Communication method, communication device, and computer-readable storage medium

Abstract

The application provides a communication method, a communication device and a computer readable storage medium. In the method, a drift detection model can be called to process first CSI information to obtain a first drift detection result, and first information indicating that a terminal updates a compression model can be sent to the terminal based on the first drift detection result. It can be seen that the network equipment can timely trigger the terminal to update the compression model based on the first drift detection result. With the update of the compression model on the terminal side, the CSI compression model in the communication system is also updated. The performance of the updated CSI compression model is better than that of the CSI compression model before the update, which can effectively avoid the performance degradation of the CSI compression model and improve the resource scheduling and transmission efficiency of the communication system.

Description

Technical Field

[0001] This application relates to the field of communication technology, and in particular to a communication method, communication device and computer-readable storage medium. Background Technology

[0002] In wireless communication systems, Channel State Information (CSI) reflects the characteristics of the wireless channel and is crucial for communication transmission. However, CSI data can be very large, consuming significant transmission resources. Therefore, a CSI compression model can be introduced to reduce the amount of CSI data transmitted.

[0003] CSI compression models can only be trained on a limited set of channels. However, a trained CSI compression model will be exposed to a wide variety of channels. Due to changes in atmospheric scattering and / or terminal movement, the real-time channel distribution may not match the channel distribution during training, resulting in data drift. This data drift can degrade the performance of the CSI compression model, thereby affecting the resource scheduling and transmission efficiency of the communication system. Summary of the Invention

[0004] This application provides a communication method, a communication device, and a computer-readable storage medium, which can effectively avoid the performance degradation of the CSI compression model and improve the resource scheduling and transmission efficiency of the communication system.

[0005] In a first aspect, embodiments of this application provide a communication method. Optionally, this method can be executed by a network device or by a device compatible with the network device, such as a processor, chip, or chip system. Taking a network device as an example, the method may include: the network device decompressing CSI information reported by a terminal based on a compression model using a decompression model to obtain first CSI information, and then calling a drift detection model to process the first CSI information to obtain a first drift detection result. Finally, the network device sends first information to the terminal based on the first drift detection result. Wherein, the first CSI information represents the channel distribution at the current time; the first drift detection result characterizes the degree of deviation of the channel distribution corresponding to the first CSI information from the channel distribution corresponding to the second CSI information; the second CSI information is CSI information used to train the CSI compression model; the CSI compression model includes a decompression model and a compression model; and the first information instructs the terminal to update the compression model.

[0006] The method provided in the first aspect allows network devices to detect the deviation of the current channel distribution from the channel distribution during CSI compression model training using a drift detection model. Based on this deviation, the network device can promptly trigger the terminal to update the compression model. Compared to schemes that trigger terminal updates when the performance of the CSI compression model deteriorates (e.g., when a performance degradation is detected based on posterior metrics such as reconstruction error), the method described in the first aspect allows network devices to promptly trigger terminal updates based on changes in channel distribution before the performance of the CSI compression model deteriorates. As the terminal-side compression model is updated, the CSI compression model within the communication system is also updated accordingly. The performance of the updated CSI compression model is better than that of the unupdated CSI compression model, effectively preventing performance degradation of the CSI compression model and improving resource scheduling and transmission efficiency of the communication system.

[0007] In one possible implementation, the step of sending first information to the terminal is performed when the drift probability in the first drift detection result is within a first probability range. The drift probability refers to the probability that the channel distribution corresponding to the first CSI information deviates from the channel distribution corresponding to the second CSI information. Therefore, in this implementation, the network device can detect data drift when the drift probability is within the first probability range and promptly trigger the terminal to update the compression model, effectively avoiding performance degradation of the CSI compression model due to data drift.

[0008] In another possible implementation, the network device sends first information to the terminal based on the first drift detection result, including: the network device determining, from the first drift detection result and multiple drift detection results within a first time period prior to the current time, the number M of drift detection results with drift probabilities within a second probability range; if the number M is greater than or equal to N, the network device sends the first information to the terminal; M is a natural number, and N is an integer greater than 1. It can be seen that in this implementation, the network device can detect data drift when the number M of drift detection results with drift probabilities within the second probability range is greater than or equal to N, and promptly trigger the terminal to update the compression model. Compared to schemes that trigger terminal updates to the compression model simply because a drift detection result with a drift probability within the second probability range exists, this implementation can delay triggering the terminal to update the compression model, avoiding oversensitivity of the communication system to slight drifts.

[0009] In another possible implementation, the method further includes: the network device sending second information to the terminal based on the first drift detection result, the second information being used to instruct the terminal to increase the CSI feedback period. Thus, in this implementation, the network device can instruct the terminal to increase the CSI feedback period. When a drift detection result is obtained within one CSI feedback period, as the CSI feedback period increases, the interval at which the network device obtains the drift detection result increases. Therefore, the interval at which the network device triggers the terminal to update the compressed model based on the drift detection result also increases accordingly, providing more buffer time for model updates. Simultaneously, the frequency of CSI information exchange between the terminal and the network device decreases as the CSI feedback period increases, effectively reducing signaling overhead.

[0010] In another possible implementation, the network device sends second information to the terminal based on the first drift detection result. This includes: the network device determining, from the first drift detection result and multiple drift detection results within a first time period prior to the current time, the number M of drift detection results with drift probabilities falling within a second probability range; if the number M is less than N, the network device sends the second information to the terminal; M is a natural number, and N is an integer greater than 1. Therefore, in this implementation, the network device can instruct the terminal to increase the CSI feedback period when the number M of drift detection results with drift probabilities falling within the second probability range is less than N. With one drift detection result obtained within one CSI feedback period, as the CSI feedback period increases, the interval at which the network device obtains the drift detection result increases. Consequently, the interval at which the network device triggers the terminal to update the compression model based on the drift detection result also increases, which is beneficial for providing more buffer time for updating the compression model.

[0011] In another possible implementation, the drift detection model includes a feature extraction layer and a domain discrimination layer. The network device calls the drift detection model to process the first CSI information to obtain a first drift detection result, including: the network device extracting features from the first CSI information based on the feature extraction layer to obtain the feature vector corresponding to the first CSI information; and the network device discriminating the feature vector corresponding to the first CSI information based on the domain discrimination layer to obtain the drift probability. It can be seen that in this implementation, the drift detection model can obtain the drift probability through the feature extraction layer and the domain discrimination layer.

[0012] In another possible implementation, the drift detection model is a trained domain-adaptive adversarial network (AAN) model. Before the network device processes the first CSI information using the drift detection model to obtain the first drift detection result, the method further includes: the network device acquiring multiple second CSI information pieces and the performance values ​​of each second CSI information piece; the network device determining the domain label of each second CSI information piece based on the performance values ​​and performance thresholds of each second CSI information piece; and the network device training the domain-adaptive AAN model based on the multiple second CSI information pieces and their domain labels to obtain the trained domain-adaptive AAN model. Therefore, in this implementation, the network device can use the trained domain-adaptive AAN model as the drift detection model, enabling high-precision real-time monitoring of CSI information.

[0013] In another possible implementation, the first CSI information includes the CSI matrix and / or the precoding matrix corresponding to the CSI matrix.

[0014] Secondly, embodiments of this application provide another communication method. Optionally, this method can be executed by a terminal or by a device matched with the terminal, such as a processor, chip, or chip system. Taking a terminal as an example, the method may include: the terminal predicting CSI information for a second time period based on a prediction model and the CSI information measured in a first time period; determining the distribution deviation degree based on the CSI information in the first time period and the CSI information in the second time period; and sending first information to a network device based on the distribution deviation degree; wherein the CSI information is used to describe the channel distribution in the corresponding time period; the distribution deviation degree is used to characterize the degree of deviation of the channel distribution described by the predicted CSI information in the second time period relative to the channel distribution described by the CSI information in the first time period; the first information includes the distribution deviation degree and / or warning information, the warning information being used to indicate that the channel distribution has a distribution deviation degree greater than or equal to a distribution deviation threshold.

[0015] The distribution deviation determined by the terminal through the method described in the second aspect characterizes the channel distribution described by the CSI information in the predicted second time period, relative to the degree of deviation of the channel distribution described by the CSI information in the first time period. When the terminal sends the distribution deviation and / or warning information indicating that the distribution deviation is greater than the distribution deviation threshold to the network device, it is beneficial for the network device to determine whether to instruct the terminal to update the compression model based on the distribution deviation and / or warning information.

[0016] In one possible implementation, the distribution deviation threshold is related to multiple distribution deviations determined within a historical time period. Therefore, in this implementation, the distribution deviation threshold is an adaptive threshold set based on multiple distribution deviations determined within a historical time period. As the historical time period changes, the adaptive threshold also changes, making it more adaptable to the needs of the scenario.

[0017] In another possible implementation, the distribution deviation threshold is a percentile value of multiple distribution deviations determined within a historical time period. By setting different percentiles, different levels of stringency can be met. For example, the higher the percentile, the larger the distribution deviation threshold, and the more lenient the drift detection; the lower the percentile, the smaller the distribution deviation threshold, and the more stringent the drift detection.

[0018] Thirdly, embodiments of this application provide another communication method. Optionally, this method can be executed by a network device or by a device matched with the network device, such as a processor, chip, or chip system. Taking a network device as an example, the method may include: the network device receiving first information from a terminal, and determining, based on the first information, whether to send second information to the terminal; the first information includes a distribution deviation degree, and / or, warning information, the warning information being used to indicate that the channel distribution has a distribution deviation degree greater than or equal to a distribution deviation threshold; the distribution deviation degree is used to characterize the degree of deviation of the channel distribution described by the CSI information in the predicted second time period relative to the channel distribution described by the CSI information in the first time period; the second information is used to instruct the terminal to update the compression model.

[0019] The method described in the third aspect allows network devices to determine whether to instruct terminals to update their compression models based on distribution deviation and / or early warning information indicating that the distribution deviation exceeds a distribution deviation threshold. Compared to schemes that only trigger terminal updates to the compression model when its performance deteriorates (e.g., when a performance degradation is detected based on posterior metrics such as reconstruction error), the method described in the third aspect allows network devices to trigger terminal updates before performance degradation occurs, based on predicted distribution deviation. As the terminal-side compression model is updated, the CSI compression model within the communication system is also updated accordingly. The updated CSI compression model exhibits better performance than the unupdated model, effectively preventing performance degradation and improving resource scheduling and transmission efficiency of the communication system.

[0020] In one possible implementation, the distribution deviation threshold is related to multiple distribution deviations determined within a historical time period.

[0021] In another possible implementation, the distribution deviation threshold is the percentile value of multiple distribution deviations determined within a historical time period.

[0022] Fourthly, a communication device is provided, comprising a processing module. The processing module is used to decompress CSI information reported by a terminal based on a compression model using a decompression model to obtain first CSI information, which represents the channel distribution at the current time. The processing module is also used to call a drift detection model to process the first CSI information to obtain a first drift detection result. The first drift detection result characterizes the degree of deviation of the channel distribution corresponding to the first CSI information from the channel distribution corresponding to the second CSI information. The second CSI information is CSI information used to train the CSI compression model. The CSI compression model includes a decompression model and a compression model. The processing module is also used to send first information to the terminal based on the first drift detection result, which instructs the terminal to update the compression model.

[0023] In one possible implementation, if the drift probability in the first drift detection result is within a first probability range, the processing module performs the step of sending first information to the terminal. This drift probability refers to the probability that the channel distribution corresponding to the first CSI information deviates from the channel distribution corresponding to the second CSI information.

