A beam prediction method and apparatus

Abstract

The embodiment of the application provides a kind of beam prediction method and device, it is related to communication field, can reduce resource overhead on the basis of meeting the accuracy requirement of AI model.The method of its method is applied to terminal device, comprising: terminal device sends first information to network device, first information is used to indicate first quantity, first quantity is the minimum value determined based on first model that terminal device obtains the beam quantity of first beam set, and first model is deep neural network model;Terminal device receives the beam information of first beam set from network device, and the beam quantity of first beam set is greater than or equal to first quantity;Terminal device measures the signal quality of first beam set, inputs the signal quality of first beam set into second model, and predicts the signal quality of second beam set based on second model.

Description

Technical Field

[0001] This application relates to the field of communications, and more particularly to a beam prediction method and apparatus. Background Technology

[0002] To ensure communication link quality, massive MIMO (Massively Multi-Size Antenna) deployment and beam management have become indispensable technologies. However, with further increases in frequency bands and antenna size, beam management under traditional frameworks incurs significant signaling overhead. One important means to reduce the signaling overhead of traditional beam management frameworks is artificial intelligence (AI) models. In AI-assisted beam management, the AI ​​model can predict the beam quality of a larger beam set (e.g., Set A) based on measurements from a smaller beam set (e.g., Set B). This eliminates the need for the base station to transmit the entire candidate beam set, thus significantly reducing signaling overhead.

[0003] For the AI ​​model on the UE side, if the number of beams in Set B is too large, it will cause unnecessary resource overhead; if the number of beams in Set B is too small, the AI ​​model cannot accurately predict the beam quality of Set A through Set B, making the AI ​​model unable to meet the accuracy requirements. Summary of the Invention

[0004] This application provides a beam prediction method and apparatus that can reduce resource consumption while meeting the accuracy requirements of AI models.

[0005] To achieve the above objectives, the embodiments of this application adopt the following technical solutions:

[0006] In a first aspect, a beam prediction method is provided, applied to a terminal device, comprising: the terminal device sending first information to a network device, the first information indicating a first quantity, the first quantity being determined by the terminal device based on the minimum number of beams in a first beam set obtained by the terminal device based on a first model, the first model being a deep neural network model; the terminal device receiving beam information of the first beam set from the network device, the number of beams in the first beam set being greater than or equal to the first quantity; the terminal device measuring the signal quality of the first beam set, inputting the signal quality of the first beam set into a second model, and predicting the signal quality of a second beam set based on the second model.

[0007] Based on the method provided in this application embodiment, the terminal device can determine a first quantity based on the minimum value of the number of beams in the first beam set obtained after the deep neural network model training convergence, and send first information (indicating the first quantity) to the network device. The network device can determine the number of beams in the first beam set based on the first quantity indicated by the first information (e.g., such that the number of beams in the first beam set is greater than or equal to the first quantity indicated by the first information). Then, the terminal device can input the signal quality of the first beam set into the second model, and predict the signal quality of the second beam set based on the second model. Since the first quantity is determined by the minimum value of the number of beams in the first beam set obtained by the terminal device based on the deep neural network model, it can ensure that the number of beams in the first beam set is as small as possible, which can minimize resource consumption and meet the accuracy requirements of the AI ​​model (e.g., the second model).

[0008] In one possible implementation, before the terminal device sends the first information to the network device, the method further includes: the terminal device training a first model and a second model based on self-supervised learning. Self-supervised learning is a machine learning method that does not rely on manually labeled "labels" for learning. The first model and the second model can automatically mine supervision signals from unlabeled raw data (e.g., the signal quality of the second beam set) (e.g., continuously outputting a mask to mask the second beam set, using the unmasked beams of the second beam set as beams of the first beam set, thereby determining the minimum number of beams in the first beam set), thereby completing feature learning or task training.

[0009] In one possible implementation, before the terminal device trains the first and second models based on self-supervised learning, the method further includes: the terminal device receiving second information from the network device, the second information including beam information of the second beam set and / or performance metrics of the second model; the terminal device training the first and second models based on self-supervised learning includes: the terminal device measuring the signal quality of the second beam set, using the signal quality of the second beam set as training data for the first model, using the output of the first model as input for the second model, and iteratively training the first and second models until the second model meets the performance metrics of the second model. Thus, by iteratively training the first and second models, the stability, generalization ability, and task adaptability of the output results of the first and second models can be improved.

[0010] In one possible implementation, the first model includes a feature extraction layer and a latent layer. The above-mentioned method uses the signal quality of the second beam set as the training data of the first model, and the output of the first model as the input of the second model. Iterative training of the first and second models continues until the second model meets the performance metrics. This includes: Step 1: Inputting the beam matrix into the feature extraction layer to obtain the feature map output by the feature extraction layer. The beam matrix includes the signal quality of each beam in the second beam set. Step 2: Using the feature map as the input to the latent layer to obtain the quantization vector output by the latent layer. Step 3: Binarizing the quantization vector to obtain a binary vector. Step 4: Converting the binary vector into a mask matrix, where the order of each mask in the mask matrix is ​​the same as the order of the corresponding beam in the beam matrix. Step 5: Using the Hadamard product of the mask matrix and the beam matrix as the input to the second model to obtain the signal quality of the second beam set predicted by the second model. Steps 1-5 are repeated until the overall loss function corresponding to the first and second models converges. The overall loss function is determined by a weighted sum of the loss functions of the first and second models, and the loss function of the second model is determined based on the performance metrics of the second model. The convergence of the overall loss function indicates that the second model meets its performance metrics. It should be understood that when the overall loss functions of the first and second models converge, the error (loss value) between the prediction results of the first and second models and the actual results is small, which meets the prediction accuracy requirements.

[0011] The feature extraction layer is used to extract key information from the beam matrix, and the latent layer (hidden feature layer) is used to perform in-depth processing on the extracted information. Together, the feature extraction layer and the latent layer (hidden feature layer) constitute the transformation link from raw data (such as beam matrix) to effective features.

[0012] For example, the feature extraction layer could be a U-Net network, and the latent layer could refer to a hidden feature layer, such as a hash layer. The quantization vector can be a hash vector, and the binary vector can be a hash code (hash value). The U-Net network is a segmentation model that can be used to extract and analyze features from a beam matrix to obtain feature maps. Compared to other complex segmentation models (such as Mask R-CNN and Transformer-based segmentation models), the U-Net network has a smaller parameter size, lower computational requirements during training, and faster training speed. The hash layer can be used to map the feature maps output by the U-Net network into fixed-length hash vectors. Hash vectors are more concise than the original data, which can improve the model's processing efficiency.

[0013] In one possible implementation, the loss function of the first model includes a first function and a second function. The first function is used to incentivize the hash code to have either a first value or a second value, and the second function is used to incentivize the sparse number of first values ​​in the hash code. The first value can be used to indicate whether to retain (without masking / filtering) the beam corresponding to the hash code (i.e., when the hash code is the first value, the beam corresponding to that hash code is included in the first beam set), and the second value can be used to indicate whether to mask (filter) the beam corresponding to the hash code (i.e., when the hash code is the second value, the beam corresponding to that hash code is not included in the first beam set). This ensures that the first model masks as many beams as possible in the beam matrix, thereby minimizing the number of beams in the first beam set.

[0014] In one possible implementation, the performance metrics of the second model include mean squared error (MSE). MSE can be used to measure the difference between predicted and actual values, ensuring that the difference between the predicted values ​​of the second model (e.g., the beam quality of Set A predicted by the second model) and the actual values ​​(e.g., the actual measured beam quality of Set A) is as small as possible, thus determining the accuracy of the model prediction.

[0015] In one possible implementation, if the overall loss functions corresponding to the first and second models converge, the value of the second function is the minimum number of beams in the first beam set obtained by the first model. The second function is used to excite the sparse number of first values ​​in the hash code, which ensures that the first model masks as many beams as possible in the beam matrix, thereby obtaining the minimum number of beams in the first beam set.

