WO2025255752A1 - 基于ai模型的通信方法、装置、设备、介质和程序产品 - Google Patents

基于ai模型的通信方法、装置、设备、介质和程序产品

Info

Publication number
WO2025255752A1
WO2025255752A1 PCT/CN2024/098789 CN2024098789W WO2025255752A1 WO 2025255752 A1 WO2025255752 A1 WO 2025255752A1 CN 2024098789 W CN2024098789 W CN 2024098789W WO 2025255752 A1 WO2025255752 A1 WO 2025255752A1
Authority
WO
WIPO (PCT)
Prior art keywords
model
uplink
models
information
terminal device
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
PCT/CN2024/098789
Other languages
English (en)
French (fr)
Inventor
陈文洪
曹建飞
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangdong Oppo Mobile Telecommunications Corp Ltd
Original Assignee
Guangdong Oppo Mobile Telecommunications Corp Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Guangdong Oppo Mobile Telecommunications Corp Ltd filed Critical Guangdong Oppo Mobile Telecommunications Corp Ltd
Priority to PCT/CN2024/098789 priority Critical patent/WO2025255752A1/zh
Publication of WO2025255752A1 publication Critical patent/WO2025255752A1/zh
Pending legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04WWIRELESS COMMUNICATION NETWORKS
    • H04W8/00Network data management
    • H04W8/22Processing or transfer of terminal data, e.g. status or physical capabilities
    • H04W8/24Transfer of terminal data

Definitions

  • This application relates to the field of wireless communication, and in particular to a communication method, apparatus, device, medium, and program product based on an AI model.
  • AI Artificial intelligence
  • This application provides a communication method, apparatus, device, medium, and program product based on an AI model, the technical solution of which includes at least:
  • Report model information of one or more AI models which are used by network devices for uplink CSI processing or uplink signal processing;
  • a communication method based on an AI model is provided, the method being executed by a network device, the method comprising:
  • Send scheduling information which is used to schedule the first terminal device to perform uplink transmission, and the uplink transmission is associated with at least one AI model among the one or more AI models.
  • a communication device based on an AI model comprising:
  • the sending module is used to report model information of one or more AI models, which are used by network devices for uplink CSI processing or uplink signal processing.
  • the sending module is further configured to perform uplink transmission according to the scheduling information of the network device, wherein the uplink transmission is associated with at least one AI model among the one or more AI models.
  • an AI model-based communication device comprising:
  • a receiving module is used to receive model information of one or more AI models, wherein the AI models are used by the device for uplink CSI processing or uplink signal processing;
  • the sending module is used to send scheduling information, which is used to schedule the first terminal device to perform uplink transmission, and the uplink transmission is associated with at least one AI model among the one or more AI models.
  • an AI model-based communication device comprising: a processor; a transceiver connected to the processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to load and execute the executable instructions to implement the AI model-based communication method as described in the foregoing aspects.
  • a computer-readable storage medium which stores at least one program that is loaded and executed by a processor to implement the AI model-based communication method as described in the foregoing aspects.
  • a computer program product or computer program including computer instructions stored in a computer-readable storage medium, a processor retrieving the computer instructions from the computer-readable storage medium, and the processor executing the computer instructions to implement the AI model-based communication method as described in the foregoing aspects.
  • a chip including a programmable logic circuit and/or at least a program, the chip being used to implement the AI model-based communication method as described in the foregoing aspects based on the programmable logic circuit and/or the at least a program.
  • the first terminal device It supports the reporting of AI model information by the first terminal device, making the AI model used by the network device for uplink CSI processing and/or uplink signal processing more compatible with the first terminal device. Regardless of the hardware architecture of the first terminal device, it ensures that the results of uplink CSI processing or uplink signal processing obtained by the network device using the AI model are consistent with the capabilities of the first terminal device, and avoids negative impacts of the AI model on the uplink transmission of the first terminal device.
  • Figure 1 shows a schematic diagram of the structure of a neuron provided in an exemplary embodiment of this application
  • Figure 2 shows a schematic diagram of the structure of a neural network provided in an exemplary embodiment of this application
  • Figure 3 shows a schematic diagram of a wireless communication system provided in an exemplary embodiment of this application
  • Figure 4 shows a flowchart of an AI model-based communication method provided in an exemplary embodiment of this application
  • Figure 5 shows a flowchart of an AI model-based communication method provided in an exemplary embodiment of this application
  • Figure 6 shows a flowchart of an AI model-based communication method provided in an exemplary embodiment of this application
  • Figure 7 shows a flowchart of an AI model-based communication method provided in an exemplary embodiment of this application.
  • Figure 8 shows a flowchart of an AI model-based communication method provided in an exemplary embodiment of this application
  • Figure 9 shows a flowchart of an AI model-based communication method provided in an exemplary embodiment of this application.
  • Figure 10 shows a flowchart of an AI model-based communication method provided in an exemplary embodiment of this application
  • Figure 11 shows a structural block diagram of an AI model-based communication device provided in an exemplary embodiment of this application.
  • Figure 12 shows a structural block diagram of an AI model-based communication device provided in an exemplary embodiment of this application
  • Figure 13 shows a schematic diagram of the structure of an AI model-based communication device provided in an exemplary embodiment of this application.
  • first, second, third, etc. may be used in this application to describe various information, this information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another.
  • first information may also be referred to as second information
  • second information may also be referred to as first information.
  • AI Artificial Intelligence
  • AI is a theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results.
  • AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence.
  • AI studies the design principles and implementation methods of various intelligent machines, enabling them to possess the functions of perception, reasoning, and decision-making.
  • AI technology is a comprehensive discipline encompassing a wide range of fields, including both hardware and software technologies.
  • Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating/interactive systems, and mechatronics.
  • AI software technologies primarily include computer vision, speech processing, natural language processing, and machine learning (ML)/deep learning.
  • AI models are an important component of AI technology, used for tasks such as processing, analysis, and prediction.
  • Common AI models include various types, such as neural networks, decision trees, support vector machines, random forests, and mean-based algorithms.
  • training can be divided into offline training and online training.
  • Offline training allows the network device to obtain a static training result by training offline using a dataset.
  • the network device can continue to collect more data for real-time online training to optimize the neural network model's parameters, achieving better inference and prediction results.
  • the corresponding model output can be obtained by inputting the currently obtained information into the model.
  • AI models can be introduced, with corresponding inputs and outputs defined.
  • channel information such as feature vectors, beam information, and delay information
  • CSI Channel State Information
  • a corresponding neural network model takes the CSI quantization bits as input to infer the corresponding channel information.
  • the terminal device takes the reference signal received power (RSRP) corresponding to multiple beams (CSI-RS resources) in the second beam set as input, infers the best beam (CSI-RS resource index) in the first beam set and its corresponding RSRP based on the AI model, and reports the inference result to the network device.
  • RSRP reference signal received power
  • CSI-RS resources multiple beams
  • CSI-RS resource index the best beam
  • AI models can also be used for other processes such as localization, channel coding, data channel decoding, modulation and demodulation, and channel estimation.
  • terminal devices can be configured with a large number of antenna elements to improve uplink transmission performance.
  • terminal devices can perform multiple-input multiple-output (MIMO) transmission across a large number of antenna ports, uplink multi-beamforming, and other technologies, thereby improving uplink transmission performance.
  • MIMO multiple-input multiple-output
  • massive MIMO requires a large amount of uplink reference signal for uplink channel detection or uplink beam management, significantly increasing uplink resource overhead and impacting uplink throughput.
  • network devices can use AI models for assisted reception, the required uplink reference signal can be greatly reduced, achieving the desired performance with only a small amount of uplink reference signal, thus reducing uplink resource overhead.
  • this application proposes a communication method based on an AI model, which ensures good compatibility between the AI model used for uplink transmission and the terminal device, thereby avoiding negative impacts of the AI model on uplink transmission performance.
  • FIG 3 illustrates a schematic diagram of a wireless communication system 100 provided in an exemplary embodiment of this application.
  • the wireless communication system 100 includes terminal devices with terminal devices, or terminal devices with network devices, or stations (STAs) with stations; this application does not limit the specific examples.
  • Figure 1 illustrates an example where the wireless communication system 100 includes network device 110 and terminal device 120.
  • the network device 110 in this application supports providing wireless communication functions, including but not limited to: Node B (NB), Evolved Node B (eNB), Next Generation Node B (gNB), Radio Network Controller (RNC), Base Station (BS), Base Station Controller (BSC), Base Transceiver Station (BTS), Home Evolved Node B (or Home Node B, HNB), Baseband Unit (BBU), Distributed Unit (DU), Wireless Relay Node, Wireless Backhaul Node, Transmission Point (TP), Transmission and Reception Point (TRP), Antenna Panel, Router, etc.
  • Node B Node B
  • eNB Evolved Node B
  • gNB Next Generation Node B
  • RNC Radio Network Controller
  • Base Station Base Station Controller
  • BSC Base Station Controller
  • BTS Base Transceiver Station
  • HNB Baseband Unit
  • BBU Baseband Unit
  • DU Distributed Unit
  • Wireless Relay Node Wireless Backhaul Node
  • TP Transmission Point
  • TRP Transmission and Reception Point
  • the terminal device 120 in this application also referred to as user equipment (UE), includes, but is not limited to: mobile phones, tablets, e-book readers, laptops, desktop computers, televisions, virtual reality (VR) devices, augmented reality (AR) devices, mixed reality (MR) devices, extended reality (XR) devices, remote terminals, set-top boxes, vehicle communication equipment, handheld devices, wearable devices, wireless devices in industrial control, wireless devices in self-driving, wireless devices in remote medical care, and smart grids.
  • VR virtual reality
  • AR augmented reality
  • MR mixed reality
  • XR extended reality
  • Wireless devices in Smart Grid, Transportation Safety, Smart City, and Smart Home can also be computing devices with wireless communication capabilities or other processing devices connected to wireless modems, etc.
  • both network device 110 and terminal device 120 support the 3rd Generation Partnership Project (3GPP) protocol, but are not limited to the 3GPP protocol.
  • 3GPP 3rd Generation Partnership Project
  • the frequency bands supported by the wireless communication system 100 include, but are not limited to: centimeter wave bands (such as bands in the range of 450MHz-6GHz, also called Sub-6GHz bands), millimeter wave (mmWave) bands (such as 45GHz, 60GHz, etc., belonging to the range of 30-300GHz), and low-frequency bands.
  • centimeter wave bands such as bands in the range of 450MHz-6GHz, also called Sub-6GHz bands
  • millimeter wave (mmWave) bands such as 45GHz, 60GHz, etc., belonging to the range of 30-300GHz
  • low-frequency bands include Sub-7GHz bands (such as 2.4GHz, 5GHz, 6GHz, etc., belonging to the range of 1-7.25GHz). (Frequency bands within the GHz range).
  • This application mainly involves two communication scenarios: one is the uplink transmission scenario, which refers to the scenario where the terminal device sends signals/data to the network device; the other is the downlink transmission scenario, which refers to the scenario where the network device sends signals/data to the terminal device.
  • LTE Long Term Evolution
  • LTE-A Advanced Long Term Evolution
  • NR New Radio
  • evolution systems of NR systems LTE-based access to unlicensed spectrum
  • LTE-U LTE-based access to unlicensed spectrum
  • NR-U NR-based access to unlicensed spectrum
  • 5G 5th Generation
  • B5G 5th Generation
  • 6G 6G and subsequent evolution systems
  • Wireless Local Area Networks (WLAN) systems Wireless Fidelity
  • Wi-Fi Global System of Mobile communication
  • CDMA Code Division Multiple Access
  • WCDMA Wideband Code Division Multiple Access
  • GPRS General Packet Radio Service
  • TN Non-Terrestrial Networks
  • UMTS Universal Mobile Telecommunication System
  • WiMAX Worldwide Interoperability for Microwave Access
  • Figure 4 illustrates a flowchart of an AI model-based communication method provided in an exemplary embodiment of this application. The method is executed by a first terminal device and includes at least some of the following steps:
  • Step 420 Report model information for one or more AI models.
  • the AI models are used by network devices for uplink CSI processing or uplink signal processing.
  • the AI models in this application include one or more of the following types: neural networks, decision trees, support vector machines, random forests, mean algorithms, etc.
  • the AI model may also be referred to as an AI rule, an AI function, or other names.
  • Uplink CSI processing refers to the processing of CSI for uplink signals. If an AI model is used by network devices to perform uplink CSI processing, the network devices can use the AI model to obtain/predict/infer the processing results of uplink CSI.
  • Uplink signal processing refers to the processing of uplink signals at the receiving end. If an AI model is used in network devices for uplink signal processing, the network devices can use the AI model to obtain/predict/infer the processing results of the uplink signals.
  • Step 440 Perform an uplink transmission based on the scheduling information of the network device, which is associated with at least one AI model among one or more AI models.
  • At least one includes both “equal to one” and “more than one”. That is, at least one AI model can be considered as one AI model or multiple AI models. The use of “at least one” below will be understood in accordance with this explanation.
  • uplink transmissions based on network device scheduling information, associated with at least one of one or more AI models may include the following three scenarios:
  • the first terminal device reports an AI model, and the uplink transmission performed by the first terminal device based on the scheduling information is associated with this AI model.
  • the first terminal device reports multiple AI models, and the uplink transmission performed by the first terminal device based on the scheduling information is associated with these multiple AI models.
  • the first terminal device reports multiple AI models, and the uplink transmission performed by the first terminal device according to the scheduling information is associated with a portion of these multiple AI models. For example, the first terminal device reports z AI models, where z is an integer greater than 1, and the uplink transmission is associated with z’ AI models out of the z AI models, where 1 ⁇ z’ ⁇ z.
  • the first terminal device performing uplink transmission refers to the first terminal device sending signals and/or data to the network device.
  • the method provided in this application embodiment supports the reporting of AI model information by the first terminal device, making the AI model used by the network device for uplink CSI processing or uplink signal processing more compatible with the first terminal device. Regardless of the hardware structure of the first terminal device, it ensures that the results of uplink CSI processing or uplink signal processing obtained by the network device using the AI model are consistent with the capabilities of the first terminal device, and avoids negative impacts of the AI model on the uplink transmission of the first terminal device.
  • step 420 can be further implemented as step 520
  • step 440 can be further implemented as step 540, as shown in FIG5.
  • the first terminal device may also execute one or more of the following optional steps: step 560 and step 580.
  • Figure 5 illustrates a flowchart of an AI model-based communication method provided in an exemplary embodiment of this application. The method is executed by a first terminal device and includes at least some of the following steps:
  • Step 520 Report model information for one or more AI models.
  • the AI models are used by network devices for uplink CSI processing or uplink signal processing.
  • model information includes one or more of the following: model dataset, model structure, and model parameters.
  • the model structure can be any one of several candidate model structures, such as: transformation. Transformer architecture, Residual Neural Network (ResNet), Recurrent Neural Network (RNN), Convolutional Neural Networks (CNN), Support Vector Machine (SVM), Long Short-Term Memory (LSTM), DNN, etc.
  • the model architecture may also include the number of neural network layers, quantization methods, etc.
  • Model parameters include the parameters used in each layer of the neural network.
  • a neural network typically includes at least an input layer, hidden layers, and an output layer.
  • the model dataset may include multiple different datasets, each corresponding to different input information and/or different output information.
  • different datasets may correspond to different combinations of values, such as combinations of N and M.
  • the model information may further include at least one of the following: model functionality, model identifier (ID), input information, output information, and dataset ID.
  • model functionality model functionality
  • ID model identifier
  • input information output information
  • dataset ID dataset ID
  • the model functions are related to uplink CSI processing or uplink signal processing.
  • the model functions include at least one of the following: uplink partial CSI recovery, uplink beam information recovery, and uplink nonlinearity compensation.
  • Uplink CSI recovery and uplink beam information recovery are related functions of uplink CSI processing, while uplink nonlinearity compensation is related to uplink signal processing.
  • uplink CSI recovery may also include CSI prediction functionality (i.e., uplink CSI recovery and prediction).
  • uplink beam information recovery may also include beam prediction functionality (i.e., uplink beam information recovery and prediction).
  • one or more AI models are trained by a first terminal device. That is, the first terminal device reports the model information of one or more AI models trained by itself to the network device.
  • one or more AI models are trained by a second terminal device. That is, the first terminal device reports the model information of one or more AI models trained by the second terminal device to the network device.
  • the second terminal device sends the one or more AI models to the first terminal device.
  • the second terminal device sends the model information of the one or more AI models to the first terminal device.
  • the second terminal device uses the same hardware configuration as the first terminal device.
  • This hardware configuration includes, for example, one or more of the following: antenna configuration, analog beamforming method, and power amplifier performance.
  • the antenna configuration includes, for example, one or more of the following parameters: number of antennas, antenna arrangement, whether the antennas are fully coherent, whether the antennas are partially coherent, and the number of antenna ports.
  • the analog beamforming method includes, for example, one or more of the following parameters: number of analog beams, analog beam direction, analog beam bandwidth, and analog beamforming matrix.
  • the power amplifier performance includes, for example, one or more of the following parameters: gain, bandwidth, linearity, efficiency, and distortion.
  • the AI model Since the AI model is trained on the terminal device side, it does not rely on data from network devices for model training, which helps improve the matching degree between the AI model and the terminal device. Because model information is mainly related to hardware configuration, and the first and second terminal devices have the same hardware configuration, this training method does not affect the training results and can greatly improve the efficiency of obtaining training datasets.
  • one or more AI models are trained by a network device, and the dataset used to train these AI models is reported by a terminal device.
  • the network device can train the AI models based on the dataset reported by one or more terminal devices (e.g., a first terminal device and/or a second terminal device).
  • the AI models are trained on the network device side, the dataset used for training is provided by the terminal device side. Therefore, the training of the AI models still does not depend on the data from the network device, and the trained AI models have a high degree of compatibility with the terminal devices.
  • training the AI models on the network device side can reduce the complexity and power consumption pressure on the terminal device side and improve the efficiency of AI model training.
  • the AI model is used by the network device to perform uplink CSI processing, and the uplink CSI processing includes uplink partial CSI recovery; the input information of the AI model includes uplink CSI corresponding to M antenna ports, and the output information of the AI model includes uplink CSI corresponding to N antenna ports; where M is an integer greater than or equal to 1, N is an integer greater than 1, and M is less than N.
  • the input information of the AI model includes uplink CSI corresponding to M antenna ports and K sub-bands
  • the output information of the AI model includes uplink CSI corresponding to N antenna ports and L sub-bands; where K is an integer greater than or equal to 1, L is an integer greater than 1, and K is less than L.
  • Subband is a granularity for dividing frequency domain resources.
  • This application uses subband as an example for illustrative purposes.
  • frequency domain resources can also be divided according to other granularities, such as: frequency band, bandwidth part (BWP), carrier, physical resource block (PRB), subchannel, subcarrier, etc.
  • K subbands can also be replaced by K frequency bands, or K BWPs, or K PRBs, or K subchannels, or K carriers, or K subcarriers, etc.
  • the antenna of the first terminal device is a partially coherent antenna
  • the first terminal device includes h coherent antenna port groups, where h is an integer greater than 1; the M antenna ports include at least one antenna port of each of the h antenna port groups.