[0024] In another possible implementation, the processing module is used to send first information to the terminal based on the first drift detection result, including: the processing module is used to determine the number M of drift detection results whose drift probability is within the second probability range from the first drift detection result and multiple drift detection results within a first time period before the current time; the processing module is used to send the first information to the terminal when the number M is greater than or equal to N; M is a natural number and N is an integer greater than 1.

[0025] In another possible implementation, the processing module is also used to send second information to the terminal based on the first drift detection result, the second information being used to instruct the terminal to increase the CSI feedback cycle.

[0026] In another possible implementation, the processing module is used to send second information to the terminal based on the first drift detection result, including: the processing module is used to determine the number M of drift detection results whose drift probability is within the second probability range from the first drift detection result and multiple drift detection results in the first time period before the current time; the processing module is used to send the second information to the terminal when the number M is less than N; M is a natural number and N is an integer greater than 1.

[0027] In another possible implementation, the drift detection model includes a feature extraction layer and a domain discrimination layer. This processing module is used to call the drift detection model to process the first CSI information and obtain the first drift detection result, including: the processing module is used to extract features from the first CSI information based on the feature extraction layer to obtain the feature vector corresponding to the first CSI information; the processing module is used to discriminate the feature vector corresponding to the first CSI information based on the domain discrimination layer to obtain the drift probability.

[0028] In another possible implementation, the drift detection model is a trained domain-adaptive adversarial network model. Before the processing module calls the drift detection model to process the first CSI information and obtain the first drift detection result, the processing module is used to: acquire multiple second CSI information and the performance value of each second CSI information; determine the domain label of each second CSI information based on the performance value and performance threshold of each second CSI information; and train the domain-adaptive adversarial network model based on the multiple second CSI information and the domain labels of each second CSI information to obtain the trained domain-adaptive adversarial network model.

[0029] In another possible implementation, the first CSI information includes the CSI matrix and / or the precoding matrix corresponding to the CSI matrix.

[0030] Fifthly, embodiments of this application provide another communication device, which includes a processing module. The processing module is used to predict CSI information for a second time period based on a prediction model and actual measured CSI information for a first time period; the CSI information describes the channel distribution within the corresponding time period; the processing module is used to determine a distribution deviation based on the CSI information in the first time period and the CSI information in the second time period; the distribution deviation characterizes the degree of deviation of the predicted channel distribution described by the CSI information in the second time period relative to the channel distribution described by the CSI information in the first time period; the processing module is used to send first information to a network device based on the distribution deviation, the first information including the distribution deviation and / or warning information, the warning information indicating that the channel distribution has a distribution deviation greater than or equal to a distribution deviation threshold.

[0031] In one possible implementation, the distribution deviation threshold is related to multiple distribution deviations determined within a historical time period.

[0032] In another possible implementation, the distribution deviation threshold is the percentile value of multiple distribution deviations determined within a historical time period.

[0033] In a sixth aspect, embodiments of this application provide another communication device, which includes a transceiver module and a processing module. The transceiver module is used to receive first information from a terminal, and the processing module is used to determine, based on the first information, whether to send second information to the terminal. The first information includes a distribution deviation degree and / or warning information. The warning information is used to indicate that the channel distribution has a distribution deviation degree greater than or equal to a distribution deviation threshold. The distribution deviation degree is used to characterize the degree of deviation of the channel distribution described by the CSI information in the predicted second time period from the channel distribution described by the CSI information in the first time period. The second information is used to instruct the terminal to update the compression model.

[0034] In one possible implementation, the distribution deviation threshold is related to multiple distribution deviations determined within a historical time period.

[0035] In another possible implementation, the distribution deviation threshold is the percentile value of multiple distribution deviations determined within a historical time period.

[0036] The fourth, fifth, and sixth aspects are the implementations on the device side corresponding to the first, second, and third aspects. The explanations, supplements, and descriptions of the beneficial effects of the first, second, and third aspects also apply to the fourth, fifth, and sixth aspects, and will not be repeated here.

[0037] A seventh aspect provides a communication device including a processor. The processor is coupled to a memory and can be used to execute instructions or data in the memory to implement the method in any possible implementation of the first aspect or the method in any possible implementation of the third aspect. Optionally, the communication device further includes a memory. Optionally, the communication device further includes a communication interface, and the processor is coupled to the communication interface.

[0038] In one implementation, the communication interface may be a transceiver, or an input / output interface.

[0039] In another implementation, the communication device is a chip configured in a network device. When the communication device is a chip configured in a network device, the communication interface can be an input / output interface.

[0040] Eighthly, a communication device is provided, including a processor. The processor is coupled to a memory and can be used to execute instructions or data in the memory to implement the method in any possible implementation of the second aspect described above. Optionally, the communication device further includes a memory. Optionally, the communication device further includes a communication interface, and the processor is coupled to the communication interface.

[0041] In one implementation, the communication interface may be a transceiver, or an input / output interface.

[0042] In another implementation, the communication device is a chip configured in the terminal. When the communication device is a chip configured in the terminal, the communication interface can be an input / output interface.

[0043] A ninth aspect provides a processor, comprising: an input circuit, an output circuit, and a processing circuit. The processing circuit is configured to receive signals through the input circuit and transmit signals through the output circuit, causing the processor to execute a method in any possible implementation of any aspect.

[0044] In specific implementation, the processor can be one or more chips, the input circuit can be input pins, the output circuit can be output pins, and the processing circuit can be transistors, gate circuits, flip-flops, and various logic circuits. The input signal received by the input circuit can be received and input by, for example, but not limited to, a receiver, and the signal output by the output circuit can be, for example, but not limited to, output to and transmitted by a transmitter. Furthermore, the input circuit and the output circuit can be the same circuit, which is used as both the input circuit and the output circuit at different times. This application does not limit the specific implementation of the processor and various circuits.

[0045] In a tenth aspect, a communication device is provided, including a processor and a memory. The processor is configured to read instructions stored in the memory, receive signals via a receiver, and transmit signals via a transmitter to execute the method in any possible implementation of any of the preceding aspects.

[0046] Optionally, the processor may be one or more, and the memory may be one or more.

[0047] Eleventhly, a computer program product is provided, the computer program product comprising: a computer program (also referred to as code or instructions), which, when the computer program is run, causes a computer to perform the method in any possible implementation of any of the preceding aspects.

[0048] In a twelfth aspect, a computer-readable storage medium is provided that stores a computer program (also referred to as code or instructions) that, when run on a computer, causes the computer to perform the methods in any possible implementation of any of the above aspects.

[0049] In a thirteenth aspect, embodiments of this application provide a chip system including one or more processors for calling and executing instructions stored in memory, causing the methods in any of the above aspects or any possible implementations of the above aspects to be executed. The chip system may be composed of chips or may include chips and other discrete devices.

[0050] The chip system may include input circuits or interfaces for transmitting information or data, and output circuits or interfaces for receiving information or data.

[0051] In a fourteenth aspect, a communication system is provided, including the aforementioned terminal device and network device. Optionally, the communication system may further include other devices that communicate with the terminal device and / or network device. Attached Figure Description

[0052] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the embodiments of this application will be described below.

[0053] Figure 1 This is a schematic diagram of a communication system 100 used in an embodiment of this application;

[0054] Figure 2 This is a schematic diagram of a communication method 200 according to an embodiment of this application;

[0055] Figure 3 This is a schematic diagram of the domain-adaptive adversarial network model according to an embodiment of this application;

[0056] Figure 4 This is a schematic diagram of another communication method 400 according to an embodiment of this application;

[0057] Figure 5 A schematic diagram of a process for calculating the Wasserstein distance is shown;

[0058] Figure 6 This is a schematic block diagram of the communication device 600 provided in an embodiment of this application;

[0059] Figure 7 This is another schematic block diagram of the communication device 700 provided in the embodiments of this application. Detailed Implementation

[0060] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0061] The technical solutions provided in this application can be applied to various communication systems, such as: Global System for Mobile Communications (GSM) systems, General Packet Radio Service (GPRS), Wireless Local Area Network (WLAN), Long Term Evolution (LTE) systems, LTE Frequency Division Duplex (FDD) systems, LTE Time Division Duplex (TDD) systems, sidelink communication systems, Universal Mobile Telecommunication System (UMTS), Worldwide Interoperability for Microwave Access (WiMAX) communication systems, non-terrestrial network (NTN) communication systems, 5th generation (5G) mobile communication systems, or new radio access technology (NR). Among these, 5G mobile communication systems can include non-standalone (NSA) and / or standalone (SA) networking. The technical solutions provided in this application can also be applied to future communication systems. This application does not limit the scope of these applications.

[0062] Figure 1 This is a schematic diagram of a communication system 100 used in an embodiment of this application. The communication system 100 may include network devices, such as... Figure 1 The network device 110 is shown. The communication system 100 may also include a terminal, such as... Figure 1 The terminal 120 shown. Network device 110 and terminal 120 can communicate via a wireless link.

[0063] Figure 1 Network device 110 and terminal 120 are illustrated as examples. Optionally, the communication system 100 may also include multiple network devices and / or multiple terminals.

[0064] The network equipment in this application can be network-side equipment such as access network equipment and core network equipment. Access network equipment is sometimes also called access node. Access network equipment has wireless transceiver capabilities and is used to communicate with terminals. Access network equipment includes, but is not limited to, base stations, evolved NodeBs (eNodeBs), transmission reception points (TRPs) in the above-mentioned communication systems, next-generation NodeBs (gNBs) in 5G mobile communication systems, access network equipment or modules of access network equipment in open RAN (ORAN) systems, satellites in NTN communication systems, base stations in future mobile communication systems, or access nodes in WiFi systems. Access network equipment can also be modules or units that can implement some of the functions of a base station. Access network equipment can be macro base stations, micro base stations, or indoor stations, relay nodes or donor nodes, or wireless controllers in cloud radioaccess network (CRAN) scenarios. Optionally, access network equipment can also be servers, wearable devices, or vehicle-mounted equipment, etc. For example, the access network equipment in vehicle-to-everything (V2X) technology can be a roadside unit (RSU). Multiple access network devices in a communication system can be base stations of the same type or different types. Base stations can communicate with terminals directly or via relay stations. Terminals can communicate with multiple base stations using different access technologies. The embodiments of this application do not limit the specific technology or device form used in the access network equipment. In this application, the access network equipment is referred to as a network device.

[0065] In this application, the means for implementing the functions of a network device can be a network device itself, or a means capable of supporting the network device in implementing those functions, such as a processor, circuit, chip, or chip system. This means can be installed in or connected to the network device. In the technical solutions provided in this application, the example of a network device being used to implement the functions of a network device is used to describe the technical solutions provided in this application.

[0066] The terminal in this application can be a wireless terminal capable of receiving network device scheduling and instruction information. A wireless terminal can be a device that provides voice and / or data connectivity to a user, a handheld device with wireless connectivity, or other processing devices connected to a wireless modem. For example, the terminal can communicate with one or more core networks or the Internet via a radio access network (RAN). The terminal can also be referred to as a terminal device, user equipment (UE), mobile station, mobile device, etc. Terminals can be widely used in various scenarios, such as device-to-device (D2D), vehicle-to-everything (V2X) communication, machine-type communication (MTC), Internet of Things (IoT), ultra-reliable low-latency communication (URLLC), virtual reality, augmented reality, industrial control, autonomous driving, telemedicine, smart grids, smart furniture, smart offices, smart wearables, smart transportation, smart cities, or satellite communication, etc. The terminal can be a mobile phone, tablet computer, computer with wireless transceiver capabilities, wearable device, vehicle, aircraft (such as drone, helicopter, airplane), hot air balloon, ship, robot, robotic arm, or smart home device, etc. The embodiments of this application do not limit the form of the terminal.