[0016] In one possible implementation, the second information further includes the number of beams in a first configurable beam set. This number includes at least one value: if the minimum number of beams in the first beam set obtained by the first model is less than the minimum of the at least one value, the first number is the minimum of the at least one value; if the minimum number of beams in the first beam set obtained by the first model is greater than or equal to the minimum of the at least one value, the first number is the minimum of a target set. The target set includes one or more values ​​that are greater than or equal to the minimum number of beams in the first beam set obtained by the first model. This prevents the lower bound of the Set B beam count obtained by the first model from exceeding the configuration capability of the network device.

[0017] In one possible implementation, the terminal device sending the first information to the network device includes: the terminal device sending uplink control information (UCI) signaling to the network device, wherein the first information is carried in at least one field of the UCI signaling. For example, taking the first quantity as B, a first quantity reporting field can be added to the UCI signaling, and the length of this field can be [missing information]. .

[0018] In one possible implementation, the terminal device receiving the second information from the network device includes: the terminal device receiving downlink control information (DCI) signaling from the network device, wherein the second information is carried in at least one field of the DCI signaling. That is, the second information is carried in at least one field of the DCI signaling. The at least one field used to carry the second information can be a newly added field or a reused existing field.

[0019] Secondly, a beam prediction method is provided, applied to a network device, comprising: the network device receiving first information from a terminal device, the first information indicating a first quantity, the first quantity being determined by the minimum number of beams in a first beam set obtained by the terminal device based on a first model; the network device sending beam information of the first beam set to the terminal device, the number of beams in the first beam set being greater than or equal to the first quantity.

[0020] Based on the method provided in the embodiments of this application, a network device can determine the number of beams in a first beam set based on the information predicted after the first model training of a terminal device converges (e.g., such that the number of beams in the first beam set is greater than or equal to the first number indicated by the first information). In this way, resource overhead can be minimized while meeting the accuracy requirements of the AI ​​model (e.g., the second model) of the terminal device.

[0021] In one possible implementation, before the network device receives the first information from the terminal device, the method further includes: the network device sending second information to the terminal device, the second information including beam information of a second beam set and / or performance indicators of a second model.

[0022] In one possible implementation, the second information further includes the number of beams in a network-configurable first beam set, which includes at least one value. If the minimum value of the number of beams in the first beam set obtained by the first model is less than the minimum value among the at least one values, the first number is the minimum value among the at least one values. If the minimum value of the number of beams in the first beam set obtained by the first model is greater than or equal to the minimum value among the at least one values, the first number is the minimum value in a target set, which includes one or more values ​​among the at least one values ​​that are greater than or equal to the minimum value of the number of beams in the first beam set obtained by the first model.

[0023] In one possible implementation, the network device receiving first information from the terminal device includes: the network device receiving uplink control information (UCI) signaling from the terminal device, wherein the first information is carried in at least one field of the UCI signaling.

[0024] In one possible implementation, the network device sending the second information to the terminal device includes: the network device sending downlink control information (DCI) signaling to the terminal device, wherein the second information is carried in at least one field of the DCI signaling.

[0025] Thirdly, a communication device is provided, comprising a transceiver module and a processing module. The transceiver module is configured to send first information to a network device, the first information indicating a first quantity, the first quantity being determined by a terminal device based on a first model, which is the minimum number of beams in a first beam set, and the first model being a deep neural network model. The transceiver module is further configured to receive beam information of the first beam set from the network device, wherein the number of beams in the first beam set is greater than or equal to the first quantity. The processing module is configured to measure the signal quality of the first beam set, input the signal quality of the first beam set into a second model, and predict the signal quality of a second beam set based on the second model.

[0026] In one possible implementation, before sending the first information to the network device, the processing module is also used to: train the first model and the second model based on self-supervised learning.

[0027] In one possible implementation, before training the first and second models based on self-supervised learning, the transceiver module is further configured to: receive second information from the network device, the second information including beam information of the second beam set and / or performance metrics of the second model; the processing module is further configured to: measure the signal quality of the second beam set, use the signal quality of the second beam set as training data for the first model, use the output of the first model as input for the second model, and iteratively train the first and second models until the second model meets the performance metrics of the second model.

[0028] In one possible implementation, the first model includes a feature extraction layer and a latent layer. The above-mentioned method uses the signal quality of the second beam set as the training data of the first model, and the output of the first model as the input of the second model. Iterative training of the first and second models continues until the second model meets the performance metrics. This includes: Step 1: Inputting the beam matrix into the feature extraction layer to obtain the feature map output by the feature extraction layer. The beam matrix includes the signal quality of each beam in the second beam set. Step 2: Using the feature map as the input to the latent layer to obtain the quantization vector output by the latent layer. Step 3: Binarizing the quantization vector to obtain a binary vector. Step 4: Converting the binary vector into a mask matrix, where the order of each mask in the mask matrix is ​​the same as the order of the corresponding beam in the beam matrix. Step 5: Using the Hadamard product of the mask matrix and the beam matrix as the input to the second model to obtain the signal quality of the second beam set predicted by the second model. Steps 1-5 are repeated until the overall loss function corresponding to the first and second models converges. The overall loss function is determined by a weighted sum of the loss functions of the first and second models, and the loss function of the second model is determined based on the performance metrics of the second model.

[0029] In one possible implementation, the loss function of the first model includes a first function and a second function, wherein the first function is used to incentivize the hash code to be a first value or a second value, and the second function is used to incentivize the sparse number of first values ​​in the hash code.

[0030] In one possible implementation, the performance metrics of the second model include mean squared error (MSE).

[0031] In one possible implementation, if the overall loss functions corresponding to the first model and the second model converge, the value of the second function is the minimum number of beams in the first beam set obtained by the first model.

[0032] In one possible implementation, the second information further includes the number of beams in a network-configurable first beam set, which includes at least one value. If the minimum value of the number of beams in the first beam set obtained by the first model is less than the minimum value among the at least one values, the first number is the minimum value among the at least one values. If the minimum value of the number of beams in the first beam set obtained by the first model is greater than or equal to the minimum value among the at least one values, the first number is the minimum value in a target set, which includes one or more values ​​among the at least one values ​​that are greater than or equal to the minimum value of the number of beams in the first beam set obtained by the first model.

[0033] In one possible implementation, the transceiver module is used to: send uplink control information (UCI) signaling to the network device, wherein the first information is carried in at least one field of the UCI signaling.

[0034] In one possible implementation, the transceiver module is used to: receive downlink control information (DCI) signaling from the network device, wherein second information is carried in at least one field of the DCI signaling.

[0035] Fourthly, a communication device is provided, the communication device including a transceiver module, the transceiver module being configured to receive first information from a terminal device, the first information being used to indicate a first quantity, the first quantity being determined by the terminal device based on the minimum value of the number of beams in a first beam set obtained based on a first model; the transceiver module is further configured to send beam information of the first beam set to the terminal device, the number of beams in the first beam set being greater than or equal to the first quantity.

[0036] In one possible implementation, before receiving the first information from the terminal device, the transceiver module is further configured to send the second information to the terminal device, the second information including beam information of the second beam set and / or performance indicators of the second model.

[0037] In one possible implementation, the second information further includes the number of beams in a network-configurable first beam set, which includes at least one value. If the minimum value of the number of beams in the first beam set obtained by the first model is less than the minimum value among the at least one values, the first number is the minimum value among the at least one values. If the minimum value of the number of beams in the first beam set obtained by the first model is greater than or equal to the minimum value among the at least one values, the first number is the minimum value in a target set, which includes one or more values ​​among the at least one values ​​that are greater than or equal to the minimum value of the number of beams in the first beam set obtained by the first model.

[0038] In one possible implementation, the transceiver module is used to receive uplink control information (UCI) signaling from the terminal device, wherein the first information is carried in at least one field of the UCI signaling.

[0039] In one possible implementation, the transceiver module is used to send downlink control information (DCI) signaling to the terminal device, wherein the second information is carried in at least one field of the DCI signaling.