  • the M antenna ports include M Sounding Reference Signal (SRS) ports, the first...
  • SRS Sounding Reference Signal
  • SRS is an uplink reference signal, that is, a reference signal sent by a terminal device to a network device.
  • This application primarily uses SRS as an example when discussing uplink reference signals.
  • SRS can be replaced by other uplink reference signals, such as a demodulation reference signal (DMRS), a phase tracking reference signal (PT-RS), or an uplink reference signal specified in the future communication protocol.
  • DMRS demodulation reference signal
  • PT-RS phase tracking reference signal
  • uplink reference signal specified in the future communication protocol such as a demodulation reference signal (DMRS), a phase tracking reference signal (PT-RS), or an uplink reference signal specified in the future communication protocol.
  • DMRS demodulation reference signal
  • PT-RS phase tracking reference signal
  • N and M There are two possible values for N and M:
  • the input information of the AI model also includes the values of N and M.
  • Scenario 2 In some embodiments, when there are multiple AI models, that is, when the first terminal device reports the model information of multiple AI models: different AI models correspond to different value combinations, which are combinations of N and M; or, the input information of the AI model also includes the values of N and M.
  • the values of N and M can also be part of the model information of the AI model.
  • the model functions include the recovery of uplink beam information.
  • the input information of the AI model includes the received signal measurement values corresponding to M SRS resources or M beams
  • the output information of the AI model includes at least one SRS Resource Indicator (SRI), where each SRI is used to indicate one of the N SRS resources; wherein M is an integer greater than or equal to 1, N is an integer greater than 1, and M is less than N.
  • SRI SRS Resource Indicator
  • At least one SRI means that the number of SRIs can be equal to or greater than 1. That is, the output information of the AI model includes one or more SRIs. Each SRI is used to indicate one of the N SRS resources, and different SRIs indicate different SRS resources. Optionally, one SRS resource corresponds to one or more antenna ports, and the antenna ports corresponding to different SRS resources are completely different or partially the same.
  • uplink beam information recovery may also include beam prediction functionality, meaning that the SRI may correspond to the current best beam or the predicted best beam for a future time.
  • the uplink transmission scheduling information is used to configure N SRS resources, which include M SRS resources, and the uplink transmission includes SRS transmission on the M SRS resources.
  • the scheduling information configures N SRS resources, and the first terminal device performs SRS transmission on M of the N SRS resources according to the scheduling information.
  • the input information for the AI model also includes the values of N and M.
  • the first terminal device reports model information of multiple AI models: different AI models correspond to different combinations of values, which are combinations of values of N and M; or, the input information of the AI model also includes the values of N and M.
  • N and M There are two possible values for N and M:
  • the input information of the AI model also includes the values of N and M.
  • Scenario 2 In some embodiments, when there are multiple AI models, that is, when the first terminal device reports the model information of multiple AI models: different AI models correspond to different value combinations, which are combinations of N and M; or, the input information of the AI model also includes the values of N and M.
  • the values of N and M can also be part of the model information of the AI model.
  • the model functions include uplink nonlinear compensation.
  • the AI model is used by the network device for uplink signal processing.
  • the input information of the AI model includes the modulated signal detected by the network device, and the output information of the AI model includes the modulated signal after nonlinear compensation.
  • the uplink signal processing includes uplink nonlinear compensation.
  • model information can be reported via UE capabilities, or via Medium Access Control (MAC) layer signaling, or via Uplink Control Information (UCI). This design helps ensure the flexibility of model information reporting.
  • MAC Medium Access Control
  • UCI Uplink Control Information
  • step 520 Other related content in step 520 can be found in step 420, and will not be repeated here.
  • Step 540 Perform an uplink transmission based on the scheduling information of the network device, which is associated with at least one AI model among one or more AI models.
  • the uplink transmission is associated with at least one of one or more AI models, which is reflected in one or more of the following aspects: the antenna port used for the uplink transmission is the same as the antenna port used for the input information in the training dataset of at least one AI model; the transmit beam used for the uplink transmission is the same as the transmit beam used for the input information in the training dataset of at least one AI model; and the power amplifier used for the uplink transmission is the same as the power amplifier used for the input information in the training dataset of at least one AI model.
  • the association between the uplink transmission and at least one of the AI models can also be reflected through other physical layer technologies.
  • These include modulation schemes, coding schemes, waveforms, bandwidth, transmission rate, communication methods (e.g., simplex, half-duplex, full-duplex), information transmission methods (e.g., serial transmission, parallel transmission), and resource mapping methods (e.g., centralized resource allocation, distributed resource allocation).
  • communication methods e.g., simplex, half-duplex, full-duplex
  • information transmission methods e.g., serial transmission, parallel transmission
  • resource mapping methods e.g., centralized resource allocation, distributed resource allocation.
  • resource allocation methods e.g., resource allocation methods.
  • the AI model associated with the uplink transmission among one or more AI models is determined based on first indication information in the scheduling information.
  • the scheduling information includes the first indication information
  • the first terminal device determines at least one AI model associated with the uplink transmission among one or more AI models based on the first indication information.
  • the scheduling information sent by the network device is associated with at least one AI model among one or more AI models. This is reflected in one or more of the following aspects: the scheduling information is associated with the input information of at least one AI model; the scheduling information is associated with the output information of at least one AI model; the scheduling information is associated with the prediction/inference results of at least one AI model; and the uplink transmission scheduled by the scheduling information is associated with at least one AI model.
  • the first terminal device before performing step 540, receives scheduling information sent by the network device.
  • step 540 Other relevant information in step 540 can be found in step 440, and will not be repeated here.
  • Step 560 Receive the second instruction information.
  • the second instruction information is used to instruct the first terminal device to update one or more AI models, or to instruct the first terminal device to adopt a transmission method not based on AI models.
  • the first terminal device adopting a transmission method not based on AI models can also be understood as the first terminal device's uplink transmission being unrelated to AI models.
  • Step 580 Report updated model information for one or more AI models.
  • the updated model information includes: the updated model structure, and the updated model parameters.
  • the updated model information may also include one or more of the following: updated model functionality, updated model ID, updated input information, updated output information, updated model dataset, and updated dataset ID.
  • the first terminal device reports complete updated model information. That is, after updating one or more AI models, the first terminal device reports the model information of the updated one or more AI models.
  • the updated model information reported by the first terminal device is not complete model information, which helps reduce resource overhead. Three possibilities are described here:
  • the first terminal device only reports the model information that has changed after the update
  • the first terminal device reports the model structure, model parameters, model functions, model ID, input information, and output information.
  • the first terminal device updates one or more AI models. After the update, only the input information and output information have changed. Therefore, in step 580, the updated model information reported by the first terminal device will only include the updated input information and the updated output information.
  • the first terminal device reports the model structure, model function, model ID, input information and output information.
  • the first terminal device updates one or more AI models. After the update, only the model parameters have changed. Therefore, in step 580, the updated model information reported by the first terminal device will only include the updated model parameters.
  • the first terminal device reports the model information of the AI model that has changed after the update.
  • the first terminal device only reports the differential information of the model information, which is used to represent the difference between the updated model information and the model information before the update.
  • the first terminal device reports the model structure, model parameters, model functions, model ID, input information, and output information.
  • the first terminal device updates one or more AI models.
  • the updated model information reported by the first terminal device includes: differential information of model structure, differential information of model parameters, differential information of model functions, differential information of model ID, differential information of input information, and differential information of output information.
  • the first terminal device reports the model structure, model parameters, model functions, input information, and output information.
  • the first terminal device updates one or more AI models.
  • the updated model information reported by the first terminal device includes: differential information of the model structure, differential information of the model parameters, differential information of the model functions, differential information of the input information, and differential information of the output information.
  • the first terminal device only reports the differential information of the model information that has changed after the update.
  • the updated model information can be any one or more of the following: model structure, model parameters, model functionality, model ID, input information, output information, model dataset, and dataset ID. Not all possibilities are listed here. However, it should be understood that which model information changes after an update is influenced by factors such as the actual communication environment, communication services, communication requirements, and the characteristics of the AI model itself. This application supports the possibility that any one or more types of model information may change after an update.
  • the update of the AI model by the first terminal device may be based on the second instruction information, or it may be unrelated to the second instruction information. That is, the update of the AI model by the first terminal device may occur after receiving the second instruction information instructing the updating of one or more AI models, or it may occur without receiving the second instruction information.
  • the first terminal device may periodically update one or more AI models.
  • the first terminal device may update one or more AI models under certain conditions, which may be related to factors such as communication quality, service requirements, model information, the capabilities of the first terminal device itself, the battery level of the first terminal device, and the memory of the first terminal device.
  • this application supports the first terminal device updating the AI model under the instruction or trigger of the second instruction information, and also supports the first terminal device spontaneously updating the AI model based on predefined rules or conditions. Therefore, step 560 is not a necessary condition for step 580. It should be considered an optional situation.
  • the first terminal device's adoption of a transmission method not based on the AI model may be based on the second instruction information, or it may be unrelated to the second instruction information. That is, the first terminal device may adopt a transmission method not based on the AI model after receiving the second instruction information indicating the adoption of such a method, or it may occur without receiving the second instruction information.
  • the first terminal device may maintain a timer (for distinction, this can be called the first timer; the existence of the first timer is not necessarily related to the existence of the second timer), and the AI model-based transmission method may stop after the timer starts counting or expires.
  • the first terminal device may adopt a transmission method not based on the AI model under certain conditions, which may be related to communication quality, service requirements, model information, the capabilities of the first terminal device itself, the battery level of the first terminal device, the memory of the first terminal device, etc.
  • this application supports the first terminal device adopting a transmission method not based on the AI model under the instruction or trigger of the second instruction information, and also supports the first terminal device spontaneously adopting a transmission method not based on the AI model based on predefined rules or conditions. Therefore, step 560 is not a necessary condition for the first terminal device to adopt a transmission method not based on the AI model, but should be considered as an optional situation.
  • steps 560 and 580 are optional.
  • the first terminal device may execute only steps 520 and 540, or the first terminal device may execute steps 520, 540 and 560, or the first terminal device may execute steps 520, 540 and 580, or the first terminal device may execute steps 520, 540 and 560 and use a transmission method not based on the AI model, or the first terminal device may execute steps 520, 540 and 560 and use a transmission method not based on the AI model.
  • the method provided in this application embodiment uses an AI model corresponding to the model information reported by the first terminal device, which is trained on the terminal device side or by the network device based on the model dataset reported by the terminal device. Since the second terminal device has the same hardware configuration as the first terminal device, regardless of whether the AI model is trained by the first terminal device, the second terminal device, or even the network device, the AI model's compatibility with the first terminal device is extremely high, perfectly matching the hardware capabilities and uplink transmission requirements of the first terminal device. The network device's use of the AI model will not negatively impact uplink transmission performance. Furthermore, it supports the first terminal device updating the AI model, allowing the AI model to better adapt to changing communication environments and requirements, providing a more flexible AI model-based communication method.
  • Figure 6 illustrates a flowchart of an AI model-based communication method provided in an exemplary embodiment of this application. The method is executed by a network device and includes at least some of the following steps:
  • Step 620 Receive model information of one or more AI models, which are used by network devices for uplink CSI processing or uplink signal processing.
  • the AI models in this application include one or more of the following types: neural networks, decision trees, support vector machines, random forests, mean algorithms, etc.
  • the AI model may also be referred to as an AI rule, an AI function, or other names.
  • Uplink CSI processing refers to the processing of CSI for uplink signals. If an AI model is used by network devices to perform uplink CSI processing, the network devices can use the AI model to obtain/predict/infer the processing results of uplink CSI.
  • Uplink signal processing refers to the processing of uplink signals at the receiving end. If an AI model is used in network devices for uplink signal processing, the network devices can use the AI model to obtain/predict/infer the processing results of the uplink signals.
  • Step 640 Send scheduling information, which is used to schedule the first terminal device to perform uplink transmission, and the uplink transmission is associated with at least one AI model among one or more AI models.
  • At least one AI model which can be considered as one AI model or multiple AI models.
  • the first terminal device performs uplink transmission, which means that the first terminal device sends signals and/or data to the network device.
  • the method provided in this application embodiment supports network devices in using received AI models for uplink CSI processing or uplink signal processing. Since the model information of the AI model is reported by the first terminal device, and the uplink transmission scheduled by the network device is associated with the AI model, regardless of the hardware structure of the first terminal device, it can ensure that the results of uplink CSI processing or uplink signal processing conform to the capabilities of the first terminal device, and can avoid negative impacts of the AI model on the uplink transmission of the first terminal device.
  • step 620 can be further implemented as step 710
  • step 640 can be further implemented as step 720, as shown in FIG7.
  • the first terminal device can also execute one or more of the following optional steps: step 730, step 740, and step 750.
  • Figure 7 illustrates a flowchart of an AI model-based communication method provided in an exemplary embodiment of this application. The method is executed by a network device and includes at least some of the following steps:
  • Step 710 Receive model information of one or more AI models, which are used by network devices for uplink CSI processing or uplink signal processing.
  • step 710 please refer to steps 520 and 620; they will not be repeated here.
  • Step 720 Send scheduling information, which is used to schedule the first terminal device to perform uplink transmission, which is associated with at least one AI model among one or more AI models.
  • step 720 please refer to steps 540 and 640; they will not be repeated here.
  • Step 730 Based on the target AI model in at least one AI model and the uplink transmission, perform uplink CSI processing or uplink signal processing.
  • the network device performs at least one of the following processes:
  • the received signal measurement value corresponding to the uplink partial beam is obtained.
  • the received signal measurement value corresponding to the uplink partial beam is used as the input information of the target AI model to obtain the SRI corresponding to the optimal beam in the complete uplink beam.
  • the modulated signal detected by the network device is used as the input information of the target AI model to obtain the modulated signal after nonlinear compensation.
  • the received signal measurement value can be represented by any one or more of the following parameters: RSRP, Received Signal Strength Indication (RSSI), Reference Signal Receiving Quality (RSRQ), Signal to Interference plus Noise Ratio (SINR), and Signal-to-Noise Ratio (SNR).
  • RSSI Received Signal Strength Indication
  • RSSRQ Reference Signal Receiving Quality
  • SINR Signal to Interference plus Noise Ratio
  • SNR Signal-to-Noise Ratio
  • Step 740 Send the second instruction message.
  • the second instruction information is used to instruct the first terminal device to update one or more AI models, or to instruct the first terminal device to adopt a transmission method not based on AI models.
  • the first terminal device adopting a transmission method not based on AI models can also be understood as the first terminal device's uplink transmission being unrelated to AI models.
  • the network device sends a second indication message based on the performance monitoring results of one or more AI models. For example, the network device monitors the performance of one or more AI models, and when the performance of one or more AI models deteriorates (this could be a deterioration in the performance of all AI models or only a portion of the AI models), it sends the second indication message to the first terminal device. For example, the network device monitors the performance of one or more AI models, and when the model information of one or more AI models reaches a threshold, it sends the second indication message to the first terminal device.
  • the network device periodically sends a second indication message.
  • the updated model information includes: the updated model structure, and the updated model parameters.
  • the updated model information may also include one or more of the following: updated model functionality, updated model ID, updated input information, updated output information, updated model dataset, and updated dataset ID.
  • the network device receives complete updated model information. That is, the network device receives updated model information for one or more AI models.
  • the updated model information received by the network device is not the complete model information, which helps reduce resource overhead. There are three possibilities: first, the network device receives the updated model information that has changed; second, the network device receives differential information of the model information; and third, the network device receives differential information of the updated model information that has changed.
  • step 750 can be found in step 580, and will not be repeated here.
  • steps 730, 740, and 750 are all optional.
  • the network device may execute only steps 710 and 720, or execute steps 710, 720, and 730, or execute steps 710, 720, and 740, or execute steps 710, 720, and 750, or execute steps 710, 720, 730, and 740, or execute steps 710, 720, 730, and 750, or execute step 7... 10.
  • Steps 720, 740, and 750 or the network device performs steps 710, 720, 730, 740, and 750, or the network device performs steps 710 and 720 and receives uplink transmissions not based on the AI model, or the network device performs steps 710, 720, and 730 and receives uplink transmissions not based on the AI model, or the network device performs steps 710, 720, and 740 and receives uplink transmissions not based on the AI model.
  • the method provided in this application embodiment uses an AI model trained on the terminal device side, or trained by the network device based on the model dataset reported by the terminal device. Since the second terminal device has the same hardware configuration as the first terminal device, regardless of whether the AI model is trained by the first terminal device, the second terminal device, or even the network device, the AI model's compatibility with the first terminal device is extremely high, perfectly matching the hardware capabilities and uplink transmission requirements of the first terminal device. Using the AI model will not negatively impact uplink transmission performance. Furthermore, it supports updating the AI model on the first terminal device, allowing the AI model to better adapt to changing communication environments and requirements, providing a more flexible AI model-based communication method.
  • Step 820 The first terminal device reports model information of one or more AI models to the network device.
  • the AI models are used by the network device for...
  • Uplink CSI processing includes uplink partial CSI recovery.
  • model information includes one or more of the following: model dataset, model structure, and model parameters.
  • a dataset may contain several data samples.
  • each data sample may contain multiple features and one or more labels.
  • Features describe the attributes of the data sample, and labels indicate the category or outcome to which the data sample belongs.
  • the input information may include one or more of the following: the number of input vectors, the size of each vector, and the specific meaning of the input.
  • the output information may include one or more of the following: the number of output vectors, the size of each vector, and the specific meaning of the output.
  • one or more AI models are trained by a first terminal device.
  • one or more AI models are trained by a second terminal device.
  • one or more AI models are trained by a network device, and the dataset used to train these AI models is reported by a terminal device.
  • the network device can train the AI models based on the dataset reported by one or more terminal devices (e.g., a first terminal device and/or a second terminal device).
  • the training dataset is provided by the terminal device side. Therefore, the training of the AI models does not depend on the data from the network device, and the trained AI models have a high degree of compatibility with the terminal devices.
  • training the AI models on the network device side can reduce the complexity and power consumption pressure on the terminal device side and improve the efficiency of AI model training.
  • the input information of the AI model includes uplink CSI corresponding to M antenna ports, and the output information includes uplink CSI corresponding to N antenna ports; M is an integer greater than or equal to 1, N is an integer greater than 1, and M is less than N.
  • M represents the number of uplink antenna ports currently being measured
  • N represents the total number of uplink antenna ports of the terminal device.
  • the CSI can be represented as a channel matrix, a channel covariance matrix, a channel eigenvector, or a precoding matrix, etc. Furthermore, the CSI can be a sub-band CSI or a wideband CSI.
  • the M antenna ports include at least one antenna port from each of the several coherent antenna port groups.
  • the antenna ports of the first terminal device contain P (P is an integer greater than 1) coherent antenna port groups
  • the M antenna ports include at least P antenna ports, and each of the P antenna ports comes from a different antenna port group. Since the phases of incoherent antenna ports are independent, the AI model cannot estimate the CSI corresponding to an antenna port in another antenna port group based on the CSI corresponding to an antenna port in one antenna port group. Therefore, it is necessary to include at least one port from each antenna port group to ensure the performance of inference/prediction.