[0067] In this application, the device for implementing the terminal's functions can be a terminal itself, or any device capable of supporting the terminal in implementing those functions, such as a processor, circuit, chip, or chip system. This device can be installed in the terminal or connected to the terminal for use. In the technical solutions provided in this application, the terminal is used as an example to illustrate the technical solutions provided in this application.

[0068] Network devices and / or terminals can be fixed or mobile. They can be deployed on land, including indoors or outdoors, handheld or vehicle-mounted; on water; or in the air on aircraft, balloons, and satellites. This application does not limit the application scenarios of the network devices and terminals. They can be deployed in the same or different scenarios; for example, both can be deployed on land simultaneously; or the network device can be deployed on land while the terminal is deployed on water, etc., and so on.

[0069] In practical applications, multiple network devices can collaborate to assist terminals in achieving wireless access, with different network devices each implementing a portion of the base station's functions. For example, network devices can be central units (CUs), distributed units (DUs), CU-control plane (CPs), CU-user plane (UPs), or radio units (RUs), etc. CUs and DUs can be set up separately or included in the same network element, such as a baseband unit (BBU). RUs can be included in radio frequency equipment or radio frequency units, such as remote radio units (RRUs), active antenna units (AAUs), or remote radio heads (RRHs).

[0070] 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 and hardware modules. CU (or CU-CP and CU-UP), DU, and RU can implement different protocol layer functions.

[0071] To facilitate understanding of the embodiments of this application, the terminology used in this application will be briefly explained first. Optionally, the explanation of some terms may also refer to the explanations in the 3rd Generation Partnership Project (3GPP) standard protocol.

[0072] 1. Channel State Information (CSI)

[0073] CSI refers to information used in a communication system to describe channel characteristics, including the effects of distance, scattering, fading, etc., on the signal. Optionally, the CSI involved in this application can be a downlink CSI, which can be used by the terminal to report downlink channel quality to the network device, so that the network device can perform resource scheduling, beam management, mobility management, and other processing based on the CSI. The CSI sent by the terminal to the network device can be carried in the CSI report. For example, the CSI can include at least one of the following: CSI Reference Signal Resource Indicator (CSI-RS Resource Indicator, CRI), Rank Indicator (RI), Channel Quality Indicator (CQI), Precoding Matrix Indicator (PMI), Layer Indicator (LI), L1-RSRP, and L1-SINR.

[0074] 2. CSI Feedback

[0075] CSI feedback includes CSI feedback method one (or codebook-based CSI feedback) and CSI feedback method two (or AI-based CSI feedback).

[0076] When using CSI feedback method one, the terminal can select the entry closest to the actual channel from a predefined codebook and feed back the CSI to the network device. Commonly used codebooks include Type-I and Type-II codebooks. This codebook is a set of vectors or matrices used for quantizing CSI in wireless communication. Each entry in the codebook represents a possible channel state (such as beam direction, amplitude / phase combination, etc.). Optionally, the CSI feedback scheme based on Type-I codebooks is suitable for low-to-medium resolution requirements (e.g., single-user MIMO or low-rank computation), while the CSI feedback scheme based on Type-II codebooks is suitable for high-resolution requirements (e.g., multi-antenna, multiple-input multiple-output (MIMO) scenarios and more complex channel quantization scenarios). Optionally, when the terminal uses CSI feedback method one, the CSI feedback period can be a default value, such as 20 milliseconds.

[0077] In the second CSI feedback method, the terminal and network device perform CSI feedback through an Artificial Intelligence (AI) model. This AI model can be called a CSI compression model, which includes a compression model (or encoder) and a decompression model (or decoder). In one implementation, the terminal compresses and quantizes the measured raw CSI using the compression model to obtain the compressed and quantized CSI, and then sends the compressed and quantized CSI to the network device. On the network device side, the network device can use the decompression model to decompress the compressed and quantized CSI reported by the terminal to obtain the reconstructed CSI. Optionally, the original CSI and the reconstructed CSI can be high-dimensional matrices, while the compressed and quantized CSI can be a low-dimensional matrix. The CSI compression model can more flexibly utilize the sparsity of CSI, resulting in better compression performance and recovery accuracy. In the second CSI feedback method, the compression model and the decompression model work collaboratively. Optionally, the compression and decompression models can be jointly trained end-to-end; alternatively, they can be trained separately, with the network device training the decompression model independently and the terminal training the compression model independently. During the training of the compression and decompression models, the parameters of the compression and decompression models are optimized so that the reconstructed CSI is as close as possible to the CSI before compression and quantization (i.e., the original CSI).

[0078] Optionally, CSI can be periodically fed back. For example, the terminal feeds back CSI to the network device based on a CSI feedback period. This CSI feedback period can be a fixed time interval (such as 20 milliseconds, 30 milliseconds) or a symbol period, without limitation.

[0079] Optionally, CSI can also be non-periodic feedback. For example, the terminal feeds back CSI to the network device based on triggering conditions. These triggering conditions include, but are not limited to, a change in channel quality, or the terminal receiving a request message from the network device, etc. For ease of explanation, the following embodiments will use the terminal feeding back CSI to the network device based on a CSI feedback period as an example for illustrative purposes.

[0080] 3. Performance of the CSI compression model

[0081] Optionally, the performance of the CSI compression model may include reconstruction performance, which refers to the ability of the CSI compression model to restore the compressed and quantized CSI to the original or near-original CSI. Reconstruction performance can be indicated by at least one of the following evaluation metrics, including but not limited to: Square Generalized Cosine Similarity (SGCS), Normalized Mean Square Error (NMSE), or Peak Signal-to-Noise Ratio (PSNR). SGCS refers to the similarity between the reconstructed CSI from the network device and the original CSI from the terminal. NMSE refers to the mean square error between the reconstructed CSI from the network device and the original CSI from the terminal. PSNR refers to the quality difference between the reconstructed CSI from the network device and the original CSI from the terminal.

[0082] Furthermore, in other implementations, CSI performance may also include at least one of compression ratio, computational complexity, or noise robustness. Compression ratio refers to the ratio between the amount of data in the original CSI and the amount of data in the compressed CSI. For example, if the original CSI data is 100 units and the compressed CSI data is 10 units, the compression ratio is 10:1. A higher compression ratio allows the CSI compression model to more effectively reduce data transmission and storage, thus reducing resource consumption in the communication system. Computational complexity refers to the computational and time costs involved in compression and decompression. Lower computational complexity means the model can complete compression and decompression operations in a shorter time, making it suitable for scenarios with high real-time requirements. Noise robustness reflects the CSI compression model's ability to compress and reconstruct CSI under noise interference. A good compression model should have a certain degree of noise robustness, accurately recovering the key information of the original CSI even in noisy environments, ensuring the performance stability of the communication system.

[0083] 4. Data drift

[0084] Optionally, data drift refers to the deviation of the current channel distribution (characterized by the first CSI information) from the channel distribution during the training of the CSI compression model (characterized by the second CSI information). This data drift may lead to a degrade in the performance of the CSI compression model. For example, data drift may cause a decrease in the reconstruction performance of the CSI compression model (such as data drift leading to an increase in the NMSE and / or a decrease in the SGCS of the CSI compression model), thereby affecting the resource scheduling and transmission efficiency of the communication system.

[0085] Channel distribution refers to the probability distribution of channel characteristics in a communication system, which can be characterized by the statistical properties (such as mean, variance, or higher-order statistics) of CSI information (including the CSI matrix or the precoding matrix corresponding to the CSI matrix). Optionally, in this application, the channel distribution can be quantized by feature vectors extracted by a feature extraction layer. For example, features can be extracted from the CSI information (including the CSI matrix or the precoding matrix corresponding to the CSI matrix) by a feature extraction layer to obtain the corresponding feature vectors, which quantify the channel distribution.

[0086] Currently, CSI compression models can only be trained on a limited set of channels. However, the trained CSI compression model will be exposed to a wide variety of channels. Due to changes in atmospheric scattering and / or terminal movement, the real-time channel distribution may deviate from the channel distribution during training, i.e., data drift. This data drift can lead to performance degradation of the CSI compression model, thereby affecting the resource scheduling and transmission efficiency of the communication system. Therefore, how to avoid performance degradation of the CSI compression model is an urgent problem to be solved.

[0087] Based on this, embodiments of this application provide a communication method in which a network device can invoke a drift detection model to process first CSI information, obtain a first drift detection result, and send first information instructing the terminal to update the compression model based on the first drift detection result. The first drift detection result characterizes the degree of deviation of the channel distribution corresponding to the first CSI information from the channel distribution corresponding to the second CSI information. It is evident that the network device can promptly trigger the terminal to update the compression model based on the drift detection result. With the update of the compression model on the terminal side, the CSI compression model within the communication system is also updated accordingly. The performance of the updated CSI compression model is better than that of the previous CSI compression model, effectively avoiding performance degradation of the CSI compression model and improving the resource scheduling and transmission efficiency of the communication system.

[0088] The solution provided in this application will be described in detail below with reference to the corresponding flowcharts. It is understood that the illustrative flowcharts provided in this application use different devices (e.g., terminals, network devices) as examples of the execution entities for interactive illustration, but this application does not limit the illustrative execution entities. The devices (e.g., terminals, network devices) in the illustrative flowcharts can also be chips, chip systems, or processors that support the implementation of this method on the device, or logic modules or software that can implement all or part of the functions of the device.

[0089] As a general statement, the message or signaling interactions involved in the process of this application embodiment can adopt standard messages or signaling, or they can be newly introduced messages or signaling. This application embodiment does not make specific limitations on this.

[0090] Figure 2 This is a schematic diagram of a communication method 200 according to an embodiment of this application. It can be understood that... Figure 2 The terminal in the middle can be Figure 1 Any terminal in the context can also refer to a device within that terminal (such as a processor, chip, or chip system). Network devices can be... Figure 1 Any access network device, or a component within an access network device (such as a processor, chip, or chip system). Figure 2 As shown, the communication method 200 includes the following steps:

[0091] S201, the network device decompresses the CSI information reported by the terminal based on the compression model based on the decompression model to obtain the first CSI information.

[0092] The first CSI information is used to represent the channel distribution at the current moment.

[0093] Among them, the CSI information reported by the terminal based on the compression model (or the compressed and quantized CSI information) is low-dimensional CSI information. The first CSI information is the CSI information obtained by the network device decompressing and reconstructing the compressed and quantized CSI information based on the decompression model. The first CSI information is high-dimensional CSI information.

[0094] In one optional implementation, step S200 is included before step S201: the terminal obtains the original CSI information through actual measurement, and compresses and quantizes the original CSI information based on a compression model to obtain compressed and quantized CSI information. The terminal sends the compressed and quantized CSI information to the network device, and the network device receives the compressed and quantized CSI information accordingly.

[0095] Optionally, the terminal can send compressed and quantized CSI information to the network device via the Physical Uplink Control Channel (PUCCH) or the Physical Uplink Shared Channel (PUSCH). For example, if the amount of compressed and quantized CSI information is small, the terminal can send it via PUCCH; if the amount of compressed and quantized CSI information is large, the terminal can send it via PUSCH.

[0096] Optionally, the terminal can send compressed and quantized CSI information to the network device based on the CSI feedback period, which can be 20 milliseconds, 30 milliseconds, etc.