[0040] Fifthly, a communication device is provided, including a wireless communication module, a memory, and one or more processors; the wireless communication module, the memory, and the processor are coupled; wherein the memory is used to store computer program code, the computer program code including computer instructions; when the computer instructions are executed by the processor, the communication device performs the method in any possible implementation of the first aspect.

[0041] In a sixth aspect, a communication device is provided, including a wireless communication module, a memory, and one or more processors; the wireless communication module, the memory, and the processor are coupled; wherein the memory is used to store computer program code, the computer program code including computer instructions; when the computer instructions are executed by the processor, the communication device performs the method in any possible implementation of the second aspect above.

[0042] In a seventh aspect, a processor is provided, comprising: an input circuit, an output circuit, and a processing circuit. The processing circuit is configured to receive signals through the input circuit and transmit signals through the output circuit, causing the processor to execute a method in any possible implementation of any of the preceding aspects.

[0043] In specific implementation, the processor can be one or more chips, the input circuit can be input pins, the output circuit can be output pins, and the processing circuit can be transistors, gate circuits, flip-flops, and various logic circuits. The input signal received by the input circuit can be received and input by, for example, but not limited to, a receiver, and the signal output by the output circuit can be, for example, but not limited to, output to and transmitted by a transmitter. Furthermore, the input circuit and the output circuit can be the same circuit, which is used as both the input circuit and the output circuit at different times. This application does not limit the specific implementation of the processor and various circuits.

[0044] Eighthly, a communication device is provided, including a processor and a memory. The processor is used to read instructions stored in the memory, receive signals via a receiver, and transmit signals via a transmitter to execute the method in any possible implementation of any of the preceding aspects.

[0045] Optionally, the processor may be one or more, and the memory may be one or more.

[0046] Ninthly, a computer program product is provided, the computer program product comprising: a computer program (also referred to as code or instructions) that, when the computer program is run, causes a computer to perform a method in any possible implementation of any of the above aspects.

[0047] In a tenth aspect, a computer-readable storage medium is provided that stores a computer program (also referred to as code or instructions) that, when run on a computer, causes the computer to perform the methods in any possible implementation of any of the preceding aspects.

[0048] Eleventhly, embodiments of this application provide a chip system including one or more processors for calling and executing instructions stored in memory, causing the methods in any possible implementation of any of the above aspects to be executed. The chip system may be composed of chips or may include chips and other discrete devices.

[0049] The chip system may include input circuits or interfaces for transmitting information or data, and output circuits or interfaces for receiving information or data.

[0050] In a twelfth aspect, a communication system is provided, including the aforementioned terminal device and network device. The terminal device performs the method in any possible implementation of the first aspect, and the network device performs the method in any possible implementation of the second aspect. Optionally, the communication system may further include other devices that communicate with the terminal device and / or the network device. Attached Figure Description

[0051] Figure 1 A schematic diagram of spatial beam prediction provided for an embodiment of this application;

[0052] Figure 2 A timing relationship diagram for beam prediction provided in an embodiment of this application;

[0053] Figure 3 A schematic diagram of a time-domain beam prediction method provided in an embodiment of this application;

[0054] Figure 4 A schematic diagram of a communication system provided in an embodiment of this application;

[0055] Figure 5 A schematic flowchart of a beam prediction method provided in an embodiment of this application;

[0056] Figure 6 A schematic flowchart illustrating another beam prediction method provided in an embodiment of this application;

[0057] Figure 7 A schematic diagram of a masking module and a deduction module provided in an embodiment of this application;

[0058] Figure 8 This is a schematic diagram illustrating a specific training process for an occlusion module provided in an embodiment of this application.

[0059] Figure 9 A schematic block diagram of a communication device provided in an embodiment of this application;

[0060] Figure 10 This is a schematic block diagram of another communication device provided in the embodiments of this application. Detailed Implementation

[0061] To ensure clarity and conciseness in the description of the following embodiments, a brief introduction to the relevant concepts or technologies is given first:

[0062] Beam: A beam is a communication resource. A beam can be wide, narrow, or other types of beam. Beamforming technology can be beamforming technology or other techniques. Beamforming technology can specifically be digital beamforming technology, analog beamforming technology, or hybrid digital / analog beamforming technology. Different beams can be considered different resources. The same or different information can be transmitted through different beams. Optionally, multiple beams with the same or similar communication characteristics can be considered as a single beam. A beam can include one or more antenna ports for transmitting data channels, control channels, and detection signals, etc. For example, a transmit beam can refer to the distribution of signal strength in different directions in space after a signal is transmitted through an antenna, and a receive beam can refer to the distribution of signal strength in different directions in space of the wireless signal received from the antenna. It is understood that one or more antenna ports forming a beam can also be considered as a set of antenna ports. When using low-frequency or mid-frequency bands, signals can be transmitted omnidirectionally or through a wide angle. When using high-frequency bands, thanks to the smaller carrier wavelength of high-frequency communication systems, antenna arrays consisting of many antenna elements can be arranged at both the transmitting and receiving ends. The transmitting end transmits signals with a certain beamforming weight, forming a spatially directional beam. At the same time, using an antenna array with a certain beamforming weight at the receiving end can improve the signal reception power and counteract path loss.

[0063] Beam management encompasses a series of operations including beam scanning, beam measurement, beam reporting, and beam selection. Beam scanning refers to covering a spatial area with a set of transmit and receive beams according to pre-specified time intervals and directions. Beam measurement refers to the evaluation of the received signal quality at the BS or UE. Signal quality metrics may include, for example, reference signal receiving power (RSRP), reference signal receiving quality (RSRQ), signal-to-interference plus noise ratio (SINR), or signal-to-noise ratio (SNR). Beam reporting is the process by which the UE transmits beam quality and beam decision information to the Radio Access Network (RAN). Beam selection refers to selecting one or more suitable beams at the BS or UE based on the measurement results obtained through the beam measurement process.

[0064] Channel state information (CSI) feedback: This is the process where the network-side device sends a reference signal to the terminal device, and the terminal device measures the reference signal and then feeds back the measured value (i.e., CSI) to the network-side device. One reference signal can correspond to one beam. The network-side device can adjust its modulation and / or coding based on the measured value (i.e., CSI) fed back by the terminal device to ensure communication rate. Therefore, CSI feedback is an important means of ensuring the communication quality of wireless communication links. Currently, with the further upgrading of communication frequency bands and the expansion of antenna size, network-side devices need to send more reference signals, and / or the terminal feedback overhead (e.g., the overhead of CSI feedback) is greater. To address this, in 5G New Radio (NR), AI technology is introduced to reduce various overheads (e.g., the overhead of sending reference signals). For example, AI-based beam management has been introduced in existing 5G NR AI air interface applications.

[0065] AI-based beam management: An AI model predicts the beam quality of a larger beam set (e.g., Set A) based on measurements from a smaller beam set (e.g., Set B). This reduces resource overhead by eliminating the need for the network to send excessive reference signals (e.g., beam information for the entire Set A) to the terminal device. The AI ​​model is a function trained on statistical data (e.g., beam measurement results) collected in a specific scenario (e.g., beam management scenario), applicable to that specific scenario. AI models can also be called machine learning (ML) models; this application uses AI models for illustration. AI models include terminal-side models and network-side models. The AI ​​models involved in this application are terminal-side models. A terminal-side model refers to an AI model deployed on the terminal device, where inference operations are performed on the terminal device.

[0066] The terminal device side model can include two use cases in beam management, namely use case 1 (BM-Case 1) and use case 2 (BM-Case 2).

