to sort the spatially multiplexed users in the user candidate set to obtain a user scheduling result.

[0135] The user scheduling system provided by the present invention trains a first neural network and a second neural network using sample channel state data labeled with a beam strength label between a user terminal and a base station terminal to obtain a trained user selection model. By introducing multiple tasks, the tasks corresponding to the two neural networks are integrated into an overall user selection model, thereby reducing network parameters and computational complexity. The channel state data between the base station terminal and the user terminal to be scheduled is input into the trained user selection model to obtain a user candidate set that meets the maximum channel gain, and the spatially multiplexed users in the user candidate set are sorted to obtain a user scheduling result, thereby improving the performance of user scheduling.

[0136] Based on the above embodiment, the system further includes:

[0137] A first training sample set construction module is used to construct a first training sample set based on sample channel state data marked with a beam strength label between the user terminal and the base station;

[0138] a virtual user channel acquisition module, configured to input the first training sample set into a first neural network for training, and acquire a virtual user channel and a pre-trained first neural network model, wherein the virtual user channel is a channel corresponding to a user group using the same beam;

[0139] A second training sample set construction module, configured to construct a second training sample set according to the virtual user channel;

[0140] A second neural network model pre-training module, configured to input the second training sample set into a second neural network for training, and obtain a pre-trained second neural network model;

[0141] The user selection model training module is used to obtain a trained user selection model based on the pre-trained first neural network and the pre-trained second neural network.

[0142] Based on the above embodiment, the user selection model training module further includes:

[0143] A user selection model pre-training unit, configured to obtain a pre-trained user selection model based on the pre-trained first neural network and the pre-trained second neural network;

[0144] The user selection model training unit is used to use the pre-trained user selection model to predict candidate user terminals; compare the beam intensity labels corresponding to the predicted candidate user terminals with the beam intensity labels of the candidate user terminals in the actual training set, and perform backpropagation based on the comparison results to fine-tune the pre-trained user selection model to obtain a trained user selection model.

[0145] Based on the above embodiment, the user scheduling module further includes:

[0146] an index set construction unit, configured to arrange the spatially multiplexed users in the user candidate set in descending order according to the channel gains of the user terminals in the user candidate set, and construct an index set;

[0147] A user index set construction unit, configured to construct a user index set according to the maximum intensity beam index of each spatial multiplexing user in the index set;

[0148] a resource allocation prediction value determining unit, configured to determine a frequency resource allocation prediction value of a spatially multiplexed user of a maximum intensity beam according to the user index set;

[0149] a descending order arrangement unit, configured to arrange each resource block classification probability value in the frequency resource allocation prediction value in descending order;

[0150] The user scheduling result determination unit is used to continue to construct the user index set, determine the frequency resource allocation prediction value of the spatial multiplexing user of the maximum intensity beam, and arrange the classification probability value of each resource block in the frequency resource allocation prediction value in descending order until all user terminals are allocated the corresponding frequency resources, and then determine the user scheduling result.

[0151] Based on the above embodiment, the system further includes:

[0152] The simulated precoding acquisition module is used to perform inverse fast Fourier transform on the precoding codebook and obtain the simulated precoding according to the user scheduling result;

[0153] A digital precoding acquisition module, configured to acquire digital precoding according to the analog precoding and the user scheduling result;

[0154] The hybrid precoding module is used to perform user scheduling on uplink and downlink channels according to the analog precoding and the digital precoding.

[0155] Based on the above embodiment, the hybrid precoding module further includes:

[0156] a system rate sum obtaining unit, configured to obtain a system rate sum according to the analog precoding and the digital precoding;

[0157] The user scheduling unit is used to perform user scheduling on the uplink and downlink channels according to the maximum signal-to-noise ratio criterion and the total system rate.

[0158] The system provided by the present invention is used to execute the above-mentioned method embodiments. Please refer to the above-mentioned embodiments for the specific processes and detailed contents, which will not be repeated here.

[0159] Figure 3 A schematic diagram of the structure of an electronic device provided by the present invention, such as Figure 3As shown, the electronic device may include: a processor 301, a communication interface 302, a memory 303, and a communication bus 304, wherein the processor 301, the communication interface 302, and the memory 303 communicate with each other via the communication bus 304. The processor 301 may call a computer program in the memory 303 to execute the steps of the user scheduling method, for example, including: obtaining channel state data between a base station and a user terminal to be scheduled; inputting the channel state data into a trained user selection model to obtain a user candidate set that satisfies the maximum channel gain; wherein the trained user selection model is obtained by training a neural network using sample channel state data labeled with beam strength labels between the user terminal and the base station, and the trained user selection model includes a first neural network and a second neural network; and sorting the spatially multiplexed users in the user candidate set to obtain a user scheduling result.

[0160] In addition, the logic instructions in the above-mentioned memory 303 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0161] On the other hand, the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the steps of the user scheduling method provided by the above methods, for example, including: obtaining channel state data between the base station and the user terminal to be scheduled; inputting the channel state data into a trained user selection model to obtain a user candidate set that meets the maximum channel gain; wherein the trained user selection model is obtained by training a neural network with sample channel state data marked with a beam strength label between the user terminal and the base station, and the trained user selection model includes a first neural network and a second neural network; sorting the spatially multiplexed users in the user candidate set to obtain a user scheduling result.

[0162] On the other hand, an embodiment of the present application also provides a processor-readable storage medium, which stores a computer program, and the computer program is used to enable the processor to execute the steps of the methods provided in the above embodiments, for example, including: obtaining channel state data between the base station and the user terminal to be scheduled; inputting the channel state data into a trained user selection model to obtain a user candidate set that meets the maximum channel gain; wherein the trained user selection model is obtained by training a neural network with sample channel state data marked with a beam strength label between the user terminal and the base station, and the trained user selection model includes a first neural network and a second neural network; sorting the spatially multiplexed users in the user candidate set to obtain a user scheduling result.