  • the first terminal device can report multiple AI models to the network device, with different AI models corresponding to different combinations of N and M values. This allows the network device to flexibly configure different numbers of antenna ports for uplink CSI measurements, saving uplink reference signal overhead.
  • different AI models can correspond to different values of M but the same value of N, because the total number of antenna ports on the first terminal device is generally constant.
  • the values of N and M are also used as input information for the AI model. Based on this, the AI model can match the most suitable model parameters according to the values of N and M, thereby achieving optimal performance.
  • the input information of the AI model includes uplink CSI corresponding to M antenna ports and K sub-bands, and the output information includes...
  • the uplink CSI corresponds to N antenna ports and L subbands; K is an integer greater than or equal to 1, L is an integer greater than 1, and K is less than L.
  • the AI model can be used for CSI recovery in both the spatial domain (antenna ports) and the frequency domain (subbands).
  • subbands can also be replaced by PRBs, subcarriers, or subchannels, etc.
  • the input information of the AI model contains uplink CSI corresponding to M antenna ports and K PRBs, and the output information contains uplink CSI corresponding to N antenna ports and L PRBs.
  • model information can be reported via UE capabilities, MAC layer signaling, or UCI. This design helps ensure the flexibility of model information reporting.
  • step 820 Other relevant content in step 820 can be found in steps 520 and/or 710, and will not be repeated here.
  • Step 840 The network device schedules the first terminal device to perform uplink transmission through scheduling information, and the uplink transmission is associated with at least one AI model among one or more AI models.
  • the scheduling information includes first indication information, which is used to indicate the AI model associated with the uplink transmission from one or more AI models.
  • the first indication information may indicate one or more of the following: model ID, function ID, dataset ID, model index (i.e., the sequence number of the AI model among all models reported by the first terminal device), etc.
  • model ID, function ID, dataset ID, and model index may be included in the model information reported by the first terminal device, so that the first terminal device can determine the AI model associated with the uplink transmission from the reported AI models based on the first indication information.
  • the association between uplink transmission and the AI model is reflected in the fact that the antenna port (antenna) used for uplink transmission is the same as the antenna port (antenna) corresponding to the input information in the training dataset of the AI model. That is, the first terminal device needs to use the antenna port (antenna) used when training the AI model for uplink transmission, thereby ensuring the consistency between inference and training.
  • the uplink transmission is an SRS transmission on M SRS antenna ports, and the SRS transmission is associated with one of one or more AI models.
  • scheduling information is used to schedule/configure the transmission on the M SRS ports.
  • the network device only needs to configure M SRS antenna ports to obtain the CSI of N SRS antenna ports, thereby saving the overhead of reference signals. If the network device configures M SRS ports, it needs to indicate the AI model associated with the M SRS ports through first indication information.
  • the input of the AI model is the CSI of the M antenna ports.
  • the M antenna ports (i.e., physical antennas) of the SRS transmission need to be the same as the M antenna ports (i.e., physical antennas) used during the training of the associated AI model.
  • step 840 can be found in steps 540 and/or 720, and will not be repeated here.
  • Step 860 The first terminal device performs uplink transmission according to the scheduling information, which is associated with at least one AI model among one or more AI models.
  • the first terminal device determines the AI model associated with the uplink transmission from one or more AI models based on the first indication information in the scheduling information.
  • the uplink transmission is an SRS transmission on M SRS ports, and the SRS transmission is associated with one of the AI models in one or more AI models.
  • step 860 can be found in steps 540 and/or 840, and will not be repeated here.
  • Step 880 The network device performs uplink CSI processing based on the target AI model in at least one AI model and the uplink transmission.
  • the uplink CSI processing includes uplink partial CSI recovery.
  • the target AI model when at least one AI model contains only one AI model, that is, when the uplink transmission is associated with only one AI model, the target AI model is that AI model.
  • the target AI model is an AI model selected by the network device from the several AI models associated with the uplink transmission.
  • the target AI model can be randomly selected by the network device or selected by the network device according to a certain rule.
  • the network device obtains the uplink partial CSI (i.e., the CSI corresponding to the M antenna ports) based on the uplink transmission (e.g., SRS transmission of M ports), and uses the uplink partial CSI as input to the target AI model to obtain the complete uplink CSI (i.e., the CSI corresponding to the N antenna ports).
  • the uplink partial CSI i.e., the CSI corresponding to the M antenna ports
  • the uplink transmission e.g., SRS transmission of M ports
  • the network device also monitors the performance of the AI model and sends a second indication message to the first terminal device when the model performance deteriorates.
  • the second indication message instructs the first terminal device to update the AI model, or to adopt a transmission method not based on the AI model. For example, it may employ a transmission method not based on AI/ML reception.
  • the first terminal device receives the second instruction information from the network device.
  • the network device can instruct the first terminal device to transmit SRS from N antenna ports at a lower frequency to obtain complete channel information.
  • the obtained channel information is then compared with the inference results of the AI model to determine the performance of the AI model.
  • the performance monitoring results can be represented by one or more of the following: Generalized Cosine Similarity (GCS), Squared Generalized Cosine Similarity (SGCS), Block Error Rate (BLER), Spectral Efficiency, etc.
  • GCS Generalized Cosine Similarity
  • SGCS Squared Generalized Cosine Similarity
  • BLER Block Error Rate
  • Spectral Efficiency Spectral Efficiency
  • the first terminal device when the second indication information is used to instruct the first terminal device to update the AI model, can report the updated model information to the network device so that the network device can update the AI model. For example, the first terminal device can do so through MAC layer signaling.
  • the UCI can report updated model information, such as model parameters, to the network device, allowing the network device to update the model using the updated model parameters and replace the existing AI model for uplink CSI recovery.
  • the first terminal device when the indication information is used to instruct the first terminal device to adopt a transmission method not based on an AI model, the first terminal device can send SRS for N antenna ports, and the network device can directly obtain the CSI corresponding to the N antenna ports without relying on an AI model.
  • step 880 Other relevant content in step 880 can be found in steps 730, 740, and 750, and will not be repeated here.
  • steps 820, 840, 860, and 880 can be executed individually or in combination.
  • executing step 820 alone implements an AI model reporting method on the terminal device side.
  • executing step 840 alone implements a scheduling/uplink transmission method on the network device side.
  • combining steps 820 and 840 implements a communication method based on an AI model.
  • Another example is the combined execution of steps 820, 840, and 860.
  • Yet another example is the combined execution of steps 820, 840, and 880.
  • it supports executing the network device-side steps individually or in combination to implement network device-side embodiments; it also supports executing the terminal device-side steps individually or in combination to implement terminal device-side embodiments.
  • the method provided in this application supports terminal devices training AI models based on their own antenna configurations and reporting the trained models to network devices for uplink CSI recovery.
  • the network device can train an AI model based on the model dataset reported by the terminal device to facilitate uplink CSI recovery.
  • the AI model used by the network device has a very high degree of compatibility with the first terminal device, perfectly matching the hardware capabilities and uplink transmission requirements of the first terminal device, thus helping to ensure the performance of uplink CSI recovery.
  • the network device can configure only a small amount of SRS signal to measure uplink CSI, significantly reducing uplink pilot overhead.
  • Step 920 The first terminal device reports model information of one or more AI models to the network device.
  • the AI models are used by the network device for uplink CSI processing, which includes uplink beam recovery.
  • step 920 takes the example of the model function including at least uplink beam recovery.
  • one or more AI models are trained by a first terminal device.
  • one or more AI models are trained by a second terminal device.
  • the second terminal device and the first terminal device use the same hardware configuration, the details of which can be found in step 520.
  • the second terminal device and the first terminal device may use the same analog beamforming method.
  • one or more AI models are trained by a network device, and the dataset used to train the one or more AI models is reported by a terminal device (such as a first terminal device and/or a second terminal device).
  • model information includes the model dataset.
  • the input information of the AI model includes the received signal measurement values corresponding to M SRS resources or M beams (the embodiment of this application takes RSRP as an example), and the output information includes one or more SRIs, each SRI being used to indicate one SRS resource among N SRS resources; M is an integer greater than or equal to 1, N is an integer greater than 1, and M is less than N.
  • the M SRS resources are a subset of the N SRS resources. That is, the N SRS resources contain the M SRS resources.
  • the first terminal device can report multiple AI models to the network device, with different AI models corresponding to different combinations of N and M values, thereby supporting the network device to flexibly configure different numbers of SRS resources for uplink beam management and saving uplink reference signal overhead.
  • different AI models can correspond to different values of M and the same value of N.
  • the values of N and M are also used as input information for the AI model. Based on this, the AI model can match the most suitable model parameters according to the values of N and M, thereby achieving optimal performance.
  • model information can be reported via UE capabilities, MAC layer signaling, or UCI. This design helps ensure the flexibility of model information reporting.
  • step 920 Other relevant content in step 920 can be found in steps 520 and/or 710, and will not be repeated here.
  • Step 940 The network device schedules the first terminal device to perform uplink transmission through scheduling information, and the uplink transmission is associated with at least one AI model among one or more AI models.
  • the scheduling information includes first indication information, which is used to indicate the AI model associated with the uplink transmission from one or more AI models.
  • the first indication information may indicate one or more of the following: model ID, function ID, dataset ID, model index (i.e., the sequence number of the AI model among all models reported by the first terminal device), etc.
  • model ID, function ID, dataset ID, and model index may be included in the model information reported by the first terminal device, so that the first terminal device can determine the AI model associated with the uplink transmission from the reported AI models based on the first indication information.
  • the association between uplink transmission and the AI model is reflected in the fact that the transmission beam used for uplink transmission is the same as the transmission beam corresponding to the input information in the training dataset of the AI model. That is, the first terminal device needs to use the transmission beam used when training the AI model for uplink transmission to ensure consistency between inference and training.
  • the first terminal device uses the beam used for the M SRS resources during training to transmit the M SRS resources in the uplink transmission, or uses the M beams used during training to transmit the M SRS resources in the uplink transmission, thereby ensuring the performance of inference/prediction.
  • the uplink transmission is an SRS transmission over M SRS resources, which are associated with one of one or more AI models.
  • scheduling information is used to schedule/configure the transmission of the M SRS resources.
  • the network device only needs to configure a first resource set containing the M SRS resources to determine the optimal beam-corresponding SRS resource in a second resource set containing N SRS resources, thereby saving reference signal overhead.
  • the network device configures M SRS resources, it needs to indicate the AI model associated with the M SRS resources through first indication information.
  • the input information of this AI model includes M RSRP values.
  • the transmission beam used by the M SRS resources needs to be the same as the beam corresponding to the M RSRP values used during the training of the associated AI model.
  • step 940 Other relevant information for step 940 can be found in steps 540 and/or 720, and will not be repeated here.
  • Step 960 The first terminal device performs uplink transmission according to the scheduling information, which is associated with at least one AI model among one or more AI models.
  • step 960 please refer to steps 540 and 940, which will not be repeated here.
  • Step 980 The network device performs uplink CSI processing based on the target AI model in at least one AI model and the uplink transmission.
  • the uplink CSI processing includes uplink beam recovery.
  • the network device obtains the RSRP of the uplink portion of the beams (i.e., M SRS resources) based on uplink transmission measurements, and uses the RSRP of the uplink portion of the beams as input to the target AI model to obtain the SRI corresponding to the optimal beam in the complete beam (i.e., the optimal resource/beam among N SRS resources/beams).
  • the AI model can output multiple SRIs, corresponding to multiple optimal beams. Furthermore, the AI model can also output the RSRP corresponding to the SRIs.
  • the network device also monitors the performance of the AI model and sends a second indication message to the first terminal device when the model performance deteriorates.
  • the second indication message instructs the first terminal device to update the AI model, or to adopt a transmission method not based on the AI model. For example, it may employ a transmission method not based on AI/ML reception.
  • the first terminal device receives the second instruction information from the network device.
  • the network device can instruct the first terminal device to transmit N SRS resources at a lower frequency to obtain the optimal beam and corresponding RSRP in the complete beam set.
  • the obtained information is then compared with the results of AI model inference to determine the performance of the AI model.
  • Performance monitoring results can be represented by one or more of the following: reliability, estimation accuracy, RSRP difference, etc.
  • the network device can send a second indication message to the terminal device.
  • the first terminal device when the second indication information is used to instruct the first terminal device to update the AI model, can report the updated model information to the network device, thereby updating the AI model.
  • the first terminal device can report the updated model information, such as model parameters, to the network device via MAC layer signaling or UCI, so that the network device can use the updated model parameters to update the model and replace the existing AI model for uplink beam recovery.
  • the first terminal device when the indication information is used to instruct the first terminal device to adopt a transmission method not based on the AI model, the first terminal device can send N SRS resources, and the network device can directly obtain the RSRPs corresponding to the N SRS resources and determine the optimal waveform among them.
  • the bundle does not require the aid of AI models.
  • step 980 Other relevant content in step 980 can be found in steps 730, 740, and 750, and will not be repeated here.
  • steps 920, 940, 960, and 980 can be executed individually or in combination.
  • executing step 920 alone implements an AI model reporting method on the terminal device side.
  • executing step 940 alone implements a scheduling/uplink transmission method on the network device side.
  • combining steps 920 and 940 implements a communication method based on an AI model.
  • Another example is the combination of steps 920, 940, and 960.
  • Yet another example is the combination of steps 920, 940, and 980.
  • it supports executing the network device-side steps individually or in combination to implement network device-side embodiments; it also supports executing the terminal device-side steps individually or in combination to implement terminal device-side embodiments.
  • the method provided in this application supports terminal devices training AI models based on their own antenna configurations and reporting the trained models to network devices for uplink beam recovery.
  • the network device can train an AI model based on the model dataset reported by the terminal device to facilitate uplink beam recovery.
  • the AI model used by the network device is highly compatible with the first terminal device, perfectly matching its hardware capabilities and uplink transmission requirements, thus helping to ensure the performance of uplink beam recovery.
  • the network device can configure only a small amount of SRS signals for uplink beam management, significantly reducing uplink pilot overhead.
  • Figure 10 shows a flowchart of an AI model-based communication method provided in an exemplary embodiment of this application. The method is executed by a first terminal device and a network device, and includes at least some of the following steps:
  • Step 1020 The first terminal device reports model information of one or more AI models to the network device.
  • the AI models are used by the network device for uplink signal processing, which includes uplink nonlinear compensation.
  • step 1020 takes the example of the model functionality including at least uplink nonlinear compensation.
  • one or more AI models are trained by a first terminal device.
  • one or more AI models are trained by a second terminal device.
  • the second terminal device and the first terminal device use the same hardware configuration, the details of which can be found in step 520; for example, the second terminal device and the first terminal device use power amplifiers with the same performance.
  • one or more AI models are trained by a network device, and the dataset used to train the one or more AI models is reported by a terminal device (such as a first terminal device and/or a second terminal device).
  • model information includes the model dataset.
  • the input information of the AI model includes the modulated signal detected by the receiver (such as a network device), and the output information includes the modulated signal after nonlinear compensation.
  • model information can be reported via UE capabilities, MAC layer signaling, or UCI. This design helps ensure the flexibility of model information reporting.
  • step 1020 Other relevant content in step 1020 can be found in steps 520 and/or 710, and will not be repeated here.
  • Step 1040 The network device schedules the first terminal device to perform uplink transmission through scheduling information, and the uplink transmission is associated with at least one AI model among one or more AI models.
  • the scheduling information includes first indication information, which is used to indicate the AI model associated with the uplink transmission from one or more AI models.
  • the first indication information may indicate one or more of the following: model ID, function ID, dataset ID, model index (i.e., the sequence number of the AI model among all models reported by the first terminal device), etc.
  • model ID, function ID, dataset ID, and model index may be included in the model information reported by the first terminal device, so that the first terminal device can determine the AI model associated with the uplink transmission from the reported AI models based on the first indication information.
  • the association between uplink transmission and the AI model is reflected in the fact that the power amplifier used for uplink transmission is the same as the power amplifier used for the input information in the training dataset of the AI model.
  • the performance of the power amplifier used for uplink transmission is the same as the performance of the power amplifier used for the input information in the training dataset of the AI model.
  • the uplink transmission is an uplink data transmission associated with the AI model.
  • scheduling information is used to schedule/configure the uplink data transmission.
  • the power amplifier used for the uplink data transmission is the same as the power amplifier used to generate the input signals in the training dataset of the AI model.
  • step 1040 Other relevant information for step 1040 can be found in steps 540 and/or 720, and will not be repeated here.
  • Step 1060 The first terminal device performs uplink transmission according to the scheduling information, and the uplink transmission is associated with at least one AI model among one or more AI models.
  • the first terminal device determines the AI model associated with the uplink transmission from one or more AI models based on the first indication information in the scheduling information.
  • the scheduling information is used to schedule uplink data transmission, and the first terminal device performs uplink data transmission according to the scheduling information.
  • Uplink data transmission is associated with at least one of one or more AI models.
  • step 1060 For other details in step 1060, please refer to steps 540 and 940, which will not be repeated here.
  • Step 1080 The network device performs uplink signal processing based on the target AI model in at least one AI model and the uplink transmission.
  • the uplink signal processing includes uplink nonlinear compensation.
  • the network device uses the modulated signal obtained from uplink transmission detection (i.e., the signal to be demodulated after MIMO reception is completed) as input to the AI model to obtain a modulated signal after nonlinear compensation (which can then be used as input to the demodulation module).
  • the network device also monitors the performance of the AI model and sends a second indication message to the first terminal device when the model performance deteriorates.
  • the second indication message instructs the first terminal device to update the AI model, or to adopt a transmission method not based on the AI model. For example, it may employ a transmission method not based on AI/ML reception.
  • the first terminal device receives the second instruction information from the network device.
  • the network device continuously monitors demodulation performance and sends a second indication message to the first terminal device when the demodulation performance deteriorates.
  • the first terminal device when the second indication information is used to instruct the first terminal device to adopt a transmission method not based on an AI model, the first terminal device needs to perform pre-distortion processing at the transmitting end to offset the influence of power amplifier nonlinearity, thereby ensuring the detection performance of the receiving end (such as a network device).
  • step 1080 Other relevant content in step 1080 can be found in steps 730, 740, and 750, and will not be repeated here.
  • steps 1020, 1040, 1060, and 1080 can be executed individually or in combination.
  • executing step 1020 alone implements an AI model reporting method on the terminal device side.
  • executing step 1040 alone implements a scheduling/uplink transmission method on the network device side.
  • combining steps 1020 and 1040 implements a communication method based on an AI model.
  • Another example is the combination of steps 1020, 1040, and 1060.
  • Yet another example is the combination of steps 1020, 1040, and 1080.
  • it supports executing network device-side steps individually or in combination to implement network device-side embodiments; it also supports executing terminal device-side steps individually or in combination to implement terminal device-side embodiments.
  • the method provided in this application supports terminal devices in training AI models based on their own antenna configurations and reporting the trained models to network devices for corresponding uplink signal processing.
  • the network device can train an AI model based on the model dataset reported by the terminal device, enabling the network device to perform corresponding uplink signal processing.
  • the AI model used by the network device has a very high degree of compatibility with the first terminal device, perfectly matching the hardware capabilities and uplink transmission requirements of the first terminal device, thus helping to ensure the performance of uplink signal processing.