[0067] Use Case 1 (BM-Case 1): The terminal device-side model can be used for downlink beam prediction in the spatial domain (or airspace domain) during beam management. For example... Figure 1 As shown, the terminal device can perform spatial beam prediction on beam set A based on the measurement results of beam set Set B. Specifically, the terminal device can use the measured signal quality values ​​of the beams in Set B as input or part of the input to the AI ​​model to predict the signal quality of the beams in Set A. The signal quality of the beams can be, for example, Layer 1 (L1) RSRP. Set B can be the beam set corresponding to the synchronization signal / PBCH block (SSB) (i.e., the beam set scanned according to the SSB), or Set B can be the beam set corresponding to the channel state information-reference signal (CSI-RS) (i.e., the beam set scanned according to the CSI-RS). Set A is typically the beam set corresponding to the CSI-RS.

[0068] like Figure 2The diagram illustrates a timing relationship for beam prediction. The network (NW) can send beam information for Set B to the user equipment (UE). The UE can then measure the signal quality of Set B (the signal quality of each beam within Set B) based on this information. The UE inputs the signal quality of Set B into an AI model, which predicts the signal quality of the top K beams (Top-K beams) of Set A. The UE then reports the prediction results; specifically, it can report the predictions of the top K beams with the best signal quality in a single CSI report. The NW can then select the optimal beam for data transmission based on the prediction results reported by the UE.

[0069] For BM-Case 1, the number of beams in Set B can be less than the number of beams in Set A. For example, Set A contains S1 beams, and Set B contains S2 beams from Set A. Here, S1 is an integer greater than 1, and S2 is an integer less than S1.

[0070] Taking Set A and Set B as examples where both are sets of beams corresponding to CSI-RS, assume Set B contains CSI-RS#[2,4,6,8], meaning Set B contains CSI-RS beams identified as 2, 4, 6, and 8 respectively, and Set A contains CSI-RS#[1,2,3,4,5,6,7,8], meaning Set A contains CSI-RS beams identified as 1, 2, 3, 4, 5, 6, 7, and 8 respectively. After the UE measures Set B, it obtains RSRP#[2,4,6,8], which is the RSRP of the CSI-RS beams identified as 2, 4, 6, and 8. Then, RSRP#[2,4,6,8] is used as input to the AI ​​model, outputting the predicted RSRP values ​​of the 8 CSI-RS beams in Set A, or the probability (predicted probability) that each CSI-RS beam in Set A is the CSI-RS beam with the largest measured RSRP value in Set A. Clearly, without using the AI ​​model, the network would need to send the entire Set A to the terminal, increasing its reference resource overhead.

[0071] Use Case 2 (BM-Case 2): Time-domain downlink beam prediction for Set A based on historical measurements of Set B. For example, the UE can measure Set B over the past M transmission occasions, using the historical measurements of Set B as input or part of the input to an AI model for prediction. The AI ​​model outputs the predicted beams within Set A for the next N time moments. Figure 3As shown, the UE can measure Set B at time t-T1, ..., time t-2, and time t-1 respectively. The measurement results of Set B at time t-T1, ..., time t-2, and time t-1 are input into the AI ​​model. The AI ​​model can output a prediction of Set A at time t (i.e., the prediction result of the beam within Set A at time t), a prediction of Set A at time t+1 (i.e., the prediction result of the beam within Set A at time t+1), ..., a prediction of Set A at time t+T2 (i.e., the prediction result of the beam within Set A at time t+T2). After the UE measures multiple Set Bs and performs model inference, it can report the prediction results for the next few times in a single CSI report.

[0072] In AI-based beam management, the network side needs to transmit configuration information for Set A and Set B to the UE side to specify the training method of the model (the model on the terminal device side). In this way, the model can reconstruct the beams in Set A through Set B, thereby reducing signaling overhead and prediction latency.

[0073] However, for terminal devices, too many beams in Set B can cause unnecessary resource overhead; while too few beams in Set B can prevent the model (the model on the terminal device side) from accurately reconstructing the beams in Set A from Set B, thus reducing the robustness and accuracy of the model.

[0074] This application provides a beam prediction method applicable to the beam management process in the 3rd Generation Partnership Project (3GPP). It can reasonably configure the Set B beam set, ensure the robustness and accuracy of the model, and minimize resource overhead.

[0075] The technical solutions of this application can be applied to various communication systems. For example, long term evolution (LTE) systems, LTE frequency division duplex (FDD) systems, LTE time division duplex (TDD) systems, worldwide interoperability for microwave access (WiMAX) communication systems, 5G mobile communication systems, or new radio (NR), etc. The 5G mobile communication system described in this application includes non-standalone (NSA) 5G mobile communication systems and / or standalone (SA) 5G mobile communication systems. The technical solutions provided in this application can also be applied to future communication systems, such as sixth-generation mobile communication systems. The communication system can also be a future evolved public land mobile network (PLMN) network, device-to-device (D2D) network, machine-to-machine (M2M) network, Internet of Things (IoT) network, or other networks.

[0076] Figure 4 A schematic diagram of a communication system to which the technical solutions provided in the embodiments of this application are applicable is given. The communication system may include a network device 200 and one or more terminal devices 100 connected to the network device 200. Figure 4 (Only one is shown). Data transmission can occur between network devices (also known as network-side devices) and terminal devices (also known as terminal-side devices).

[0077] In this application, network device 200 can be a device capable of communicating with terminal device 100. For example, network device 200 can be a base station, which can be an evolved Node B (eNB or eNodeB) in LTE, a base station in NR, a relay station or access point, or a base station in a future network, etc., without limitation in this embodiment. In NR, a base station can also be called a transmission reception point (TRP) or gNB. In this embodiment, the network device can be a separately sold network device, such as a base station, or a chip within the network device that implements the corresponding functions. In this embodiment, the chip system can be composed of chips or can include chips and other discrete components. In the technical solutions provided in this embodiment, the network device itself is used as an example to describe the technical solutions provided in this embodiment.

[0078] In this application embodiment, the terminal device 100 can also be referred to as a terminal device, which can be a device with wireless transceiver capabilities. The terminal device can be deployed on land, including indoors or outdoors, handheld or vehicle-mounted; it can also be deployed on water (such as on ships); and it can also be deployed in the air (e.g., on airplanes, balloons, and satellites). The terminal device can be user equipment (UE). The UE includes handheld devices, vehicle-mounted devices, wearable devices, or computing devices with wireless communication capabilities. For example, the UE can be a mobile phone, tablet computer, or computer with wireless transceiver capabilities. The terminal device can also be a virtual reality (VR) terminal device, an augmented reality (AR) terminal device, a wireless terminal device in industrial control, a wireless terminal device in autonomous driving, a wireless terminal device in telemedicine, a wireless terminal device in a smart grid, a wireless terminal device in a smart city, a wireless terminal device in a smart home, etc. In this application embodiment, the terminal device can be a separately sold terminal device or a chip within a terminal device. In the technical solutions provided in the embodiments of this application, the terminal device is used as an example to describe the technical solutions provided in the embodiments of this application.

[0079] Examples of this application Figure 4The network device 200 or terminal device 100 can be implemented by a single device or as a functional module within a single device; this application embodiment does not specifically limit this. It is understood that the aforementioned functions can be network elements in hardware devices, software functions running on dedicated hardware, virtualization functions instantiated on a platform (e.g., a cloud platform), or chip systems. In this application embodiment, the chip system can be composed of chips or can include chips and other discrete devices.

[0080] The technical solutions of the embodiments of this application will now be described with reference to the accompanying drawings. In the description of this application, unless otherwise stated, "at least one" refers to one or more, and "multiple" refers to two or more. Furthermore, to facilitate a clear description of the technical solutions of the embodiments of this application, the terms "first," "second," etc., are used in the embodiments of this application to distinguish identical or similar items with substantially the same function and effect. Those skilled in the art will understand that the terms "first," "second," etc., do not limit the quantity or execution order, and that "first," "second," etc., do not necessarily imply differences.

[0081] For ease of understanding, the beam prediction method provided in the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0082] like Figure 5 As shown, this application provides a beam prediction method, using Set B as the first beam set and Set A as the second beam set as an example. Set A has more beams than Set B. Set A and Set B can be referenced to the relevant descriptions in the BM-Case 1 use case, but the application of this application is not limited to the BM-Case 1 use case. The method includes:

[0083] 501. The terminal device sends first information to the network device, and correspondingly, the network device receives the first information from the terminal device. The first information is used to indicate a first quantity, which is determined by the minimum number of beams of Set B obtained by the terminal device based on the first model.