[0163] The processor-readable storage medium can be any available medium or data storage device that can be accessed by the processor, including but not limited to magnetic storage (such as floppy disks, hard disks, magnetic tapes, magneto-optical disks (MO)), optical storage (such as CDs, DVDs, BDs, HVDs, etc.), and semiconductor storage (such as ROMs, EPROMs, EEPROMs, non-volatile memories (NANDFLASH), solid-state drives (SSDs)), etc.

[0164] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0165] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, or of course, by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the existing technology can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0166] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A user scheduling method, characterized in that: include: Acquire channel state data between the base station and the user terminal to be scheduled; Inputting the channel state data into a trained user selection model to obtain a user candidate set that satisfies a maximum channel gain; wherein the trained user selection model is obtained by training a neural network using sample channel state data labeled with a beam strength label between a user terminal and a base station, and the trained user selection model includes a first neural network and a second neural network; Sort the spatial multiplexing users in the user candidate set to obtain a user scheduling result; The trained user selection model is obtained by training through the following steps: Constructing a first training sample set based on sample channel state data labeled with a beam strength label between the user terminal and the base station; Inputting the first training sample set into a first neural network for training to obtain a virtual user channel and a pre-trained first neural network model, wherein the virtual user channel is a channel corresponding to a user group using the same beam; Constructing a second training sample set according to the virtual user channel; Inputting the second training sample set into a second neural network for training to obtain a pre-trained second neural network model; A trained user selection model is obtained based on the pre-trained first neural network and the pre-trained second neural network.

2. The user scheduling method according to claim 1, characterized in that: The step of obtaining a trained user selection model based on the pre-trained first neural network and the pre-trained second neural network includes: Obtaining a pretrained user selection model based on the pretrained first neural network and the pretrained second neural network; The pre-trained user selection model is used to predict candidate user terminals; the beam intensity labels corresponding to the predicted candidate user terminals are compared with the beam intensity labels of the candidate user terminals in the actual training set, and backpropagation is performed based on the comparison results to fine-tune the pre-trained user selection model to obtain a trained user selection model.

3. The user scheduling method according to claim 1, characterized in that: Sorting the spatially multiplexed users in the user candidate set to obtain a user scheduling result includes: S1, arranging the spatially multiplexed users in the user candidate set in descending order according to the channel gains of the user terminals in the user candidate set, and constructing an index set; S2, constructing a user index set according to the maximum intensity beam index of each spatially multiplexed user in the index set; S3, determining a frequency resource allocation prediction value of a spatially multiplexed user of a maximum intensity beam according to the user index set; S4, arranging the classification probability value of each resource block in the frequency resource allocation prediction value in descending order; S5, repeat S2 to S4 until all user terminals are allocated corresponding frequency resources, and then determine the user scheduling result.

4. The user scheduling method according to claim 1, characterized in that: After sorting the users in the user candidate set and obtaining the user scheduling result, the method further includes: Perform inverse fast Fourier transform on the precoding codebook and obtain simulated precoding based on the user scheduling result; Acquire digital precoding according to the analog precoding and the user scheduling result; According to the analog precoding and the digital precoding, user scheduling is performed on uplink and downlink channels.

5. The user scheduling method according to claim 4, characterized in that: The performing user scheduling on uplink and downlink channels according to the analog precoding and the digital precoding includes: Acquire a total system rate according to the analog precoding and the digital precoding; Performing user scheduling on uplink and downlink channels according to a maximum signal-to-noise ratio criterion and the total system rate; The system rate sum calculation formula includes: ； ； in, Indicates the k User's n Resource blocks can achieve a rate of Indicates the total system rate, Indicates the number of resource blocks, Indicates the number of RF links, Indicates the dimension mmWave massive MIMO channels, Indicates the dimension The transmitted signal vector, Indicates the dimension The digital precoder, Indicates the dimension The analog precoder, Indicates the number of transmitting antennas, Indicates the number of data stream communications between the base station and the user terminal, represents the variance, represents the simulation domain, Represents a numeric domain, Indicates the j data streams.

6. A user scheduling system, characterized in that: include: The channel state data acquisition module is used to obtain the channel state data between the base station and the user terminal to be scheduled; A user candidate set acquisition module is configured to input the channel state data into a trained user selection model to obtain a user candidate set that satisfies a maximum channel gain; wherein the trained user selection model is obtained by training a neural network using sample channel state data labeled with a beam strength label between a user terminal and a base station, and the trained user selection model includes a first neural network and a second neural network; A user scheduling module is used to sort the spatial multiplexing users in the user candidate set and obtain a user scheduling result; The system further comprises: A first training sample set construction module is configured to construct a first training sample set based on sample channel state data labeled with a beam strength label between a user terminal and a base station; A virtual user channel acquisition module is configured to input the first training sample set into a first neural network for training, and acquire a virtual user channel and a pre-trained first neural network model, wherein the virtual user channel is a channel corresponding to a user group using the same beam; A second training sample set construction module is configured to: construct a second training sample set according to the virtual user channel; A second neural network model pre-training module is used to: input the second training sample set into the second neural network for training, and obtain a pre-trained second neural network model; The user selection model training module is used to obtain a trained user selection model based on the pre-trained first neural network and the pre-trained second neural network.

7. An electronic device comprising a processor and a memory storing a computer program, characterized in that: When the processor executes the computer program, the steps of the user scheduling method according to any one of claims 1 to 5 are implemented.

8. 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, the steps of the user scheduling method according to any one of claims 1 to 5 are implemented.

Images