Information processing method and information processing apparatus

Abstract

This application provides an information processing method and an information processing apparatus, applicable to the field of communications. In the technical solution of this application, information at a previous time step is first compressed based on a compression model, and then information at a later time step is predicted based on a prediction model and the compressed information at the previous time step, which helps to improve the robustness of the predicted information.

Description

Technical Field

[0001] This application relates to the field of communications, and more particularly to information processing methods and information processing apparatus. Background Technology

[0002] In communication systems, terminals and base stations typically need to exchange communication-related information. The following explanation will use the example of a terminal reporting communication-related information to the base station. One solution for a terminal to report relevant information is as follows: A prediction model and a compression model are deployed on the terminal, while a decompression model is deployed on the base station. The terminal uses historical relevant information to predict future relevant information using the prediction model, and then compresses the predicted information using the compression model before reporting the compressed information to the base station. The base station decompresses the received information (optionally, it also performs inverse quantization) to recover the terminal's predicted information and manages it based on this information.

[0003] For example, due to environmental changes, relative motion, and other factors, the channel between the base station and the terminal becomes time-varying. The scheduling configuration determined based on historical channel state information (CSI) reference signal (RS) may not be suitable for the actual channel during data transmission, which is known as channel aging. One solution to address the channel aging problem is as follows: deploy a CSI prediction model and a CSI compression model on the terminal, and deploy a CSI decoding model on the base station side; the terminal uses historical channel state information (CSI) to predict the CSI for future data transmission using the CSI prediction model, compresses the predicted CSI using the CSI compression model, and reports the compressed CSI to the base station; the base station decompresses the received CSI, recovers the terminal's predicted CSI, and performs scheduling configuration based on this CSI.

[0004] However, the above technical solutions suffer from poor robustness of the predicted information, which leads to poor accuracy of the information transmitted between the terminal and the base station, and consequently poor communication performance. Summary of the Invention

[0005] This application provides an information processing method and an information processing apparatus, which help improve the robustness of prediction results, thereby improving the accuracy of information transmitted between the terminal and the base station, and thus improving communication performance.

[0006] Firstly, this application provides an information processing method, which can be executed by a first device. The first device can be a communication device, or a device within the communication device (e.g., a module, communication module, circuit or chip responsible for communication functions (such as a modem chip, also known as a baseband chip, or a system-on-chip (SoC) chip containing a modem core or a system-in-package (SIP) chip), a chip system, or a processor), or a logical node, logical module, or software capable of implementing all or part of the functions of the communication device. As an example, the communication device is an access network device or a terminal.

[0007] This method includes: acquiring first information at a first time step; compressing the first information based on a first model to obtain compressed first information; and predicting first information at a second time step based on a second model and the compressed first information, wherein the second time step is located after the first time step in the time domain.

[0008] It is understood that the term "first information" in this application is a general term for a type of information, not referring to any specific piece of information. For example, the first information may be CSI information. In this application, the first model may also be called a compression model or encoder, and the second model may also be called a prediction model or predictor.

[0009] Analysis reveals that the compressed model is essentially a mapping from the original, uncompressed first information to a low-dimensional latent space (or representation space). This mapping preserves the structured characteristics of the first information (such as sparsity and low rank) while eliminating redundant information and unpredictable random factors like noise. Therefore, the first information output by the compressed model in the low-dimensional latent space is more robust than the original first information. The key to the prediction model is characterizing the time-varying properties of the first information. Predicting in the low-dimensional latent space allows the prediction model to focus on the time-varying patterns of the structured characteristics of the first information, resulting in predictions that are more robust than those in the original high-dimensional tensor space.

[0010] In existing technologies, the first step is to predict the first information of the original high-dimensional tensor space at a later time step based on the first information inherent in the original high-dimensional tensor space at a earlier time step. Then, this first information is compressed. Because the prediction model utilizes information beyond the time-varying structured features of the first information, including redundant information and unpredictable random factors such as noise, the prediction results suffer from low robustness. The prediction results are then input into a compression model for further compression. Since the prediction results are predictive and cannot reproduce the random factors contained in the original first information, the compression model cannot eliminate these random factors from the predicted first information, thus failing to ensure the robustness of the compressed predicted first information.

[0011] In the method of this application, the first information is first compressed, which can retain the structured features of the first information (such as sparsity, low rank, etc.) while discarding redundant information and unpredictable random factors such as noise. In this way, the first information in the low-dimensional latent space output by the compressed model is more robust than the original first information. Then, the compressed first information is used for prediction. Since the first information output by the compressed model is more robust than the original first information, the first information predicted by the prediction model based on the first information output by the compressed model is also more robust.

[0012] In some possible implementations, predicting the first information at the second time step based on the second model and compressed first information includes: predicting the first information at the second time step based on the first system state parameters of the second model and compressed first information. The first system state parameters include the system state parameters obtained after the second model is sequentially input with m compressed first information. These m compressed first information are the information obtained by the first model compressing the m first information corresponding to each of the m time steps. These m time steps are located before the first time step in the time domain, and m is a positive integer.

[0013] In this method, when predicting the first information at the second time based on the second model and the compressed first information, the first system state parameters include the system state parameters obtained after the second model is sequentially input with m compressed first information. These m compressed first information are obtained by compressing the first information at m historical time points. Therefore, the second model can make full use of the time-varying pattern of the structured features between the first information at multiple time points to predict more accurate first information, thereby improving the accuracy of the predicted first information.

[0014] In some possible implementations, this method further includes: obtaining a third model, which is used to decompress the compressed first information; and training the first and second models based on the third model.

[0015] In this application, the third model can be referred to as the decompression model. In this implementation, the first and second models are trained based on the third model, which can reduce the limitations of the training datasets for the first and second models, or in other words, increase the breadth of the training datasets for the first and second models, thereby enhancing the performance of the trained first and second models.

[0016] In some possible implementations, training the first and second models based on the third model includes: obtaining a training dataset containing first information at the third time step and first information at the fourth time step, the fourth time step being located after the third time step in the time domain; training the first and second models based on the third model and the training dataset, wherein the input of the first model contains the first information at the third time step, the input of the second model contains the output of the first model, the input of the third model contains the output of the second model, and the label corresponding to the output of the third model contains the first information at the fourth time step.

[0017] In this implementation, during the training of the first and second models, the loss only needs to be calculated once for each iteration. Compared with training the first and second models separately, which requires calculating the loss for each model separately, this reduces the computational load and improves training efficiency.

[0018] In some possible implementations, training the first and second models based on the third model includes: obtaining a training dataset containing first information at the third time step, first information at the fourth time step, and first information at the fifth time step, where the fifth time step is after the fourth time step in the time domain; training the first model based on the training dataset and the third model, wherein the input of the first model contains the first information at the third time step, the input of the third model contains the output of the first model, and the label of the output of the third model contains the first information at the third time step; and training the second model based on the training dataset, the third model, and the first model, wherein the input of the first model contains the first information at the fourth time step, the input of the second model contains the output of the first model, the input of the third model contains the output of the second model, and the label corresponding to the output of the third model contains the first information at the fifth time step.

[0019] Optionally, the fourth time step and the third time step can be the same time step. This can reduce the amount of data in the training dataset, thereby reducing the storage overhead of the first device. In addition, in scenarios where the training dataset is received from the second device, the transmission overhead of the training dataset can be reduced.

[0020] In this implementation, the first model is first trained based on the third model and the training dataset, and then the second model is trained based on the first model, the third model, and the training dataset.

[0021] Compared to the approach of simultaneously training the first and second models based on a third model, training the first and second models separately allows the first device to flexibly select their respective training datasets. It also facilitates the first device's monitoring and management of the training performance of the first and second models separately, ultimately contributing to improved performance. Furthermore, it allows the first device to reuse this training method in scenarios where only the first model needs to be used (i.e., compression without prediction).

[0022] In some possible implementations, the method further includes: acquiring a training dataset containing first information at a third time step and compressed first information at a fourth time step, the fourth time step being located after the third time step in the time domain; training a first model and a second model based on the training dataset, the first model receiving the first information at the third time step as input, the second model receiving the output of the first model as input, and the label of the output of the second model containing the compressed first information at the fourth time step.

[0023] This method, compared to training the first and second models based on the third model, can reduce the amount of computation and complexity.

[0024] In some possible implementations, this method further includes: acquiring a training dataset containing first information at a third time step, compressed first information at a third time step, compressed first information at a fourth time step, and compressed first information at a fifth time step, wherein the fifth time step is located after the fourth time step in the time domain; training a first model based on the training dataset, wherein the input of the first model contains the first information at a third time step, and the label corresponding to the output of the first model contains the compressed first information at a third time step; and training a second model based on the training dataset, wherein the input of the second model contains the compressed first information at a fourth time step, and the label corresponding to the output of the second model contains the compressed first information at a fifth time step.

[0025] In this implementation, optionally, the third and fourth time points can be the same time point, which can reduce the amount of data in the training dataset.

[0026] In some possible implementations, acquiring the training dataset includes receiving the training dataset. The first device obtains the training dataset by receiving it from other devices, which helps save resources consumed by the first device in collecting the training dataset itself and reduces the complexity of the first device.

[0027] In some possible implementations, the training dataset contains the first information at time 6 and the first differential quantization information, where the first differential quantization information is obtained by quantizing the first differential information, and the first differential information is the difference between the first information at time 7 and the first information at time 6.

[0028] In some possible implementations, the training dataset contains compressed first information and first differential quantization information at the eighth time step. The first differential quantization information is obtained by quantizing the first differential information. The first differential information is the difference between the compressed first information at the ninth time step and the compressed first information at the eighth time step.

[0029] In both implementation methods described above, the training data in the training dataset is represented using differential quantization, which reduces the storage space requirements of the training data and saves storage resources. It also saves transmission overhead in scenarios where the training dataset is obtained through transmission from other devices.

[0030] In some possible implementations, the training dataset includes a first sequence, a second sequence, and indication information. The first sequence contains t pieces of first information corresponding to t time points, and the second sequence contains s predicted and compressed pieces of first information corresponding to s time points. The indication information indicates at least one of the following: the length of the first sequence, the length of the second sequence, the correlation between the time points corresponding to the first information in the first sequence, the correlation between the time points corresponding to the predicted and compressed first information in the second sequence, and the correlation between the first information in the first sequence and the predicted and compressed first information in the second sequence, where s and t are both positive integers.

[0031] Secondly, this application provides an information processing method, which can be executed by a second device. The second device can be a communication device, or a device within the communication device (e.g., a module, communication module, circuit or chip responsible for communication functions (such as a modem chip, also known as a baseband chip, or a system-on-a-chip (SoC) chip containing a modem core, or a system-in-package (SIP) chip), chip system, or processor), or a logical node, logical module, or software capable of implementing all or part of the functions of the communication device. As an example, the communication device is an access network device or a terminal.

[0032] The method includes: determining a training dataset, wherein the training dataset contains first information and first differential quantization information at a sixth time step, the first differential quantization information being obtained by quantizing the first differential information, and the first differential information being the difference between the first information at a seventh time step and the first information at a sixth time step; and / or, the training dataset contains compressed first information and second differential quantization information at an eighth time step, the second differential quantization information being obtained by quantizing the second differential information, and the second differential information being the difference between compressed first information at a ninth time step and compressed first information at an eighth time step.

[0033] In some possible implementations, the training dataset includes a first sequence, a second sequence, and indication information. The first sequence contains t pieces of first information corresponding to t time points, and the second sequence contains s predicted and compressed pieces of first information corresponding to s time points. The indication information indicates at least one of the following: the length of the first sequence, the length of the second sequence, the correlation between the time points corresponding to the first information in the first sequence, the correlation between the time points corresponding to the predicted and compressed first information in the second sequence, and the correlation between the first information in the first sequence and the predicted and compressed first information in the second sequence, where s and t are both positive integers.

[0034] Thirdly, this application provides an information processing apparatus. This apparatus may be a communication device, or a device within a communication device (e.g., a module, communication module, circuit or chip responsible for communication functions (such as a modem chip, also known as a baseband chip, or a SoC chip or SIP chip containing a modem core), chip system, or processor), or a logical node, logical module, or software capable of implementing all or part of the functions of the communication device. As an example, the communication device is a terminal or access network device.