[0084] The terminal device may deploy a first model, which may be a deep neural network model. This first model can be used to obtain the minimum number of beams in Set B (i.e., the lower bound of the number of beams in Set B). The first quantity indicated by the first information may be greater than or equal to the minimum number of beams in Set B obtained by the first model.

[0085] After receiving the first information from the terminal device, the network device can determine the beam pattern / pattern of Set B based on the first quantity indicated by the first information. The beam pattern of Set B is used to indicate all beams contained in Set B. The number of beams in Set B determined by the network device is greater than or equal to the first quantity.

[0086] In some embodiments, the terminal device sends UCI signaling to the network device, and the first information is carried in at least one field of the UCI signaling. That is, the terminal device can report the first information (i.e., the first quantity) based on the UCI signaling.

[0087] For example, taking setA as an example, which includes A beams and has a first quantity of B, some fields included in the UCI signaling can be as shown in Table 1.

[0088] Table 1

[0089]

[0090] As shown in Table 1, in addition to the traditional beam measurement result reporting fields (such as CRI reporting field and RSRP reporting field), the UCI signaling also includes a newly added first quantity reporting field, the length of which can be... Where B is greater than 0 and less than A.

[0091] 502. The network device sends the beam information of Set B to the terminal device, and the terminal device receives the beam information of Set B from the network device accordingly, wherein the number of beams of Set B is greater than or equal to a first number.

[0092] Network devices can send / distribute beam information of Set B to terminal devices according to the beam pattern of Set B. Correspondingly, terminal devices can receive the beam information of Set B sent / distributed by the network device. The beam information of Set B sent / distributed by the network device may include the beam identifier of each beam in Set B. Optionally, the beam information of Set B sent / distributed by the network device may include the number of beams in Set B.

[0093] 503. The terminal device measures the signal quality of Set B, inputs the signal quality of Set B into the second model, and predicts the signal quality of Set A based on the second model.

[0094] The terminal device measures the signal quality of each beam in Set B based on the beam information of Set B sent by the network device, inputs the measured signal quality of each beam in Set B into the second model, and predicts the signal quality of Set A based on the second model. The second model is an AI model.

[0095] Based on the method provided in the embodiments of this application, the terminal device can determine a first quantity based on the minimum value of the number of beams in the first beam set obtained by the first model, and send the first information (used to indicate the first quantity) to the network device. The network device can determine the number of beams in the first beam set based on the first information (e.g., such that the number of beams in the first beam set is greater than or equal to the first quantity indicated by the first information). In this way, resource overhead can be reduced as much as possible while meeting the accuracy requirements of the AI ​​model (e.g., the second model).

[0096] In some embodiments, such as Figure 6 As shown, before step 501, the following steps may also be included:

[0097] 5001. The network device sends second information to the terminal device, and correspondingly, the terminal device receives the second information from the network device. The second information includes at least one of the following: beam information of Set A, the number of beams of Set B that the network can configure, and model performance indicators.

[0098] The beam information of Set A is used to assist the terminal device in training the first model so that the terminal device can obtain the lower bound of the number of beams in Set B (the minimum number of beams in Set B) based on the trained first model.

[0099] The terminal device can measure the signal quality of all beams in Set A and use the measured signal quality of all beams in Set A as training data for the first model. The signal quality measurement methods may include, for example, reference signal receiving power (RSRP), reference signal receiving quality (RSRQ), signal-to-interference plus noise ratio (SINR), or signal-to-noise ratio (SNR). For instance, the terminal device can use the measured RSRP of all beams in Set A as training data for the first model (i.e., the training data for the first model includes the RSRP of all beams in Set A).

[0100] Among them, the model performance metrics (performance metrics of the second model) are used to assist the terminal device in training the second model so that the terminal device can predict the beam quality of Set A based on the trained second model. For example, the performance metrics of the second model include MSE. MSE is used to measure the difference between the predicted value (e.g., the beam quality of Set A predicted by the second model) and the actual value (e.g., the actual measured beam quality of Set A).

[0101] The configurable Set B beam count is used to prevent the lower bound of the Set B beam count obtained by the first model from exceeding the configuration capability of the network device. The configurable Set B beam count can include at least one value. For example, the configurable Set B beam count can be {4}, meaning the configurable Set B beam count can be 4; or, the configurable Set B beam count can be {4, 6}, meaning the configurable Set B beam count can be 4 or 6; or, the configurable Set B beam count can be {4, 8, 6}, meaning the configurable Set B beam count can be 4, 8, or 6.

[0102] If the minimum number of beams in the first beam set obtained by the first model is less than the minimum of at least one value, the first quantity is the minimum of at least one value; if the minimum number of beams in the first beam set obtained by the first model is greater than or equal to the minimum of at least one value, the first quantity is the minimum of the target set. The target set includes one or more values ​​that are greater than or equal to the minimum number of beams in the first beam set obtained by the first model.

[0103] For example, suppose the number of configurable Set B beams in the network is {4, 8, 6}, where 4 is the minimum value among the configurable Set B beams. If the lower bound of the Set B beam count obtained by the first model is 3, since 3 is less than 4, the lower bound of the Set B beam count is considered to be 4. If the lower bound of the Set B beam count obtained by the first model is 5, since 5 is greater than 4, and 6 and 8 are greater than 5, the lower bound of the Set B beam count is considered to be the minimum value (i.e., 6) in the target set {6, 8}, that is, the lower bound of the Set B beam count is considered to be 6.

[0104] In other embodiments, if the number of Set B beams that can be configured by the network is not received, the lower bound of the number of Set B beams obtained by the first model can be reported directly.

[0105] In some embodiments, the network device can send DCI signaling to the terminal device, and correspondingly, the terminal device can receive DCI signaling from the network device. At least one field in the DCI signaling can be used to carry second information. That is, the second information is carried in at least one field of the DCI signaling. The at least one field used to carry the second information can be a newly added field or a reused existing field.

[0106] For example, taking DCI signaling format DCI Format 1_1 as an example, the field design of DCI Format 1_1 can be as shown in Table 2.

[0107] Table 2

[0108]

[0109] As shown in Table 2, two new fields can be added to DCI Format 1_1: a first field (e.g., Set BBeam Num Config) and a second field (e.g., Model Performance Ref). The first field (e.g., Set B BeamNum Config) indicates the number of Set B beams that the network can configure, and the second field (e.g., Model Performance Ref) indicates the model performance metric (the performance metric of the second model).

[0110] The TCI-State configuration in the TCI-State List can be used to indicate the direction information of the downlink beam (e.g., beams in beam set Set A) that the UE should use. Therefore, in implementation, a mapping relationship can be established with Set A. That is, the TCI-State configuration in the TCI-State List can be used to indicate the beam information of Set A.

[0111] Additionally, Rel-19 and later enhanced versions can enhance the multi-TCI (Multi-Transmission Configuration Indicator) of DCI Format 1_1 to support more flexible beam management and multi-stream transmission.

[0112] In some embodiments, the measurement report configuration signaling of RRC can be adjusted so that the terminal device reports information that meets the requirements (e.g., reports first information).

[0113] For example, a network device sends RRC signaling (e.g., RRC measurement report configuration signaling) to a terminal device, and the terminal device receives RRC signaling from the network device accordingly. The RRC signaling includes a first variable and a second variable. The first variable indicates the first piece of information to be reported by the terminal device, and the second variable indicates a first quantity, which is determined based on the minimum value of the number of beams in setB (the lower bound of the number of beams in setB). For example, the minimum value of the number of beams in setB (the lower bound of the number of beams in setB) obtained after the first model training of the terminal device converges can be B, where B is greater than 0.

[0114] For example, the format of the RRC measurement report configuration signaling is as follows:

[0115] CSI-ReportConfig ::= {

[0116] reportConfigId = 1,

[0117] ...