[0097] In this application, the CSI information includes a CSI matrix and / or the corresponding precoding matrix. The CSI matrix can also be replaced by a channel matrix. Taking the CSI information as a CSI matrix as an example, the original CSI information is the CSI matrix corresponding to a frequency domain unit determined by the terminal through channel estimation or based on channel reciprocity. Optionally, this application does not limit the frequency domain unit. For example, the frequency domain length of a frequency domain unit can be R times the subband, where R <= 1, and the value of R can be 1 or 1 / 2, a resource block (RB), etc., without limitation. This application does not limit the method by which the terminal obtains the original CSI matrix; it is only for illustrative purposes. In this case, the CSI information sent by the terminal to the network device is a compressed and quantized CSI matrix. The network device decompresses the compressed and quantized CSI matrix based on a decompression model to obtain the reconstructed CSI matrix. The higher the similarity between the reconstructed CSI matrix and the original CSI matrix on the terminal side, the better the reconstruction performance of the CSI compression model.

[0098] Taking the CSI information as an example, where the CSI information is the precoding matrix corresponding to the CSI matrix (which can be simply referred to as the precoding matrix), optionally, the precoding matrix can be obtained by the terminal performing Singular Value Decomposition (SVD) on the original CSI matrix or the covariance matrix of the original CSI matrix, or it can be obtained by the terminal performing Eigen Value Eecomposition (EVD) on the covariance matrix of the CSI matrix. It should be understood that this application does not limit the way the terminal obtains the precoding matrix corresponding to the CSI matrix, and it is only an illustrative example. The precoding matrix can be directly used for downlink data transmission; or it can be processed by some beamforming methods to obtain the final precoding matrix used for downlink data transmission. The beamforming method may include, but is not limited to, at least one of Zero Forcing (ZF), Regularized Zero-Forcing (RZF), Minimum Mean-Squared Error (MMSE), or Maximizing Signal-to-Leakage-and-Noise Ratio (SLNR), and this application does not limit it. In this scenario, the CSI information sent by the terminal to the network device is a compressed and quantized precoding matrix. The network device decompresses the compressed and quantized precoding matrix using a decompression model to obtain the reconstructed precoding matrix. The higher the similarity between the reconstructed precoding matrix and the original precoding matrix on the terminal side, the better the reconstruction performance of the CSI compression model. Optionally, the terminal can also indicate the compressed and quantized precoding matrix using a Precoding Matrix Indicator (PMI), and the network device can recover this precoding matrix based on the PMI. The higher the similarity between the precoding matrix recovered by the network device based on the PMI and the aforementioned precoding matrix to be fed back, the better the reconstruction performance of the CSI compression model.

[0099] S202, the network device calls the drift detection model to process the first CSI information and obtain the first drift detection result.

[0100] The first drift detection result is used to characterize the degree of deviation of the channel distribution corresponding to the first CSI information from the channel distribution corresponding to the second CSI information. The first CSI information is the CSI information at the current time, and the second CSI information is the CSI information used to train the CSI compression model. In one optional implementation, if the features of the first CSI information and the second CSI information are successfully aligned, the channel distribution represented by the first CSI information is consistent with the channel distribution represented by the second CSI information, and the first drift detection result indicates that the deviation of the channel distribution corresponding to the first CSI information from the channel distribution corresponding to the second CSI information is small. If the features of the first CSI information and the second CSI information cannot be aligned, the channel distribution represented by the first CSI information differs significantly from the channel distribution represented by the second CSI information, and the first drift detection result indicates that the deviation of the channel distribution corresponding to the first CSI information from the channel distribution corresponding to the second CSI information is large. This application does not describe in detail the process of training a CSI compression model based on the second CSI information for network devices.

[0101] Optionally, the drift detection result may include a drift probability, which refers to the probability that the channel distribution corresponding to the first CSI information deviates from the channel distribution corresponding to the second CSI information. If the drift probability is large, the degree of deviation indicated by the drift detection result is large; if the drift probability is small, the degree of deviation indicated by the drift detection result is small; the drift probability can be [0,1]. Optionally, the drift detection result may include a drift degree, which refers to the degree to which the channel distribution corresponding to the first CSI information deviates from the channel distribution corresponding to the second CSI information. If the drift degree is high, the degree of deviation indicated by the drift detection result is large; if the drift degree is low, the degree of deviation indicated by the drift detection result is small; the drift degree can be set as needed, for example, the drift degree may include high drift and low drift, or for example, the drift degree may include high drift, medium drift and low drift, etc. For ease of explanation, the following embodiments use the drift detection result including the drift probability as an example for illustrative purposes. The specific implementation of the drift degree can be referred to the relevant embodiments of the drift probability, which will not be described in detail.

[0102] In one implementation, the drift detection model can be a classification model trained based on a classification algorithm, including but not limited to at least one of the following: Decision Tree (DT), Rocchio algorithm, Extreme Gradient Boosting (XGBoost), Naive Bayes (NB), Linear Discriminant Analysis (LDA), Support Vector Machine (SVM), or Random Forest (RF). Specifically, the DT algorithm is a machine learning algorithm for constructing a tree structure. Each non-leaf node of the constructed decision tree represents a test on a feature attribute, each branch of the constructed decision tree represents the output of that feature attribute, and each leaf node of the constructed decision tree represents the classification result. The Rocchio algorithm is a text classification algorithm that primarily determines the classification result based on the cosine similarity between vectors. The GBoost algorithm is an ensemble machine learning algorithm that primarily determines the classification result through multiple interrelated classification and regression trees. In this context, interrelated classification and regression trees refer to the (i+1)th classification and regression tree being correlated with the training and testing results of the ith classification and regression tree, where i is a positive integer greater than 0. The NB algorithm is a classification algorithm based on Bayes' theorem, primarily determining the classification result using probabilities obtained through maximum likelihood estimation. The LDA algorithm primarily determines the classification result using distances obtained through distance estimation. The SVM algorithm is a binary classification model algorithm that determines the classification result through the optimal hyperplane constructed using a kernel function. The RF algorithm is also an ensemble machine learning algorithm that primarily determines the classification result using multiple independent decision trees.

[0103] In another implementation, the drift detection model can be a trained domain-adaptive adversarial network model. Domain adaptation is an important branch of transfer learning, aiming to address how to apply a model trained on the source domain to the target domain when the distributions of the source and target domains differ. Traditional machine learning typically requires training on a large amount of labeled data, and it's essential to ensure that the training set (or source domain) and test set (or target domain) have the same data distribution. However, in practical applications, the training and test sets often have different distributions, which can lead to a decrease in the performance of the machine learning model. Domain adaptation can effectively address the impact of inconsistent data distributions on the performance of machine learning models.

[0104] Optionally, the drift detection model may include a feature extraction layer and a domain discrimination layer. Step S202 may specifically include: the network device extracts features from the first CSI information based on the feature extraction layer to obtain the feature vector corresponding to the first CSI information, and discriminates the feature vector corresponding to the first CSI information based on the domain discrimination layer to obtain the drift probability. This drift probability refers to the probability that the channel distribution corresponding to the first CSI information deviates from the channel distribution corresponding to the second CSI information. For example, when the drift detection model is a trained classification model, this drift probability refers to the probability that the channel distribution corresponding to the first CSI information deviates from the CSI information used to train the classification model; or, for example, when the drift detection model is a trained domain adaptive adversarial network device, this drift probability refers to the probability that the channel distribution corresponding to the first CSI information deviates from the channel distribution corresponding to the source domain CSI information.

[0105] The following embodiments are described exemplarily using a drift detection model that is a trained domain-adaptive adversarial network model.

[0106] Optionally, the drift detection model may include, but is not limited to, at least one of a feature extraction layer, a gradient reversal layer, or a domain discriminant layer. The feature extraction layer is used to extract features from the CSI information, mapping the CSI information to feature vectors in the feature space. The gradient reversal layer (GRL) is used to reverse the gradient direction of the feature extraction layer during backpropagation, ensuring that the update direction of the feature extraction layer's parameters is opposite to the target of the domain discriminant layer. The domain discriminant layer is used to discriminate the feature vectors to determine whether the CSI information input to the feature extraction layer belongs to the source domain or the target domain.

[0107] Optionally, the network architecture of the feature extraction layer is not limited in this application embodiment. For example, the feature extraction layer can be a convolutional neural network (CNN) including three hidden layers. For instance, the first hidden layer can be a two-dimensional convolutional layer, which may include a convolutional layer with 32 filters and 3×3 convolutional kernels, and uses the ReLU activation function; the second hidden layer can also be a two-dimensional convolutional layer, which may include a convolutional layer with 64 filters and 3×3 convolutional kernels, and uses the ReLU activation function; the third hidden layer can be a global average pooling layer, which can be used to output a 128-dimensional feature vector.

[0108] Optionally, the network architecture of the domain discrimination layer is not limited in this application embodiment. For example, the domain discrimination layer can be a fully connected network including two fully connected layers. For instance, the first fully connected layer may include 64 neurons and use the ReLU activation function; the second fully connected layer may include 1 neuron and use the Sigmoid activation function, and the second fully connected layer can be used to output the drift probability.

[0109] In one possible implementation, before executing step S202, the network device also needs to train the domain-adaptive adversarial network model. Optionally, the CSI information used to train the domain-adaptive adversarial network model is also second CSI information. This can include: the network device acquiring multiple pieces of second CSI information and the performance values ​​of each piece of second CSI information; the network device determining the domain label of each piece of second CSI information based on the performance values ​​and performance thresholds of each piece of second CSI information; and the network device training the domain-adaptive adversarial network model based on the multiple pieces of second CSI information and the domain labels of each piece of second CSI information to obtain the trained domain-adaptive adversarial network model.

[0110] Optionally, this performance value can be SGCS. The SGCS of the second CSI information refers to the similarity between the original CSI information corresponding to the second CSI information and the reconstructed CSI information corresponding to the second CSI information.

[0111] In one implementation, the network device determines the domain label of each second CSI information based on its performance value and a performance threshold. This includes: identifying second CSI information with performance values ​​greater than or equal to the performance threshold as source domain CSI information and setting its domain label to a first value; and identifying second CSI information with performance values ​​less than the performance threshold as target domain CSI information and setting its domain label to a second value. The first and second values ​​are two different probability values. For example, the first value can be "0" and the second value can be "1" to allow the domain discrimination layer to successfully distinguish between source and target domain CSI information. For source domain CSI information, the domain discrimination layer considers a lower probability that the source domain feature originates from the target domain as better. Conversely, for target domain CSI information, the domain discrimination layer should correctly identify the target domain feature.

[0112] In one implementation, the network device trains a domain-adaptive adversarial network model based on multiple second CSI information and the domain labels of each second CSI information to obtain the trained domain-adaptive adversarial network model. The process includes: the network device calling the domain-adaptive adversarial network model to detect the second CSI information and obtain the drift probability of the second CSI; determining a loss function value based on the drift probability and the domain label of the second CSI information; if the loss function value is greater than a preset value, updating the parameters of the domain-adaptive adversarial network model; repeating the above steps until the loss function value is less than or equal to the preset value, thereby obtaining the trained domain-adaptive adversarial network model.

[0113] Optionally, the network device may determine the loss function value based on the first loss function expression.

[0114] For example, the expression for the first loss function can be:

[0115]

[0116] Among them, L domain It is the value of the loss function. Let d be the drift probability derived from the i-th second CSI information; i Let be the domain label of the i-th second CSI information. This second CSI information can be source domain CSI information or target domain CSI information. The domain label of the source domain CSI information is the first value, and the domain label of the target domain CSI information is the second value.