  • the first terminal device does not need to perform pre-distortion processing on the power amplifier, significantly reducing the processing complexity and overhead on the terminal side.
  • Figure 11 shows a structural block diagram of an AI model-based communication device provided in an exemplary embodiment of this application.
  • This device can be implemented as the first terminal device described above, or as part of the first terminal device described above.
  • the device includes a transmitting module 1110.
  • the device also includes a processing module 1130 and/or a receiving module 1150.
  • the sending module 1110 is used to report model information of one or more AI models, which are used by the network device for uplink CSI processing or uplink signal processing; the sending module 1110 is also used to perform uplink transmission according to the scheduling information of the network device, which is associated with at least one AI model among the one or more AI models.
  • the sending module 1110 is further configured to report updated model information of the one or more AI models.
  • the sending module 1110 is further configured to perform uplink transmission using a transmission method not based on an AI model.
  • the sending module 1110 is configured to perform one or more of the following steps: step 420, step 440, step 520, step 540, step 580, step 820, step 860, step 920, step 960, step 1020, and step 1060.
  • the processing module 1130 is used to train one or more AI models.
  • the processing module 1130 is used to train a portion of the multiple AI models.
  • the processing module 1130 is configured to determine, based on the first indication information in the scheduling information, one or more AI models associated with the uplink transmission.
  • the processing module 1130 is used to update one or more AI models.
  • the receiving module 1150 is configured to receive one or more AI models trained by the second terminal device, or to receive model information of one or more AI models trained by the second terminal device.
  • the receiving module 1150 is configured to receive a portion of the AI models trained by the second terminal device from among multiple AI models, or to receive model information of a portion of the AI models trained by the second terminal device from among multiple AI models.
  • the receiving module 1150 is used to receive the scheduling information.
  • the receiving module 1150 is configured to receive second indication information, the second indication information being used to instruct the device to update the one or more AI models, or to instruct the device to adopt a transmission method not based on AI models.
  • the apparatus provided in this application supports reporting model information from AI models trained on the terminal device side.
  • the AI model is highly compatible with this apparatus, perfectly matching its hardware capabilities and uplink transmission requirements. Using the AI model on the network device will not negatively impact uplink transmission performance. Furthermore, it helps reduce uplink resource overhead and lowers the processing complexity of this apparatus. Additionally, it supports updating the AI model, allowing it to more flexibly adapt to changing communication environments and requirements.
  • Figure 12 shows a structural block diagram of an AI model-based communication device provided in an exemplary embodiment of this application.
  • This device can be implemented as a network device as described above, or as part of a network device as described above.
  • the device includes a receiving module 1210 and a transmitting module 1250.
  • the device also includes a processing module 1230.
  • the receiving module 1210 is used to receive model information of one or more AI models, which are used by the device for uplink CSI processing or uplink signal processing.
  • the sending module 1250 is used to send scheduling information, which is used to schedule the first terminal device to perform uplink transmission, and the uplink transmission is associated with at least one AI model among the one or more AI models.
  • the processing module 1230 is configured to perform uplink CSI processing or uplink signal processing based on the target AI model in the at least one AI model and the uplink transmission.
  • the processing module 1230 is configured to perform one or more processes:
  • the uplink partial CSI is obtained, and the uplink partial CSI is used as the input information of the target AI model to obtain the complete uplink CSI;
  • the received signal measurement value corresponding to the uplink partial beam is obtained, and the received signal measurement value corresponding to the uplink partial beam is used as the input information of the target AI model to obtain the SRI corresponding to the optimal beam in the uplink complete beam.
  • the modulation signal detected by the device is used as the input information of the target AI model to obtain a modulation signal after nonlinear compensation.
  • the processing module 1230 is used to monitor the performance of the one or more AI models.
  • the processing module 1230 is configured to instruct the first terminal device to transmit SRS of N antenna ports at a lower frequency to obtain complete channel information, and then compare the obtained channel information with the results of AI model inference to determine the performance of the AI model.
  • the processing module 1230 is configured to instruct the first terminal device to send N SRS resources at a lower frequency, thereby obtaining the optimal beam and corresponding RSRP in the complete beam set, and then comparing the obtained information with the results of AI model inference to determine the performance of the AI model.
  • the processing module 1230 is used to continuously monitor demodulation performance.
  • the sending module 1250 is further configured to send second indication information, the second indication information being used to instruct the first terminal device to update the one or more AI models, or to instruct the first terminal device to adopt a transmission method not based on AI models.
  • the sending module 1250 is further configured to send the second indication information based on the performance monitoring results of the one or more AI models. For example, the second indication information is sent when the performance of one or more AI models deteriorates.
  • the receiving module 1210 is further configured to receive updated model information of the one or more AI models.
  • the receiving module 1210 is configured to perform one or more of the following steps: step 620, step 710, step 750.
  • the sending module 1250 is configured to perform one or more of the following steps: step 640, step 720, step 740, step 840, step 940, and step 1040.
  • the processing module 1230 is configured to perform one or more of the following steps: step 730, step 880, step 980, and step 1080.
  • receiving module 1210, processing module 1230, and sending module 1250 are described in reference to one or more steps performed by the network device in the embodiments shown in Figures 6, 7, 8, 9, and 10.
  • the relevant content described in the previous embodiments also applies to the device shown in Figure 12, and will not be repeated here.
  • the apparatus provided in this application supports the use of model information from AI models reported by a first terminal device.
  • the AI models are trained on the terminal device side or by this apparatus based on the model dataset reported by the terminal device.
  • the AI models are highly compatible with the first terminal device, perfectly matching its hardware capabilities and uplink transmission requirements. Using the AI models will not negatively impact uplink transmission performance. Furthermore, it helps reduce uplink resource overhead and lowers the processing complexity of this apparatus. Additionally, it supports monitoring the AI models to indicate updates or discontinuation of their use, allowing for more flexible adaptation to changing communication environments and needs.
  • the apparatus provided in the above embodiments is only illustrated by the division of the above functional modules.
  • the above functions can be assigned to different functional modules as needed, that is, the internal structure of the communication device can be divided into different functional modules to complete all or part of the functions described above.
  • the apparatus and method embodiments provided in the above embodiments belong to the same concept.
  • Figure 13 shows a schematic diagram of the structure of an AI model-based communication device provided in an exemplary embodiment of this application, including at least one of the following: a receiver 1301, a transmitter 1302, a processor 1303, a memory 1304, and a bus (not shown in the figure).
  • the communication device 1300 is used to perform some or all of the steps performed by the first terminal device described above.
  • the communication device 1300 is used to perform some or all of the steps performed by the network device described above.
  • receiver 1301 is used to implement the receiving function
  • transmitter 1302 is used to implement the transmitting function.
  • receiver 1301 and transmitter 1302 can be implemented as a communication component, which can be a communication chip, and can be referred to as a transceiver.
  • receiver 1301 and transmitter 1302 can be implemented as a wireless communication component and/or a wired communication component.
  • the wireless communication component includes a wireless communication chip and/or a radio frequency antenna.
  • the wired communication component includes a wired communication chip and/or a wired interface.
  • receiver 1301 can be used to implement the functions and steps of receiving module 1150, and transmitter 1302 can be used to implement the functions and steps of sending module 1110.