[0035] This device may include modules that perform the methods / operations / steps / actions described in the first aspect or any of the possible implementations thereof. These modules may be hardware circuits, software, or a combination of hardware circuits and software.

[0036] In one design, the device may include a processing module; further, it may also include a communication module. The communication module is used to perform the sending and receiving actions in the method described in the first aspect or any of its possible implementations above, while the processing module is used to perform the processing actions involved in the method described in the first aspect or any of its possible implementations above.

[0037] Fourthly, this application provides an information processing apparatus. This apparatus may be a communication device, or a device within a communication device (e.g., a module, communication module, circuit or chip responsible for communication functions (such as a modem chip, also known as a baseband chip, or a SoC chip or SIP chip containing a modem core), a chip system, or a processor), or a logical node, logical module, or software capable of implementing all or part of the functions of the communication device. As an example, the communication device is a terminal or access network device.

[0038] This device may include modules that perform the methods / operations / steps / actions described in the second aspect or any of the possible implementations thereof. These modules may be hardware circuits, software, or a combination of hardware circuits and software.

[0039] In one design, the device may include a processing module and a communication module. The communication module is used to perform the sending and receiving actions in the method described in the second aspect above or any possible implementation thereof, while the processing module is used to perform the processing actions involved in the method described in the second aspect above or any possible implementation thereof.

[0040] Fifthly, this application provides an information processing apparatus, including a processor, wherein instructions are executed by the processor to cause the method as described in the first aspect or any possible implementation thereof to be implemented.

[0041] Optionally, the device may further include a storage medium that stores the instructions executed by the processor.

[0042] In some implementations, the storage medium is integrated with the processor, for example, the storage medium is integrated into the processor.

[0043] In a sixth aspect, this application provides an information processing apparatus, including a processor, wherein instructions are executed by the processor to cause the method as described in the second aspect or any possible implementation thereof to be implemented.

[0044] Optionally, the device may further include a storage medium that stores the instructions executed by the processor.

[0045] In some implementations, the storage medium is integrated with the processor, for example, the storage medium is integrated into the processor.

[0046] In a seventh aspect, this application provides a chip including processing circuitry for running programs or instructions to implement methods as described in the first aspect or any possible implementation thereof.

[0047] Optionally, the chip may further include a memory for storing programs or instructions.

[0048] Optionally, the chip may also include the transceiver circuit, or an input / output interface.

[0049] Eighthly, this application provides a chip including processing circuitry for running programs or instructions to implement methods as described in the second aspect or any possible implementation thereof.

[0050] Optionally, the chip may further include a memory for storing programs or instructions.

[0051] Optionally, the chip may also include the transceiver circuit, or an input / output interface.

[0052] Ninthly, this application provides a computer-readable storage medium including instructions that, when executed by a processor, cause the method as described in the first aspect or any possible implementation thereof to be implemented.

[0053] In a tenth aspect, this application provides a computer-readable storage medium including instructions that, when executed by a processor, cause the method as described in the second aspect or any possible implementation thereof to be implemented.

[0054] In one aspect, this application provides a computer program product comprising computer program code or instructions that, when executed, cause the method as described in the first aspect or any of its possible implementations to be implemented.

[0055] In a twelfth aspect, this application provides a computer program product comprising computer program code or instructions that, when executed, cause the method as described in the second aspect or any of its possible implementations to be implemented.

[0056] In a thirteenth aspect, this application provides a communication system for performing the methods described in the first aspect above or any possible implementation thereof and for performing the methods described in the second aspect above or a corresponding implementation thereof.

[0057] It is understandable that the technical effects in any of the second to thirteenth aspects can be referenced from the technical effects in the corresponding aspect. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of a neuron structure according to an embodiment of this application;

[0059] Figure 2 This is a schematic diagram of an FNN structure according to an embodiment of this application;

[0060] Figure 3 This is a schematic diagram of a CSI feedback method according to an embodiment of this application;

[0061] Figure 4 This is a schematic diagram illustrating the pairing of two end models according to an embodiment of this application;

[0062] Figure 5 This is a schematic diagram illustrating AI-CSI prediction according to an embodiment of this application;

[0063] Figure 6 This is a schematic diagram illustrating the prediction and compression of CSI according to one embodiment of this application;

[0064] Figures 7 to 10This is a schematic diagram of a communication system according to several embodiments of this application;

[0065] Figure 11 This is a schematic diagram of a chip system architecture according to an embodiment of this application;

[0066] Figures 12 to 13 This is a flowchart illustrating the information processing method of several embodiments of this application;

[0067] Figure 14 This is a schematic diagram illustrating the feedback of first information according to one embodiment of this application;

[0068] Figures 15 to 18 This is a schematic diagram illustrating the model training methods of several embodiments of this application;

[0069] Figure 19 This is a schematic diagram illustrating the feedback of the first information according to another embodiment of this application;

[0070] Figures 20 to 21 This is a schematic diagram illustrating the model training methods of several embodiments of this application;

[0071] Figure 22 and Figure 23 This is a schematic diagram of the structure of an information processing apparatus according to several embodiments of this application. Detailed Implementation

[0072] To better introduce the technical solution of this application, some related concepts and technologies involved in this application will be introduced below.

[0073] Machine learning (ML) is an important technological approach to achieving artificial intelligence (AI). Machine learning can be divided into supervised learning, unsupervised learning, and reinforcement learning.

[0074] Supervised learning, based on collected sample values ​​and labels, uses machine learning algorithms to learn the mapping relationship between sample values ​​and labels, and expresses this learned mapping relationship using a machine learning model. The process of training the machine learning model is the process of learning this mapping relationship. For example, in signal detection, the noisy received signal is the sample, and the corresponding real constellation point is the label. Machine learning aims to learn the mapping relationship between samples and labels through training, that is, to enable the machine learning model to learn a signal detector. During training, the model parameters are optimized by calculating the error between the model's predicted values ​​and the real labels. Once the mapping relationship is learned, it can be used to predict the sample label of each new sample. The mapping relationship learned in supervised learning can include linear mappings and nonlinear mappings. Based on the type of label, the learning task can be divided into classification tasks and regression tasks.

[0075] Unsupervised learning relies solely on collected sample values, using algorithms to discover inherent patterns within the samples. One type of unsupervised learning algorithm uses the samples themselves as supervisory signals; that is, the model learns the mapping relationship from sample to sample, which is called self-supervised learning. During training, model parameters are optimized by calculating the error between the model's predictions and the samples themselves. Self-supervised learning can be used for signal compression and decompression recovery applications; common algorithms include autoencoders and generative adversarial networks.

[0076] Reinforcement learning, unlike supervised learning, is a type of algorithm that learns problem-solving strategies through interaction with the environment. Unlike supervised and unsupervised learning, reinforcement learning problems do not have explicit "correct" action labels. The algorithm needs to interact with the environment to obtain reward signals from the environment, and then adjust its decision actions to obtain a larger reward signal value. For example, in downlink power control, the reinforcement learning model adjusts the downlink transmission power of each user based on the total system throughput feedback from the wireless network, aiming to achieve a higher system throughput. The goal of reinforcement learning is also to learn the mapping relationship between the environment state and the optimal decision action. However, because the label of the "correct action" cannot be obtained in advance, the network cannot be optimized by calculating the error between the action and the "correct action." Reinforcement learning training is achieved through iterative interaction with the environment.

[0077] Deep neural networks (DNNs) are a specific implementation of machine learning. According to the general approximation theorem, neural networks can theoretically approximate any continuous function, thus enabling them to learn arbitrary mappings. Traditional communication systems rely on extensive expert knowledge to design communication modules, while DNN-based deep learning communication systems can automatically discover hidden pattern structures from large datasets, establish mapping relationships between data, and achieve performance superior to traditional modeling methods.

[0078] The idea behind DNNs originates from the neuronal structure of the brain. Figure 1 This is a schematic diagram of a neuron structure according to an embodiment of this application. Figure 1 As shown, each neuron performs a weighted summation of its input values, and the result of this weighted summation is used to generate the output through a nonlinear function. Specifically, assume the neuron's input is x = [x0, ..., x...]. n The weights corresponding to the inputs are w = [w0, ..., w0]. n The bias of the weighted summation is b. The nonlinear function can take many forms; one example is the max {0, x} maximum value function. The effect of a neuron's execution can be...

[0079] DNNs typically have a multi-layered structure, with each layer containing multiple neurons. The input layer processes the received values ​​through neurons and then passes them to the hidden layers. Similarly, the hidden layers then pass the computation results to the final output layer, producing the final output of the DNN.

[0080] DNNs typically have more than one hidden layer, and these hidden layers often directly affect the ability to extract information and fit functions. Increasing the number of hidden layers or widening the width of each layer can improve the function fitting ability of a DNN. The weights in each neuron are the parameters of the DNN network model. The model parameters are optimized through the training process, enabling the DNN network to extract data features and express mapping relationships. DNNs generally use supervised or unsupervised learning strategies to optimize model parameters.

[0081] Based on the way the network is constructed, DNNs can be divided into feedforward neural networks (FNN), convolutional neural networks (CNN), and recurrent neural networks (RNN). Figure 2 This is a schematic diagram of an FNN structure according to an embodiment of this application. Its characteristic is that the neurons in adjacent layers are completely connected to each other, which makes FNN usually require a large amount of storage space and result in high computational complexity.

[0082] CNNs are neural networks specifically designed to process data with a grid-like structure. For example, time-series data (discrete sampling along the time axis) and image data (two-dimensional discrete sampling) can both be considered grid-like data. CNNs do not use all the input information at once for computation; instead, they use a fixed-size window to extract a portion of the information for convolution operations, which significantly reduces the computational cost of model parameters. Furthermore, depending on the type of information extracted by the window (such as people and objects in an image representing different types of information), each window can use different convolution kernels, allowing CNNs to better extract features from the input data.

[0083] Recurrent Neural Networks (RNNs) are a type of distributed neural network (DNN) that utilizes feedback time-series information. Their inputs include the current input value and the system state at the previous time step. RNNs are well-suited for acquiring temporally correlated sequence features, and are particularly applicable to applications such as speech recognition and channel coding / decoding.

[0084] The FNN, CNN, and RNN mentioned above are common neural network structures, all built upon neurons. As introduced above, each neuron performs a weighted summation operation on its input values, and the result is passed through a nonlinear function to produce the output. We call the weights of the weighted summation operation and the nonlinear function in the neural network the parameters of the neural network. Taking a neuron with max{0,x} as the nonlinear function as an example, we perform... The parameters of the operated neuron are weights w = [w0, ..., w n The weighted summation bias is b, and the nonlinear function is max{0,x}. The parameters of all neurons in a neural network constitute the parameters of that neural network.

[0085] A training dataset is a collection of training samples, each serving as an input to the neural network. It's used for model training. The training dataset is one of the most crucial parts of machine learning; the training process essentially involves learning certain features from the training dataset to minimize the difference between the neural network's output and the ideal target value. Typically, even with the same network structure, neural networks trained on different training datasets will have different weights and outputs. Therefore, the composition and selection of the training dataset, to a certain extent, determine the performance of the trained neural network.

[0086] When AI models are deployed in an over-the-air (OTA) architecture, whether for offline or online model updates / training, data from the actual network deployment is required to form the dataset needed for model updates / training. A high-quality training dataset helps wireless communication AI algorithm design achieve greater performance gains and improves the generalization ability and robustness of the final algorithm across various scenarios. Conversely, a flawed training dataset can easily lead to inaccurate gain evaluation, model overfitting, weak generalization ability, and poor scenario adaptability, among other problems.

[0087] In communication systems, terminals and base stations typically need to exchange communication-related information. The following explanation will use the example of a terminal reporting communication-related information to the base station. One solution for a terminal to report relevant information is as follows: A prediction model and a compression model are deployed on the terminal, while a decompression model is deployed on the base station. The terminal uses historical relevant information to predict future relevant information using the prediction model, and then compresses the predicted information using the compression model before reporting the compressed information to the base station. The base station decompresses the received information (optionally, it also performs inverse quantization) to recover the terminal's predicted information and manages it based on this information.