[0117] Furthermore, this application can also employ batch training. In this scenario, the network device can determine the loss function value based on the second loss function expression.

[0118] For example, the expression for the second loss function can be:

[0119]

[0120] Among them, L domain It represents the loss function value, where L is the number of batches for batch training. Let d be the drift probability derived from the i-th second CSI information; i Let be the domain label of the i-th second CSI information. This second CSI information can be source domain CSI information or target domain CSI information. The domain label of the source domain CSI information is the first value, and the domain label of the target domain CSI information is the second value.

[0121] The following is combined with Figure 3 The schematic diagram of the domain-adaptive adversarial network model illustrates in detail the forward and backward propagation of the model. For example... Figure 3As shown, a domain-adaptive adversarial network model may include a feature extraction layer, a gradient inversion layer, and a domain discriminant layer.

[0122] During the forward propagation of the domain-adaptive adversarial network model, the network device extracts features from the second CSI information based on the feature extraction layer to obtain the feature vector corresponding to the second CSI information, and then discriminates the feature vector corresponding to the second CSI information based on the domain discriminant layer to obtain the drift probability of the second CSI information.

[0123] During backpropagation of the domain-adaptive adversarial network (DAN) model, the network device updates the parameters of the DAN model based on the loss function value, or it updates the parameters using gradient descent. Figure 3 As shown, the network device calculates the gradient of the loss function expression (either the first loss function expression or the second loss function expression), updates the parameters of the domain discrimination layer based on the gradient, reverses the gradient through the gradient reversal layer to obtain the reversed gradient, and updates the parameters of the feature extraction layer based on the reversed gradient.

[0124] Optionally, when updating the parameters of the domain-adaptive adversarial network (DAN) model, the network device can alternately update the parameters of the feature extraction layer and the domain discriminant layer. For example, during odd-numbered parameter updates, the network device keeps the parameters of the feature extraction layer unchanged and updates the parameters of the domain discriminant layer; during even-numbered parameter updates, the network device keeps the parameters of the domain discriminant layer unchanged and updates the parameters of the feature extraction layer. It can be understood that the parameters of the feature extraction layer and the domain discriminant layer need to be initialized before updating the parameters of the domain-adaptive DAN model.

[0125] Optionally, the trained domain-adaptive adversarial network model can also be obtained through pre-training followed by incremental learning. For example, the network device can complete pre-training after training the domain-adaptive adversarial network model for a preset number of steps. After pre-training, incremental training is used to dynamically update the parameters of the domain-adaptive adversarial network model, thereby obtaining the trained domain-adaptive adversarial network model. The method may further include: the network device updating the parameters of the domain-adaptive adversarial network model based on the first drift detection result and / or the first CSI information. Optionally, during incremental training, the first CSI information is the target domain CSI information, and its corresponding domain label can be set to a second value.

[0126] S203, the network device sends first information to the terminal based on the first drift detection result, and the terminal receives the first information accordingly.

[0127] The first information is used to instruct the terminal to update the compression model. Optionally, this first information can be carried in Radio Resource Control (RRC) signaling, without limitation. In other implementations, this first information can also be replaced with "model update information".

[0128] In the first approach, step S203 is performed when the drift probability in the first drift detection result is within the first probability range. In other words, when the drift probability in the first drift detection result is within the first probability range, the network device determines that data drift exists and sends first information to the terminal instructing the terminal to update the compression model. Optionally, this application does not limit the specific value of the first probability range; for example, the first probability range can be [0.7, 1], etc.

[0129] In the second approach, step S203 is performed when the number M of drift detection results with drift probabilities within the second probability range is greater than or equal to N. Specifically, the network device determines the number M of drift detection results with drift probabilities within the second probability range from the first drift detection result and multiple drift detection results within a first time period before the current time; if the number M is greater than or equal to N, it sends first information to the terminal; where M is a natural number and N is an integer greater than 1. Optionally, this application does not limit the specific value of the second probability range; for example, the second probability range can be (0.4, 0, 7), etc. Optionally, this application does not limit the first time period; the first time period can be multiple CSI feedback periods before the current CSI feedback period, 5 minutes before the current time, etc. Optionally, the first time period can be replaced by a drift probability sliding window. As can be seen, in this implementation, when the number M of drift detection results with drift probabilities within the second probability range is greater than or equal to N, the network device determines that data drift exists and sends the first information to the terminal instructing the terminal to update the compression model. Compared with the scheme that triggers the update of the compression model as soon as there is a drift detection result with a drift probability within the second probability range, this implementation can delay the triggering of the terminal to update the compression model and avoid the communication system being overly sensitive to slight drifts.

[0130] In the third method, step S203 is performed when the drift probability in the first drift detection result is within the first probability range, and / or when the number M of drift detection results including those with drift probabilities within the second probability range is greater than or equal to N. For related explanations, please refer to the first and second methods, which will not be elaborated further.

[0131] Furthermore, if the first drift detection result does not meet the above three conditions, the network device can repeat steps S201 and S202 to continuously obtain the first drift detection result. This is beneficial for monitoring data drift based on the first drift detection result, so as to update the CSI compression model in a timely manner when drift detection is detected or drift detection is aggravated, that is, to update the compression model and / or decompression model in the CSI compression model.

[0132] S204, the terminal responds to the first message and updates the compression model.

[0133] In one possible implementation, the terminal can update the parameters of the compressed model based on the raw CSI information obtained from actual measurements, thus obtaining the updated compressed model. For example, the terminal can update the parameters of the compressed model based on the raw CSI information actually measured within a second time period. This application does not limit the second time period; for example, the second time period can be the current CSI feedback cycle, the K nearest CSI feedback cycles to the current CSI feedback cycle, the 5 minutes nearest to the current time, etc. Taking the second time period as the K nearest CSI feedback cycles to the current CSI feedback cycle as an example, if the terminal receives the first information in the Tth CSI feedback cycle, the terminal can update the parameters of the compressed model based on the raw CSI information within these K CSI feedback cycles from the (T-K+1)th to the Tth CSI feedback cycle, where K is a positive integer. Optionally, in other embodiments, the terminal can also update the parameters of the compressed model based on the compressed and quantized CSI information, without limitation.

[0134] In another possible implementation, the terminal can select a corresponding compressed model from the model library based on at least one of the following: model version number, terminal mobility status, or original CSI information within the second time period, and use the selected compressed model as the updated compressed model. Optionally, in this implementation, the network device can inform the terminal of the model version number. For example, the model version number can be sent separately; the network device sends the model version number to the terminal, and the terminal receives the model version number. Alternatively, the model version number can be carried in the first information. This application does not limit the length of the model version number; for example, the length of the model version number can be 16 bits.

[0135] Furthermore, the updated compression model also has a corresponding effective time. This application does not limit the specific type of the effective time; for example, it can be Coordinated Universal Time (UTC), Beijing Time, etc. Optionally, one time type can correspond to one time precision; for example, the time precision corresponding to UTC time can be 1 millisecond (ms), and the time precision corresponding to Beijing time can be 1 second (s). The network device can inform the terminal of the effective time of the updated compression model. For example, the effective time of the updated compression model can be sent separately; the network device sends the effective time of the updated compression model to the terminal, and the terminal receives the effective time of the updated compression model. Alternatively, the effective time of the updated compression model can be carried in the first piece of information.

[0136] In some embodiments, the communication method may further include step S205: the network device updates the decompression model, which is a compression model corresponding to the compression model on the terminal side, and the decompression model and the compression model on the terminal side constitute the CSI compression model.

[0137] Optionally, the network device can update the parameters of the decompression model based on the decompressed and reconstructed CSI information to obtain the updated decompression model. For example, the network device can update the parameters of the decompression model based on the decompressed and reconstructed CSI information within a third time period. For a detailed explanation of the third time period, please refer to the relevant embodiments regarding the second time period in step S204, which will not be elaborated further. In other implementations, the network device can also update the parameters of the decompression model based on the compressed and quantized CSI information, which will not be detailed further.

[0138] In another possible implementation, the network device can select a corresponding decompression model from the model library based on at least one of the following: model version number, terminal mobility status, or decompressed and reconstructed CSI information within a third time period, and use the selected decompression model as the updated compression model. Optionally, the model version number can be pre-configured. Optionally, in this implementation, the terminal can inform the network device of its mobility status. For example, the terminal's mobility status can be sent separately; the terminal sends its mobility status to the network device, and the network device receives the terminal's mobility status. Alternatively, the terminal's mobility status can be carried in the CSI information, and so on. The terminal's mobility status may include the terminal's location and / or speed. The updated decompression model may also have a corresponding effective time. The effective time of the updated decompression model may be the same as or different from the effective time of the updated compression model. For an explanation of the effective time, please refer to the preceding text; it will not be elaborated upon here.

[0139] Optionally, the communication method may include only steps S203 to S204, excluding step S205. In this case, the compression model in the CSI compression model is updated, but the decompression model is not updated. As the terminal-side compression model is updated, the CSI compression model of the communication system is also updated accordingly. Alternatively, it may include only step S205, excluding steps S203 to S204. In this case, the decompression model in the CSI compression model is updated, but the compression model is not updated. As the network device-side decompression model is updated, the CSI compression model of the communication system is also updated accordingly. Or, it may include steps S203 to S205. In this case, the compression model in the CSI compression model is updated, and the decompression model is also updated. As the terminal-side compression model and the network device-side decompression model are updated, the CSI compression model of the communication system is also updated accordingly.

[0140] In scenarios where the communication method includes steps S203 to S205, this embodiment does not limit the execution order of steps S203 and S205 by the network device. For example, the network device may execute step S203 first, then step S205; or it may execute step S205 first, then step S203; or it may execute steps S203 and S205 simultaneously. In the scenario where the network device executes step S205 first, then step S203, if the network device selects a decompression model from the model library, the network device can inform the terminal of the compression model corresponding to the updated decompression model. For example, the network device can indicate the updated compression model through the model version number, allowing the terminal to quickly determine the updated compression model.

[0141] Optionally, to avoid CSI feedback interruption during the update process of the compression or decompression model, before step S203, the method further includes: the network device and the terminal switching from CSI feedback mode two to CSI feedback mode one. To instruct the terminal to switch CSI feedback modes, the network device may send first feedback adjustment information to the terminal, and the terminal receives the first feedback adjustment information. The first feedback adjustment information is used to instruct the terminal to switch from CSI feedback mode two to CSI feedback mode one.

[0142] Accordingly, once the compression model on the terminal side and / or the decompression model on the network device side have been updated, the network device and the terminal switch from CSI feedback mode one to CSI feedback mode two. Similarly, to instruct the terminal to switch CSI feedback modes, the network device can send second feedback adjustment information to the terminal, and the terminal receives the second feedback adjustment information. The second feedback adjustment information is used to instruct the terminal to switch from CSI feedback mode one to CSI feedback mode two. Optionally, before instructing the terminal to switch from CSI feedback mode one to CSI feedback mode two, the network device needs to know that the compression model on the terminal side has been updated. Then, if the compression model has been updated, the terminal can send third information to the network device, and the network device receives the third information, which indicates that the compression model has been updated. In other implementations, the third information can be replaced with "model update confirmation information".

[0143] Furthermore, the method may also include step S206, in which the network device sends second information to the terminal based on the first drift detection result, and the terminal receives the second information accordingly.