  • receiver 1301 can be used to implement the functions and steps of receiving module 1210
  • transmitter 1302 can be used to implement the functions and steps of sending module 1250.
  • the processor 1303 includes one or more processing cores.
  • the processor 1303 executes various functional applications and information processing by running software programs and modules.
  • the processor 1303 can be used to implement the functions and steps of the processing module 1130 or processing module 1230 described above.
  • the memory 1304 can be used to store a computer program executed by the processor 1303, which executes the computer program to implement the various steps in the above method embodiments.
  • the memory 1304 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, including but not limited to: magnetic disks or optical disks, electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), static random access memory (SRAM), read-only memory (ROM), magnetic storage, flash memory, and programmable read-only memory (PROM).
  • EEPROM electrically erasable programmable read-only memory
  • EPROM erasable programmable read-only memory
  • SRAM static random access memory
  • ROM read-only memory
  • magnetic storage magnetic storage
  • flash memory programmable read-only memory
  • the memory 1304 may be connected to the processor 1303, the receiver 1301, and the transmitter 1302.
  • the receiver 1301 independently receives signals/data, or the processor 1303 controls the receiver 1301 to receive signals/data, or the processor 1303 requests the receiver 1301 to receive signals/data, or the processor 1303 cooperates with the receiver 1301 to receive signals/data.
  • the transmitter 1302 independently transmits signals/data, or the processor 1303 controls the transmitter 1302 to transmit signals/data, or the processor 1303 requests the transmitter 1302 to transmit signals/data, or the processor 1303 cooperates with the transmitter 1302 to transmit signals/data.
  • a chip is also provided, the chip including programmable logic circuits and/or program instructions, which, when the chip is run on a communication device, is used to implement the AI model-based communication method provided in the above-described method embodiments.
  • the chip includes a transmitting module 1110.
  • the chip further includes a processing module 1130 and/or a receiving module 1150. Related details can be found above and will not be repeated here.
  • each module can be implemented as a circuit structure.
  • the chip includes a receiving module 1210 and a transmitting module 1250.
  • the chip further includes a processing module 1230.
  • each module can be implemented as a circuit structure.
  • a computer-readable storage medium which stores at least one program, which is loaded and executed by a processor to implement the AI model-based communication method provided in the above-described method embodiments.
  • a computer program product which includes computer instructions stored in a computer-readable storage medium.
  • a processor retrieves the computer instructions from the computer-readable storage medium and executes the computer instructions to implement the AI model-based communication method provided in the above-described method embodiments.
  • a computer program including computer instructions, the computer instructions being stored in a computer-readable storage medium, a processor retrieving the computer instructions from the computer-readable storage medium, and the processor executing the computer instructions to implement the AI model-based communication method provided in the above-described method embodiments.

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

本申请公开了一种基于AI模型的通信方法、装置、设备、介质和程序产品,属于无线通信领域。该方法由第一终端设备执行,该方法包括:上报一个或多个AI模型的模型信息,AI模型用于网络设备进行上行CSI处理或上行信号处理;根据网络设备的调度信息进行上行传输,上行传输与一个或多个AI模型中的至少一个AI模型关联。有助于避免AI模型对第一终端设备的上行传输产生负面影响。

Description

基于AI模型的通信方法、装置、设备、介质和程序产品 技术领域
本申请涉及无线通信领域,特别涉及一种基于AI模型的通信方法、装置、设备、介质和程序产品。
背景技术
人工智能(Artificial Intelligence,AI)模型可以被应用于无线通信系统中,以实现更为智能的、更为质量更高的上下行传输。
但是,不同的通信设备的硬件配置存在差异,用于无线通信的AI模型都是基于某些硬件配置训练得到的,无法与各种不同硬件配置的通信设备都匹配。尤其是当网络设备使用AI模型执行一些针对上行信号的处理时,AI模型与终端设备进行上行传输所用的硬件配置的不匹配很可能导致上行传输性能变差。
发明内容
本申请提供了一种基于AI模型的通信方法、装置、设备、介质和程序产品,该技术方案至少包括:
根据本申请实施例的一个方面,提供了一种基于AI模型的通信方法,该方法由第一终端设备执行,该方法包括:
上报一个或多个AI模型的模型信息,所述AI模型用于网络设备进行上行CSI处理或上行信号处理;
根据所述网络设备的调度信息进行上行传输,所述上行传输与所述一个或多个AI模型中的至少一个AI模型关联。
根据本申请实施例的另一个方面,提供了一种基于AI模型的通信方法,该方法由网络设备执行,该方法包括:
接收一个或多个AI模型的模型信息,所述AI模型用于所述网络设备进行上行CSI处理或上行信号处理;
发送调度信息,所述调度信息用于调度第一终端设备进行上行传输,所述上行传输与所述一个或多个AI模型中的至少一个AI模型关联。
根据本申请实施例的一个方面,提供了一种基于AI模型的通信装置,该装置包括:
发送模块,用于上报一个或多个AI模型的模型信息,所述AI模型用于网络设备进行上行CSI处理或上行信号处理;
所述发送模块,还用于根据所述网络设备的调度信息进行上行传输,所述上行传输与所述一个或多个AI模型中的至少一个AI模型关联。
根据本申请实施例的另一个方面,提供了一种基于AI模型的通信装置,该装置包括:
接收模块,用于接收一个或多个AI模型的模型信息,所述AI模型用于所述装置进行上行CSI处理或上行信号处理;
发送模块,用于发送调度信息,所述调度信息用于调度第一终端设备进行上行传输,所述上行传输与所述一个或多个AI模型中的至少一个AI模型关联。
根据本申请实施例的一个方面,提供了一种基于AI模型的通信设备,该通信设备包括:处理器;与所述处理器相连的收发器;用于存储所述处理器的可执行指令的存储器;其中,所述处理器被配置为加载并执行所述可执行指令以实现如上述各个方面所述的基于AI模型的通信方法。
根据本申请实施例的一个方面,提供了一种计算机可读存储介质,该计算机可读存储介质中存储有至少一段程序,该至少一段程序由处理器加载并执行以实现如上述各个方面所述的基于AI模型的通信方法。
根据本申请实施例的一个方面,提供了一种计算机程序产品或计算机程序,所述计算机程序产品或所述计算机程序包括计算机指令,所述计算机指令存储在计算机可读存储介质中,处理器从所述计算机可读存储介质中获取所述计算机指令,所述处理器执行所述计算机指令以实现如上述各个方面所述的基于AI模型的通信方法。
根据本申请实施例的一个方面,提供了一种芯片,所述芯片包括可编程逻辑电路和/或至少一段程序,所述芯片用于基于所述可编程逻辑电路和/或所述至少一段程序,以实现如上述各个方面所述的基于AI模型的通信方法。
本申请实施例提供的技术方案可以包括以下有益效果:
支持第一终端设备上报AI模型的模型信息,使得网络设备进行上行CSI处理和/或上行信号处理时采用的AI模型与第一终端设备更加匹配。无论第一终端设备具备什么样的硬件结构,都能保障网络设备使用AI模型获取的上行CSI处理的结果或上行信号处理的结果符合第一终端设备的能力,且避免AI模型对第一终端设备的上行传输产生负面影响。
附图说明
为了更清楚地说明本申请实施例中的技术方案,下面将对实施例描述中所需要使用的附图作简单地介 绍,显而易见地,下面描述中的附图仅仅是本申请的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动的前提下,还可以根据这些附图获得其它的附图。
图1示出了本申请一个示例性实施例提供的神经元的结构示意图;
图2示出了本申请一个示例性实施例提供的神经网络的结构示意图;
图3示出了本申请一个示例性实施例提供的无线通信系统的示意图;
图4示出了本申请一个示例性实施例提供的基于AI模型的通信方法的流程示意图;
图5示出了本申请一个示例性实施例提供的基于AI模型的通信方法的流程示意图;
图6示出了本申请一个示例性实施例提供的基于AI模型的通信方法的流程示意图;
图7示出了本申请一个示例性实施例提供的基于AI模型的通信方法的流程示意图;
图8示出了本申请一个示例性实施例提供的基于AI模型的通信方法的流程示意图;
图9示出了本申请一个示例性实施例提供的基于AI模型的通信方法的流程示意图;
图10示出了本申请一个示例性实施例提供的基于AI模型的通信方法的流程示意图;
图11示出了本申请一个示例性实施例提供的基于AI模型的通信装置的结构框图;
图12示出了本申请一个示例性实施例提供的基于AI模型的通信装置的结构框图;
图13示出了本申请一个示例性实施例提供的基于AI模型的通信设备的结构示意图。
具体实施方式
为使本申请的目的、技术方案和优点更加清楚,下面将结合附图对本申请实施方式作进一步地详细描述。这里将详细地对示例性实施例进行说明,其示例表示在附图中。下面的描述涉及附图时,除非另有表示,不同附图中的相同数字表示相同或相似的要素。以下示例性实施例中所描述的实施方式并不代表与本申请相一致的所有实施方式。相反,它们仅是与如所附权利要求书中所详述的、本申请的一些方面相一致的装置和方法的例子。
在本申请使用的术语是仅仅出于描述特定实施例的目的,而非旨在限制本申请。在本申请和所附权利要求书中所使用的单数形式的“一种”、“所述”和“该”也旨在包括多数形式,除非上下文清楚地表示其他含义。还应当理解,本文中使用的术语“和/或”是指并包含一个或多个相关联的列出项目的任何或所有可能组合。
应当理解,尽管在本申请可能采用术语第一、第二、第三等来描述各种信息,但这些信息不应限于这些术语。这些术语仅用来将同一类型的信息彼此区分开。例如,在不脱离本申请范围的情况下,第一信息也可以被称为第二信息,类似地,第二信息也可以被称为第一信息。取决于语境,如在此所使用的词语“如果”可以被解释成为“在……时”或“当……时”或“响应于确定”。在本说明书中,在表述Boolean Value(布尔型值)所表达的含义时,会表达为“0”表示“第一含义”,“1”表示“第二含义”,不失一般性地,本领域技术人员能够理解,其代表地含义是可以对调的,即“1”表示“第一含义”,“0”表示“第二含义”。
人工智能(Artificial Intelligence,AI):
AI是利用数字计算机或者数字计算机控制的机器模拟、延伸和扩展人的智能,感知环境、获取知识并使用知识获得最佳结果的理论、方法、技术及应用系统。换句话说,AI是计算机科学的一个综合技术,它企图了解智能的实质,并生产出一种新的能以人类智能相似的方式做出反应的智能机器。AI也就是研究各种智能机器的设计原理与实现方法,使机器具有感知、推理与决策的功能。
AI技术是一门综合学科,涉及领域广泛,既有硬件层面的技术也有软件层面的技术。AI基础技术一般包括如传感器、专用AI芯片、云计算、分布式存储、大数据处理技术、操作/交互系统、机电一体化等技术。AI软件技术主要包括计算机视觉技术、语音处理技术、自然语言处理技术以及机器学习(Machine Learning,ML)/深度学习等几大方向。
AI模型是AI技术中的重要组成部分,可以用于处理、分析及预测等任务。常见的AI模型有多种类型,如神经网络(Neural Network)、决策树、支持向量机、随机森林、均值算法等。
神经网络(Neural Network):
神经网络是一种由多个神经元(即节点)相互连接构成的运算模型。神经元结构如图1所示,神经元之间的连接(w1、w2、…、wn、b)代表从输入信号(a1、a2、…、an、1)到输出信号的加权值,简称为权重(Weight),权重可以表示不同输入信号对神经元输出信号的影响大小。每个神经元对不同的输入信号进行加权求和(Sum),并通过特定的激活函数(Activation Function,图中表示为f)输出t。
如图2所示,一个简单的神经网络的基本结构包括:输入层(10),隐藏层(20)和输出层(30)。输入层(10)负责接收数据,隐藏层(20)对数据进行处理,最后的结果在输出层(30)产生。通过多个神经元之间不同的连接方式、权重和激活函数,可以产生不同的输出,进而拟合从输入到输出的映射关系。每一个上一级节点都与其全部的下一级节点相连,此全连接模型也可以称为深度神经网络(Deep Neural Network,DNN)。
通过数据集的构建、训练、验证和测试等过程,可以训练并得到一个AI模型,如神经网络模型。
当神经网络模型用于终端设备和网络设备之间的无线通信时,训练可以分为离线训练和在线训练。若采用离线训练,网络设备可以通过数据集离线训练的方式得到一个静态的训练结果。在网络设备或终端设备使用神经网络模型的过程中,随着终端设备的进一步测量和/或上报,网络设备可以继续收集更多的数据,进行实时的在线训练来优化神经网络模型的参数,达到更好的推断和预测结果。在得到神经网络模型后,通过将当前得到的信息输入到模型中,可以推理得到相应的模型输出。
神经网络模型用于无线通信时,可以分为单端模型和双端模型。其中,单端模型只在终端设备或网络设备单侧部署即可使用,模型的训练也可以只在单侧进行;双端模型需要在终端设备和网络设备两侧部署,两侧的模型需要一起进行训练,也就是说两侧部署的模型是对应的或成对的,不能单独使用或单独更新。
为了实现不同的无线通信功能,可以引入不同的AI模型,定义相应的输入和输出。例如,将AI模型用于信道状态信息(Channel State Information,CSI)反馈时,可以将基于参考信号测量得到的信道信息(如特征向量,波束信息,时延信息等)作为模型的输入,从而推理出相应的CSI量化比特。在网络侧会有一个对应的神经网络模型,将CSI量化比特作为输入,就可以推理得到相应的信道信息。例如,将AI模型用于波束管理时,终端设备将测量得到的第二波束集合中多个波束(CSI-RS资源)对应的参考信号接收功率(Reference Signal Received Power,RSRP)作为输入,基于AI模型推理得到第一波束集合中最好的波束(CSI-RS资源索引)以及相应的RSRP,并将推理的结果上报给网络设备。此外,AI模型还可以用于定位,信道编码,数据信道解码,调制解调,信道估计等其他过程。
在一些场景中,终端设备可以配置大量的天线单元,用于提高上行传输性能。基于大规模天线,终端设备可以进行大量天线端口的多输入多输出(Multiple-Input Multiple-Output,MIMO)传输、上行多波束模拟赋形等,从而提高上行传输性能。但是,大规模MIMO需要大量的上行参考信号用于上行信道探测,或者用于上行波束管理,导致上行资源开销大大增加,影响了上行吞吐量。如果网络设备可以基于AI模型进行辅助接收,则可以大大减少需要的上行参考信号,只用少量的上行参考信号就可以达到预期的性能,从而降低上行资源开销。但是,不同终端设备的硬件配置(如天线架构,模拟波束赋形能力,功率放大器(Power Amplifier,PA)性能等)不同,如果采用的AI模型与终端设备的硬件不匹配,则会对上行传输性能产生负面影响,甚至导致采用AI模型的上行传输性能反而不如未采用AI模型的情况。
为此,本申请提出一种基于AI模型的通信方法,使得上行传输所使用的AI模型与终端设备具备较好的匹配性,避免AI模型对上行传输性能造成负面影响。
图3示出了本申请一个示例性实施例提供的无线通信系统100的示意图。该无线通信系统100中包括终端设备与终端设备,或终端设备与网络设备,或站点(Station,STA)与站点,本申请对此不作限定。图1以无线通信系统100包括网络设备110和终端设备120为例。
本申请中的网络设备110,支持提供无线通信功能,包括但不限于:节点B(Node B,NB)、演进型节点B(Evolved Node B,eNB)、下一代节点B(Next Generation Node B,gNB)、无线网络控制器(Radio Network Controller,RNC)、基站(Base Station,BS)、基站控制器(Base Station Controller,BSC)、基站收发台(Base Transceiver Station,BTS)、家庭基站(Home Evolved Node B或Home Node B,HNB)、基带单元(Baseband Unit,BBU)、分布式单元(Distributed Unit,DU)、无线中继节点、无线回传节点、传输点(Transmission Point,TP)、发射接收点(Transmission and Reception Point,TRP)、天线面板、路由器等。
本申请中的终端设备120,也可以称为用户设备(User Equipment,UE),包括但不限于:手机、平板电脑、电子书阅读器、膝上便携计算机、台式计算机、电视机、虚拟现实(Virtual Reality,VR)设备、增强现实(Augmented Reality,AR)设备、混合现实(Mediated Reality,MR)设备、扩展现实(Extended Reality,XR)设备、远程终端、机顶盒、车载通信设备、手持设备、可穿戴设备、工业控制(Industrial Control)中的无线设备、无人驾驶(Self Driving)中的无线设备、远程医疗(Remote Medical)中的无线设备、智能电网(Smart Grid)中的无线设备、运输安全(Transportation Safety)中的无线设备、智慧城市(Smart City)中的无线设备、智慧家庭(Smart Home)中的无线设备(比如智能摄像头、智能遥控器、智能水表电表等)、无线通信芯片、专用集成电路(Application Specific Integrated Circuit,ASIC)、片上系统(System on Chip,SoC)、物联网(Internet of Things,IoT)节点、车联网(Internet of Vehicles,IoV)节点、传感器等,还可以是具备无线通信功能的计算设备或连接到无线调制解调器的其它处理设备等。