[0088] However, the above technical solutions suffer from poor robustness of the predicted information, which leads to poor accuracy of the information transmitted between the terminal and the base station, and consequently poor communication performance.

[0089] Analysis reveals that a compressed model is essentially a mapping from the original information (i.e., uncompressed information) to a low-dimensional latent space (or representation space). This mapping preserves the structured characteristics of the information (such as sparsity and low rank) while eliminating redundant information and unpredictable random factors like noise. Therefore, the information output from the compressed model in the low-dimensional latent space is more robust than the original information. The key to predictive models is characterizing the time-varying properties of information. Predicting in the low-dimensional latent space allows the model to focus on the time-varying patterns of the information's structured characteristics, resulting in more robust predictions compared to those made in the original high-dimensional tensor space.

[0090] In the aforementioned technique, information about future time steps within the original high-dimensional tensor space is first predicted based on information inherent in earlier time steps within the original high-dimensional tensor space. Then, this information is compressed. Because the prediction model, in addition to utilizing the time-varying nature of the structured features of the information, also includes unpredictable random factors such as redundancy and noise, the prediction results suffer from low robustness. The prediction results are then input into a compression model for further compression. Since the prediction results are predictive and cannot reproduce the random factors contained in the original information, the compression model cannot eliminate these random factors from the predicted information, thus failing to ensure the robustness of the compressed prediction information.

[0091] The following example illustrates how a terminal reports CSI feedback to a base station in a communication system.

[0092] In communication systems, base stations need to acquire the CSI (Continuous Search Indicator) to determine the resources, MCS (Multi-Channel System), precoding, and other configurations for scheduling the downlink data channel of the UE. In TDD (Transient Directional Control) systems, due to the reciprocity of uplink and downlink channels, the base station can obtain the uplink CSI by measuring the uplink reference signal and then infer a relatively accurate downlink CSI, for example, using the uplink CSI as the downlink CSI. In FDD (Fulfilled Directional Control) systems, uplink and downlink reciprocity cannot be guaranteed. The downlink CSI is obtained by the UE measuring the downlink reference signal, such as CSI-RS or SSB. Therefore, the UE needs to generate a CSI report according to the protocol predefined method or the base station configuration and feed the CSI back to the base station so that it can acquire the downlink CSI.

[0093] In AI-CSI feedback, when the model is deployed at the base station, the base station needs to obtain the CSI-RS estimation results at the UE side as labels for training. The AE model consists of two sub-models: an encoder and a decoder. AE can generally refer to a network structure composed of two sub-models. The AE model can also be called a two-sided model, a dual-end model, or a collaborative model. The encoder and decoder of the AE are usually trained together and can be used in a mutually compatible manner.

[0094] CSI feedback can be implemented based on the AE model. Figure 3 This is a schematic diagram of a CSI feedback method according to an embodiment of this application. Figure 3 As shown, the terminal compresses the measured CSI (which may also include quantization) using an encoder to obtain the feedback CSI; the base station decompresses the terminal's feedback CSI using a decoder to obtain the recovered CSI. For the base station, the model's input is the feedback CSI, and its output is the recovered CSI; the model's training requires the terminal's measured CSI as the ground truth label for the recovered CSI.

[0095] In some implementations, model pairing between the base station and the terminal is achieved through model transfer. Figure 4 This is a schematic diagram illustrating the pairing of two end models according to one embodiment of this application. Figure 4 As shown, the base station, based on historically collected CSI data, including a first input and a second output, jointly trains the encoder and decoder locally. During the pairing process, the base station constructs a training dataset based on the encoder's input (i.e., the first input) and output (i.e., the first output) and transmits this training dataset to the terminal; the terminal uses this training dataset to train the encoder, enabling pairing of the base station and terminal's pairing models.

[0096] In communication systems, due to environmental changes, relative motion, and other factors, the channel between the base station and the terminal becomes time-varying. Scheduling configurations based on historical CSI (Continuous Channel Identity) may not be suitable for the actual channel during data transmission, a phenomenon known as channel aging. One solution to address channel aging is for the terminal to use historically collected CSI information to predict future CSI changes during data transmission and report the predicted CSI to the base station to guide scheduling configuration. This method is called AI-CSI prediction.

[0097] Figure 5 This is a schematic diagram illustrating AI-CSI prediction according to one embodiment of this application. Figure 5 As shown, the terminal acquires the historical CSI sequence, denoted as H. t-k ,…,H t The historical CSI sequence is input into the CSI predictor to obtain the CSI at the next time step, denoted as H. t+1, where k is a non-negative integer, t is an integer greater than or equal to k, and k and t can represent time.

[0098] In AI-CSI prediction, in some implementations, the prediction model is deployed on the terminal. In order for the terminal to obtain the prediction model, the base station can use data transmission, that is, build a training dataset based on the historical CSI sequence and the next time step (or the model prediction value), and pass the training dataset to the terminal; the terminal uses the training dataset to train the prediction model.

[0099] Information prediction and compression can be combined. Figure 6 This is a schematic diagram illustrating the predicted and compressed CSI of one embodiment of this application. Figure 6 As shown, the terminal first bases its data on historical CSI (H). t-k ,…,H t-2 H t-1 H t The next time step CSI (H) is obtained by predicting CSI using a predictor. t+1 ), and then H t+1 Or V obtained after singular value decomposition (SVD) t+1 Input encoder to obtain compressed CSI(C t+1 ), and compress the CSI(C t+1 The data is uploaded to the base station. The base station uses a decoder to recover the compressed CSI data. or

[0100] In the transmission of the corresponding training dataset, for the prediction model, the base station needs to transmit the complete CSI sequence (H) to the terminal. t-k ,…,H t H t+1 Information such as sequence length; for the encoder, the base station needs to transmit the measured CSI(H) to the UE. t ) and compressed CSI (C t The terminal uses the aforementioned training dataset to train the prediction model and encoder, respectively, and completes model deployment. However, the prediction model trained in this way has low robustness in predicting CSI; and the overhead of transmitting the training dataset is relatively large.

[0101] To address at least one of the aforementioned technical problems, this application proposes a new technical solution. The following sections describe the system architecture, method, and apparatus of this application.

[0102] In the description of the embodiments of this application, unless otherwise stated, " / " indicates that the objects before and after are in an "or" relationship. For example, A / B can mean A or B. "And / or" in this application is merely a description of the relationship between the related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. A and B can be singular or plural.

[0103] In the description of the embodiments of this application, unless otherwise stated, "a plurality of" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0104] In the description of the embodiments of this application, the words "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design options. Specifically, the use of the words "exemplary" or "for example" is intended to present the relevant concepts in a specific manner to facilitate understanding.

[0105] In the description of the embodiments of this application, the terms "information", "signal", "message", "channel", and "signaling" may sometimes be used interchangeably. It should be noted that when their distinctions are not emphasized, their intended meanings are matched.

[0106] In the description of the embodiments of this application, the terms "of", "corresponding (relevant)" and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing their distinction, their intended meanings are matched.

[0107] In the description of the embodiments of this application, the order of the process numbers 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.

[0108] In the description of the embodiments of this application, "preset," "predefined," or "preconfigured" can be implemented by pre-storing corresponding codes, tables, or other means that can be used to indicate relevant information in the device (e.g., including terminals and wireless access network devices), or by being pre-defined in a protocol. This application does not limit the specific implementation method. "Stored" can refer to storing in one or more memories. The one or more memories can be separate settings or integrated into an encoder or decoder, processor, or communication device. The one or more memories can also be partially separate settings and partially integrated into a decoder, processor, or communication device. The type of memory can be any form of storage medium, and this application does not limit this.

[0109] In the description of the embodiments of this application, "protocol" may refer to standard protocols in the field of communications, such as 3GPP LTE protocols (such as technical specification (TS) 36, i.e., the TS36 series of technical specifications), NR protocols (such as the TS38 series of technical specifications), and related protocols applied to future communication systems. This application does not limit this.

[0110] It is understood that in this application, "...when" and "if" both refer to the corresponding processing that will be carried out under certain objective circumstances, and are not limited to a specific time, nor do they require a judgment action to be performed during implementation.

[0111] It is understood that some optional features in the embodiments of this application can be implemented independently in certain scenarios without relying on other features, such as the current solution on which they are based, to solve the corresponding technical problems and achieve the corresponding effects. Alternatively, they can be combined with other features as needed in certain scenarios. Correspondingly, the apparatus given in the embodiments of this application can also implement these features or functions, which will not be elaborated here.

[0112] In this application, unless otherwise specified, the same or similar parts between the various embodiments can be referred to each other. In the various embodiments of this application, unless otherwise specified or there is a logical conflict, the terminology and / or descriptions between different embodiments are consistent and can be mutually referenced. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships. The following descriptions of the embodiments of this application do not constitute a limitation on the scope of protection of this application.

[0113] In this application, entity A sends information to entity B, either directly or indirectly through other entities. Similarly, entity B receives information from entity A, either directly or indirectly through other entities. Entities A and B can be RAN nodes or terminals, or modules within RAN nodes or terminals. Information transmission and reception can be between RAN nodes and terminals, such as between a base station and a terminal; between two RAN nodes, such as between a CU and a DU; or between different modules within a single device, such as between a terminal chip and other modules of the terminal, or between a base station chip and other modules of the base station.

[0114] It is understood that the network architecture and business scenarios described in the embodiments of this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided in the embodiments of this application. As those skilled in the art will know, with the evolution of network architecture and the emergence of new business scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0115] The technical solutions provided in this application can be applied to various communication systems, such as: 5th generation (5G) or new radio (NR) systems, long term evolution (LTE) systems, LTE frequency division duplex (FDD) systems, LTE time division duplex (TDD) systems, wireless local area network (WLAN) systems, satellite communication systems, future communication systems, or integrated systems of multiple systems. The technical solutions provided in this application can also be applied to device-to-device (D2D) communication, vehicle-to-everything (V2X) communication, machine-to-machine (M2M) communication, machine-type communication (MTC), and Internet of Things (IoT) communication systems or other communication systems. Furthermore, the terms "system" and "network" are interchangeable.

[0116] In a communication system, a network element can send signals to or receive signals from another network element. These signals can include information, signaling, or data. The term "network element" can also be replaced by an entity, network entity, device, communication equipment, communication module, node, communication node, etc. This disclosure uses a network element as an example. For instance, a communication system can include at least one terminal device and at least one network device. The network device can send downlink signals to the terminal device, and / or the terminal device can send uplink signals to the network device.

[0117] In the embodiments of this application, the terminal device may also be referred to as user equipment (UE), access terminal, user unit, user station, mobile station, mobile station, remote station, remote terminal, mobile device, user terminal, terminal, wireless communication device, user agent, or user apparatus.

[0118] Terminal devices can be devices that provide voice / data, such as handheld devices with wireless connectivity, in-vehicle devices, etc. Currently, examples of terminals include: mobile phones, tablets, laptops, PDAs, mobile internet devices (MIDs), wearable devices, virtual reality (VR) devices, augmented reality (AR) devices, wireless terminals in industrial control, wireless terminals in self-driving, wireless terminals in remote medical surgery, wireless terminals in smart grids, wireless terminals in transportation safety, wireless terminals in smart cities, wireless terminals in smart homes, cellular phones, cordless phones, session initiation protocol (SIP) phones, wireless local loop (WLL) stations, personal digital assistants (PDAs), handheld devices with wireless communication capabilities, computing devices or other processing devices connected to a wireless modem, wearable devices, terminal devices in 5G networks, or future public land mobile communication networks. Terminal devices in a network (PLMN), etc., are not limited to this in the embodiments of this application.