[0144] The second information is used to instruct the terminal to increase the CSI feedback period. The second information may include the increased CSI feedback period or the amount of increase. For example, if the current CSI feedback period is 20ms, the second information may include an increased CSI feedback period of 30ms or an increase of 10ms. Optionally, this second information can be carried in the RRC signaling, and the increased CSI feedback period or the amount of increase can be indicated through the `reportSlotConfig` field of `CSI-ReportConfig` in the RRC signaling, without limitation.

[0145] Referring to the first method in step S203, if the drift probability in the first drift detection result is not within the first probability range, the network device executes step S206. That is, if the drift probability in the first drift detection result is within the first probability range, the network device executes step S203; if the drift probability in the first drift detection result is not within the first probability range, the network device executes step S206.

[0146] Referring to the second or third method in step S203, the network device determines the number M of drift detection results with drift probabilities within the second probability range from the first drift detection result and multiple drift detection results within the first time period before the current time. If the number M is less than N, the network device executes step S206. In other words, if the number M of drift detection results with drift probabilities within the second probability range is greater than or equal to N, data drift is detected, and the network device promptly triggers the terminal to update the compression model. If the number M of drift detection results with drift probabilities within the second probability range is less than N, the network device instructs the terminal to increase the CSI feedback period. The frequency of CSI interaction between the terminal and the network device decreases as the CSI feedback period increases, reducing signaling overhead to some extent. When a drift detection result is obtained within one CSI feedback period, the interval at which the network device obtains the drift detection result increases with the increase of the CSI feedback period. Therefore, the interval at which the network device triggers the terminal to update the compression model based on the drift detection result also increases accordingly, providing more buffer time for model updates. Meanwhile, the frequency of CSI information exchange between the terminal and network equipment decreases as the CSI feedback cycle increases, which can effectively reduce signaling overhead.

[0147] S206, the terminal responds to the second information, increasing the CSI feedback cycle.

[0148] In one alternative implementation, after step S206, steps S201 to S206 are repeated to monitor data drift in real time.

[0149] As can be seen, in this embodiment, the network device can detect the deviation of the current channel distribution from the channel distribution during CSI compression model training using a drift detection model, and can promptly trigger an update of the CSI compression model based on the deviation (including triggering the terminal to update the compression model and / or triggering the network device to update the decompression model). Compared to the scheme that triggers an update of the CSI compression model when its performance deteriorates (e.g., a performance deterioration is detected based on posterior indicators such as reconstruction error), this method can directly detect the channel distribution. It can trigger an update of the CSI compression model in a timely manner based on changes in the channel distribution before the performance deteriorates. The updated CSI compression model has better performance than the unupdated model, effectively avoiding performance degradation and improving the resource scheduling and transmission efficiency of the communication system. Furthermore, this embodiment applies a domain-adaptive adversarial network model to drift detection. The network device can quantify the distribution deviation between the real-time first CSI information and the training second CSI information in real time through the domain discriminant layer of the domain-adaptive adversarial network model, enabling high-precision real-time monitoring of CSI information.

[0150] The foregoing embodiments illustrate the process of network devices detecting data drift. In other embodiments, the existence of data drift can also be predicted in advance. Based on this, this application also provides another communication method, which determines the distribution deviation degree based on the CSI information in a first time period and the predicted CSI information in a second time period. The distribution deviation degree is used to characterize the degree of deviation of the channel distribution described by the CSI information in the predicted second time period from the channel distribution described by the CSI information in the first time period, so that the probability of data drift is higher when the distribution deviation degree is greater than the distribution deviation threshold.

[0151] Figure 4 This is a schematic diagram of another communication method 400 according to an embodiment of this application. It can be understood that... Figure 4 The terminal in the middle can be Figure 1 Any terminal in the context can also refer to a device within that terminal (such as a processor, chip, or chip system). Network devices can be... Figure 1 Any access network device can refer to any component within the access network device (such as a processor, chip, or chip system). The following section uses the example of communication method 400 being executed by a terminal to illustrate the specific implementation of communication method 400.

[0152] like Figure 4 As shown, the communication method 400 includes the following steps:

[0153] S401, the terminal predicts CSI information for a second time period based on the prediction model and the CSI information measured in the first time period.

[0154] The CSI information is used to describe the channel distribution within a corresponding time period. For example, the CSI information within a first time period describes the channel distribution within that first time period, and the CSI information within a second time period describes the channel distribution within that second time period. Optionally, the CSI information includes a CSI matrix and / or the precoding matrix corresponding to the CSI matrix. For a detailed explanation of the CSI information, please refer to the relevant embodiments of the communication method 200, which will not be elaborated upon here.

[0155] In this context, the first time period and the second time period are two distinct time periods, with the times in the second time period being later than those in the first time period. In some implementations, the first time period can be understood as the time period before the current time, and the second time period as the time period after the current time. For example, if the current time is 9:00, the first time period could be 8:00-9:00, and the second time period could be 9:00-10:00. Based on this, the CSI information within the first time period is used to describe the historical or current channel distribution, while the CSI information within the second time period is used to describe the future channel distribution.

[0156] Furthermore, in CSI feedback, the first time period may include the A CSI feedback periods closest to the current CSI feedback period, and the second time period may include the B CSI feedback periods following the current CSI feedback period, where A and B are integers, and A > B > 1. For example, with A = 64 and B = 3, the terminal can predict CSI information for the 65th to 67th CSI feedback periods based on the prediction model and the CSI information measured in the 1st to 64th CSI feedback periods; or, the terminal can predict CSI information for the 66th to 68th CSI feedback periods based on the prediction model and the CSI information measured in the 2nd to 65th CSI feedback periods; and so on.

[0157] In one optional implementation, the prediction model may be, but is not limited to, at least one or more of the following: a linear regression model, a recurrent neural network (RNN) model, a gated recurrent unit (GRU) model, or a long short-term memory (LSTM) network model. Optionally, the prediction model may be a lightweight prediction model to ensure successful operation on the terminal side. For example, the prediction model may be a lightweight LSTM model. This application does not limit the structure of the LSTM model; for example, the LSTM model may include 64 hidden layers, and the number of parameters in the LSTM model may not exceed 10 kilobytes (KB). The prediction model used for prediction may be a trained prediction model. The training process of the prediction model is not detailed in this application; relevant implementations can be found in related materials.

[0158] S402, the terminal determines the distribution deviation based on the CSI information in the first time period and the CSI information in the second time period.

[0159] In one alternative implementation, the distribution deviation can be characterized by at least one of Euclidean distance, Manhattan distance, or Wasserstein distance. Subsequent embodiments use Wasserstein distance as an example. Optionally, the Wasserstein distance can also be referred to as the Earth Mover's Distance.

[0160] Please see Figure 5 , Figure 5 A schematic diagram of a process for calculating the Wasserstein distance is shown. Figure 5 As shown, this process may include, but is not limited to, the following steps:

[0161] S501, the terminal performs feature extraction on the CSI information in the first time period to obtain the corresponding first feature vector, and performs feature extraction on the CSI information in the second time period to obtain the corresponding second feature vector.

[0162] Taking the first time period, which includes A CSI feedback cycles, as an example, the terminal can extract features from the CSI information within the A CSI feedback cycles to obtain A first feature vectors. Similarly, taking the second time period, which includes B CSI feedback cycles, as an example, the terminal can extract features from the CSI information within the B CSI feedback cycles to obtain B second feature vectors.

[0163] Optionally, the terminal can implement feature extraction based on the feature extraction layer. This application does not limit the specific implementation of the feature extraction layer.

[0164] S502, the terminal determines the cost matrix based on the first feature vector and the second feature vector.

[0165] Optionally, the terminal can input the first feature vector and the second feature vector into the following formula to obtain the cost matrix.

[0166] C i,j =‖x i -y j || 2

[0167] Among them, C i,j Used to represent the cost matrix; x i Used to represent the i-th first eigenvector, i = 1, 2, ..., A; y j Used to represent the j-th second eigenvector, j = 1, 2, ..., B.

[0168] The cost matrix in this embodiment is calculated using Euclidean distance. In other embodiments, other distance metrics, such as Manhattan distance, may also be used.

[0169] S503, the terminal determines the Gibbs kernel matrix based on the cost matrix.

[0170] Optionally, the terminal can input the cost matrix into the following formula to obtain the Gibbs kernel matrix.

[0171]

[0172] Among them, K i,j Used to represent the Gibbs kernel matrix; C i,j Used to represent the cost matrix, i = 1, 2, ..., A; j = 1, 2, ..., B; λ is used to represent a preset constant.

[0173] S504, the terminal iteratively updates the dual vector until the dual vector satisfies the convergence condition.

[0174] Optionally, the convergence condition can be expressed as:

[0175] Here, u and v are dual vectors, and v is initialized to a vector of all 1s before the iterative update.

[0176] Taking the cost matrix as the Euclidean distance as an example, in Euclidean space, v = (1, 1, ..., 1); then the dual vector u of v can be expressed as m*v = m + m + ... + m.

[0177] Optionally, the value of ò can be e. -4 .

[0178] S505: The terminal determines the optimal transmission matrix based on the Gibbs kernel matrix and dual vectors.

[0179] Optionally, the terminal can use the Gibbs kernel matrix K i,j and dual vector u i and v j Enter the following formula to obtain the optimal transfer matrix.

[0180] γ i,j =u i K i,j v j

[0181] Where, γ i,j Used to represent the optimal transfer matrix; u i It is the dual vector u that satisfies the convergence condition; v j It is the dual vector v that satisfies the convergence condition; i = 1, 2, ..., A; j = 1, 2, ..., B.

[0182] In S506, the terminal introduces an entropy regularization term and replaces the traditional linear programming in the Wasserstein distance with iterative matrix scaling to calculate the Wasserstein distance.

[0183] The Wasserstein distance is a metric used to measure the difference between two probability distributions. It is insensitive to the shape of the distribution, exhibits continuity, and is sensitive to low-probability events, making it suitable for dynamically changing channel environments. The Wasserstein distance involves the optimal transmission problem, i.e., finding the minimum "transfer cost" to transform one distribution into another, which is essentially a linear programming problem. However, in high-dimensional data, directly calculating the optimal transmission matrix using linear programming algorithms is very complex. Therefore, approximation algorithms (such as the Sinkhorn algorithm) are needed. The Sinkhorn algorithm transforms the calculation of the Wasserstein distance into a convex optimization problem by introducing entropy regularization, replacing traditional linear programming with iterative matrix scaling (Sinkhorn iteration), significantly reducing computational complexity while maintaining reasonable accuracy.

[0184] Optionally, the Wasserstein distance approximation is:

[0185]

[0186] Where, γ i,j Used to represent the optimal transfer matrix; C i,j Used to represent the cost matrix.

[0187] S403, the terminal sends the first information to the network device based on the distribution deviation, and the corresponding terminal receives the first information.

[0188] The first information includes at least one of the following: distribution deviation; or, warning information, which indicates that the channel distribution has a distribution deviation greater than or equal to a distribution deviation threshold. Optionally, the warning information may include a warning type identifier, which indicates that the warning information is related to CSI data drift. Further, the first information may also include CSI information within a first time period and / or CSI information within a second time period, so that the network device can decide whether to update the CSI compression model based on the CSI information in the first information. The first information may be sent separately or carried in the uplink RRC signaling "AI-PredictionReport", without limitation.