[0118] reportQuantity = CRI-RSRP-Setbmin, / / Report the CRI-RSRP measured in setA and the lower bound of the number of beams in setB.

[0119] groupBasedBeamReporting = TRUE, / / Enable multi-beam reporting

[0120] numberOfSetABeams = A, / / Assuming setA has A beams

[0121] numberOfSetBBeamsMin = B, / / Assuming the lower bound of the number of setB beams is B

[0122] ...

[0123] }

[0124] Among them, reportQuantity (an example of the first variable) can be a newly added variable. The enumeration type of reportQuantity can be CRI-RSRP-Setbmin, which is used to indicate the measurement results of setA returned by the terminal device (e.g., CRI-RSRP of each beam in setA) and the lower bound of the number of beams in setB. numberOfSetBBeamsMin (an example of the second variable) can be a newly added variable. numberOfSetBBeamsMin is used to indicate the minimum value of the number of beams in setB (the lower bound of the number of beams in setB).

[0125] 5002. The terminal device trains the first and second models based on self-supervised learning.

[0126] Self-supervised learning is a machine learning method that does not rely on manually labeled "labels" for learning. The model can automatically extract supervision signals (such as the structure of the data itself and feature associations) from raw data without manual labeling, thereby completing feature learning or task training.

[0127] In some embodiments, the terminal device can measure the signal quality of Set A (the signal quality of each beam in Set A), use the signal quality of Set A (the signal quality of each beam in Set A, such as RSRP) as the training data of the first model, use the output of the first model as the input of the second model, and iteratively train the first model and the second model until the second model meets the performance index of the second model.

[0128] In some embodiments, the first model includes a feature extraction layer and a latent layer. The process of iteratively training the first and second models may include:

[0129] Step 1: Input the beam matrix into the feature extraction layer to obtain the feature map output by the feature extraction layer. The beam matrix includes the signal quality of each beam in the second beam set.

[0130] Step 2: Use the feature map as input to the latent layer to obtain the quantized vector of the latent layer output;

[0131] Step 3: Binarize the quantization vector to obtain a binary vector;

[0132] Step 4: Convert the binary vector into a mask matrix. The order of each mask in the mask matrix is ​​the same as the order of the corresponding beams in the beam matrix.

[0133] Step 5: Use the Hadamard product of the mask matrix and the beam matrix as input to the second model to obtain the signal quality of the second beam set predicted by the second model.

[0134] Repeat steps 1-5 until the overall loss function for both the first and second models converges. The overall loss function is determined by a weighted sum of the loss functions of the first and second models, while the loss function for the second model is determined based on its performance metrics. Convergence of the overall loss function indicates that the second model meets its performance metrics. It should be understood that when the overall loss functions for both models converge, the error (loss value) between the predictions of the first and second models and the actual results is small, thus meeting the prediction accuracy requirements.

[0135] The feature extraction layer is used to extract key information from the beam matrix, and the latent layer (hidden feature layer) is used to perform in-depth processing on the extracted information. Together, the feature extraction layer and the latent layer (hidden feature layer) constitute the transformation link from raw data (such as beam matrix) to effective features.

[0136] For example, the feature extraction layer could be a U-Net network, and the latent layer could refer to a hidden feature layer. The latent layer (hidden feature layer) could be a hash layer, for example. The quantization vector could be a hash vector, and the binary vector could be a hash code (hash value). The U-Net network is a segmentation model used to extract and analyze features from a beam matrix to obtain feature maps. Compared to other complex segmentation models, the U-Net network has a smaller parameter size, requires less computational power during training, and trains faster. The hash layer can be used to map the feature maps output by the U-Net network into fixed-length hash vectors. Hash vectors are more concise than the original data, improving the model's processing efficiency.

[0137] The overall loss function for the first and second models is determined by a weighted sum of the loss functions of the first and second models. The loss function of the first model includes a first function and a second function. The first function is used to activate the hash code with either a first value or a second value. For example, the first value can be non-zero (e.g., 1), and the second value can be zero. The first value indicates the beam corresponding to the hash code that is preserved (unmasked / unfiltered), and the second value indicates the beam corresponding to the hash code that is masked (filtered). The second function is used to ensure that the number of first values ​​(e.g., 1) in the hash code is sparse. The loss function of the second model is determined based on the performance metric of the second model. Convergence of the overall loss function indicates that the second model meets the performance metric of the second model. For example, the performance metric of the second model is MSE.

[0138] If the overall loss functions corresponding to the first and second models converge, the value of the second function (i.e., the number of first values ​​(e.g., 1) in the hash code) is the minimum number of beams in the first beam set obtained by the first model.

[0139] In this embodiment, the first model can be called a masking model or a masking module (or a masking module). The second model can be called a deduction model or a reasoning model (or a deduction module or a reasoning module). The following description uses the first model as the masking module and the second model as the deduction module as an example.

[0140] During the model training phase, both the masking module and the inference module can be trained simultaneously. Both modules are deployed on the UE side. The masking module is a masked autoencoder (MAE) designed to output a mask that can mask Set A (i.e., predict Set B based on Set A). The inference module is a reconstruction model designed to predict the signal quality of Set A.

[0141] For example, such as Figure 7 As shown, during the training phase, the input to the masking module (first model) is a matrix. The output of the masking module is a binary mask matrix. binary mask matrix With matrix Same type (i.e., two matrices have the same number of rows and the same number of columns). Where, matrix... This includes the signal quality of each beam in Set A, as measured. Binary mask matrix. It is a matrix composed of 0s and 1s, which retains the elements corresponding to "1" (e.g., ...). The "1" in the matrix (corresponding beam in the image), filtering out elements corresponding to "0" (e.g., ... The "0" in the matrix (The corresponding beam in the middle). and The product of Hadama is notated as This can be used as input to the inference module (second model). Inputting data into the inference model... Afterwards, the inference module can be based on Reconstructed . This indicates the signal quality of Set A predicted by the inference module.

[0142] Subsequently, in the derivation phase (model usage phase), referring to step 503, after the UE receives the beam information of Set B sent by the network device, the UE can measure the signal quality of each beam in Set B based on the beam information of Set B sent by the network device, and input the measured signal quality of each beam in Set B into the derivation module. The derivation module can reconstruct the model based on the signal quality of each beam in Set B. That is, the signal quality of Set A can be predicted through the deduction module. The masking module is no longer involved in this process.

[0143] For example, such as Figure 8 As shown, the specific training process for the occlusion module is as follows:

[0144] (I) Constructing the beam matrix corresponding to beam set Set A ,Right now It is an H-row, W-column matrix, defined as follows:

[0145] Equation (1);

[0146] in, express The Middle The RSRP of the nth beam, which is located within the matrix of the nth beam. Line number Column. Among them, The value range is 1 to For example, such as Figure 8 As shown, H can be 5 and W can be 7, i.e., the beam matrix. It can include 5 rows and 7 columns, with a total of 35 beams corresponding to RSRP.

[0147] (ii) Beam matrix The input to the U-net network is shown below, and the input-output relationship of the U-net network is as follows:

[0148] Equation (2);

[0149] in, This represents the output feature map of the U-net network. ,Right now It is an H-row, W-column matrix. The function representing the overall U-net network. These represent the parameters of the U-net network.

[0150] (iii) Latent layer, such as hash layer, receives features from the previous high-level layer (e.g., the output of U-net network). ) as input, through a having ( A fully connected structure with 10 neurons outputs a value of length 1. The vector is compressed using the sigmoid activation function to... The intervals form a continuous representation of the hash vector, as follows:

[0151] Equation (3);

[0152] in, for Activation vector in the latent layer, , This represents a K-dimensional binary array where each element takes the value 0 or 1. For the global function of the hash layer, These are the parameters for the hash layer.

[0153] The weight parameters between the U-net network and the potential layer (e.g., the hash layer) can be... .