[0119] By way of example and not limitation, in this embodiment, the terminal device can also be a wearable device. Wearable devices, also known as wearable smart devices, are a general term for devices that utilize wearable technology to intelligently design and develop everyday wearables, such as glasses, gloves, watches, clothing, and shoes. Wearable devices are portable devices that are worn directly on the body or integrated into the user's clothing or accessories. Wearable devices are not merely hardware devices, but also achieve powerful functions through software support, data interaction, and cloud interaction. Broadly speaking, wearable smart devices include those that are feature-rich, large in size, and can achieve complete or partial functions without relying on a smartphone, such as smartwatches or smart glasses, as well as those that focus on a specific type of application function and require the use of other devices such as smartphones, such as various smart bracelets and smart jewelry for vital sign monitoring.

[0120] In this embodiment, the device for implementing the functions of the terminal device can be the terminal device itself, or it can be any device capable of supporting the terminal device in implementing those functions, such as a chip system. This device can be installed in or used in conjunction with the terminal device. In this embodiment, the chip system can be composed of chips or may include chips and other discrete components. This embodiment only uses the terminal device as an example to illustrate the device for implementing the functions of the terminal device, and does not constitute a limitation on the solution of this embodiment.

[0121] The network device in this application embodiment can be a device for communicating with a terminal device. This network device can also be called an access network device or a wireless access network device, such as a base station. In this application embodiment, the network device can refer to a radio access network (RAN) node (or device) that connects the terminal device to the wireless network. A base station can broadly encompass, or be replaced by, various names including: NodeB, evolved NodeB (eNB), next-generation NodeB (gNB), relay station, access point, transmitting and receiving point (TRP), transmitting point (TP), master station, auxiliary station, motor slide retainer (MSR) node, home base station, network controller, access node, wireless node, access point (AP), transmission node, transceiver node, baseband unit (BBU), remote radio unit (RRU), active antenna unit (AAU), remote radio head (RRH), central unit (CU), distributed unit (DU), radio unit (RU), positioning node, etc. A base station can be a macro base station, micro base station, relay node, donor node, or similar entities, or combinations thereof. A base station can also refer to a communication module, modem, or chip installed within the aforementioned equipment or apparatus. A base station can also be a mobile switching center, equipment performing base station functions in D2D, V2X, and M2M communications, network-side equipment in future communication networks, or equipment performing base station functions in future communication systems. A base station can support networks using the same or different access technologies. Optionally, a RAN node can also be a server, wearable device, vehicle, or in-vehicle equipment. For example, the access network equipment in vehicle-to-everything (V2X) technology can be a roadside unit (RSU). The embodiments of this application do not limit the specific technologies or equipment forms used in the network equipment.

[0122] Base stations can be fixed or mobile. For example, a helicopter or drone can be configured to act as a mobile base station, and one or more cells can move depending on the location of the mobile base station. In other examples, a helicopter or drone can be configured as a device to communicate with another base station.

[0123] In some deployments, the network devices mentioned in the embodiments of this application may be devices including CU, DU, or CU and DU, or devices with control plane CU nodes (central unit-control plane (CU-CP)) and user plane CU nodes (central unit-user plane (CU-UP)) and DU nodes. For example, the network devices may include gNB-CU-CP, gNB-CU-UP, and gNB-DU.

[0124] In some deployments, multiple RAN nodes collaborate to assist terminals in achieving wireless access, with different RAN nodes each implementing some of the base station's functions. For example, RAN nodes can be CUs, DUs, CU-CPs, CU-UPs, or RUs. CUs and DUs can be configured separately or included in the same network element, such as a BBU. RUs can be included in radio frequency equipment or radio frequency units, such as RRUs, AAUs, or RRHs.

[0125] RAN nodes can support one or more types of fronthaul interfaces. Different fronthaul interfaces correspond to DUs and RUs with different functions. If the fronthaul interface between the DU and RU is a common public radio interface (CPRI), the DU is configured to implement one or more baseband functions, and the RU is configured to implement one or more radio frequency functions. If the fronthaul interface between the DU and RU is another type of interface, relative to the CPRI, some downlink and / or uplink baseband functions, such as, for downlink, precoding, digital beamforming (BF), or one or more of inverse fast Fourier transform (IFFT) / cyclic prefix addition (CP), are moved from the DU to the RU; and for uplink, digital beamforming (BF), or one or more of fast Fourier transform (FFT) / cyclic prefix removal (CP), are moved from the DU to the RU. In one possible implementation, the interface can be an enhanced common public radio interface (eCPRI). Under the eCPRI architecture, the segmentation between DU and RU differs, corresponding to different categories (Cat) of eCPRI, such as eCPRI CatA, B, C, D, E, F.

[0126] Taking eCPRI CatA as an example, for downlink transmission, layer mapping is used as the dividing line. DU is configured to implement one or more functions preceding layer mapping (i.e., coding, rate matching, scrambling, modulation, and layer mapping), while other functions following layer mapping (e.g., resource element (RE) mapping, digital beamforming (BF), or one or more inverse fast Fourier transform (IFFT) / cyclic prefix (CP) addition) are moved to RU. For uplink transmission, de-RE mapping is used as the dividing line. DU is configured to implement one or more functions preceding de-mapping (i.e., decoding, rate matching de-matching, descrambling, demodulation, inverse discrete Fourier transform (IDFT), channel equalization, and de-RE mapping), while other functions following de-mapping (e.g., digital BF or fast Fourier transform (FFT) / CP removal) are moved to RU. It is understandable that the functional descriptions of the DU and RU corresponding to various types of eCPRI can be found in the eCPRI protocol, and will not be elaborated here.

[0127] In one possible design, the processing unit in the BBU used to implement baseband functions is called the baseband high (BBH) unit, and the processing unit in the RRU / AAU / RRH used to implement baseband functions is called the baseband low (BBL) unit.

[0128] In different systems, CU (or CU-CP and CU-UP), DU, or RU may have different names, but those skilled in the art will understand their meaning. For example, in an ORAN system, CU can also be called O-CU (open CU), DU can also be called O-DU, CU-CP can also be called O-CU-CP, CU-UP can also be called O-CU-UP, and RU can also be called O-RU. Any of the units among CU (or CU-CP, CU-UP), DU, and RU in this application can be implemented through software modules, hardware modules, or a combination of software modules and hardware modules.

[0129] In this embodiment, the apparatus for implementing the functions of a network device can be a network device itself; it can also be an apparatus capable of supporting the network device in implementing those functions, such as a chip system, hardware circuit, software module, or a hardware circuit plus a software module. This apparatus can be installed in the network device or used in conjunction with the network device. In this embodiment, the example of a network device being used to implement the functions of a network device is provided only and does not constitute a limitation on the solutions described in this embodiment.

[0130] Network devices and / or terminal devices can be deployed on land, including indoors or outdoors, handheld or vehicle-mounted; they can also be deployed on water; and they can also be deployed in the air on airplanes, balloons, and satellites. This application does not limit the scenario in which the network devices and terminal devices are located. Furthermore, terminal devices and network devices can be hardware devices, or software functions running on dedicated hardware or general-purpose hardware, such as virtualization functions instantiated on a platform (e.g., a cloud platform), or entities that include dedicated or general-purpose hardware devices and software functions. This application does not limit the specific form of the terminal devices and network devices.

[0131] In wireless communication networks, such as mobile communication networks, the services supported by the networks are becoming increasingly diverse, thus requiring increasingly diverse demands. For example, networks need to support ultra-high speeds, ultra-low latency, and / or massive connectivity. This characteristic makes network planning, network configuration, and / or resource scheduling increasingly complex. Furthermore, as network functions become more powerful, such as supporting higher spectrum, supporting higher-order multiple-input multiple-output (MIMO) technologies, supporting beamforming, and / or supporting beam management, network energy efficiency has become a hot research topic. These new demands, new scenarios, and new characteristics bring unprecedented challenges to network planning, operation, and efficient operation. To meet these challenges, artificial intelligence (AI) technology can be introduced into wireless communication networks to achieve network intelligence. To support AI technology in wireless networks, AI nodes may also be introduced.

[0132] Optionally, the AI ​​node can be deployed in one or more of the following locations within the communication system: access network devices, terminal devices, or core network devices, etc. Alternatively, the AI ​​node can be deployed independently, for example, in a location other than any of the aforementioned devices, such as in the host or cloud server of an over-the-top (OTT) system. The AI ​​node can communicate with other devices in the communication system, which can be, for example, one or more of the following: network devices, terminal devices, or core network elements, etc.

[0133] It is understood that this application does not limit the number of AI nodes. For example, when there are multiple AI nodes, these nodes can be divided based on function, such as different AI nodes being responsible for different functions.

[0134] It can also be understood that AI nodes can be independent devices, or they can be integrated into the same device to achieve different functions. Alternatively, they can be network elements in hardware devices, software functions running on dedicated hardware, or virtualization functions instantiated on a platform (e.g., a cloud platform). This application does not limit the specific form of the aforementioned AI nodes.

[0135] AI nodes can be AI network elements or AI modules.

[0136] AI modules are used to implement corresponding AI functions. AI modules deployed in different network elements can be the same or different. Depending on the parameter configuration, the AI ​​module can achieve different functions. The AI ​​module model can be configured based on one or more of the following parameters: structural parameters (e.g., at least one of the following: number of neural network layers, neural network width, inter-layer connections, neuron weights, neuron activation function, or biases in the activation function), input parameters (e.g., the type and / or dimension of the input parameters), or output parameters (e.g., the type and / or dimension of the output parameters). The biases in the activation function can also be referred to as the neural network biases.

[0137] An AI module can have one or more models. A model can infer an output, which includes one or more parameters. The learning, training, or inference processes of different models can be deployed on different nodes or devices, or they can be deployed on the same node or device.

[0138] Figure 7 This is a schematic diagram of a communication system according to an embodiment of this application. Figure 7 As shown, the communication system 100 may include at least one network device, such as Figure 7 The network device 110 shown; the communication system 100 may also include at least one terminal device, such as Figure 7 The terminal device 120 is shown. The network device 110 and the terminal device 120 can communicate via a wireless link. The various communication devices in this communication system, for example, the network device 110 and the terminal device 120, can communicate via multi-antenna technology.

[0139] In practical applications, this communication system may include multiple network devices or multiple terminal devices. This application does not limit the number of network devices and terminal devices included in the communication system.

[0140] Figure 8 This is a schematic diagram of a communication system according to another embodiment of this application. Compared to Figure 7 Regarding the communication system 100 shown, Figure 8 The communication system 200 shown also includes an AI network element 140. The AI ​​network element 140 is used to perform AI-related operations, such as building training datasets or training AI models.

[0141] In some implementations, network device 110 can send data related to the training of the AI ​​model to AI network element 140, which then constructs a training dataset and trains the AI ​​model. For example, the data related to the training of the AI ​​model may include data reported by terminal device 120. AI network element 140 can send the results of operations related to the AI ​​model to network device 110, which then forwards them to terminal device 120. For example, the results of operations related to the AI ​​model may include at least one of the following: a trained AI model, model evaluation results, or test results. Exemplarily, a portion of the trained AI model may be deployed on network device 110, and another portion on terminal device 120. Alternatively, the trained AI model may be deployed on network device 110. Or, the trained AI model may be deployed on terminal device 120.

[0142] Understandable. Figure 8 This illustration only takes the direct connection between AI network element 140 and network device 110 as an example. In other scenarios, AI network element 140 can also be connected to terminal device 120. Alternatively, AI network element 140 can be connected to both network device 110 and terminal device 120 simultaneously. Alternatively, AI network element 140 can also be connected to network device 110 through a third-party network element. This application embodiment does not limit the connection relationship between AI network element and other network elements.

[0143] In some implementations, the AI ​​network element 140 can be configured as a module in network devices and / or terminal devices, for example, configured in... Figure 7 In the network device 110 or terminal device 120 shown.

[0144] It should be noted that, Figure 7 and Figure 8 This is a simplified diagram for ease of understanding only. For example, the communication system may also include other devices, such as wireless relay devices and / or wireless backhaul devices, as well as core network devices. Figure 7 and Figure 8 It was not drawn in the middle.

[0145] Figure 9 This is a schematic diagram of a communication system according to yet another embodiment of this application. Figure 9As shown, network elements in a communication system are connected via interfaces (e.g., NG, Xn) or air interfaces. These network element nodes, such as core network equipment, access network nodes (RAN nodes), network equipment, terminals, or one or more devices in operation administration and maintenance (OAM), are equipped with one or more AI modules (for clarity, ...). Figure 9 Only one is shown in the image.