在一些实施例中,网络设备110和终端设备120均支持第三代合作伙伴计划(3rd Generation Partnership Project,3GPP)协议,但不限于3GPP协议。
在一些实施例中,无线通信系统100可支持的频段包括但不限于:厘米波频段(比如450MHz-6GHz范围内的频段,又叫Sub-6GHz频段)、毫米波(mmWave)频段(比如45GHz、60GHz等属于30~300GHz范围内的频段)、低频频段。其中,低频频段包括Sub-7GHz频段(比如2.4GHz、5GHz、6GHz等属于1~7.25 GHz范围内的频段)。
本申请主要涉及两种通信场景,其一为上行传输场景,指终端设备向网络设备发送信号/数据的场景;其二为下行传输场景,指网络设备向终端设备发送信号/数据的场景。
本申请的一些实施例中描述的技术方案可以适用于各种通信系统,例如:长期演进(Long Term Evolution,LTE)系统、先进的长期演进(Advanced long term evolution,LTE-A)系统、新空口(New Radio,NR)系统、NR系统的演进系统、非授权频谱上的LTE(LTE-based access to unlicensed spectrum,LTE-U)系统、非授权频谱上的NR(NR-based access to unlicensed spectrum,NR-U)系统、第五代通信(5th-Generation,5G)系统、蜂窝物联网系统、蜂窝无源物联网系统、NR系统后续的演进系统、超5G(Beyond 5th-Generation,B5G)系统、6G及后续的演进系统、无线局域网(Wireless Local Area Networks,WLAN)系统、无线保真(Wireless Fidelity,Wi-Fi)系统、全球移动通讯(Global System of Mobile communication,GSM)系统、码分多址(Code Division Multiple Access,CDMA)系统、宽带码分多址(Wideband Code Division Multiple Access,WCDMA)系统、通用分组无线业务(General Packet Radio Service,GPRS)、地面通信网络(Terrestrial Networks,TN)系统、非地面通信网络(Non-Terrestrial Networks,NTN)系统、通用移动通信系统(Universal Mobile Telecommunication System,UMTS)、全球互联微波接入(Worldwide Interoperability for Microwave Access,WiMAX)通信系统等。
图4示出了本申请一个示例性实施例提供的基于AI模型的通信方法的流程示意图,该方法由第一终端设备执行,该方法包括如下至少部分步骤:
步骤420:上报一个或多个AI模型的模型信息,AI模型用于网络设备进行上行CSI处理或上行信号处理。
本申请实施例中的AI模型,包括如下一种或多种类型:神经网络、决策树、支持向量机、随机森林、均值算法等。可选地,AI模型还可以称为AI规则或AI函数或其他名称。
上行CSI处理,指针对上行信号的CSI的处理。若AI模型用于网络设备进行上行CSI处理,则网络设备可以采用AI模型获取/预测/推断上行CSI的处理结果。
上行信号处理,指针对上行信号的接收端处理。若AI模型用于网络设备进行上行信号处理,则网络设备可以采用AI模型获取/预测/推断上行信号的处理结果。
步骤440:根据网络设备的调度信息进行上行传输,该上行传输与一个或多个AI模型中的至少一个AI模型关联。
本申请中的“至少一个”,包含“等于一个”和“大于一个”两种情况。也就是说,至少一个AI模型,可以认为是一个AI模型或多个AI模型。下文使用“至少一个”的情况均可参照此处的解释来理解。
因此,根据网络设备的调度信息进行的上行传输与一个或多个AI模型中的至少一个AI模型关联,可能包含以下三种情况:
情况1.第一终端设备上报一个AI模型,第一终端设备根据调度信息进行的上行传输与这一个AI模型关联。
情况2.第一终端设备上报多个AI模型,第一终端设备根据调度信息进行的上行传输与这多个AI模型关联。
情况3.第一终端设备上报多个AI模型,第一终端设备根据调度信息进行的上行传输与这多个AI模型中的部分AI模型关联。示例性的,第一终端设备上报z个AI模型,z为大于1的整数,上行传输与z个AI模型中的z’个AI模型关联,1≤z’<z。
本申请中,第一终端设备进行上行传输,指第一终端设备向网络设备发送信号和/或数据。
综上所述,本申请实施例中提供的方法,支持第一终端设备上报AI模型的模型信息,使得网络设备进行上行CSI处理或上行信号处理时采用的AI模型与第一终端设备更加匹配。无论第一终端设备具备什么样的硬件结构,都能保障网络设备使用AI模型获取的上行CSI处理的结果或上行信号处理的结果符合第一终端设备的能力,且避免AI模型对第一终端设备的上行传输产生负面影响。
在一些实施例中,在图4所示实施例的基础上,步骤420可以进一步实现为步骤520,步骤440可以进一步实现为步骤540,如图5所示。可选的,除步骤520、步骤540外,第一终端设备还可以执行如下可选步骤中的一个或多个:步骤560、步骤580。
图5示出了本申请一个示例性实施例提供的基于AI模型的通信方法的流程示意图,该方法由第一终端设备执行,该方法包括如下至少部分步骤:
步骤520:上报一个或多个AI模型的模型信息,AI模型用于网络设备进行上行CSI处理或上行信号处理。
在一些实施例中,模型信息包括如下一项或多项:模型数据集,模型结构,模型参数。
其中,模型结构可以是若干种候选的模型结构中的任意一种,候选的模型结构比如包括:转换 (Transformer)结构、残差神经网络(Residual Neural Network,ResNet)、循环神经网络(Recurrent Neural Network,RNN)、卷积神经网络(Convolutional Neural Networks,CNN)、支持向量机(Support Vector Machine,SVM)、长短时记忆网络(Long Short-Term Memory,LSTM)、DNN等。可选的,模型结构还可以包括神经网络的层数、量化方法等。
模型参数包括神经网络各层所用的参数。神经网络比如至少包括输入层、隐藏层和输出层。
可选的,模型数据集包括多个不同的数据集,不同的数据集对应不同的输入信息和/或不同的输出信息。示例性,不同的数据集对应不同的取值组合,该取值组合为N和M的取值组合。
在一些实施例中,模型信息还包括如下至少之一:模型功能,模型标识(Identifier,ID),输入信息,输出信息,数据集ID。
在一些实施例中,模型功能与上行CSI处理或上行信号处理相关,具体地,模型功能包括如下至少一种:上行部分CSI恢复,上行波束信息恢复,上行非线性补偿。
其中,上行部分CSI恢复和上行波束信息恢复是上行CSI处理的相关功能,上行非线性补偿是上行信号处理的相关功能。在一些实施例中,上行CSI恢复也可以包含CSI预测的功能(即上行部分CSI恢复及预测)。在一些实施例中,上行波束信息恢复也可以包含波束预测的功能(即上行波束信息恢复及预测)。
在一些实施例中,一个或多个AI模型由第一终端设备训练得到。也就是说,第一终端设备将自身训练得到的一个或多个AI模型的模型信息上报给网络设备。
在一些实施例中,一个或多个AI模型由第二终端设备训练得到。也就是说,第一终端设备将第二终端设备训练得到的一个或多个AI模型的模型信息上报给网络设备。可选的,第二终端设备训练得到一个或多个AI模型后,将一个或多个AI模型发送给第一终端设备。可选的,第二终端设备训练得到一个或多个AI模型后,将一个或多个AI模型的模型信息发送给第一终端设备。
在一些实施例中,AI模型的数量为多个,多个AI模型中的一部分AI模型由第一终端设备训练得到,另一部分AI模型由第二终端设备训练得到。
第二终端设备与第一终端设备采用相同的硬件配置。其中,硬件配置比如包括如下一种或多种配置:天线配置、模拟波束赋形方式、功率放大器性能。其中,天线配置比如包括如下一项或多项参数:天线数量、天线排列方式、天线是否为全相干、天线是否为部分相干、天线端口数量等。模拟波束赋形方式比如包括如下一项或多项参数:模拟波束数量、模拟波束方向、模拟波束带宽、模拟波束赋形矩阵。功率放大器性能比如包括如下一项或多项参数:增益、带宽、线性度、效率、失真。
由于AI模型是在终端设备侧训练得到的,并不依赖于网络设备的数据进行模型训练,且有助于提高AI模型与终端设备的匹配程度。由于模型信息主要与硬件配置相关,且第一终端设备与第二终端设备具备相同的硬件配置,这样的训练方式并不会影响训练结果,且可以大大提高获取训练数据集的效率。
在一些实施例中,一个或多个AI模型由网络设备训练得到,用于训练该一个或多个AI模型的数据集由终端设备上报。示例性的,模型信息包括模型数据集,则网络设备可以根据一个或多个终端设备(比如是第一终端设备和/或第二终端设备)上报的数据集进行训练,从而得到一个或多个AI模型。这种情况下,虽然AI模型是在网络设备侧训练得到的,但用于训练的数据集是由终端设备侧提供的,AI模型的训练仍然无需依赖于网络设备的数据,训练得到的AI模型与终端设备的匹配程度较高。并且,在网络设备侧训练AI模型,能够减轻终端设备侧的复杂度和功耗压力,提高AI模型训练的效率。
(1)模型功能包括上行部分CSI恢复的情况。
在一些实施例中,AI模型用于网络设备进行上行CSI处理,且上行CSI处理包括上行部分CSI恢复;AI模型的输入信息包括M个天线端口对应的上行CSI,AI模型的输出信息包括N个天线端口对应的上行CSI;其中,M为大于或等于1的整数,N为大于1的整数,M小于N。
在一些实施例中,AI模型的输入信息包括M个天线端口和K个子带对应的上行CSI,AI模型的输出信息包括N个天线端口和L个子带对应的上行CSI;其中,K为大于或等于1的整数,L为大于1的整数,K小于L。
子带是一种频域资源的划分粒度,本申请实施例以子带为例进行示意性说明。实际上,频域资源还可以按照其他粒度划分,比如:频带、带宽部分(Bandwidth Part,BWP)、载波、物理资源块(Physical Resource Block,PRB)、子信道、子载波等等。因此,K个子带也可以替换为K个频带,或K个BWP,或K个PRB,或K个子信道,或K个载波,或K个子载波等。
在一些实施例中,上行CSI恢复也可以包含CSI预测的功能,即所述N个天线端口对应的上行CSI可以是当前CSI的,也可以是预测的未来时刻的完整CSI。
在一些实施例中,第一终端设备的天线为部分相干天线,且第一终端设备包括相干的h个天线端口组,h为大于1的整数;M个天线端口包括h个天线端口组中每个天线端口组的至少一个天线端口。
在一些实施例中,M个天线端口包括M个探测参考信号(Sounding Reference Signal,SRS)端口,第 一终端设备的上行传输包括M个SRS端口上的SRS传输。
SRS是一种上行参考信号,即由终端设备向网络设备发送的参考信号。本申请在涉及上行参考信号时,主要以SRS为例,实际上,SRS还可以替换为其他上行参考信号,比如解调参考信号(Demodulation Reference Signal,DMRS),或相位追踪参考信号(Phase Tracking Reference Signal,PT-RS),或通信协议未来规定的上行参考信号。
关于N和M的取值的两种情况:
情况1:在一些实施例中,AI模型的输入信息还包括N和M的取值。
情况2:在一些实施例中,在AI模型的数量为多个的情况下,也即,第一终端设备上报多个AI模型的模型信息的情况下:不同的AI模型对应不同的取值组合,该取值组合为N和M的取值组合;或者,AI模型的输入信息还包括N和M的取值。
可选的,在这两种情况下,N和M的取值也可以是AI模型的模型信息的一部分。
(2)模型功能包括上行波束信息恢复的情况。
在一些实施例中,AI模型的输入信息包括M个SRS资源或M个波束对应的接收信号测量值,AI模型的输出信息包括至少一个SRS资源指示(SRS Resource Indicator,SRI),至少一个SRI中的每个SRI用于指示N个SRS资源中的一个SRS资源;其中,M为大于或等于1的整数,N为大于1的整数,M小于N。
“至少一个SRI”,意味着SRI的数量可以等于1或大于1,也即,AI模型的输出信息包括一个SRI或多个SRI。而每个SRI用于指示N个SRS资源中的一个SRS资源,不同的SRI指示不同的SRS资源。可选的,一个SRS资源对应一个或多个天线端口,不同的SRS资源所对应的天线端口完全不同或部分相同。
在一些实施例中,上行波束信息恢复也可以包含波束预测的功能,即所述SRI可以对应当前的最佳波束,也可以是预测的未来时刻的最佳波束。
在一些实施例中,上行传输的调度信息用于配置N个SRS资源,N个SRS资源包括M个SRS资源,上行传输包括M个SRS资源上的SRS传输。换句话说,调度信息配置了N个SRS资源,第一终端设备根据调度信息在N个SRS资源中的M个SRS资源上进行SRS传输。
在一些实施例中,AI模型的输入信息还包括N和M的取值。
在一些实施例中,在AI模型的数量为多个的情况下,也即,第一终端设备上报多个AI模型的模型信息的情况下:不同的AI模型对应不同的取值组合,该取值组合为N和M的取值组合;或者,AI模型的输入信息还包括N和M的取值。
关于N和M的取值的两种情况:
情况1:在一些实施例中,AI模型的输入信息还包括N和M的取值。
情况2:在一些实施例中,在AI模型的数量为多个的情况下,也即,第一终端设备上报多个AI模型的模型信息的情况下:不同的AI模型对应不同的取值组合,该取值组合为N和M的取值组合;或者,AI模型的输入信息还包括N和M的取值。
可选的,在这两种情况下,N和M的取值也可以是AI模型的模型信息的一部分。
(3)模型功能包括上行非线性补偿的情况。
在一些实施例中,AI模型用于网络设备进行上行信号处理,AI模型的输入信息包括网络设备检测到的调制信号,AI模型的输出信息包括经过非线性补偿后的调制信号。其中,上行信号处理包括上行非线性补偿。
在一些实施例中,模型信息可以通过UE能力进行上报,或者,通过媒体访问控制(Medium Access Control,MAC)层信令上报,或者,通过上行控制信息(Uplink Control Information,UCI)进行上报。这一设计有助于保障模型信息的上报灵活性。
步骤520的其他相关内容可参考步骤420,此处不再赘述。
步骤540:根据网络设备的调度信息进行上行传输,该上行传输与一个或多个AI模型中的至少一个AI模型关联。
在一些实施例中,上行传输与一个或多个AI模型中的至少一个AI模型关联,体现在如下一个或多个方面:上行传输采用的天线端口,与至少一个AI模型的训练数据集中的输入信息所采用的天线端口相同;上行传输采用的发送波束,与至少一个AI模型的训练数据集中的输入信息所采用的发送波束相同;上行传输采用的功率放大器,与至少一个AI模型的训练数据集中的输入信息所采用的功率放大器相同。
可选的,上行传输与一个或多个AI模型中的至少一个AI模型的关联,还可以通过其他物理层技术体现。比如,调制方式、编码方式、波形、带宽、传输速率、通信方式(比如单工通信、半双工通信、全双工通信等)、信息传输方式(比如串行传输、并行传输)、资源映射方式(比如集中式资源分配方式、分布 式资源分配方式)等物理层技术中的一项或多项。
在一些实施例中,一个或多个AI模型中与上行传输关联的AI模型,基于调度信息中的第一指示信息确定。示例性的,调度信息包括第一指示信息,第一终端设备基于第一指示信息确定一个或多个AI模型中与上行传输关联的至少一个AI模型。
在一些实施例中,网络设备发送的调度信息与一个或多个AI模型中的至少一个AI模型关联。体现在如下一个或多个方面:调度信息与至少一个AI模型的输入信息关联;调度信息与至少一个AI模型的输出信息关联;调度信息与至少一个AI模型的预测/推理结果关联;调度信息调度的上行传输与至少一个AI模型关联。
在一些实施例中,在执行步骤540前,第一终端设备接收网络设备发送的调度信息。
步骤540的其他相关内容可参考步骤440,此处不再赘述。
步骤560:接收第二指示信息。
第二指示信息用于指示第一终端设备更新一个或多个AI模型,或者,用于指示第一终端设备采用不基于AI模型的传输方式。其中,第一终端设备采用不基于AI模型的传输方式,也可以理解为,第一终端设备的上行传输与AI模型不关联。
步骤580:上报一个或多个AI模型的更新模型信息。
可选的,更新模型信息包括:更新后的模型结构,和,更新后的模型参数。可选的,更新模型信息还包括如下一项或多项:更新后的模型功能,更新后的模型ID,更新后的输入信息,更新后的输出信息,更新后的模型数据集,更新后的数据集ID。
在一些实施例中,第一终端设备上报完整的更新模型信息。也就是说,更新一个或多个AI模型后,第一终端设备上报更新后的一个或多个AI模型的模型信息。
在一些实施例中,第一终端设备上报的更新模型信息并不是完整的模型信息,这有助于减少资源开销。此处介绍三种可能:
其一,第一终端设备只上报更新后产生变化的模型信息;
比如,在步骤520中,第一终端设备上报了模型结构、模型参数、模型功能、模型ID、输入信息和输出信息,第一终端设备对一个或多个AI模型进行更新,更新后只有输入信息和输出信息产生了变化,那么,在步骤580中,第一终端设备上报的更新模型信息就只包括更新后的输入信息和更新后的输出信息。
又比如,在步骤520中,第一终端设备上报了模型结构、模型功能、模型ID、输入信息和输出信息,第一终端设备对一个或多个AI模型进行更新,更新后只有模型参数产生了变化,那么,在步骤580中,第一终端设备上报的更新模型信息就只包括更新后的模型参数。
又比如,第一终端设备上报更新后产生变化的AI模型的模型信息。
其二,第一终端设备只上报模型信息的差分信息,差分信息用于表示更新后的模型信息与更新前的模型信息之间的差异。
比如,在步骤520中,第一终端设备上报了模型结构、模型参数、模型功能、模型ID、输入信息和输出信息,第一终端设备对一个或多个AI模型进行更新,那么,在步骤580中,第一终端设备上报的更新模型信息包括:模型结构的差分信息、模型参数的差分信息、模型功能的差分信息、模型ID的差分信息、输入信息的差分信息和输出信息的差分信息。
又比如,在步骤520中,第一终端设备上报了模型结构、模型参数、模型功能、输入信息和输出信息,第一终端设备对一个或多个AI模型进行更新,那么,在步骤580中,第一终端设备上报的更新模型信息包括:模型结构的差分信息、模型参数的差分信息、模型功能的差分信息、输入信息的差分信息和输出信息的差分信息。
其三,第一终端设备只上报更新后产生变化的模型信息的差分信息。
上述三种情况中,更新后产生变化的模型信息,可以是模型结构、模型参数、模型功能、模型ID、输入信息、输出信息、模型数据集和数据集ID中的任意一种或多种,此处不一一列举所有可能性。但应当理解,哪些模型信息会在更新后产生变化,受实际的通信环境、通信业务、通信需求、AI模型自身情况等因素的影响,本申请支持任意一种或多种模型信息在更新后产生变化的可能。
另外,第一终端设备对AI模型的更新,可能是基于第二指示信息进行的,或与第二指示信息无关。也就是说,第一终端设备对AI模型的更新,可能发生在接收到指示更新一个或多个AI模型的第二指示信息后,也可能发生在未接收到第二指示信息的情况。比如,第一终端设备周期性地更新一个或多个AI模型。又比如,在满足某个或某些条件的情况下,第一终端设备更新一个或多个AI模型,此处的条件可能与通信质量、业务需求、模型信息、第一终端设备自身的能力、第一终端设备的电量、第一终端设备的内存等等因素相关。总之,本申请支持第一终端设备在第二指示信息的指示或触发下更新AI模型,也支持第一终端设备基于预定义的规则或条件自发地更新AI模型。因此,步骤560并不是步骤580的必要条件, 而应当作为一种可选情况。
类似地,第一终端设备采用不基于AI模型的传输方式,可能是基于第二指示信息进行的,或与第二指示信息无关。也就是说,第一终端设备采用不基于AI模型的传输方式,可能发生在接收到指示采用不基于AI模型的传输方式的第二指示信息后,也可能发生在未接收到第二指示信息的情况。比如,第一终端设备侧维持一个计时器(为了区分,可以将其称为第一计时器,第一计时器的存在与第二计时器是否存在并非一定关联),计时器开始计时或超时后停止基于AI模型的传输方式。又比如,在满足某个或某些条件的情况下,第一终端设备采用不基于AI模型的传输方式,此处的条件可能与通信质量、业务需求、模型信息、第一终端设备自身的能力、第一终端设备的电量、第一终端设备的内存等等因素相关。总之,本申请支持第一终端设备在第二指示信息的指示或触发下采用不基于AI模型的传输方式,也支持第一终端设备基于预定义的规则或条件自发地采用不基于AI模型的传输方式。因此,步骤560并不是第一终端设备采用不基于AI模型的传输方式的必要条件,而应当作为一种可选情况。
需要强调的是,步骤560和步骤580均为可选步骤。第一终端设备可以只执行步骤520和步骤540,或者,第一终端设备执行步骤520、步骤540和步骤560,或者,第一终端设备执行步骤520、步骤540和步骤580,或者,第一终端设备执行步骤520、步骤540、步骤560和步骤580,或者,第一终端设备执行步骤520、步骤540和采用不基于AI模型的传输方式,或者,第一终端设备执行步骤520、步骤540、步骤560和采用不基于AI模型的传输方式。
综上所述,本申请实施例中提供的方法,第一终端设备上报的模型信息所对应的AI模型,是在终端设备侧训练得到的,或由网络设备根据终端设备上报的模型数据集训练得到的,由于第二终端设备与第一终端设备具有相同的硬件配置,无论AI模型是由第一终端设备还是由第二终端设备甚至是由网络设备训练的,AI模型与第一终端设备的契合度都是极高的,十分匹配第一终端设备的硬件能力和上行传输需求,网络设备使用AI模型后不会对上行传输性能造成负面影响。并且,还支持第一终端设备更新AI模型,以便于AI模型更适应变化的通信环境、变化的通信需求,提供一种更为灵活的基于AI模型的通信方法。
图6示出了本申请一个示例性实施例提供的基于AI模型的通信方法的流程示意图,该方法由网络设备执行,该方法包括如下至少部分步骤:
步骤620:接收一个或多个AI模型的模型信息,AI模型用于网络设备进行上行CSI处理或上行信号处理。
本申请实施例中的AI模型,包括如下一种或多种类型:神经网络、决策树、支持向量机、随机森林、均值算法等。可选地,AI模型还可以称为AI规则或AI函数或其他名称。
上行CSI处理,指针对上行信号的CSI的处理。若AI模型用于网络设备进行上行CSI处理,则网络设备可以采用AI模型获取/预测/推断上行CSI的处理结果。
上行信号处理,指针对上行信号的接收端处理。若AI模型用于网络设备进行上行信号处理,则网络设备可以采用AI模型获取/预测/推断上行信号的处理结果。
步骤640:发送调度信息,调度信息用于调度第一终端设备进行上行传输,该上行传输与一个或多个AI模型中的至少一个AI模型关联。
至少一个AI模型,可以认为是一个AI模型或多个AI模型。
第一终端设备进行上行传输,指第一终端设备向网络设备发送信号和/或数据。