[0189] Optionally, the distribution deviation threshold is related to multiple distribution deviations determined within a historical time period. For example, the distribution deviation threshold can be the average of multiple distribution deviations determined within a historical time period. Another example is that the distribution deviation threshold is the percentile value of multiple distribution deviations determined within a historical time period. Taking the 94th percentile as an example, the percentile value xp is a number less than 94% of these multiple distribution deviations and greater than or equal to 6% of these multiple distribution deviations. Therefore, in this implementation, an adaptive distribution deviation threshold can be set based on percentiles. By setting different percentiles, different drift detection requirements can be met. For example, the higher the percentile, the larger the distribution deviation threshold, and the more lenient the drift detection; the lower the percentile, the smaller the distribution deviation threshold, and the more stringent the drift detection. The number of distribution deviations determined within the historical time period can be C, where C is a positive integer. This application does not limit the value of C; for example, C = 50. The historical time period is the period before the first time period. For example, the historical time period is D CSI feedback cycles before the first time period, where D is a positive integer. The specific implementation of each distribution deviation determined within the historical time period can be found in steps S401 and S402, and will not be described in detail here.

[0190] S404: Based on the first information, the network device determines whether to send the second information to the terminal.

[0191] The second piece of information is used to instruct the terminal to update the compression model.

[0192] In one optional implementation, the network device determines whether to send second information to the terminal based on the first information and multiple first information from other terminals within the terminal's area. This application does not limit the specific implementation of the area; for example, the area where the terminal is located can refer to the cell, sector, etc., where the terminal is located. Optionally, taking the first information including warning information as an example, the network device can determine the number P of first information including warning information from multiple first information; if the number P is greater than or equal to Q, it determines to send second information to the terminal, where P is a natural number and Q is an integer greater than 1. Optionally, taking the first information including distribution deviation as an example, the network device can determine the number Z of first information including distribution deviation greater than or equal to a distribution deviation threshold from multiple first information; if the number Z is greater than or equal to Q, it determines to send second information to the terminal, where Z is a natural number and Q is an integer greater than 1. In other embodiments, the network device can also determine whether to send second information to the terminal based on distribution deviation and warning information, which will not be detailed further. As can be seen, in this embodiment, when there are Q distribution deviations greater than or equal to the distribution deviation threshold, the network device determines that there is data drift and sends second information to the terminal instructing the terminal to update the compression model.

[0193] In another alternative implementation, the network device determines whether to send second information to the terminal based on at least one of the first information, multiple first information messages from other terminals within the terminal's area, or the network device's load. Thus, in this implementation, when deciding whether to trigger a terminal update of its compression model, the network device considers not only the first information reported by the terminal but also the network device's load. The network device comprehensively considers multiple aspects to determine whether data drift exists and whether to trigger a terminal update of its compression model.

[0194] Furthermore, after step S404, the terminal can update the compression model in response to the second information. The relevant implementation can be found in communication method 200, and will not be detailed further.

[0195] Optionally, the network device may also determine whether to update the decompression model based on the first information. For the relevant implementation, please refer to step S404, which will not be described in detail here.

[0196] In other embodiments, the network device may also determine to adjust the CSI feedback period based on the first information. Based on this, the network device may send an instruction to the terminal to increase the CSI feedback period. For further details, please refer to [link to relevant documentation]. Figure 2 The relevant embodiments are not described in detail.

[0197] As can be seen, this application embodiment introduces a data drift warning mechanism for channel distribution. The terminal can determine the distribution deviation based on the CSI information in the first time period and the predicted CSI information in the second time period. This distribution deviation characterizes the degree of deviation of the channel distribution described by the predicted CSI information in the second time period relative to the channel distribution described by the CSI information in the first time period. When the terminal sends the distribution deviation and / or warning information indicating that the distribution deviation is greater than the distribution deviation threshold to the network device, the network device can determine whether to update the CSI compression model (including updating the compression model on the terminal side and / or updating the decompression model on the network device side) based on the distribution deviation or warning information. Compared with the scheme that triggers the update of the CSI compression model when the performance of the CSI compression model deteriorates (such as when the performance of the CSI compression model is detected based on posterior indicators such as reconstruction error), this method can trigger the update of the CSI compression model in advance by using the predicted distribution deviation before the performance of the CSI compression model deteriorates. The performance of the updated CSI compression model is better than that of the unupdated CSI compression model, which can effectively avoid the performance degradation of the CSI compression model and improve the resource scheduling and transmission efficiency of the communication system. In addition, the distribution deviation threshold is an adaptive threshold set based on multiple distribution deviations determined within a historical time period. As the historical time period changes, the adaptive threshold set based on multiple distribution deviations determined within the historical time period also changes. There are no prior assumptions about the statistical distribution of Wasserstein distance, which can meet different stringency requirements for drift detection.

[0198] It should be noted that communication method 400 and communication method 200 are two relatively independent embodiments. In some embodiments, the communication system may only include communication method 200 and not communication method 400; in other embodiments, the communication system may only include communication method 400 and not communication method 200; in still other embodiments, the communication system may include both communication method 200 and communication method 400. When the communication system includes both communication method 200 and communication method 400, the communication method includes a three-layer closed-loop mechanism of detection-early warning-update, covering real-time detection and forward-looking early warning, covering the entire lifecycle of drift detection.

[0199] Optionally, the first information in communication method 200 and the first information in communication method 400 are also relatively independent. The definition of the first information in communication method 200 and the definition of the first information in communication method 400 are different. When the communication system includes both communication method 200 and communication method 400, other names can be introduced to distinguish the first information in communication method 200 and the first information in communication method 400. For example, the first information in communication method 400 can be replaced with the fourth information. Similarly, the second information in communication method 200 and the second information in communication method 400 are also relatively independent. The first time period in communication method 200 and the first time period in communication method 400 are also relatively independent. The second time period in communication method 200 and the second time period in communication method 400 are also relatively independent. These will not be described in detail.

[0200] It should be understood that Figures 1 to 5 The flowcharts or scene diagrams shown are for illustrative purposes only and are not intended to limit the embodiments of this application to the examples illustrated. In fact, those skilled in the art can interpret the embodiments based on... Figures 1 to 5 The examples in the document can be transformed into equivalent ways to obtain more implementations.

[0201] The above text combined Figures 1 to 5 This document describes in detail the communication method provided in the embodiments of this application. The following will combine... Figures 6 to 7 The device embodiments of this application are described in detail below. It should be understood that the communication device of this application embodiment can execute the various communication methods of the foregoing embodiments of this application, that is, the specific working processes of the various products below can be referred to the corresponding processes in the foregoing method embodiments.

[0202] In the embodiments described above, the terminal may execute some or all of the steps in each embodiment; the network device may execute some or all of the steps in each embodiment. These steps or operations are merely examples, and the embodiments of this application may also perform other operations or variations thereof. Furthermore, the steps may be executed in different orders as presented in the embodiments, and it is not necessary to execute all the operations in the embodiments of this application. Moreover, the sequence number of each step does not imply the order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0203] Figure 6 This is a schematic block diagram of a communication device provided in an embodiment of this application. Figure 6As shown, the communication device 600 may include a communication module 620. The communication module 620 can implement corresponding communication functions, which can be internal communication functions of the communication device 600 or communication functions between the communication device 600 and other devices. Optionally, the communication module 620 may also be referred to as a communication interface or transceiver module. Optionally, the communication device 600 further includes a processing module 610. The processing module 610 can implement corresponding processing functions.

[0204] Optionally, the communication device 600 further includes a storage module, which can be used to store instructions and / or data; the processing module 610 can read the instructions and / or data in the storage module so that the communication device 600 can implement the aforementioned method embodiments.

[0205] In one possible design, the communication device 600 may correspond to the network device in the above method embodiments, or to a component (such as a circuit, chip, or chip system) configured in the network device. The communication device 600 can be used to perform the steps or processes performed by the network device in any of the above method embodiments.

[0206] For example, the processing module 610 is used to decompress the CSI information reported by the terminal based on the compression model based on the decompression model to obtain first CSI information, which represents the channel distribution at the current time; the processing module 610 is used to call the drift detection model to process the first CSI information to obtain a first drift detection result; the first drift detection result is used to characterize the degree of deviation of the channel distribution corresponding to the first CSI information from the channel distribution corresponding to the second CSI information; the second CSI information is CSI information used to train the CSI compression model; the CSI compression model includes a decompression model and a compression model; the processing module 610 is used to send first information to the terminal based on the first drift detection result, which instructs the terminal to update the compression model.

[0207] In one possible implementation, if the drift probability in the first drift detection result is within a first probability range, the step of sending first information to the terminal is performed. The drift probability refers to the probability that the channel distribution corresponding to the first CSI information deviates from the channel distribution corresponding to the second CSI information.

[0208] In another possible implementation, the processing module 610 is used to send first information to the terminal based on the first drift detection result, including: the processing module 610 is used to determine the number M of drift detection results whose drift probability is within the second probability range from the first drift detection result and multiple drift detection results in the first time period before the current time; the processing module 610 is used to send the first information to the terminal when the number M is greater than or equal to N; M is a natural number and N is an integer greater than 1.

[0209] In another possible implementation, the processing module 610 is used to send second information to the terminal based on the first drift detection result. The second information is used to instruct the terminal to increase the CSI feedback cycle.

[0210] In another possible implementation, the processing module 610 is used to send second information to the terminal based on the first drift detection result, including: the processing module 610 is used to determine the number M of drift detection results whose drift probability is within the second probability range from the first drift detection result and multiple drift detection results in the first time period before the current time; the processing module 610 is used to send the second information to the terminal when the number M is less than N; M is a natural number and N is an integer greater than 1.

[0211] In another possible implementation, the drift detection model includes a feature extraction layer and a domain discrimination layer. The processing module 610 is used to call the drift detection model to process the first CSI information and obtain the first drift detection result, including: the processing module 610 extracting features from the first CSI information based on the feature extraction layer to obtain the feature vector corresponding to the first CSI information; and the processing module 610 discriminating the feature vector corresponding to the first CSI information based on the domain discrimination layer to obtain the drift probability.

[0212] In another possible implementation, the drift detection model is a trained domain-adaptive adversarial network model. Before the processing module 610 calls the drift detection model to process the first CSI information and obtains the first drift detection result, the processing module 610 is used to acquire multiple second CSI information and the performance value of each second CSI information. The processing module 610 is used to determine the domain label of each second CSI information based on the performance value and performance threshold of each second CSI information. The processing module 610 is used to train the domain-adaptive adversarial network model based on the multiple second CSI information and the domain label of each second CSI information to obtain the trained domain-adaptive adversarial network model.

[0213] In another possible implementation, the first CSI information includes the CSI matrix and / or the precoding matrix corresponding to the CSI matrix.

[0214] The above are merely examples; for detailed steps or procedures, please refer to the descriptions in the foregoing embodiments.

[0215] In one possible design, the communication device 600 may correspond to the terminal in the above method embodiments, or a component (such as a circuit, chip, or chip system) configured in the terminal. The communication device 600 may be used to execute the steps or processes performed by the terminal in any of the above method embodiments.

[0216] For example, the processing module 610 is used to predict CSI information in a second time period based on the prediction model and the CSI information measured in the first time period; the CSI information is used to describe the channel distribution in the corresponding time period; the processing module 610 is used to determine the distribution deviation degree based on the CSI information in the first time period and the CSI information in the second time period; the distribution deviation degree is used to characterize the degree of deviation of the channel distribution described by the predicted CSI information in the second time period relative to the channel distribution described by the CSI information in the first time period; the processing module 610 is used to send first information to the network device based on the distribution deviation degree, the first information including the distribution deviation degree and / or warning information, the warning information being used to indicate that there is a distribution deviation degree greater than or equal to the distribution deviation threshold in the channel distribution.

[0217] In one possible implementation, the distribution deviation threshold is related to multiple distribution deviations determined within a historical time period.