[0146] Network devices can function as a single RAN node or comprise multiple RAN nodes, such as CUs and DUs. The CU and / or DU may also be configured with one or more AI modules. Optionally, the CU may be further divided into CU-CP and CU-UP. One or more AI modules are configured in the CU-CP and / or CU-UP.

[0147] Figure 10 This is a schematic diagram of a communication system according to yet another embodiment of this application. Figure 10 As shown, the communication system includes a RAN intelligent controller (RIC). For example, the RIC could be... Figure 9 The AI ​​module shown is used to implement AI-related functions.

[0148] RICs include near-real-time RICs (near-RT RICs) and non-real-time RICs (non-RT RICs). Non-real-time RICs primarily process non-real-time information, such as data that is not sensitive to latency (with latency on the order of seconds). Real-time RICs primarily process near-real-time information, such as data that is relatively sensitive to latency (with latency on the order of tens of milliseconds).

[0149] Near real-time (NRT) RICs are used for model training and inference. For example, they are used to train AI models and then use those models for inference. NRT RICs can obtain network-side and / or terminal-side information from network devices (e.g., CUs, CU-CPs, CU-UPs, DUs, and / or RUs) and / or terminals. This information can be used as training data or inference data. Optionally, the NRT RIC can deliver inference results to network devices and / or terminals. Optionally, inference results can be exchanged between CUs and DUs, and / or between DUs and RUs. For example, the NRT RIC delivers inference results to the DU, and the DU sends them to the RU.

[0150] Non-real-time RICs are also used for model training and inference. For example, they can be used to train AI models and then use those models for inference. Non-real-time RICs can obtain network-side and / or terminal-side information from network devices (e.g., CUs, CU-CPs, CU-UPs, DUs, and / or RUs) and / or terminals. This information can be used as training data or inference data, and the inference results can be delivered to the network devices and / or terminals. Optionally, inference results can be exchanged between CUs and DUs, and / or between DUs and RUs; for example, a non-real-time RIC delivers the inference result to a DU, which then forwards it to an RU.

[0151] For example, near real-time RICs are set up in network devices (e.g., CU, DU), while non-real-time RICs are set up in OAM, cloud servers, core network devices, or other network devices. RICs can be trained by obtaining subsets from multiple end devices from network devices (e.g., CU, CU-CP, CU-UP, DU, and / or RU), recombining them into a training dataset #2, and training on the training dataset #2.

[0152] For example, near real-time RIC and non-real-time RIC can also be set up separately as a network element, and the network device can be a near real-time RIC or a non-real-time RIC.

[0153] Near real-time RICs and non-real-time RICs can also be configured as separate network elements. Optionally, near real-time RICs and non-real-time RICs can also be part of other devices. For example, near real-time RICs can be set in network devices (e.g., CU, DU), while non-real-time RICs can be set in OAM, cloud servers, core network devices, or other network devices.

[0154] Figure 11This is a schematic diagram of a chip system architecture according to an embodiment of this application. This chip system architecture can be used in access network devices and / or terminal devices. Input / output control is used to manage the input and output signals of the device; for example, input / output control can be represented as a modem, keyboard, mouse, touchscreen, etc. Input / output control may also be part of a processor. Communication control is used to manage the device's reception and transmission of signals; communication control may also be part of a processor. Receiver / transmitter is used to communicate with other devices; the receiver / transmitter may include a modem for modulating information or demodulating modulated information. Antenna is used to transmit or receive signals. Storage may include random access memory (RAM) or read-only memory (ROM); storage may be used to store code that can be executed by the processor to implement corresponding functions. Processors may include intelligent hardware devices such as general-purpose processors, digital signal processors (DSPs), central processing units (CPUs), field-programmable gate arrays (FPGAs), graphics processing units (GPUs), and neural processing units (NNs).

[0155] Figure 12 This is a schematic flowchart of an information processing method according to an embodiment of this application. The method includes steps S1210, S1220, and S1230. The method is executed by a first device.

[0156] It is understood that the first device in the embodiments of this application can be a communication device, or a device within a communication device (e.g., a module, communication module, circuit or chip responsible for communication functions (such as a modem chip, also known as a baseband chip, or a SoC chip or SIP chip containing a modem core), chip system, or processor), or a logical node, logical module, or software capable of implementing all or part of the functions of the communication device. The communication device corresponding to the first device can be referred to as the first communication device. As an example, the first communication device is a terminal or access network device.

[0157] As an example, the access network device in this embodiment is the access network device in any of the aforementioned communication systems or a module capable of implementing all or part of the functions of the access network device, and the terminal in this embodiment is the terminal in any of the aforementioned communication systems or a module capable of implementing all or part of the functions of the terminal.

[0158] As an example, the first communication device is a terminal.

[0159] S1210, the first device acquires the first information at the first moment.

[0160] It's understandable that "first information" refers to a general type of information, rather than being limited to the specific content carried by a particular piece of information. For example, "first information" can refer to CSI or SSB in general.

[0161] In some implementations, the first information represents information obtained through measurement. For example, the first information represents the measured CSI (or CSI obtained through SVD) or the measured SSB.

[0162] It is understandable that the first moment can contain one or more moments.

[0163] In some implementations, the first device acquiring the first information at a first moment may include: measuring the first information at the first moment.

[0164] S1220, the first device compresses the first information at the first moment based on the first model to obtain the compressed first information at the first moment.

[0165] As an example, the first model can be called an AI model or an ML model.

[0166] As an example, the first model is a DNN model.

[0167] In the embodiments of this application, the first model may be referred to as an encoder, a compression model, or a compressor.

[0168] It is understood that the first model is used to compress the input information and output the compressed information. The first device compresses the first information based on the first model to obtain the compressed first information, including: inputting the first information at a first moment into the first model, obtaining the output of the first model, and using the output of the first model as the compressed first information at the first moment.

[0169] S1230, the first device predicts the first information at the second time based on the second model and the compressed first information at the first time, the second time being located after the first time in the time domain.

[0170] As an example, the second model can be called an AI model or an ML model.

[0171] As an example, the second model is an RNN model or a "Transformer" model. When the second model is an RNN model, for example, the second model includes an LSTM network.

[0172] In the embodiments of this application, the second model may be referred to as a prediction model or predictor.

[0173] It is understood that the second model is used to predict information at subsequent times based on information at earlier times, and outputs the predicted information. The first device predicts the first information at the second time based on the second model and the compressed first information at the first time, including: inputting the compressed first information at the first time into the second model, obtaining the output of the second model, and using the output of the second model as the compressed first information at the second time.

[0174] In the method of this application embodiment, the first information is first compressed, which can retain the structured features of the first information (such as sparsity, low rank, etc.) while discarding unpredictable random factors such as redundant information and noise. In this way, the first information in the low-dimensional latent space output by the compressed model is more robust than the original first information. Then, the compressed first information is predicted. Since the first information output by the compressed model is more robust than the original first information, the first information predicted by the prediction model based on the first information output by the compressed model is also more robust.

[0175] In the embodiments of this application, such as Figure 13 As shown, optionally, it may also include: S1240, the first device sends the compressed first information at the second time to the second device; S1250, the second device decompresses the compressed first information at the second time based on the third model to obtain the first information at the second time.

[0176] It is understandable that the first information at the second moment can characterize the first information measured at the second moment.

[0177] As an example, the third model can be referred to as an AI model or an ML model. As an example, the third model is a DNN model. In embodiments of this application, the third model can be referred to as a decompression model or a decoder.

[0178] In some implementations, the prediction of the first information at the second time step based on the second model and the compressed first information at the first time step includes: predicting the first information at the second time step based on the first system state parameters of the second model and the compressed first information at the first time step. The first system state parameters include the system state parameters obtained after the second model is sequentially input with m compressed first information. The m compressed first information are the information obtained by the first model compressing the m first information corresponding to each of the m time steps, where the m time steps are located before the first time step in the time domain, and m is a positive integer. For specific implementation details, please refer to [reference needed]. Figure 14 Related content in China.

[0179] Figure 14 This is a schematic diagram of the feedback first information according to one embodiment of this application. Figure 14 It mainly describes the model's input, output, and the data transferred between models. Figure 14 Taking the first device as the terminal and the second device as the base station as an example. As an example, V represents the CSI obtained by SVD processing after measurement.

[0180] Figure 14 In the middle, V, C, The subscript indicates the chronological order of the information. For example, a larger subscript value indicates a later time; equal subscript values ​​indicate the same time. For instance, V2 represents the first piece of information at the second time point, and C2 represents the compressed first piece of information at the second time point. This represents the predicted, second-time-bound, compressed first information. This represents the recovered first information from the second time step. The encoder is a compression model used to compress the input first information. The predictor is a prediction model used to predict the first information from a later time step based on the input first information from a previous time step. The decoder is a decompression model used to decompress the input first information.

[0181] like Figure 14 As shown, the encoder on the terminal sequentially inputs V1, V2, ..., V t-2 V t-1 and V t t is a positive integer. When the encoder input is V1, the output is recorded as C1; when the encoder input is V2, the output is recorded as C2; and so on. t When the output is C, the output is denoted as C. t This step involves compressing the first information at a certain moment to obtain the compressed first information at that moment.

[0182] like Figure 14 As shown, the predictor on the terminal is input with C1, C2, ..., C in sequence. t-2 C t-1 and C t When the predictor input is C1, the predictor's system state parameter is denoted as W″1, and the output is denoted as... When the encoder input is C2, the predictor's system state parameter is denoted as W″2, and the output is denoted as... Similarly, when the encoder inputs Ct, the predictor's system state parameter is denoted as W″. t The output is denoted as This step represents predicting a compressed first information at a later time step based on compressed first information from one or more earlier time steps. For example, compressed first information C1 at the first time step can be used to predict compressed first information at the second time step. Based on the compressed first information C1 at the first time step and the compressed first information C2 at the second time step, the compressed first information at the third time step is predicted. And so on, based on the compressed first information C1, C2, ..., Ct from the first time point to the t-th time point. t-2 C t-1 and C t The compressed first information at time t+1 is predicted.

[0183] like Figure 14 As shown, the terminal will predict the... and It was reported to the base station. Understandable. and They can be sent together or reported in batches. For example, predicted... After that, the terminal can report. Alternatively, the terminal can predict and obtain multiple compressed first information pieces, and then report these multiple compressed first information pieces all at once in a single message.

[0184] like Figure 14 As shown, the decoder on the base station receives the following inputs sequentially. and Decoder input When, the output is recorded as Decoder input When, the output is recorded as And so on, encoder input When, the output is recorded as This step involves decompressing the compressed first information at a certain moment to recover the first information at that moment.

[0185] The following describes a method for the first device to acquire the first and second models. In some implementations, the first device acquires a third model and trains the first and second models based on the third model. This implementation helps to achieve end-to-end model alignment between the first and second devices.

[0186] When the first device trains the first model and the second model based on the third model, the first device can jointly train the first model and the second model based on the third model. This method can be referred to as the first joint training method. Alternatively, the first device can train the first model and the second model separately based on the third model. This method can be referred to as the first separate training method. Compared with the separate training method, the joint training method helps to reduce the amount of loss calculation and improve training efficiency.

[0187] In some implementations, the third model can be trained by the first device based on the training dataset transmitted by the second device. This implementation can be referred to as transmitting the third model based on the training dataset.

[0188] The first joint training method can be understood as follows: Obtain a training dataset containing the first information at time step three and the first information at time step four, with time step four occurring after time step three in the time domain; train the first and second models based on the third model and the training dataset. The input of the first model contains the first information from time step three, the input of the second model contains the output of the first model, the input of the third model contains the output of the second model, and the label corresponding to the output of the third model contains the first information from time step four. For specific implementation details, please refer to [reference needed]. Figure 15 Related content in China.

[0189] Figure 15 This is a schematic diagram of a first joint training method according to an embodiment of this application. Figure 15 In the example of the first device obtaining the third model through training, the first model is represented by an encoder, the second model by a predictor, and the third model by a decoder.