综上所述,本申请实施例中提供的方法,支持网络设备采用接收到的AI模型来进行上行CSI处理或上行信号处理。由于AI模型的模型信息是第一终端设备上报的,且网络设备调度的上行传输与AI模型关联,无论第一终端设备具备什么样的硬件结构,都能保障上行CSI处理的结果或上行信号处理的结果符合第一终端设备的能力,且能够避免AI模型对第一终端设备的上行传输产生负面影响。
在一些实施例中,在图6所示实施例的基础上,步骤620可以进一步实现为步骤710,步骤640可以进一步实现为步骤720,如图7所示。可选的,除步骤710、步骤720外,第一终端设备还可以执行如下可选步骤中的一个或多个:步骤730、步骤740、步骤750。
图7示出了本申请一个示例性实施例提供的基于AI模型的通信方法的流程示意图,该方法由网络设备执行,该方法包括如下至少部分步骤:
步骤710:接收一个或多个AI模型的模型信息,AI模型用于网络设备进行上行CSI处理或上行信号处理。
步骤710的相关内容请参考步骤520、步骤620,此处不再赘述。
步骤720:发送调度信息,调度信息用于调度第一终端设备进行上行传输,该上行传输与一个或多个AI模型中的至少一个AI模型关联。
步骤720的相关内容请参考步骤540、步骤640,此处不再赘述。
步骤730:基于至少一个AI模型中的目标AI模型和上行传输,进行上行CSI处理或上行信号处理。
在一些实施例中,网络设备执行如下至少一种处理:
·基于上行传输得到上行部分CSI,将上行部分CSI作为目标AI模型的输入信息,得到上行完整CSI;
·基于上行传输得到上行部分波束对应的接收信号测量值,将上行部分波束对应的接收信号测量值作为目标AI模型的输入信息,得到上行完整波束中最优波束对应的SRI;
·将网络设备检测到的调制信号作为目标AI模型的输入信息,得到经过非线性补偿的调制信号。
其中,接收信号测量值,采用如下任意一种或多种参数表示:RSRP、接收信号强度指示(Received Signal Strength Indication,RSSI)、参考信号接收质量(Reference Signal Receiving Quality,RSRQ)、信号与干扰加噪声比(Signal to Interference plus Noise Ratio,SINR)、信噪比(Signal-to-Noise Ratio,SNR)。
步骤740:发送第二指示信息。
第二指示信息用于指示第一终端设备更新一个或多个AI模型,或者,用于指示第一终端设备采用不基于AI模型的传输方式。其中,第一终端设备采用不基于AI模型的传输方式,也可以理解为,第一终端设备的上行传输与AI模型不关联。
在一些实施例中,网络设备基于一个或多个AI模型的性能监测结果,发送第二指示信息。示例性的,网络设备进行一个或多个AI模型的性能监测,当一个或多个AI模型的模型性能恶化时(可以是全部AI模型的性能恶化,也可以仅是部分AI模型的性能恶化),向第一终端设备发送第二指示信息。示例性的,网络设备进行一个或多个AI模型的性能监测,当一个或多个AI模型的模型信息达到阈值时,向第一终端设备发送第二指示信息。
在一些实施例中,网络设备周期性地发送第二指示信息。
在一些实施例中,网络设备在某个计时器(为了区分,可以称其为第二计时器,第二计时器的存在与第一计时器是否存在并非一定关联,可以理解,这个计时器也可以称为第一计时器,本申请中的“第一”或“第二”仅用于区分而不意味着存在关联)开始计时或超时后发送第二指示信息。
步骤750:接收一个或多个AI模型的更新模型信息。
可选的,更新模型信息包括:更新后的模型结构,和,更新后的模型参数。可选的,更新模型信息还包括如下一项或多项:更新后的模型功能,更新后的模型ID,更新后的输入信息,更新后的输出信息,更新后的模型数据集,更新后的数据集ID。
在一些实施例中,网络设备接收完整的更新模型信息。也就是说,网络设备接收更新后的一个或多个AI模型的模型信息。
在一些实施例中,网络设备接收的更新模型信息并不是完整的模型信息,这有助于减少资源开销。此处存在三种可能:其一,网络设备接收的是更新后产生变化的模型信息;其二,网络设备接收的是模型信息的差分信息;其三,网络设备接收的是更新后产生变化的模型信息的差分信息。
步骤750的其他相关内容可参考步骤580,此处不再赘述。
需要强调的是,步骤730、步骤740和步骤750均为可选步骤。网络设备可以只执行步骤710和步骤720,或者,网络设备执行步骤710、步骤720和步骤730,或者,网络设备执行步骤710、步骤720和步骤740,或者,网络设备执行步骤710、步骤720和步骤750,或者,网络设备执行步骤710、步骤720、步骤730和步骤740,或者,网络设备执行步骤710、步骤720、步骤730和步骤750,或者,网络设备执行步骤710、步骤720、步骤740和步骤750,或者,网络设备执行步骤710、步骤720、步骤730、步骤740和步骤750,或者,网络设备执行步骤710、步骤720和接收不基于AI模型的上行传输,或者,网络设备执行步骤710、步骤720、步骤730和接收不基于AI模型的上行传输,或者,网络设备执行步骤710、步骤720、步骤740和接收不基于AI模型的上行传输。
另外,步骤710至步骤750的执行顺序可以根据实际情况调整。比如步骤730在步骤720之前执行。比如步骤750在步骤740之前执行。
综上所述,本申请实施例中提供的方法,网络设备采用的AI模型是在终端设备侧训练得到的,或由网络设备根据终端设备上报的模型数据集训练得到的,由于第二终端设备与第一终端设备具有相同的硬件配置,无论AI模型是由第一终端设备还是由第二终端设备甚至是由网络设备训练的,AI模型与第一终端设备的契合度都是极高的,十分匹配第一终端设备的硬件能力和上行传输需求,网络设备使用AI模型后不会对上行传输性能造成负面影响。并且,还支持第一终端设备更新AI模型,以便于AI模型更适应变化的通信环境、变化的通信需求,提供一种更为灵活的基于AI模型的通信方法。
进一步地,在图5和/或图7所示实施例的基础上,以上行CSI处理包括上行部分CSI恢复为例,图8示出了本申请一个示例性实施例提供的基于AI模型的通信方法的流程示意图,该方法由第一终端设备和网络设备执行,该方法包括如下至少部分步骤:
步骤820:第一终端设备向网络设备上报一个或多个AI模型的模型信息,AI模型用于网络设备进行 上行CSI处理,上行CSI处理包括上行部分CSI恢复。
在一些实施例中,模型信息包括如下一项或多项:模型数据集,模型结构,模型参数。
在一些实施例中,模型信息还包括如下一项或多项:模型功能,模型ID,输入信息,输出信息,数据集ID。
本申请实施例中,以模型功能至少包括上行部分CSI恢复为例。模型ID用于识别AI模型,可以是AI模型自身的ID或索引,也可以是AI模型对应的数据集的ID或索引。输入信息可以包含在模型结构中上报或单独上报,也就是说,模型结构可以包含输入信息或不包含输入信息。输出信息可以包含在模型结构中上报或单独上报,也就是说,模型结构可以包含输出信息或不包含输出信息。模型数据集即AI模型对应的数据集,指用于训练AI模型的数据集合。数据集ID用于识别AI模型对应的数据集,可以是数据集的ID或索引。
可选的,模型数据集包括多个不同的数据集,不同的数据集对应不同的输入信息和/或不同的输出信息。示例性,不同的数据集对应不同的取值组合,该取值组合为N和M的取值组合。
可选的,一个数据集包含若干个数据样本。可选的,每个数据样本都包含多个特征和一个或多个标签。特征用于描述数据样本的属性,标签指数据样本所属的类别或结果。
可选的,输入信息包括如下一项或多项:输入向量个数,每个向量的大小,输入的具体含义等。可选的,输出信息包括如下一项或多项:输出向量个数,每个向量的大小,输出的具体含义等。
在一些实施例中,一个或多个AI模型由第一终端设备训练得到。或者,一个或多个AI模型由第二终端设备训练得到。或者,AI模型的数量为多个,多个AI模型中的一部分AI模型由第一终端设备训练得到,另一部分AI模型由第二终端设备训练得到。由于AI模型可以是终端设备侧训练得到的,无需网络设备进行模型训练,降低了对网络设备的依赖性,提高AI模型与终端设备的匹配程度。并且,AI模型可以是由第一终端设备训练得到的,也可以是由第二终端设备训练得到的,由于模型信息主要与硬件配置相关,且第一终端设备与第二终端设备具备相同的硬件配置,这样的训练方式并不会影响训练结果,且可以大大提高获取训练数据集的效率。
在一些实施例中,一个或多个AI模型由网络设备训练得到,用于训练该一个或多个AI模型的数据集由终端设备上报。示例性的,模型信息包括模型数据集,则网络设备可以根据一个或多个终端设备(比如是第一终端设备和/或第二终端设备)上报的数据集进行训练,从而得到一个或多个AI模型。虽然AI模型是在网络设备侧训练得到的,但用于训练的数据集是由终端设备侧提供的,AI模型的训练仍然无需依赖于网络设备的数据,训练得到的AI模型与终端设备的匹配程度较高。并且,在网络设备侧训练AI模型,能够减轻终端设备侧的复杂度和功耗压力,提高AI模型训练的效率。
在一些实施例中,AI模型的输入信息包括M个天线端口对应的上行CSI,输出信息包括N个天线端口对应的上行CSI;M为大于或等于1的整数,N为大于1的整数,M小于N。
其中,M表示当前测量的上行天线端口的数量,N表示终端设备总的上行天线端口的数量。基于该AI模型,可以从测量的部分天线端口上的CSI,推理/预测得到完整天线端口上的CSI,也即推理/预测得到全部天线端口上的CSI。
本申请实施例中的CSI,可以表示为信道矩阵,或信道协方差矩阵,或信道特征向量,或预编码矩阵等。并且,CSI可以是子带的CSI,也可以是宽带的CSI。
在一些实施例中,当第一终端设备的天线阵列为部分相干的天线阵列,且天线端口包含若干个相干的天线端口组(各个天线端口组之间是非相干的,一个天线端口组内部是相干的),则M个天线端口包含若干个相干的天线端口组中每个天线端口组的至少一个天线端口。具体的,假设第一终端设备的天线端口中包含P(P为大于1的整数)个相干的天线端口组,则M个天线端口至少包含P个天线端口,P个天线端口中的每个天线端口来自于不同的天线端口组。由于非相干的天线端口之间相位是独立的,AI模型无法根据一个天线端口组中的天线端口对应的CSI,去估计另一个天线端口组中的天线端口对应的CSI,因此需要包含每个天线端口组中的至少一个端口,以保证推理/预测的性能。
在一些实施例中,AI模型为多个,且不同的N和M的取值组合对应多个AI模型中的不同AI模型。也就是说,第一终端设备可以向网络设备上报多个AI模型,不同的AI模型对应不同的N和M的取值组合,从而支持网络设备灵活配置不同数量的天线端口用于上行CSI测量,节约上行的参考信号开销。特别的,不同的AI模型可以对应不同的M的取值和相同的N的取值,因为第一终端设备总的天线端口数一般是不变的。例如,第一终端设备可以上报三个AI模型的模型信息,分别对应{M=8,12,16,N=32},或者对应{M=2,4,8,N=16}。
在一些实施例中,N和M的取值也作为AI模型的输入信息。基于这种方式,AI模型可以根据N和M的取值匹配最合适的模型参数,从而获得最优的性能。
在一些实施例中,AI模型的输入信息包含M个天线端口和K个子带对应的上行CSI,输出信息包含 N个天线端口和L个子带对应的上行CSI;K为大于或等于1的整数,L为大于1的整数,K小于L。也就是说,AI模型可以同时用于空域(天线端口)和频域(子带)的CSI恢复。另外,这里子带也可以用PRBs或子载波或子信道等代替,比如,AI模型的输入信息包含M个天线端口和K个PRB对应的上行CSI,输出信息包含N个天线端口和L个PRB对应的上行CSI。
在一些实施例中,模型信息可以通过UE能力进行上报,或者,通过MAC层信令上报,或者,通过UCI进行上报。这一设计有助于保障模型信息的上报灵活性。
步骤820的其他相关内容可参考步骤520和/或步骤710,此处不再赘述。
步骤840:网络设备通过调度信息调度第一终端设备进行上行传输,该上行传输与一个或多个AI模型中的至少一个AI模型关联。
在一些实施例中,调度信息中包含第一指示信息,第一指示信息用于从一个或多个AI模型中指示与上行传输关联的AI模型。可选的,第一指示信息可以指示如下信息中的一项或多项:模型ID,功能ID,数据集ID,模型索引(即AI模型在第一终端设备上报的所有模型中的序号)等。可选的,模型ID、功能ID、数据集ID和模型索引中的一项或多项可以包含在第一终端设备上报的模型信息中,这样,第一终端设备基于第一指示信息,就可以从已上报的AI模型中确定与上行传输关联的AI模型。
在一些实施例中,上行传输与AI模型的关联体现在:上行传输所用的天线端口(天线),与AI模型的训练数据集中的输入信息所对应的天线端口(天线)相同。也就是说,第一终端设备需要采用训练AI模型时使用的天线端口(天线),来进行上行传输,从而保证推理和训练的一致性。
在一些实施例中,上行传输为M个SRS天线端口上的SRS传输,SRS传输与一个或多个AI模型中的一个AI模型关联。这里,调度信息用于调度/配置M个SRS端口的传输,网络设备只要配置M个SRS天线端口来获得N个SRS天线端口的CSI,从而可以节约参考信号的开销。如果网络设备配置M个SRS端口,则需要通过第一指示信息指示与M个SRS端口关联的AI模型,该AI模型的输入为M个天线端口的CSI。进一步的,SRS传输的M个天线端口(即物理天线)需要与关联的AI模型训练时使用的M个天线端口(即物理天线)相同。
步骤840的其他相关内容可参考步骤540和/或步骤720,此处不再赘述。
步骤860:第一终端设备根据调度信息进行上行传输,该上行传输与一个或多个AI模型中的至少一个AI模型关联。
在一些实施例中,第一终端设备基于调度信息中的第一指示信息,从一个或多个AI模型中确定与上行传输关联的AI模型。
在一些实施例中,上行传输为M个SRS端口上的SRS传输,SRS传输与一个或多个AI模型中的一个AI模型关联。
步骤860的其他内容可参考步骤540和/或步骤840,此处不再赘述。
步骤880:网络设备基于至少一个AI模型中的目标AI模型和上行传输,进行上行CSI处理,上行CSI处理包括上行部分CSI恢复。
在一些实施例中,当至少一个AI模型中只包含一个AI模型时,也即,上行传输只与一个AI模型关联时,目标AI模型即为该AI模型。当至少一个AI模型中包含多个AI模型时,也即,上行传输与多于一个AI模型关联时,目标AI模型为网络设备从与上行传输关联的若干个AI模型中选出的一个AI模型。另外,目标AI模型可以是网络设备随机选择的,也可以是网络设备按照某一规则选择的。
在一些实施例中,网络设备基于上行传输(比如M个端口的SRS传输)得到上行部分CSI(即M个天线端口对应的CSI),将上行部分CSI作为目标AI模型的输入,从而得到上行的完整CSI(即N个天线端口对应的CSI)。
在一些实施例中,网络设备还进行AI模型的性能监测,并在模型性能恶化时向第一终端设备发送第二指示信息,第二指示信息用于指示第一终端设备更新AI模型,或者,用于指示第一终端设备采用不基于AI模型的传输方式。采用不基于AI模型的传输方式,比如,采用不基于AI/ML接收的传输方式。
相应地,第一终端设备接收网络设备的第二指示信息。
在一些实施例中,网络设备可以令第一终端设备以较低的频率发送N个天线端口的SRS,从而获得完整的信道信息,再将获得的信道信息与AI模型推理的结果比较,从而确定AI模型的性能。性能监测的结果可以采用如下一项或多项体现:广义余弦相似度(Generalized Cosine Similarity,GCS),平方广义余弦相似度(Squared Generalized Cosine Similarity,SGCS),误块率(Block Error Rate,BLER),频谱效率(Spectral Efficiency)等。当AI模型的性能恶化时,例如性能监测的结果持续低于门限值时,网络设备可以向第一终端设备发送第二指示信息。
在一些实施例中,当第二指示信息用于指示第一终端设备更新AI模型时,第一终端设备可以向网络设备上报更新模型信息,以便于网络设备进行AI模型的更新。例如,第一终端设备可以通过MAC层信令 或者UCI向网络设备上报更新的模型信息,如模型参数等,从而网络设备可以采用更新的模型参数来更新模型,替换现有AI模型用于上行CSI恢复。
在一些实施例中,当指示信息用于指示第一终端设备采用不基于AI模型的传输方式时,第一终端设备可以发送N个天线端口的SRS,网络设备可以直接得到N个天线端口对应的CSI,不需要借助AI模型。
步骤880的其他相关内容可参考步骤730、步骤740、步骤750,此处不再赘述。
另外,步骤820、步骤840、步骤860、步骤880可以单独执行,也可以组合执行。比如,单独执行步骤820,实现一种终端设备侧的AI模型上报方法。又比如,单独执行步骤840,实现一种网络设备侧的调度方法/上行传输方法。又比如,组合执行步骤820和步骤840,实现一种基于AI模型的通信方法。又比如,组合执行步骤820、步骤840、步骤860。又比如,组合执行步骤820、步骤840、步骤880。又比如,组合执行步骤820、步骤840、步骤860、步骤880。并且,支持单独执行或组合执行网络设备侧的步骤,以实现网络设备侧的实施例;也支持单独执行或组合执行终端设备侧的步骤,以实现终端设备侧的实施例。
综上所述,本申请实施例提供的方法,支持终端设备基于自身的天线配置进行AI模型的训练,并将训练好的模型上报给网络设备,用于网络设备进行相应的上行CSI恢复。或者,由网络设备根据终端设备上报的模型数据集训练得到AI模型,以便于网络设备进行相应的上行CSI恢复。网络设备使用的AI模型与第一终端设备的契合度是极高的,十分匹配第一终端设备的硬件能力和上行传输需求,有助于保证上行CSI恢复的性能。并且,借助于第一终端设备上报的AI模型的模型信息,网络设备可以只配置少量的SRS信号来测量上行CSI,显著降低了上行的导频开销。
在图5和/或图7所示实施例的基础上,以上行CSI处理包括上行波束恢复为例,图9示出了本申请一个示例性实施例提供的基于AI模型的通信方法的流程示意图,该方法由第一终端设备和网络设备执行,该方法包括如下至少部分步骤:
步骤920:第一终端设备向网络设备上报一个或多个AI模型的模型信息,AI模型用于网络设备进行上行CSI处理,上行CSI处理包括上行波束恢复。
模型信息的相关内容可参考步骤520、步骤820。步骤920与步骤820的区别在于,步骤920中以模型功能至少包括上行波束恢复为例。
在一些实施例中,一个或多个AI模型由第一终端设备训练得到。或者,一个或多个AI模型由第二终端设备训练得到。或者,AI模型的数量为多个,多个AI模型中的一部分AI模型由第一终端设备训练得到,另一部分AI模型由第二终端设备训练得到。其中,第二终端设备与第一终端设备采用相同的硬件配置,硬件配置的相关内容可参考步骤520,比如,第二终端设备与第一终端设备采用相同的模拟波束赋形方式。
在一些实施例中,一个或多个AI模型由网络设备训练得到,用于训练该一个或多个AI模型的数据集由终端设备(比如是第一终端设备和/或第二终端设备)上报。示例性的,模型信息包括模型数据集。
在一些实施例中,AI模型的输入信息包含M个SRS资源或M个波束对应的接收信号测量值(本申请实施例以RSRP为例),输出信息包含一个或多个SRI,每个SRI用于指示N个SRS资源中的一个SRS资源;M为大于或等于1的整数,N为大于1的整数,M小于N。
其中,M表示当前测量的SRS资源(或波束)的数量,N表示终端设备总的SRS资源(或波束)的数量。基于该AI模型,可以从测量的部分波束(或部分波束集合)的质量,推理/预测得到所有波束(或完整波束集合)中质量最好的波束。
在一些实施例中,M个SRS资源为N个SRS资源的子集。也即,N个SRS资源包含M个SRS资源。
在一些实施例中,AI模型为多个,且不同的N和M的取值组合对应多个AI模型中的不同AI模型。也就是说,第一终端设备可以向网络设备上报多个AI模型,不同的AI模型对应不同的N和M的取值组合,从而支持网络设备灵活配置不同数量的SRS资源用于上行波束管理,节约上行的参考信号开销。特别的,不同的AI模型可以对应不同的M的取值和相同的N的取值。例如,第一终端设备可以上报两个AI模型,分别对应{M=4,8,N=16},或者分别对应{(M=4,N=16)(M=8,N=32)}。
在一些实施例中,N和M的取值也作为AI模型的输入信息。基于这种方式,AI模型可以根据N和M的取值匹配最合适的模型参数,从而获得最优的性能。
在一些实施例中,AI模型的输入信息包含M个SRS资源上测量得到的RSRP,输出信息包含一个或多个SRI以及每个SRI分别对应的RSRP值,每个SRI用于指示N个SRS资源中的一个SRS资源。
在一些实施例中,模型信息可以通过UE能力进行上报,或者,通过MAC层信令上报,或者,通过UCI进行上报。这一设计有助于保障模型信息的上报灵活性。
步骤920的其他相关内容可参考步骤520和/或步骤710,此处不再赘述。
步骤940:网络设备通过调度信息调度第一终端设备进行上行传输,该上行传输与一个或多个AI模型中的至少一个AI模型关联。
在一些实施例中,调度信息中包含第一指示信息,第一指示信息用于从一个或多个AI模型中指示与上行传输关联的AI模型。可选的,第一指示信息可以指示如下信息中的一项或多项:模型ID,功能ID,数据集ID,模型索引(即AI模型在第一终端设备上报的所有模型中的序号)等。可选的,模型ID、功能ID、数据集ID和模型索引中的一项或多项可以包含在第一终端设备上报的模型信息中,这样,第一终端设备基于第一指示信息,就可以从已上报的AI模型中确定与上行传输关联的AI模型。
在一些实施例中,上行传输与AI模型的关联体现在:上行传输所用的发送波束,与AI模型的训练数据集中的输入信息所对应的发送波束相同。也就是说,第一终端设备需要采用训练AI模型时使用的发送波束,来进行上行传输,从而保证推理和训练的一致性。示例性的,AI模型的输入信息为M个RSRP值,对应M个SRS资源或者M个波束,上行传输为M个SRS资源上的SRS传输,则第一终端设备采用训练时的M个SRS资源所采用的波束来发送上行传输中的M个SRS资源,或者采用训练时的M个波束来发送上行传输中的M个SRS资源,从而保证推理/预测的性能。
在一些实施例中,上行传输为M个SRS资源上的SRS传输,M个SRS资源与一个或多个AI模型中的一个AI模型关联。这里,调度信息用于调度/配置M个SRS资源的传输,网络设备只要配置一个包含M个SRS资源的第一资源集合,就可以确定一个包含N个SRS资源的第二资源集合中的最优波束对应的SRS资源,从而可以节约参考信号的开销。其中,如果网络设备配置M个SRS资源,则需要通过第一指示信息指示与M个SRS资源关联的AI模型,该AI模型的输入信息包括M个RSRP值。进一步的,M个SRS资源所用的发送波束需要与关联的AI模型训练时使用的M个RSRP值对应的波束相同。
步骤940的其他相关内容可参考步骤540和/或步骤720,此处不再赘述。
步骤960:第一终端设备根据调度信息进行上行传输,该上行传输与一个或多个AI模型中的至少一个AI模型关联。
在一些实施例中,第一终端设备基于调度信息中的第一指示信息,从一个或多个AI模型中确定与上行传输关联的AI模型。
在一些实施例中,第一终端设备接收网络设备配置的N个SRS资源,并在其中的M个SRS资源上进行SRS传输。可选的,网络设备向第一终端设备指示N个SRS资源中哪些资源是实际传输SRS的。可选的,第一终端设备和网络设备约定或协商N个SRS资源中哪些资源是实际传输SRS的。
在一些实施例中,第一终端设备接收网络设备配置的M个SRS资源,并在这M个SRS资源上进行SRS传输。