[0154] (iv) Furthermore, the output of the latent layer is used to generate hash codes through a binarization layer:

[0155] Equation (4);

[0156] in, This represents the final generated hash code. , representing a K-dimensional binary array, It is a step function used to generate a mask that is either 0 or 1 using the extracted high-level features.

[0157] (v) Reshape into a mask matrix m, Reshaping rules and beam matrices The beam positions in the mask matrix m are kept consistent. That is, the order of each mask in the mask matrix m is consistent with the order of the beams corresponding to that mask. For example, as... Figure 8As shown, the mask matrix m can include 5 rows and 7 columns, with a total of 35 masks corresponding to the beams. The first mask (i.e., the mask in the first row and first column) is the beam matrix. The mask corresponding to the first beam (i.e., the beam corresponding to RSRP in the first row and first column); the second mask (i.e., the mask in the first row and second column) is the beam matrix. The mask corresponding to the first beam (i.e., the beam corresponding to RSRP in the first row and second column), and so on.

[0158] Wherein, the mask matrix m and the beam matrix The product of Hadama is :

[0159] Equation (5);

[0160] Then, the deduction module is used to calculate based on Reconstruct the complete The inference module can be masked by various methods. Under the conditions, according to Reconstruction (Prediction) In other words, the inference module can perform multiple calculations based on the output of the occlusion module. Iterative Refactoring .

[0161] For example, the overall loss function for the masking module and the inference module is as follows:

[0162] Equation (6);

[0163] in, and It is the loss function corresponding to the masking module. It is the loss function corresponding to the inference module. express coefficient, express coefficient, express The coefficient.

[0164] in, This represents the overall model parameters (the overall parameters of the occlusion module and the inference module). , The hash codes used to incentivize (encourage) the output of potential layers (e.g., hash layers) are close to 1 or 0. The goal is to incentivize (encourage) the output hash code of the latent layer (e.g., hash layer) to have as few 1s as possible, but not to eventually optimize to 0 (i.e., the number of 1s in the hash code will not be 0). In practice, it can be optimized to sparsity (many components are close to 0, and valid components are retained to satisfy the reconstruction constraints). The MSE error is used to ensure the robustness of the inference module during reconstruction. The three loss functions mentioned above can be used to construct the overall loss function based on a weighted summation to prevent unbalanced optimization.

[0165] At the end of training, when the overall loss function of the masking module and the inference module converges, The value of (i.e., the code weight of the output mask of the masking module) is the lower bound of the number of Set B beams acquired by the masking module (i.e., the minimum number of Set B beams acquired by the masking module). In this way, by using self-supervised learning, it is possible to mask as many beams in Set A as possible with the mask, thereby determining the lower bound of the number of Set B beams.

[0166] In some embodiments, during the training process described above (such as step (4)), the problem of nondifferentiability of the first norm (i.e., L1 norm) at 0 may occur, as shown in equation (7).

[0167] Equation (7);

[0168] in, Indicates to Take the partial derivative of the i-th element. This represents the L1 norm. hour, =1; hour, for ; When =0, Undefined. In this case, the subgradient method can be used to avoid the problem of the norm not being differentiable at 0.

[0169] Based on the method provided in this application, the terminal device can train the masking module and the inference module using self-supervised learning. The trained and converged masking module can be used to obtain a lower bound on the number of beams in Set B, and the trained and converged inference module can reconstruct the signal quality of Set A based on the output of the masking module, ensuring accuracy. The terminal device can obtain a lower bound on the number of beams in Set B based on the trained and converged masking module (i.e., the first model), determine a first quantity based on this lower bound (the first quantity is greater than or equal to the lower bound on the number of beams in Set B obtained by the masking module), and send first information (indicating the first quantity) to the network device. The network device can determine the number of beams in Set B based on the first information sent by the terminal device (e.g., the number of beams in Set B determined by the network device is greater than or equal to the first quantity indicated by the first information). Then, the terminal device receives the beam information of Set B from the network device and measures the signal quality of Set B based on this information. The signal quality of Set B is then input into the trained inference module so that the inference module can reconstruct the signal quality of Set A. This ensures that the number of beams in the Set B sent by the network device is neither too many nor too few, thereby improving model efficiency and reducing resource consumption while meeting the accuracy and robustness requirements of AI models (such as inference modules).

[0170] It should be understood that Figures 1 to 8 The illustrated diagrams are for illustrative purposes only and are not intended to limit the embodiments of this application to the examples shown. In fact, those skilled in the art can interpret the embodiments based on the examples depicted. Figures 1 to 8 The examples in the document can be transformed into equivalent ways to obtain more implementations.

[0171] The above text combined Figures 1 to 8 This document describes in detail the communication method provided in the embodiments of this application. The following will combine... Figures 9 to 10 The device embodiments of this application are described in detail below. It should be understood that the communication device of this application embodiment can execute the various communication methods of the foregoing embodiments of this application, that is, the specific working processes of the various products below can be referred to the corresponding processes in the foregoing method embodiments.

[0172] In the embodiments described above, the terminal device may execute some or all of the steps in each embodiment; the network device may execute some or all of the steps in each embodiment. These steps or operations are merely examples, and the embodiments of this application may also perform other operations or variations thereof. Furthermore, the steps may be executed in different orders as presented in the embodiments, and it is not necessary to execute all the operations in the embodiments of this application. Moreover, the sequence number of each step does not imply the order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0173] Figure 9 This is a schematic block diagram of a communication device provided in an embodiment of this application. Figure 9 As shown, the communication device 900 may include a processing module 910 and a communication module 920. The processing module 910 can read instructions and / or data from the storage module to enable the communication device 900 to implement the aforementioned method embodiments. The communication module 920 can implement corresponding communication functions, which can be internal communication functions of the communication device 900 or communication functions between the communication device 900 and other devices. Optionally, the communication module 920 may also be referred to as a communication interface or a transceiver module.

[0174] Optionally, the communication device 900 also includes a storage module that can be used to store instructions and / or data.

[0175] In one possible design, the communication device 900 may correspond to the terminal device (e.g., UE) in the above method embodiments, or a component (such as a circuit, chip, or chip system) configured in the terminal device. The communication device 900 can be used to perform the steps or processes performed by the terminal device in any of the above method embodiments.

[0176] For example, the communication module 920 is used to send first information to the network device, the first information being used to indicate a first quantity, the first quantity being determined by the terminal device based on the minimum number of beams in a first beam set obtained by a first model, the first model being a deep neural network model; the communication module 920 is also used to receive beam information of the first beam set from the network device, the number of beams in the first beam set being greater than or equal to the first quantity; the processing module 910 is used to measure the signal quality of the first beam set, input the signal quality of the first beam set into a second model, and predict the signal quality of the second beam set based on the second model.

[0177] In another possible design, the communication device 900 may correspond to a network device (e.g., a base station) in the above method embodiments, or a component (such as a circuit, chip, or chip system) configured in a network device. The communication device 900 can be used to perform the steps or processes performed by the network device in any of the above method embodiments.

[0178] For example, the communication module 920 is used to receive first information from the terminal device, the first information being used to indicate a first quantity, the first quantity being determined by the terminal device based on the minimum number of beams in the first beam set obtained by the terminal device based on the first model; the communication module 920 is also used to send beam information of the first beam set to the terminal device, the number of beams in the first beam set being greater than or equal to the first quantity.

[0179] The above are merely examples; for detailed steps or procedures, please refer to the descriptions in the foregoing embodiments.

[0180] Figure 10 This is another schematic block diagram of the communication device 1000 provided in the embodiments of this application. The communication device 1000 may be a chip, chip system, or processor, etc., used by a terminal device or network device to implement the above-described methods. The communication device 1000 can be used to implement the methods described in the above-described method embodiments; for details, please refer to the descriptions in the above-described method embodiments.

[0181] like Figure 10 As shown, the communication device 1000 may include one or more processors 1010, which may also be referred to as processing units or processing modules, and can implement certain control functions. The processor 1010 may be a general-purpose processor or a dedicated processor, such as a baseband processor or a central processing unit. The baseband processor can be used to process communication protocols and communication data, while the central processing unit can be used to control the communication device 1000 (e.g., a base station, baseband chip, user, user chip), execute software programs, and process data from the software programs.