[0190] Figure 15 In the joint training method shown, the training dataset for training the decoder includes a first training data column and a second training data column. The first training data column contains multiple time-corresponding first pieces of information before compression, denoted as (V i V i The first information after decompression at time i is represented; the second training data column contains multiple compressed first information items corresponding one-to-one at multiple times, denoted as (C i ), C i This represents the first information before decompression at the i-th time; where i is a positive integer. In the embodiments of this application, the first information after decompression can also be understood as the first information before compression, and the first information before decompression can also be called the first information after compression.

[0191] The first and second training data columns constitute multiple sets of training data. Each set of training data contains the first information after compression and the first information before compression, which are associated at the same time. For example, the i-th set of training data is denoted as (V i C i ), indicating the position or chronological order of related times among multiple times.

[0192] It is understandable that the training dataset could be the training dataset used by the second device to train the third model.

[0193] like Figure 15 As shown in the first step, C in the i-th training data i The input is the decoder, and the output of the decoder is denoted as... This represents the first information obtained by the decoder after decompression; the first device is based on and V iThe loss between these parameters is used to adjust the model weights in the decoder, thereby training the decoder. It can be understood that V... i Used as the label corresponding to the output during the decoder training process.

[0194] After training to obtain the decoder, such as Figure 15 As shown in the second step, the training dataset for jointly training the encoder and predictor contains a third training data sequence and a fourth training data sequence. The third training data sequence is denoted as (V'1, V'2, ..., V'). t-1 V' t The fourth training data sequence is denoted as (V'2, V'3, ..., V'). t V' t+1 It can be understood that the difference of 1 between the index of the initial training data V'2 in the third training data sequence and the index of the initial training data V'1 in the fourth training data sequence is merely an example. This difference represents the number of time intervals between the times corresponding to these two initial training data. If this difference is denoted as r, it means that when predicting the first information at a later time, it must be based on at least r pieces of first information from earlier times, where r is a positive integer.

[0195] like Figure 15 As shown in the second step, the encoder sequentially inputs the third training data sequence (V'1, V'2, ..., V'). t-1 V' t The information in the code is used as the input to the predictor, and the output of the predictor is used as the input to the decoder. The decoder then outputs the corresponding information. The first device is based on the first output sequence. With the fourth training data sequence (V'2, V'3, ..., V') t V' t+1 The loss between the two sequences is used to adjust the model weights in the encoder and predictor. It can be understood that the information at index i in the fourth training data sequence is used as the label corresponding to the information at index i in the first output sequence, where i is an integer, greater than 1, and less than or equal to t+1.

[0196] Understandable. Figure 15 In the second step, while training the encoder and predictor, the model weights of the decoder can remain unchanged, i.e., in a frozen state. The dashed lines on the predictor represent the transfer of system state parameters.

[0197] In some implementations, the training data in the training dataset for jointly training the encoder and predictor includes: training data sent from the second device to the first device, and / or training data collected by the first device itself. The inclusion of training data collected by the first device in the second training dataset helps improve the performance of the encoder and predictor.

[0198] The first separate training method can be understood as follows: Obtain a training dataset containing the first information at time step three, time step four, and time step five, with time step five occurring after time step four in the time domain. Train a first model based on the training dataset and the third model, where the input of the first model contains the first information at time step three, the input of the third model contains the output of the first model, and the label corresponding to the output of the third model contains the first information at time step three. Train a second model based on the training dataset, the third model, and the first model, where the input of the first model contains the first information at time step four, the input of the second model contains the output of the first model, the input of the third model contains the output of the second model, and the label corresponding to the output of the third model contains the first information at time step five. A specific implementation of the first separate training method can be found in [reference needed]. Figure 16 Related content in China.

[0199] It is understandable that in some implementations, the fourth time step and the third time step are the same time step, which can reduce the amount of training data.

[0200] Figure 16 This is a schematic diagram of a first separate training method according to an embodiment of this application. Figure 16 The separate training method shown is the same as Figure 15 One difference in the joint training method shown is that after the decoder is trained in the first step, the first device trains the encoder only based on the decoder. After the encoder is trained, the predictor is trained based on the encoder and the decoder.

[0201] Figure 16 In the second step, the training dataset for training the encoder includes a third training data column, denoted as (V'1, V'2, ..., V'). t V' t+1 The training data in the first training data column consists of multiple sets of training data, each set containing a first piece of information related to time. For example, the i-th set of training data is denoted as (V'i), where i is a positive integer less than or equal to t+1, representing the position or chronological order of the associated time within multiple time periods. V'i from the i-th set of training data is input to the encoder, the encoder's output is used as the decoder's input, and the decoder's output is denoted as... This represents the first information obtained by the decoder after decompression; the first device is based on and V' i The loss between these factors is used to adjust the model weights in the encoder, thereby training the encoder. It can be understood that V' i It is also used as the label corresponding to the output of the decoder during encoder training.

[0202] Understandable. Figure 16 In the second step, when training the encoder, the model weights of the decoder can remain unchanged, i.e., in a frozen state.

[0203] Figure 16 The third step in Figure 15 The second step is similar, except that the first device is based on the first output sequence. With the fifth training data sequence (V'2, V'3, ..., V') t V' t+1 The loss between the two only adjusts the model weights in the predictor, while the model weights in the encoder remain frozen.

[0204] In some implementations of this application, the first device does not train the first and second models based on the third model, but directly trains the first and second models based on the training dataset. Compared with the aforementioned training method based on the third model, this direct training method is simpler and suitable for scenarios where the UE does not need to perform self-enhancement on the encoder.

[0205] When the first device directly trains the first model and the second model, the first device can jointly train the first model and the second model; this method can be referred to as the second joint training method. Alternatively, the first device can train the first model and the second model separately; this method can be referred to as the second separate training method. Compared with the separate training method, the joint training method helps to reduce the amount of loss calculation and improve training efficiency.

[0206] The second joint training method may include: acquiring a training dataset containing first information at time step three and compressed first information at time step four, with time step four occurring after time step three in the time domain; training a first model and a second model based on the training dataset, where the first model's input is the first information at time step three, and the second model's input includes the first model's output, with the label corresponding to the second model's output containing the compressed first information at time step four. A specific implementation of this second joint training method can be found in [reference needed]. Figure 17 The relevant content is shown below.

[0207] like Figure 17 As shown, the training dataset for jointly training the encoder and predictor includes a first training data sequence and a second training data sequence. The first training data sequence is denoted as (V'1, V'2, ..., V'). t-1 V' t The second training data sequence is denoted as (C2, C3, ..., C...). t C t+1 The encoder is sequentially input with the first training data sequence (V'1, V'2, ..., V'). t-1 V't The information in the () is used as the corresponding output as the input to the predictor, and the predictor outputs sequentially. The first device is based on the output sequence. With the second training data sequence (C2, C3, ..., C t C t+1 The loss between the two parameters is used to adjust the model weights in the encoder and predictor. It can be understood that the information at index i in the second training data sequence is used as the label corresponding to the information at index i in the output sequence, where i is an integer, greater than 1, and less than or equal to t+1.

[0208] The second separate training method may include: acquiring a training dataset containing: first information at time 3, compressed first information at time 3, compressed first information at time 4, and compressed first information at time 5, with time 5 occurring after time 4 in the time domain; training a first model based on the training dataset, where the input of the first model contains the first information at time 3, and the label corresponding to the output of the first model contains the compressed first information at time 3; and training a second model based on the training dataset, where the input of the second model contains the compressed first information at time 4, and the label corresponding to the output of the second model contains the compressed first information at time 5. A specific implementation of this second separate training method can be found in [reference needed]. Figure 18 The relevant content is shown below.

[0209] like Figure 18 As shown, the training datasets for training the encoder and predictor separately contain a first training data sequence and a second training data sequence. The first training data sequence is denoted as (V1, V2, ..., V...). t-1 V t The second training data sequence is denoted as (C1, C2, C3, ..., C...). t C t+1 ).

[0210] In the first step, the encoder sequentially inputs the first training data sequence (V1, V2, ..., V...). t-1 V t The information in ) is output sequentially. The first device based on the output sequence With the sixth training data sequence (C1, C2, C3, ..., C... t The loss between t and t is used to adjust the model weights in the encoder. It can be understood that the information at index i in the sixth training data sequence is used as the label corresponding to the information at index i in the output sequence, where i is a positive integer less than or equal to t.

[0211] In the second step, the predictor is input sequentially (C1, C2, C3, ..., C...). t ), output sequentially The first device is based on the output sequence. With the seventh training data sequence (C2, C3, ..., C t C t+1 The loss between t and t is used to adjust the model weights in the predictor. It can be understood that the information at index i in the seventh training data sequence is used as the label corresponding to the information at index i in the output sequence, where i is an integer, greater than 1, and less than or equal to t+1.

[0212] The following is an introduction based on Table 1. Figures 15 to 18 This involves some relevant information about the training dataset related to the training methods used in the training process.

[0213] Table 1

[0214]

[0215] As can be seen from the foregoing, in some implementations, in order to construct suitable training data based on the first training data sequence and / or the second training data sequence, it is necessary to know at least one of the following information: the length of the first training data sequence, the length of the second training data sequence, the correlation between the times corresponding to the first information in the first training data sequence, the correlation between the times corresponding to the predicted and compressed first information in the second training data sequence, and the correlation between the first information in the first training data sequence and the predicted and compressed first information in the second training data sequence.

[0216] In the implementation of the second device sending the training dataset to the first device, the second device may optionally also send instruction information to the first device, which indicates at least one of the above-mentioned information.

[0217] During training, due to the inherent properties of RNNs, some implementations only require the sequence used as input to the model to be complete, while the sequence used as labels can be incomplete. Therefore, if the sequence used as labels is missing some terms, the first device can still train normally, but the corresponding loss function term will be missing; if the sequence used as input to the model is missing some terms, the first device can still train by truncating continuous subsequences.

[0218] In some implementations, the training dataset contains the first information at the sixth time step and the first differential quantization information. The first differential quantization information is obtained by quantizing the first differential information, and the first differential information is the difference between the first information at the seventh time step and the first information at the sixth time step.

[0219] In some implementations, the training dataset contains compressed first information and second differential quantization information at time eight. The second differential quantization information is obtained by quantizing the second differential information and is the difference between compressed first information at time nine and compressed first information at time eight.

[0220] The initial information corresponding to adjacent time intervals (before compression, after compression, and prediction) may be relatively stable; for example, their differences may exhibit a sparse or low-rank structure. Therefore, differential quantization of the training data sequence can further reduce the storage and transmission overhead of the model training data.

[0221] To measure CSI sequence S V = <V1,V2,…,V t For example, the difference V between any time and the next time can be expressed as... i -V i-1 Quantification And Replace V i Transmission is performed. After the terminal receives the data, it... Inverse quantization is performed with the quantization parameters, and combined with V. i-1 Can restore the measurement of CSI V i Differential quantization can be performed iteratively, for example, for measuring CSI sequences S. V The sequence constructed using the first CSI and subsequent quantized differences. Replacement; where each difference can use the same quantization method, or different quantization methods can be used separately or in groups, and the corresponding quantization parameters need to be transmitted to the terminal along with the quantized CSI sequence. For the feedback CSI sequence S C Similarly, differential quantization can be used.

[0222] Differential quantization methods can choose between scalar quantization or vector quantization. Scalar quantization maps each individual amplitude or phase value to a discrete value. Its advantage is simplicity, but it's less efficient for high-dimensional data. The quantization parameters for scalar quantization are quantization bits, scaling factor, and zero-point, or equivalently, quantization bits and truncation intervals. Vector quantization combines multiple scalar data into a single vector for overall quantization. It divides the vector space into regions and maps the input vector to the nearest codeword. Its advantage is utilizing the correlation across multiple dimensions, resulting in higher quantization efficiency, but it incurs significant overhead in transmitting quantization parameters, and the quantization and dequantization operations are relatively complex. The quantization parameters for vector quantization are quantization bits and the quantization codebook.