[0218] In another possible implementation, the distribution deviation threshold is the percentile value of multiple distribution deviations determined within a historical time period.

[0219] The above are merely examples; for detailed steps or procedures, please refer to the descriptions in the foregoing embodiments.

[0220] In one possible design, the communication device 600 may correspond to the network device in the above method embodiments, or to a component (such as a circuit, chip, or chip system) configured in the network device. The communication device 600 can be used to perform the steps or processes performed by the network device in any of the above method embodiments.

[0221] For example, the communication device includes a communication module and a processing module. The communication module 620 is used to receive first information from a terminal, and the processing module 610 is used to determine whether to send second information based on the first information. The second information is used to instruct the terminal to update the compression model. The first information includes distribution deviation and / or warning information. The distribution deviation is used to characterize the channel distribution described by CSI information in the predicted second time period, relative to the channel distribution described by CSI information in the first time period. The warning information is used to indicate that there is a distribution deviation of the channel distribution that is greater than or equal to the distribution deviation threshold.

[0222] In one possible implementation, the distribution deviation threshold is related to multiple distribution deviations determined within a historical time period.

[0223] In another possible implementation, the distribution deviation threshold is a percentile value of multiple distribution deviations determined within a historical time period.

[0224] The above are merely examples; for detailed steps or procedures, please refer to the descriptions in the foregoing embodiments.

[0225] Figure 7 This is another schematic block diagram of the communication device 700 provided in the embodiments of this application. The communication device 700 may be a chip, chip system, or processor, etc., used in a terminal or network device to implement the above-described methods. The communication device 700 can be used to implement the methods described in the above-described method embodiments; for details, please refer to the descriptions in the above-described method embodiments.

[0226] like Figure 7 As shown, the communication device 700 may include one or more processors 710, which may also be referred to as processing units or processing modules, and can implement certain control functions. The processor 710 may be a general-purpose processor or a dedicated processor, such as a baseband processor or a central processing unit. The baseband processor can be used to process communication protocols and communication data, while the central processing unit can be used to control the communication device 700 (e.g., a base station, baseband chip, user, user chip), execute software programs, and process data from the software programs.

[0227] In an alternative design, the processor 710 may also store instructions and / or data that can be executed by the processor 710 to cause the communication device 700 to perform the methods described in the above method embodiments.

[0228] In another alternative design, the communication device 700 may include a communication interface 720 for implementing receiving and transmitting functions. For example, the communication interface 720 may be a transceiver circuit, interface, interface circuit, or transceiver. The transceiver circuit, interface, interface circuit, or transceiver for implementing receiving and transmitting functions may be separate or integrated. The aforementioned transceiver circuit, interface, interface circuit, or transceiver may be used for reading and writing code / data, or it may be used for transmitting or relaying signals.

[0229] Optionally, the communication device 700 may include one or more memories 730, which may store instructions that can be executed on the processor 710, causing the communication device 700 to perform the methods described in the above method embodiments. Optionally, the memories 730 may also store data. Optionally, the processor 710 may also store instructions and / or data. The processor 710 and the memories 730 may be provided separately or integrated together.

[0230] It should be understood that, in one possible design, the steps in the method embodiments provided in this application can be implemented by integrated logic circuits in the processor's hardware or by instructions in software form. The steps of the methods disclosed in the embodiments of this application can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules in the processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method. To avoid repetition, detailed descriptions are not provided here.

[0231] In one implementation, the communication device 700 may correspond to the terminal in the above method embodiments and may be used to execute the various steps and / or processes executed by the terminal in the above method embodiments. The processor 710 may be used to execute instructions stored in the memory 730, and when the processor 710 executes the instructions stored in the memory, the processor 710 is used to execute the various steps and / or processes of the above method embodiments corresponding to the terminal.

[0232] In another implementation, the communication device 700 may correspond to the network device in the above method embodiments and may be used to execute the various steps and / or processes executed by the network device in the above method embodiments. The processor 710 may be used to execute instructions stored in the memory 730, and when the processor 710 executes the instructions stored in the memory, the processor 710 is used to execute the various steps and / or processes of the above method embodiments corresponding to the network device.

[0233] It should be understood that the aforementioned processing device can be one or more chips. For example, the processing device can be a field-programmable gate array (FPGA), an application-specific integrated circuit (ASIC), a system-on-chip (SoC), a central processor unit (CPU), a network processor (NP), a digital signal processor (DSP), a microcontroller unit (MCU), a programmable logic device (PLD), or other integrated chips.

[0234] It is understood that the memory in the embodiments of this application can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM). It should be noted that the memory used in the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0235] According to the method provided in the embodiments of this application, this application also provides a chip system, which includes one or more processors for calling and executing instructions stored in memory, thereby causing the method described in the embodiments of this application to be executed. The chip system may be composed of chips or may include chips and other discrete devices.

[0236] The chip system may include input circuits or interfaces for transmitting information or data, and output circuits or interfaces for receiving information or data.

[0237] According to the method provided in the embodiments of this application, this application also provides a communication system, which includes the aforementioned network device and terminal.

[0238] According to the method provided in the embodiments of this application, this application also provides a computer program product, which includes: computer program code, which, when run on a computer, causes the computer to execute the various steps or processes executed by the network device or terminal in any of the foregoing method embodiments.

[0239] According to the method provided in the embodiments of this application, this application also provides a computer-readable storage medium storing program code, which, when run on a computer, causes the computer to execute the various steps or processes executed by the network device or terminal in any of the foregoing method embodiments.

[0240] The computer-readable storage medium may be the aforementioned volatile memory or non-volatile memory, or it may include both volatile memory and non-volatile memory.

[0241] In the embodiments of this application, the terms and English abbreviations are exemplary examples given for ease of description and should not be construed as limiting the application in any way. This application does not preclude the possibility of defining other terms that can achieve the same or similar functions in existing or future agreements.

[0242] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When these computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated.

[0243] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0244] It should be understood that in the various embodiments of this application, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0245] In summary, the above description is merely a preferred embodiment of the technical solution of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A communication method, characterized in that, The method includes: Based on the decompression model, the CSI information reported by the terminal based on the compression model is decompressed to obtain the first CSI information, which is used to represent the channel distribution at the current time. The drift detection model is invoked to process the first CSI information to obtain a first drift detection result; the first drift detection result is used to characterize the degree of deviation of the channel distribution corresponding to the first CSI information from the channel distribution corresponding to the second CSI information; the second CSI information is CSI information used to train the CSI compression model; the CSI compression model includes the decompression model and the compression model; Based on the first drift detection result, a first message is sent to the terminal, which is used to instruct the terminal to update the compression model.

2. The method as described in claim 1, characterized in that, If the drift probability in the first drift detection result is within the first probability range, the step of sending the first information to the terminal is executed; the drift probability refers to the probability that the channel distribution corresponding to the first CSI information deviates from the channel distribution corresponding to the second CSI information.

3. The method as described in claim 1, characterized in that, The step of sending the first information to the terminal based on the first drift detection result includes: From the first drift detection result and multiple drift detection results within the first time period before the current time, determine the number M of drift detection results whose drift probability is within the second probability range; If the number M is greater than or equal to N, the first information is sent to the terminal; M is a natural number and N is an integer greater than 1.

4. The method according to any one of claims 1 to 3, characterized in that, The method further includes: Based on the first drift detection result, a second message is sent to the terminal, which instructs the terminal to increase the CSI feedback cycle.

5. The method as described in claim 4, characterized in that, The step of sending second information to the terminal based on the first drift detection result includes: From the first drift detection result and multiple drift detection results within the first time period before the current time, determine the number M of drift detection results whose drift probability is within the second probability range; If the number M is less than N, a second message is sent to the terminal; M is a natural number and N is an integer greater than 1.

6. The method according to any one of claims 2 to 5, characterized in that, The drift detection model includes a feature extraction layer and a domain discrimination layer. The process of calling the drift detection model to process the first CSI information to obtain a first drift detection result includes: Based on the feature extraction layer, feature extraction is performed on the first CSI information to obtain the feature vector corresponding to the first CSI information; The drift probability is obtained by judging the feature vector corresponding to the first CSI information based on the domain discrimination layer.

7. The method according to any one of claims 1 to 6, characterized in that, The drift detection model is a trained domain-adaptive adversarial network model. Before calling the drift detection model to process the first CSI information and obtain the first drift detection result, the method further includes: Acquire multiple second CSI information entries and the performance values ​​of each second CSI information entry; The domain label of each second CSI information is determined based on the performance value and performance threshold of each second CSI information. Based on the multiple second CSI information and the domain labels of each second CSI information, a domain-adaptive adversarial network model is trained to obtain the trained domain-adaptive adversarial network model.

8. The method according to any one of claims 1 to 7, characterized in that, The first CSI information includes the CSI matrix and / or the precoding matrix corresponding to the CSI matrix.

9. A communication method, characterized in that, The method includes: Based on the prediction model and the CSI information in the first time period measured by actual measurements, the CSI information in the second time period is predicted; the CSI information is used to describe the channel distribution in the corresponding time period. Based on the CSI information in the first time period and the CSI information in the second time period, the distribution deviation is determined; the distribution deviation is used to characterize the degree of deviation of the predicted channel distribution described by the CSI information in the second time period relative to the channel distribution described by the CSI information in the first time period. Send first information to the network device, the first information including the distribution deviation degree and / or warning information, the warning information being used to indicate that the channel distribution has a distribution deviation degree greater than or equal to the distribution deviation threshold.

10. The method as described in claim 9, characterized in that, The distribution deviation threshold is related to multiple distribution deviations determined within a historical time period.

11. The method as described in claim 10, characterized in that, The distribution deviation threshold is a percentile value of multiple distribution deviations determined within a historical time period.

12. A communication method, characterized in that, The method includes: Receive the first message, Based on the first information, it is determined whether to send second information to the terminal. The first information includes distribution deviation and / or warning information. The warning information is used to indicate that the channel distribution has a distribution deviation greater than or equal to the distribution deviation threshold. The distribution deviation is used to characterize the degree of deviation of the channel distribution described by the CSI information in the predicted second time period from the channel distribution described by the CSI information in the first time period. The second information is used to instruct the terminal to update the compression model.

13. The method as described in claim 12, characterized in that, The distribution deviation threshold is related to multiple distribution deviations determined within a historical time period.

14. The method as described in claim 13, characterized in that, The distribution deviation threshold is the percentile value of multiple distribution deviations determined within a historical time period.

15. A communication device, characterized in that, The device includes at least one processor coupled to a memory storing a program or instructions, the processor executing the program or instructions to cause the device to perform the method as claimed in any one of claims 1 to 8, or the method as claimed in any one of claims 9 to 11, or the method as claimed in any one of claims 12 to 14.

16. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed, they cause the computer to perform the method as described in any one of claims 1 to 8, or the method as described in any one of claims 9 to 11, or the method as described in any one of claims 12 to 14.

17. A communication system, characterized in that, Includes the communication device as described in claim 15.

18. A computer program product, characterized in that, The computer program product includes a computer program that, when run, causes a computer to perform the method as claimed in any one of claims 1 to 8, or the method as claimed in any one of claims 9 to 11, or the method as claimed in any one of claims 12 to 14.

19. A chip system, characterized in that, The chip system includes one or more processors, the one or more processors being configured to call and execute instructions stored in memory, such that the method as claimed in any one of claims 1 to 8 is executed, or the method as claimed in any one of claims 9 to 11 is executed, or the method as claimed in any one of claims 12 to 14 is executed.

Images