[0182] In an alternative design, the processor 1010 may also store instructions and / or data that can be executed by the processor 1010 to cause the communication device 1000 to perform the methods described in the above method embodiments.

[0183] In another alternative design, the communication device 1000 may include a communication interface 1020 for implementing receiving and transmitting functions. For example, the communication interface 1020 may be a transceiver circuit, interface, interface circuit, or transceiver. The transceiver circuit, interface, interface circuit, or transceiver for implementing receiving and transmitting functions may be separate or integrated. The aforementioned transceiver circuit, interface, interface circuit, or transceiver may be used for reading and writing code / data, or it may be used for transmitting or relaying signals.

[0184] Optionally, the communication device 1000 may include one or more memories 1030, which may store instructions that can be executed on the processor 1010, causing the communication device 1000 to perform the methods described in the above method embodiments. Optionally, the memories 1030 may also store data. Optionally, the processor 1010 may also store instructions and / or data. The processor 1010 and the memories 1030 may be provided separately or integrated together.

[0185] It should be understood that, in one possible design, the steps in the method embodiments provided in this application can be implemented by integrated logic circuits in the processor's hardware or by instructions in software form. The steps of the methods disclosed in the embodiments of this application can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules in the processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, detailed descriptions are not provided here.

[0186] In one implementation, the communication device 1000 may correspond to the terminal device in the above method embodiments and may be used to execute the various steps and / or processes executed by the terminal device in the above method embodiments. The processor 1010 may be used to execute instructions stored in the memory 1030, and when the processor 1010 executes the instructions stored in the memory, the processor 1010 is used to execute the various steps and / or processes of the above method embodiments corresponding to the terminal device.

[0187] In another implementation, the communication device 1000 may correspond to the network device in the above method embodiments and may be used to execute the various steps and / or processes executed by the network device in the above method embodiments. The processor 1010 may be used to execute instructions stored in the memory 1030, and when the processor 1010 executes the instructions stored in the memory, the processor 1010 is used to execute the various steps and / or processes of the above method embodiments corresponding to the network device.

[0188] It should be understood that the aforementioned processing device can be one or more chips. For example, the processing device can be a field-programmable gate array (FPGA), an application-specific integrated circuit (ASIC), a system-on-chip (SoC), a central processor unit (CPU), a network processor (NP), a digital signal processor (DSP), a microcontroller unit (MCU), a programmable logic device (PLD), or other integrated chips.

[0189] It is understood that the memory in the embodiments of this application can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM). It should be noted that the memory used in the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0190] According to the method provided in the embodiments of this application, this application also provides a chip system, which includes one or more processors for calling and executing instructions stored in memory, thereby causing the method described in the embodiments of this application to be executed. The chip system may be composed of chips or may include chips and other discrete devices.

[0191] The chip system may include input circuits or interfaces for transmitting information or data, and output circuits or interfaces for receiving information or data.

[0192] According to the method provided in the embodiments of this application, this application also provides a communication system, which includes the aforementioned network device and terminal device.

[0193] According to the method provided in the embodiments of this application, this application also provides a computer program product, which includes: computer program code, which, when run on a computer, causes the computer to execute the various steps or processes executed by the network device or terminal device in any of the foregoing method embodiments.

[0194] According to the method provided in the embodiments of this application, this application also provides a computer-readable storage medium storing program code, which, when run on a computer, causes the computer to execute the various steps or processes executed by the network device or terminal device in any of the foregoing method embodiments.

[0195] The computer-readable storage medium may be the aforementioned volatile memory or non-volatile memory, or it may include both volatile memory and non-volatile memory.

[0196] In the embodiments of this application, the terms and English abbreviations are exemplary examples given for ease of description and should not be construed as limiting the application in any way. This application does not preclude the possibility of defining other terms that can achieve the same or similar functions in existing or future agreements.

[0197] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When these computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated.

[0198] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0199] It should be understood that in the various embodiments of this application, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0200] In summary, the above description is merely a preferred embodiment of the technical solution of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A beam prediction method, characterized in that, Applied to terminal devices, including: The terminal device sends first information to the network device. The first information is used to indicate a first quantity. The first quantity is determined by the terminal device based on the minimum number of beams in the first beam set obtained by the first model, which is a deep neural network model. The terminal device receives beam information of a first beam set from the network device, wherein the number of beams in the first beam set is greater than or equal to the first number. The terminal device measures the signal quality of the first beam set, inputs the signal quality of the first beam set into the second model, and predicts the signal quality of the second beam set based on the second model. Before the terminal device sends the first information to the network device, the terminal device receives the second information from the network device. The second information includes the beam information of the second beam set and / or the performance indicators of the second model. The terminal device measures the signal quality of the second beam set and trains the first model and the second model based on the signal quality of the second beam set. The first model includes a feature extraction layer and a latent layer, and training the first model and the second model based on the signal quality of the second beam set includes: Step 1: Input the beam matrix into the feature extraction layer to obtain the feature map output by the feature extraction layer. The beam matrix includes the signal quality of each beam in the second beam set. Step 2: Use the feature map as input to the latent layer to obtain the quantized vector output by the latent layer; Step 3: Binarize the quantization vector to obtain a binary vector; Step 4: Convert the binary vector into a mask matrix, wherein the order of each mask in the mask matrix is ​​the same as the order of the corresponding beam in the beam matrix; Step 5: Use the Hadamard product of the mask matrix and the beam matrix as input to the second model to obtain the signal quality of the second beam set predicted by the second model; Repeat steps 1-5 until the overall loss function corresponding to the first model and the second model converges. The overall loss function is determined by a weighted sum of the loss functions of the first model and the second model. The loss function of the second model is determined based on the performance index of the second model.

2. The method according to claim 1, characterized in that, The loss function of the first model includes a first function and a second function. The first function is used to activate the binary vector to be a first value or a second value, and the second function is used to activate the number of first values ​​in the binary vector to be sparse.

3. The method according to claim 1, characterized in that, The performance metrics for the second model include mean squared error (MSE).

4. The method according to claim 2, characterized in that, If the overall loss functions corresponding to the first model and the second model converge, the value of the second function is the minimum number of beams in the first beam set obtained by the first model.

5. The method according to any one of claims 1-4, characterized in that, The second information also includes the number of beams in the network-configurable first beam set, wherein the number of beams in the network-configurable first beam set includes at least one value. If the minimum number of beams in the first beam set obtained by the first model is less than the minimum of the at least one value, then the first number is the minimum of the at least one value. If the minimum number of beams in the first beam set obtained by the first model is greater than or equal to the minimum of the at least one value, the first number is the minimum value in the target set, and the target set includes one or more values ​​among the at least one value that are greater than or equal to the minimum number of beams in the first beam set obtained by the first model.

6. The method according to any one of claims 1-4, characterized in that, The terminal device sends the first information to the network device, including: The terminal device sends uplink control information (UCI) signaling to the network device, wherein the first information is carried in at least one field of the UCI signaling.

7. The method according to any one of claims 1-4, characterized in that, The terminal device receives the second information from the network device, including: The terminal device receives downlink control information (DCI) signaling from the network device, wherein the second information is carried in at least one field of the DCI signaling.

8. A communication system, characterized in that, It includes terminal devices and network devices, wherein the terminal devices perform the method as described in any one of claims 1-7.

9. A communication device, characterized in that, The communication device is a terminal device or a network device, and the communication device includes: a wireless communication module, a memory, and one or more processors; the wireless communication module, the memory, and the processor are coupled together. The memory is used to store computer program code, which includes computer instructions; when the computer instructions are executed by the processor, the communication device performs the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, Includes computer instructions; When the computer instructions are executed on a terminal device, the terminal device causes the terminal device to perform the method as described in any one of claims 1-7.

Images