[0223] In some implementations of this application, the predictor and encoder are fused to perform prediction in the latent space. The fusion of the predictor and encoder can be understood as follows: the encoder directly outputs the latent variable associated with the first information at the next moment (i.e., the predicted and compressed first information) based on the current and historical first information. In other words, the encoder implements both compression and prediction functions.

[0224] In the implementation of predictor and encoder fusion, the model resulting from the fusion of the predictor and encoder is called the fused encoder or the fused first model. This first model is used for both compression and prediction, achieving both simultaneously. This simultaneous compression and prediction can be understood as meaning that compression and prediction cannot be strictly separated in the data processing. An example of the fused first model is an RNN. For an exemplary implementation of feeding back first information based on the fused first model, please refer to [reference needed]. Figure 19 The relevant content in [the document / document].

[0225] Figure 19 This is a schematic diagram of the first feedback information according to another embodiment of this application. Figure 19 It mainly describes the model's input, output, and the data transferred between models. Figure 19 Taking the first device as the terminal and the second device as the base station as an example. As an example, V represents the CSI obtained by SVD processing after measurement.

[0226] like Figure 19 As shown, the encoder fused on the terminal sequentially inputs V1, V2, ..., V t-2 V t-1 and V t t is a positive integer. When the fused encoder input V1, the system state parameters are denoted as W1, and the output is denoted as C2; when the encoder input V2, the system state parameters are denoted as W2, and the output is denoted as C3; and so on, the fused encoder input V t At that time, the system state parameter is denoted as W. t The output is denoted as C. t+1 This step involves compressing the first information from the current moment and the previous few moments to predict the compressed first information for the next moment.

[0227] For example, based on the first information V1 at the first time step, the compressed first information C2 at the second time step is predicted; based on the first information V2 at the second time step and the system state parameter W1 at the first time step, the compressed first information C3 at the third time step is predicted; and so on, based on the first information V1 at the t-th time step... t The system state parameters are denoted as W. t The compressed first information C at time t+1 is predicted. t+1.

[0228] like Figure 19 As shown, the terminal will predict C2, ..., C t-1 C t and C t+1 Reported to the base station. This is understandable; C2, ..., C t-1 C t and C t+1 They can be sent together or reported in batches. For example, after predicting C2, the terminal can report C2; or, after predicting multiple compressed first information, the terminal can report all of these compressed first information in one message.

[0229] like Figure 19 As shown, the decoder on the base station receives C2, ..., C in sequence. t-1 C t and C t+1 When the decoder input is C2, the system state parameter is denoted as W′2, and the output is denoted as... When the decoder input is C3, the system state parameter is denoted as W′3, and the output is denoted as... And so on, encoder input C t+1 When this happens, the system state parameter is denoted as W′. t+1 The output is denoted as This step involves decompressing the compressed first information at a certain moment to recover the first information at that moment.

[0230] In the implementation of predictor and encoder fusion, the first device can train a fused first model based on the third model, which can be referred to as the third joint training method; or, the first device can directly train a fused first model, which can be referred to as the third separate training method.

[0231] Figure 20 This is a schematic diagram of a third joint training method according to an embodiment of this application. Figure 20 In the example of the first device obtaining the third model through training, the first model is represented by an encoder and the third model is represented by a decoder.

[0232] Figure 20 In the joint training method shown, the training dataset for training the decoder includes the first training data column (C2, C3, ..., C...). t C t+1 ) and the second training data column (V2, V3, ..., V t V t+1 ).

[0233] It is understandable that the training dataset could be the training dataset used by the second device to train the third model.

[0234] like Figure 20 As shown in the first step, the decoder is sequentially input with the first training data sequence (C2, C3, ..., C...). t C t+1 The decoder outputs the information in the sequence. The first device is based on the output sequence. With the second training data sequence (V2, V3, ..., V... t V t+1 The loss between the two is used to adjust the model weights in the decoder.

[0235] like Figure 20 As shown in the second step, the fused encoder is sequentially input into the third training data sequence (V'1, V'2, ..., V'). t-1 V' t The information in the () is used as the corresponding output as the input to the decoder, and the decoder outputs in sequence. The first device is based on the output sequence. With the second training data sequence (V2, V3, ..., V... t V t+1 The model weights in the encoder are adjusted by adjusting the loss between the two.

[0236] Understandable. Figure 20 In the second step, when training the fused encoder, the model weights of the decoder can remain unchanged, i.e., in a frozen state.

[0237] In some implementations, the training data in the training dataset of the jointly trained fused encoder includes: training data sent from the second device to the first device, and / or training data collected by the first device itself. The inclusion of training data collected by the first device in the second training dataset helps to improve the performance of the fused encoder.

[0238] Figure 21 This is a schematic diagram of a third separate training method according to an embodiment of this application. Figure 21 In the joint training method shown, the training dataset for training the fusion unit includes the first training data column (V'1, V'2, ..., V'). t-1 V' t ) and the second training data column (C2, C3, ..., C t C t+1 ).

[0239] like Figure 21 As shown, the fused encoder is sequentially input with the first training data sequence (V'1, V'2, ..., V'). t-1 V' t The information in ) is output sequentially. The first device is based on the output sequence. With the second training data sequence (C2, C3, ..., C t C t+1 The model weights in the encoder are adjusted by adjusting the loss between the two.

[0240] The following is an introduction based on Table 2. Figures 20 to 21 This involves some relevant information about the training dataset related to the training methods used in the training process.

[0241] Table 2

[0242]

[0243] As can be seen from the foregoing, in some implementations, in order to construct suitable training data based on the first training data sequence and / or the second training data sequence, it is necessary to know at least one of the following information: the length of the first training data sequence, the length of the second training data sequence, the correlation between the times corresponding to the first information in the first training data sequence, the correlation between the times corresponding to the predicted and compressed first information in the second training data sequence, and the correlation between the first information in the first training data sequence and the predicted and compressed first information in the second training data sequence.

[0244] In the implementation of the second device sending the training dataset to the first device, the second device may optionally also send instruction information to the first device, which indicates at least one of the above-mentioned information.

[0245] The training dataset used to train the fusion encoder can also be represented by differential quantization. For implementation details, please refer to the previous content.

[0246] Figure 22 and Figure 23 This is a schematic diagram of the structure of an information processing apparatus according to an embodiment of this application. These apparatuses can be used to implement the functions implemented by the first or second apparatus in the above method embodiments, and therefore can also achieve the beneficial effects of the above method embodiments.

[0247] As an example, this device may be an access network device, or a device within an access network device (e.g., a module, communication module, circuit or chip responsible for communication functions (such as a modem chip, also known as a baseband chip, or a SoC chip or SIP chip containing a modem core), chip system or processor), or a logical node, logical module or software that can implement all or part of the functions of the access network device.

[0248] As an example, the device may be a terminal, or a device within the terminal (e.g., a module, a communication module, a circuit or chip responsible for communication functions (such as a modem chip, also known as a baseband chip, or a SoC chip or SIP chip containing a modem core), a chip system, or a processor), or a logical node, logical module, or software that can implement all or part of the terminal functions.

[0249] like Figure 22 As shown, the information processing apparatus 1600 includes a processing unit 1610 and a transceiver unit 1620. The apparatus 1600 is used to implement the functions implemented by the first apparatus or the second apparatus in any of the above method embodiments.

[0250] As an example, when the device 1600 is used to implement the function implemented by the first device in any of the above method embodiments, the processing unit 1610 is used to: acquire first information at a first time; compress the first information based on a first model to obtain compressed first information; predict the first information at a second time based on a second model and the compressed first information, wherein the second time is located after the first time in the time domain.

[0251] In some possible implementations, when the processing unit 1610 is used to predict the first information at the second time based on the second model and the compressed first information, it is specifically used to: predict the first information at the second time based on the first system state parameters of the second model and the compressed first information. The first system state parameters include the system state parameters obtained after the second model is sequentially input with m compressed first information. These m compressed first information are the information obtained by the first model compressing the m first information corresponding to each of the m time points. These m time points are located before the first time point in the time domain, and m is a positive integer.

[0252] In some possible implementations, the processing unit 1610 is also used to: obtain a third model, which is used to decompress the compressed first information; and to train the first model and the second model based on the third model.

[0253] In some possible implementations, when the processing unit 1610 trains the first and second models based on the third model, it specifically performs the following: acquiring a training dataset containing first information at the third time step and first information at the fourth time step, the fourth time step being located after the third time step in the time domain; training the first and second models based on the third model and the training dataset, wherein the input of the first model contains the first information at the third time step, the input of the second model contains the output of the first model, the input of the third model contains the output of the second model, and the label corresponding to the output of the third model contains the first information at the fourth time step.

[0254] In some possible implementations, when the processing unit 1610 trains the first and second models based on the third model, it is specifically used to: train the first and second models based on the third model, including: acquiring a training dataset containing first information at a third time step, first information at a fourth time step, and first information at a fifth time step, wherein the fifth time step is located after the fourth time step in the time domain; training the first model based on the training dataset and the third model, wherein the input of the first model contains the first information at the third time step, the input of the third model contains the output of the first model, and the label of the output of the third model contains the first information at the third time step; and training the second model based on the training dataset, the third model, and the first model, wherein the input of the first model contains the first information at the fourth time step, the input of the second model contains the output of the first model, the input of the third model contains the output of the second model, and the label corresponding to the output of the third model contains the first information at the fifth time step.

[0255] In some possible implementations, the processing unit 1610 is further configured to: acquire a training dataset containing first information at a third time step and compressed first information at a fourth time step, the fourth time step being located after the third time step in the time domain; train a first model and a second model based on the training dataset, the first model receiving the first information at the third time step as input, the second model receiving the output of the first model as input, and the label of the output of the second model containing the compressed first information at the fourth time step.

[0256] In some possible implementations, the processing unit 1610 is further configured to: acquire a training dataset containing first information at a third time step, compressed first information at a third time step, compressed first information at a fourth time step, and compressed first information at a fifth time step, wherein the fifth time step is located after the fourth time step in the time domain; train a first model based on the training dataset, wherein the input of the first model contains the first information at a third time step, and the label corresponding to the output of the first model contains the compressed first information at a third time step; and train a second model based on the training dataset, wherein the input of the second model contains the compressed first information at a fourth time step, and the label corresponding to the output of the second model contains the compressed first information at a fifth time step.

[0257] In some possible implementations, the transceiver unit 1620 is used to receive the training dataset.

[0258] As an example, when the device 1600 is used to implement the function implemented by the second device in any of the above method embodiments, the transceiver unit 1620 is used to: send the training dataset; the processing unit 1610 is used to: decompress the first information based on the third model.

[0259] For a more detailed description of the operations performed by the processing unit 1610 and the transceiver unit 1620, please refer to the relevant descriptions in the foregoing method embodiments.

[0260] like Figure 23 As shown, the information processing device 1700 includes a processor 1710 and an interface circuit 1720. The processor 1710 and the interface circuit 1720 are coupled to each other. It is understood that the interface circuit 1720 can be a transceiver or an input / output interface. Optionally, the device 1700 may also include a memory 1730 for storing instructions executed by the processor 1710, or storing input data required by the processor 1710 to execute instructions, or storing data generated after the processor 1710 executes instructions. Sometimes, the interface circuit 1720 can also be understood as part of the processor 1710, in which case the device 1700 includes the processor 1710.

[0261] As an example, when the device 1700 is used to implement any of the aforementioned methods, the processor 1710 is used to implement the functions of the processing unit 1610, and the interface circuit 1720 is used to implement the functions of the transceiver unit 1620.

[0262] As an example, when the above-mentioned device is a chip used in a communication device, the chip receiving information can be understood as the information being received first by other modules (such as an RF module or antenna) in the communication device, and then sent to the chip by these modules. Similarly, the chip sending information can be understood as the information being first sent to other modules (such as an RF module or antenna) in the communication device, and then sent by these modules.

[0263] In some embodiments of this application, a computer program product is also provided, which, when run on a processor, can implement the method implemented by the first device in any of the above embodiments.

[0264] In some embodiments of this application, a computer program product is also provided, which, when run on a processor, can implement the method implemented by the second device in any of the above embodiments.

[0265] In some embodiments of this application, a computer-readable storage medium is also provided, which contains computer instructions that, when executed on a processor, can implement the method implemented by the first device in any of the above embodiments.

[0266] In some embodiments of this application, a computer-readable storage medium is also provided, which contains computer instructions that, when executed on a processor, can implement the method implemented by the second device in any of the above embodiments.

[0267] In some embodiments of this application, a communication system is also provided, which can implement the methods implemented by the first device and the second device in any of the above method embodiments.

[0268] It is understood that the processor in the embodiments of this application can be a central processing unit, or other general-purpose processors, digital signal processors, application-specific integrated circuits, field-programmable gate arrays, or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. A general-purpose processor can be a microprocessor or any conventional processor.

[0269] In this embodiment of the application, the processor may include one or more of the following: a central processing unit (CPU), an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a microprocessor unit (MPU), a micro controller unit (MCU), a graphics processing unit (GPU), a field-programmable gate array (FPGA), an artificial intelligence processor (AI processor), or a neural processing unit (NPU).

[0270] In this application embodiment, the memory may include, but is not limited to, cache, read-only memory (ROM), random access memory (RAM), synchronous dynamic random access memory (SDRAM), hard disk drive (HDD) or solid-state drive (SSD), erasable programmable read-only memory (EPROM), or compact disc read-only memory (CD-ROM), etc. Memory is any other medium capable of carrying or storing desired program code having an instruction or data structure form and accessible by a computer, but is not limited thereto. The memory in this application embodiment may also be a circuit or any other device capable of implementing storage functions for storing computer programs or instructions, and / or data.

[0271] The method steps in the embodiments of this application can be implemented in hardware or in software instructions executable by a processor. The software instructions can consist of corresponding software modules, which can be stored in random access memory, flash memory, read-only memory, programmable read-only memory, erasable programmable read-only memory, electrically erasable programmable read-only memory, registers, hard disks, portable hard disks, optical discs, or any other form of storage medium well known in the art. An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. The storage medium can also be a component of the processor. The processor and the storage medium can reside in an application-specific integrated circuit (ASIC). Alternatively, the ASIC can reside in a base station or terminal. The processor and the storage medium can also exist as discrete components in the base station or terminal.

[0272] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer programs or instructions. When the computer program or instructions are loaded and executed on a computer, the processes or functions described in the embodiments of this application are performed entirely or partially. The computer can be a general-purpose computer, a special-purpose computer, a computer network, a network device, a user equipment, or other programmable device. The computer program or instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another. For example, the computer program or instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium, such as a floppy disk, hard disk, or magnetic tape; it can also be an optical medium, such as a digital video optical disc; or it can be a semiconductor medium, such as a solid-state drive. The computer-readable storage medium may be a volatile or non-volatile storage medium, or may include both types of storage media.

[0273] In the various embodiments of this application, unless otherwise specified or in case of logical conflict, the terminology and / or descriptions of different embodiments are consistent and can be referenced by each other. The technical features of different embodiments can be combined to form new embodiments according to their inherent logical relationship.

Claims

1. An information processing method characterized by comprising: include: Obtain the first information at the first moment; The first information is compressed based on the first model to obtain the compressed first information; The first information at a second time point is predicted based on the second model and the compressed first information, and the second time point is located after the first time point in the time domain.

2. The method of claim 1, wherein, The prediction of the first information at the second time step based on the second model and the compressed first information includes: Based on the first system state parameters of the second model and the compressed first information, the first information at the second time moment is predicted. The first system state parameters include the system state parameters obtained after the second model sequentially inputs m compressed first information. The m compressed first information are the information obtained by the first model compressing the m first information corresponding to each of the m time moments. The m time moments are located before the first time moment in the time domain, and m is a positive integer.

3. The method according to claim 1 or 2, characterized in that, The method further includes: Obtain a third model, which is used to decompress the compressed first information; The first and second models are trained based on the third model.

4. The method of claim 3, wherein, The training of the first model and the second model based on the third model includes: Obtain a training dataset containing the first information at a third time step and the first information at a fourth time step, wherein the fourth time step is located after the third time step in the time domain; The first model and the second model are trained based on the third model and the training dataset, wherein the input of the first model contains the first information at the third time step, the input of the second model contains the output of the first model, the input of the third model contains the output of the second model, and the label corresponding to the output of the third model contains the first information at the fourth time step.

5. The method of claim 3, wherein, The training of the first model and the second model based on the third model includes: Obtain a training dataset, which includes the first information at the third time step, the first information at the fourth time step, and the first information at the fifth time step, wherein the fifth time step is located after the fourth time step in the time domain; The first model is trained based on the training dataset and the third model, wherein the input of the first model contains the first information at the third time step, the input of the third model contains the output of the first model, and the label corresponding to the output of the third model contains the first information at the third time step; The second model is trained based on the training dataset, the third model, and the first model, wherein the input of the first model includes the first information at the fourth time step, the input of the second model includes the output of the first model, the input of the third model includes the output of the second model, and the label corresponding to the output of the third model includes the first information at the fifth time step.

6. The method of claim 1, wherein, The method further includes: Obtain a training dataset containing the first information at a third time step and the compressed first information at a fourth time step, wherein the fourth time step is located after the third time step in the time domain; The first model and the second model are trained based on the training dataset. The input of the first model is the first information at the third time step. The input of the second model includes the output of the first model. The label corresponding to the output of the second model includes the compressed first information at the fourth time step.

7. The method of claim 1, wherein, The method further includes: Obtain a training dataset, the training dataset containing the first information at a third time step, the compressed first information at the third time step, the compressed first information at a fourth time step, and the compressed first information at a fifth time step, the fifth time step being located after the fourth time step in the time domain; The first model is trained based on the training dataset. The input of the first model includes the first information at the third time step. The label corresponding to the output of the first model includes the compressed first information at the third time step. The second model is trained based on the training dataset. The input of the second model includes the compressed first information at the fourth time step, and the label corresponding to the output of the second model includes the compressed first information at the fifth time step.

8. The method according to any one of claims 4 to 7, characterized in that, The step of obtaining the training dataset includes: receiving the training dataset.

9. The method according to any one of claims 4 to 8, characterized in that, The training dataset includes the first information at the sixth time step and the first differential quantization information. The first differential quantization information is obtained by quantizing the first differential information. The first differential information is the difference between the first information at the seventh time step and the first information at the sixth time step. And / or, The training dataset includes compressed first information and second differential quantization information at the eighth time step. The second differential quantization information is obtained by quantizing the second differential information. The second differential information is the difference between compressed first information at the ninth time step and compressed first information at the eighth time step.

10. The method according to any one of claims 4 to 9, characterized in that, The training dataset includes a first sequence, a second sequence, and indication information. The first sequence contains t pieces of first information corresponding to t time points. The second sequence contains s predicted and compressed pieces of first information corresponding to s time points. The indication information indicates at least one of the following: the length of the first sequence, the length of the second sequence, the correlation between the time points corresponding to the first information in the first sequence, the correlation between the time points corresponding to the predicted and compressed first information in the second sequence, and the correlation between the first information in the first sequence and the predicted and compressed first information in the second sequence, where s and t are both positive integers.

11. An information processing apparatus, characterized by comprising: Includes a processing module, the processing module being used for: Obtain the first information at the first moment; The first information is compressed based on the first model to obtain the compressed first information; The first information at a second time point is predicted based on the second model and the compressed first information, and the second time point is located after the first time point in the time domain.

12. The apparatus of claim 11, wherein, The processing module is specifically used for: Based on the first system state parameters of the second model and the compressed first information, the first information at the second time moment is predicted. The first system state parameters include the system state parameters after the second model sequentially inputs m compressed first information. The m compressed first information are the information obtained by the first model compressing the m first information corresponding to each of the m time moments. The m time moments are located before the first time moment in the time domain, and m is a positive integer.

13. The apparatus of claim 11 or 12, wherein, The processing module is also used for: Obtain a third model, which is used to decompress the compressed first information; The first and second models are trained based on the third model.

14. The apparatus of claim 13, wherein, The device also includes an acquisition module; The acquisition module is used to: acquire a training dataset, the training dataset containing the first information at a third time step and the first information at a fourth time step, the fourth time step being located after the third time step in the time domain; The processing module is specifically used for: The first model and the second model are trained based on the third model and the training dataset, wherein the input of the first model contains the first information at the third time step, the input of the second model contains the output of the first model, the input of the third model contains the output of the second model, and the label corresponding to the output of the third model contains the first information at the fourth time step.

15. The apparatus of claim 13, wherein, The device further includes an acquisition module: The acquisition module is used to: acquire a training dataset, the training dataset containing the first information at the third time step, the first information at the fourth time step, and the first information at the fifth time step, wherein the fifth time step is located after the fourth time step in the time domain; The processing module is specifically used for: The first model is trained based on the training dataset and the third model, wherein the input of the first model contains the first information at the third time step, the input of the third model contains the output of the first model, and the label corresponding to the output of the third model contains the first information at the third time step; The second model is trained based on the training dataset, the third model, and the first model, wherein the input of the first model includes the first information at the fourth time step, the input of the second model includes the output of the first model, the input of the third model includes the output of the second model, and the label corresponding to the output of the third model includes the first information at the fifth time step.

16. The apparatus of claim 11, wherein, The device further includes an acquisition module: The acquisition module is used to: acquire a training dataset, the training dataset containing the first information at a third time step and the compressed first information at a fourth time step, the fourth time step being located after the third time step in the time domain; The processing module is also used for: The first model and the second model are trained based on the training dataset. The first model is input to the measurement CSI at the third time step, and the output of the first model is used as the input of the second model. The label corresponding to the output of the second model contains the compressed first information at the fourth time step.

17. The apparatus of claim 11, wherein, The device further includes an acquisition module: The acquisition module is used to: acquire a training dataset, the training dataset containing the first information at a third time step, the compressed first information at the third time step, the compressed first information at a fourth time step, and the compressed first information at a fifth time step, wherein the fifth time step is located after the fourth time step in the time domain; The processing module is also used for: The first model is trained based on the training dataset. The input of the first model includes the first information at the third time step, and the label corresponding to the output of the first model includes the first information compressed at the third time step. The second model is trained based on the training dataset. The input of the second model includes the compressed first information at the fourth time step, and the label corresponding to the output of the second model includes the compressed first information at the fifth time step.

18. The apparatus of any one of claims 14-17, wherein, The acquisition module is specifically used to receive the training dataset.

19. The apparatus according to any one of claims 14 to 18, characterized in that, The training dataset includes the first information at the sixth time step and the first differential quantization information. The first differential quantization information is obtained by quantizing the first differential information. The first differential information is the difference between the first information at the seventh time step and the first information at the sixth time step. And / or, The training dataset includes compressed first information and second differential quantization information at the eighth time step. The second differential quantization information is obtained by quantizing the second differential information. The second differential information is the difference between compressed first information at the ninth time step and compressed first information at the eighth time step.

20. The apparatus according to any one of claims 14 to 19, characterized in that, The training dataset includes a first sequence, a second sequence, and indication information. The first sequence contains t pieces of first information corresponding to t time points. The second sequence contains s predicted and compressed pieces of first information corresponding to s time points. The indication information indicates at least one of the following: the length of the first sequence, the length of the second sequence, the correlation between the time points corresponding to the first information in the first sequence, the correlation between the time points corresponding to the predicted and compressed first information in the second sequence, and the correlation between the first information in the first sequence and the predicted and compressed first information in the second sequence, where s and t are both positive integers.

21. The method according to any one of claims 1 to 10 or the apparatus according to any one of claims 11 to 20, characterized in that, The first information includes Channel State Information (CSI).

22. An information processing device, characterized in that, The device includes a processor coupled to a memory for storing program instructions, and the processor for executing the program instructions in the memory to implement the method as described in any one of claims 1 to 10.

23. A computer-readable storage medium, characterized in that, Used to store a computer program, the computer program including instructions for implementing the method as described in any one of claims 1 to 10.

24. A computer program product, the computer program product comprising instructions, characterized in that, When the instructions are executed on a computer, the computer causes the computer to perform the method as described in any one of claims 1 to 10.

Images