步骤960的其他内容可参考步骤540、步骤940,此处不再赘述。
步骤980:网络设备基于至少一个AI模型中的目标AI模型和上行传输,进行上行CSI处理,上行CSI处理包括上行波束恢复。
在一些实施例中,当至少一个AI模型中只包含一个AI模型时,也即,上行传输只与一个AI模型关联时,目标AI模型即为该AI模型。当至少一个AI模型中包含多个AI模型时,也即,上行传输与多于一个AI模型关联时,目标AI模型为网络设备从与上行传输关联的若干个AI模型中选出的一个AI模型。另外,目标AI模型可以是网络设备随机选择的,也可以是网络设备按照某一规则选择的。
在一些实施例中,网络设备基于上行传输测量得到上行部分波束(即M个SRS资源)的RSRP,将上行部分波束的RSRP作为目标AI模型的输入,从而得到完整波束中最优波束对应的SRI(即N个SRS资源/波束中的最优资源/波束)。AI模型可以输出多个SRI,对应最优的多个波束。进一步的,AI模型还可以输出SRI对应的RSRP。
在一些实施例中,网络设备还进行AI模型的性能监测,并在模型性能恶化时向第一终端设备发送第二指示信息,第二指示信息用于指示第一终端设备更新AI模型,或者,用于指示第一终端设备采用不基于AI模型的传输方式。采用不基于AI模型的传输方式,比如,采用不基于AI/ML接收的传输方式。
相应地,第一终端设备接收网络设备的第二指示信息。
在一些实施例中,网络设备可以令第一终端设备以较低的频率发送N个SRS资源,从而获得完整波束集合中的最优波束及对应的RSRP,再将获得的信息与AI模型推理的结果比较,从而确定AI模型的性能。性能监测的结果可以采用如下一项或多项体现:可靠性,估计准确率,RSRP差值等。当AI模型的性能恶化时,例如性能监测的结果持续低于门限值时,网络设备可以向终端设备发送第二指示信息。
在一些实施例中,当第二指示信息用于指示第一终端设备更新AI模型时,第一终端设备可以向网络设备上报更新的模型信息,从而进行AI模型的更新。例如,第一终端设备可以通过MAC层信令或者UCI向网络设备上报更新的模型信息,如模型参数等,从而网络设备可以采用更新的模型参数来更新模型,替换现有AI模型用于上行波束恢复。
在一些实施例中,当指示信息用于指示第一终端设备采用不基于AI模型的传输方式时,第一终端设备可以发送N个SRS资源,网络设备可以直接得到N个SRS资源对应的RSRP,并确定出其中的最优波 束,不需要借助AI模型。
步骤980的其他相关内容可参考步骤730、步骤740、步骤750,此处不再赘述。
另外,步骤920、步骤940、步骤960、步骤980可以单独执行,也可以组合执行。比如,单独执行步骤920,实现一种终端设备侧的AI模型上报方法。又比如,单独执行步骤940,实现一种网络设备侧的调度方法/上行传输方法。又比如,组合执行步骤920和步骤940,实现一种基于AI模型的通信方法。又比如,组合执行步骤920、步骤940、步骤960。又比如,组合执行步骤920、步骤940、步骤980。又比如,组合执行步骤920、步骤940、步骤960、步骤980。并且,支持单独执行或组合执行网络设备侧的步骤,以实现网络设备侧的实施例;也支持单独执行或组合执行终端设备侧的步骤,以实现终端设备侧的实施例。
综上所述,本申请实施例提供的方法,支持终端设备基于自身的天线配置进行AI模型的训练,并将训练好的模型上报给网络设备,用于网络设备进行相应的上行波束恢复。或者,由网络设备根据终端设备上报的模型数据集训练得到AI模型,以便于网络设备进行相应的上行波束恢复。网络设备使用的AI模型与第一终端设备的契合度是极高的,十分匹配第一终端设备的硬件能力和上行传输需求,有助于保证上行波束恢复的性能。并且,借助于第一终端设备上报的AI模型的模型信息,网络设备可以只配置少量的SRS信号来进行上行波束管理,显著降低了上行的导频开销。
在图5和/或图7所示实施例的基础上,以上行信号处理包括上行非线性补偿为例,图10示出了本申请一个示例性实施例提供的基于AI模型的通信方法的流程示意图,该方法由第一终端设备和网络设备执行,该方法包括如下至少部分步骤:
步骤1020:第一终端设备向网络设备上报一个或多个AI模型的模型信息,AI模型用于网络设备进行上行信号处理,上行信号处理包括上行非线性补偿。
模型信息的相关内容可参考步骤520、步骤820。步骤1020与步骤820的区别在于,步骤920中以模型功能至少包括上行非线性补偿为例。
在一些实施例中,一个或多个AI模型由第一终端设备训练得到。或者,一个或多个AI模型由第二终端设备训练得到。或者,AI模型的数量为多个,多个AI模型中的一部分AI模型由第一终端设备训练得到,另一部分AI模型由第二终端设备训练得到。其中,第二终端设备与第一终端设备采用相同的硬件配置,硬件配置的相关内容可参考步骤520,比如,第二终端设备与第一终端设备采用的功率放大器性能相同。
在一些实施例中,一个或多个AI模型由网络设备训练得到,用于训练该一个或多个AI模型的数据集由终端设备(比如是第一终端设备和/或第二终端设备)上报。示例性的,模型信息包括模型数据集。
在一些实施例中,AI模型的输入信息包含接收端(比如网络设备)检测到的调制信号,输出信息包含经过非线性补偿后的调制信号。
在一些实施例中,模型信息可以通过UE能力进行上报,或者,通过MAC层信令上报,或者,通过UCI进行上报。这一设计有助于保障模型信息的上报灵活性。
步骤1020的其他相关内容可参考步骤520和/或步骤710,此处不再赘述。
步骤1040:网络设备通过调度信息调度第一终端设备进行上行传输,该上行传输与一个或多个AI模型中的至少一个AI模型关联。
在一些实施例中,调度信息中包含第一指示信息,第一指示信息用于从一个或多个AI模型中指示与上行传输关联的AI模型。可选的,第一指示信息可以指示如下信息中的一项或多项:模型ID,功能ID,数据集ID,模型索引(即AI模型在第一终端设备上报的所有模型中的序号)等。可选的,模型ID、功能ID、数据集ID和模型索引中的一项或多项可以包含在第一终端设备上报的模型信息中,这样,第一终端设备基于第一指示信息,就可以从已上报的AI模型中确定与上行传输关联的AI模型。
在一些实施例中,上行传输与AI模型的关联体现在:上行传输所用的功率放大器,与AI模型的训练数据集中的输入信息所采用的功率放大器相同。或者,体现在:上行传输所用的功率放大器性能,与AI模型的训练数据集中的输入信息所采用的功率放大器性能相同。
在一些实施例中,上行传输为上行数据传输,上行数据传输与AI模型关联。这里,调度信息用于调度/配置上行数据传输。进一步的,上行数据传输所用的功率放大器,与AI模型的训练数据集中的输入信号生成时所采用的功率放大器相同。
步骤1040的其他相关内容可参考步骤540和/或步骤720,此处不再赘述。
步骤1060:第一终端设备根据调度信息进行上行传输,该上行传输与一个或多个AI模型中的至少一个AI模型关联。
在一些实施例中,第一终端设备基于调度信息中的第一指示信息,从一个或多个AI模型中确定与上行传输关联的AI模型。
在一些实施例中,调度信息用于调度上行数据传输,第一终端设备根据调度信息进行上行数据传输, 上行数据传输与一个或多个AI模型中的至少一个AI模型关联。
步骤1060的其他内容可参考步骤540、步骤940,此处不再赘述。
步骤1080:网络设备基于至少一个AI模型中的目标AI模型和上行传输,进行上行信号处理,上行信号处理包括上行非线性补偿。
在一些实施例中,网络设备将上行传输检测得到的调制信号(即MIMO接收完成后的待解调信号)作为AI模型的输入,从而得到经过非线性补偿的调制信号(后续可以作为解调模块的输入)。
在一些实施例中,网络设备还进行AI模型的性能监测,并在模型性能恶化时向第一终端设备发送第二指示信息,第二指示信息用于指示第一终端设备更新AI模型,或者,用于指示第一终端设备采用不基于AI模型的传输方式。采用不基于AI模型的传输方式,比如,采用不基于AI/ML接收的传输方式。
相应地,第一终端设备接收网络设备的第二指示信息。
在一些实施例中,网络设备持续监测解调性能,并在解调性能恶化时向第一终端设备发送第二指示信息。
在一些实施例中,当第二指示信息用于指示第一终端设备采用不基于AI模型的传输方式时,第一终端设备需要在发送端进行预失真处理以抵消功放非线性的影响,从而保证接收端(比如网络设备)的检测性能。
步骤1080的其他相关内容可参考步骤730、步骤740、步骤750,此处不再赘述。
另外,步骤1020、步骤1040、步骤1060、步骤1080可以单独执行,也可以组合执行。比如,单独执行步骤1020,实现一种终端设备侧的AI模型上报方法。又比如,单独执行步骤1040,实现一种网络设备侧的调度方法/上行传输方法。又比如,组合执行步骤1020和步骤1040,实现一种基于AI模型的通信方法。又比如,组合执行步骤1020、步骤1040、步骤1060。又比如,组合执行步骤1020、步骤1040、步骤1080。又比如,组合执行步骤1020、步骤1040、步骤1060、步骤1080。并且,支持单独执行或组合执行网络设备侧的步骤,以实现网络设备侧的实施例;也支持单独执行或组合执行终端设备侧的步骤,以实现终端设备侧的实施例。
综上所述,本申请实施例提供的方法,支持终端设备基于自身的天线配置进行AI模型的训练,并将训练好的模型上报给网络设备,用于网络设备进行相应的上行信号处理。或者,由网络设备根据终端设备上报的模型数据集训练得到AI模型,以便于网络设备进行相应的上行信号处理。网络设备使用的AI模型与第一终端设备的契合度是极高的,十分匹配第一终端设备的硬件能力和上行传输需求,有助于保证上行信号处理的性能。并且,借助于上报的AI模型的模型信息,第一终端设备不需要进行功放的预失真处理,明显降低了终端侧的处理复杂度和开销。
图11示出了本申请一个示例性实施例提供的基于AI模型的通信装置的结构框图,该装置可以实现成为前文所述的第一终端设备,或实现成为前文所述的第一终端设备的一部分。该装置包括发送模块1110。可选地,该装置还包括处理模块1130和/或接收模块1150。
发送模块1110,用于上报一个或多个AI模型的模型信息,所述AI模型用于网络设备进行上行CSI处理或上行信号处理;所述发送模块1110,还用于根据所述网络设备的调度信息进行上行传输,所述上行传输与所述一个或多个AI模型中的至少一个AI模型关联。
在一些实施例中,所述发送模块1110,还用于上报所述一个或多个AI模型的更新模型信息。
在一些实施例中,所述发送模块1110,还用于采用不基于AI模型的传输方式来进行上行传输。
在一些实施例中,发送模块1110用于执行如下步骤中的一个或多个:步骤420、步骤440、步骤520、步骤540、步骤580、步骤820、步骤860、步骤920、步骤960、步骤1020、步骤1060。
在一些实施例中,处理模块1130,用于训练一个或多个AI模型。
在一些实施例中,处理模块1130,用于训练多个AI模型中的部分AI模型。
在一些实施例中,处理模块1130,用于基于所述调度信息中的第一指示信息,确定一个或多个AI模型中与所述上行传输关联的AI模型。
在一些实施例中,处理模块1130,用于更新一个或多个AI模型。
在一些实施例中,接收模块1150,用于接收第二终端设备训练的一个或多个AI模型,或者,用于接收第二终端设备训练的一个或多个AI模型的模型信息。
在一些实施例中,接收模块1150,用于接收多个AI模型中由第二终端设备训练的部分AI模型,或者,用于接收多个AI模型中由第二终端设备训练的部分AI模型的模型信息。
在一些实施例中,接收模块1150,用于接收所述调度信息。
在一些实施例中,接收模块1150,用于接收第二指示信息,所述第二指示信息用于指示所述装置更新所述一个或多个AI模型,或者,用于指示所述装置采用不基于AI模型的传输方式。
发送模块1110、处理模块1130、接收模块1150执行的步骤,请参考图4、图5、图8、图9、图10 所示实施例中第一终端设备执行的一个或多个步骤,前文各个实施例所述的相关内容,同样适用于图11所示装置,此处不再一一赘述。
综上所述,本申请实施例提供的装置,支持上报由在终端设备侧训练得到的AI模型的模型信息,AI模型与本装置的契合度极高,十分匹配本装置的硬件能力和上行传输需求,网络设备使用AI模型后不会对上行传输性能造成负面影响。并且,还有助于减少上行资源开销,降低本装置的处理复杂度。并且,还支持本装置更新AI模型,以便于AI模型更加灵活适应变化的通信环境、变化的通信需求。
图12示出了本申请一个示例性实施例提供的基于AI模型的通信装置的结构框图,该装置可以实现成为前文所述的网络设备,或实现成为前文所述的网络设备的一部分。该装置包括接收模块1210和发送模块1250。可选地,该装置还包括处理模块1230。
接收模块1210,用于接收一个或多个AI模型的模型信息,所述AI模型用于所述装置进行上行CSI处理或上行信号处理。
发送模块1250,用于发送调度信息,所述调度信息用于调度第一终端设备进行上行传输,所述上行传输与所述一个或多个AI模型中的至少一个AI模型关联。
在一些实施例中,处理模块1230,用于基于所述至少一个AI模型中的目标AI模型和所述上行传输,进行上行CSI处理或上行信号处理。
在一些实施例中,处理模块1230,用于执行一种或多种处理:
·基于所述上行传输得到上行部分CSI,将所述上行部分CSI作为所述目标AI模型的输入信息,得到上行完整CSI;
·基于所述上行传输得到上行部分波束对应的接收信号测量值,将所述上行部分波束对应的接收信号测量值作为所述目标AI模型的输入信息,得到上行完整波束中最优波束对应的SRI;
·将所述装置检测到的调制信号作为所述目标AI模型的输入信息,得到经过非线性补偿的调制信号。
在一些实施例中,处理模块1230,用于监测所述一个或多个AI模型的性能。
在一些实施例中,处理模块1230,用于令第一终端设备以较低的频率发送N个天线端口的SRS,从而获得完整的信道信息,再将获得的信道信息与AI模型推理的结果比较,从而确定AI模型的性能。
在一些实施例中,处理模块1230,用于令第一终端设备以较低的频率发送N个SRS资源,从而获得完整波束集合中的最优波束及对应的RSRP,再将获得的信息与AI模型推理的结果比较,从而确定AI模型的性能。
在一些实施例中,处理模块1230,用于持续监测解调性能。
在一些实施例中,所述发送模块1250,还用于发送第二指示信息,所述第二指示信息用于指示所述第一终端设备更新所述一个或多个AI模型,或者,用于指示所述第一终端设备采用不基于AI模型的传输方式。
在一些实施例中,所述发送模块1250,还用于基于所述一个或多个AI模型的性能监测结果,发送所述第二指示信息。比如,在一个或多个AI模型的性能恶化时,发送第二指示信息。
在一些实施例中,所述接收模块1210,还用于接收所述一个或多个AI模型的更新模型信息。
在一些实施例中,接收模块1210用于执行如下步骤中的一个或多个:步骤620、步骤710、步骤750。
在一些实施例中,发送模块1250用于执行如下步骤中的一个或多个:步骤640、步骤720、步骤740、步骤840、步骤940、步骤1040。
在一些实施例中,处理模块1230用于执行如下步骤中的一个或多个:步骤730、步骤880、步骤980、步骤1080。
接收模块1210、处理模块1230、发送模块1250执行的步骤,请参考图6、图7、图8、图9、图10所示实施例中网络设备执行的一个或多个步骤,前文各个实施例所述的相关内容,同样适用于图12所示装置,此处不再一一赘述。
综上所述,本申请实施例提供的装置,支持使用第一终端设备上报的AI模型的模型信息,且AI模型是在终端设备侧训练得到的或是由本装置根据终端设备上报的模型数据集训练得到的,AI模型与第一终端设备的契合度极高,十分匹配第一终端设备的硬件能力和上行传输需求,本装置使用AI模型后不会对上行传输性能造成负面影响。并且,还有助于减少上行资源开销,降低本装置的处理复杂度。并且,还支持本装置监测AI模型以指示AI模型的更新或停止使用,以便更加灵活适应变化的通信环境、变化的通信需求。
需要说明的是:上述实施例提供的装置在实现其功能时,仅以上述各功能模块的划分进行举例说明,实际应用中,可以根据需要而将上述功能分配由不同的功能模块完成,即将通信设备的内部结构划分成不同的功能模块,以完成以上描述的全部或者部分功能。另外,上述实施例提供的装置与方法实施例属于同一构思。
图13示出了本申请一个示例性实施例提供的基于AI模型的通信设备的结构示意图,包括如下至少之一:接收器1301、发射器1302、处理器1303、存储器1304、总线(图中未示出)。可选的,该通信设备1300用于执行上述第一终端设备所执行的部分或全部步骤。可选的,该通信设备1300用于执行上述网络设备所执行的部分或全部步骤。
其中,接收器1301用于实现接收功能,发射器1302用于实现发送功能。可选地,接收器1301和发射器1302可以实现为一个通信组件,该通信组件可以是一块通信芯片,该通信组件可以称为收发器。可选地,接收器1301和发射器1302可以实现为无线通信组件和/或有线通信组件。可选地,无线通信组件包括无线通信芯片和/或射频天线。可选地,有线通信组件包括有线通信芯片和/或有线接口。
在一些实施例中,接收器1301可用于实现上述接收模块1150的功能和步骤,发射器1302可用于实现上述发送模块1110的功能和步骤。
在一些实施例中,接收器1301可用于实现上述接收模块1210的功能和步骤,发射器1302可用于实现上述发送模块1250的功能和步骤。
处理器1303包括一个或者一个以上处理核心,处理器1303通过运行软件程序以及模块,从而执行各种功能应用以及信息处理。在一些实施例中,处理器1303可用于实现上述处理模块1130或处理模块1230的功能和步骤。
存储器1304可用于存储处理器1303执行的计算机程序,处理器1303用于执行该计算机程序,以实现上述方法实施例中的各个步骤。
此外,存储器1304可以由任何类型的易失性或非易失性存储设备或者它们的组合实现,易失性或非易失性存储设备包括但不限于:磁盘或光盘,电可擦除可编程只读存储器(Electrically-Erasable Programmable Read Only Memory,EEPROM),可擦除可编程只读存储器(Erasable Programmable Read Only Memory,EPROM),静态随时存取存储器(Static Random Access Memory,SRAM),只读存储器(Read-Only Memory,ROM),磁存储器,快闪存储器,可编程只读存储器(Programmable Read-Only Memory,PROM)。
在一些实施例中,存储器1304可以与处理器1303以及接收器1301、发射器1302相连。
在一些实施例中,接收器1301独立进行信号/数据的接收,或处理器1303控制接收器1301进行信号/数据的接收,或处理器1303请求接收器1301进行信号/数据的接收,或处理器1303配合接收器1301进行信号/数据的接收。
在一些实施例中,发射器1302独立进行信号/数据的发送,或处理器1303控制发射器1302进行信号/数据的发送,或处理器1303请求发射器1302进行信号/数据的发送,或处理器1303配合发射器1302进行信号/数据的发送。
对于本实施例中未详细说明的细节,可参见上文实施例,此处不再一一赘述。
在本申请的一个示例性实施例中,还提供了一种芯片,所述芯片包括可编程逻辑电路和/或程序指令,当所述芯片在通信设备上运行时,用于实现上述各个方法实施例提供的基于AI模型的通信方法。
在一些实施例中,所述芯片包括发送模块1110。可选地,所述芯片还包括处理模块1130和/或接收模块1150。相关内容可参考前文所述,此处不再赘述。可选的,各个模块可实现为电路结构。
在一些实施例中,所述芯片包括接收模块1210和发送模块1250。可选地,所述芯片还包括处理模块1230。相关内容可参考前文所述,此处不再赘述。可选的,各个模块可实现为电路结构。
在本申请的一个示例性实施例中,还提供了一种计算机可读存储介质,计算机可读存储介质中存储有至少一段程序,至少一段程序由处理器加载并执行以实现上述各个方法实施例提供的基于AI模型的通信方法。
在本申请的一个示例性实施例中,还提供了一种计算机程序产品,计算机程序产品包括计算机指令,计算机指令存储在计算机可读存储介质中,处理器从计算机可读存储介质中获取计算机指令,处理器执行计算机指令以实现上述各个方法实施例提供的基于AI模型的通信方法。
在本申请的一个示例性实施例中,还提供了一种计算机程序,计算机程序包括计算机指令,计算机指令存储在计算机可读存储介质中,处理器从计算机可读存储介质中获取计算机指令,处理器执行计算机指令以实现上述各个方法实施例提供的基于AI模型的通信方法。
本领域普通技术人员可以理解实现上述实施例的全部或部分步骤可以通过硬件来完成,也可以通过程序来指令相关的硬件完成,程序可以存储于一种计算机可读存储介质中,上述提到的存储介质可以是只读存储器,磁盘或光盘等。
以上仅为本申请的可选实施例,并不用以限制本申请,凡在本申请的精神和原则之内,所作的任何修改、等同替换、改进等,均应包含在本申请的保护范围之内。

Claims (80)

  1. 一种基于AI模型的通信方法,其特征在于,所述方法由第一终端设备执行,所述方法包括:
    上报一个或多个AI模型的模型信息,所述AI模型用于网络设备进行上行CSI处理或上行信号处理;
    根据所述网络设备的调度信息进行上行传输,所述上行传输与所述一个或多个AI模型中的至少一个AI模型关联。
  2. 根据权利要求1所述的方法,其特征在于,所述模型信息包括如下一项或多项:模型数据集,模型结构,模型参数。
  3. 根据权利要求2所述的方法,其特征在于,所述模型信息还包括如下一项或多项:模型功能,模型标识ID,输入信息,输出信息,数据集ID。
  4. 根据权利要求3所述的方法,其特征在于,所述模型功能包括如下一种或多种:上行部分CSI恢复,上行波束信息恢复,上行非线性补偿。
  5. 根据权利要求1至4任一所述的方法,其特征在于,所述一个或多个AI模型由所述第一终端设备训练得到;或者,所述一个或多个AI模型由第二终端设备训练得到,所述第二终端设备与所述第一终端设备采用相同的硬件配置;或者,所述多个AI模型中的一部分AI模型由所述第一终端设备训练得到,所述多个AI模型中的另一部分AI模型由所述第二终端设备训练得到。
  6. 根据权利要求1至5任一所述的方法,其特征在于,所述AI模型用于所述网络设备进行上行CSI处理,且所述上行CSI处理包括上行部分CSI恢复;
    所述AI模型的输入信息包括M个天线端口对应的上行CSI,所述AI模型的输出信息包括N个天线端口对应的上行CSI;其中,M为大于或等于1的整数,N为大于1的整数,M小于N。
  7. 根据权利要求6所述的方法,其特征在于,所述AI模型的输入信息包括所述M个天线端口和K个子带对应的上行CSI,所述AI模型的输出信息包括所述N个天线端口和L个子带对应的上行CSI;
    其中,K为大于或等于1的整数,L为大于1的整数,K小于L。
  8. 根据权利要求6或7所述的方法,其特征在于,所述第一终端设备的天线为部分相干天线,且所述第一终端设备包括相干的h个天线端口组,h为大于1的整数;
    所述M个天线端口包括所述h个天线端口组中每个天线端口组的至少一个天线端口。
  9. 根据权利要求6至8任一所述的方法,其特征在于,所述M个天线端口包括M个SRS端口,所述上行传输包括所述M个SRS端口上的SRS传输。
  10. 根据权利要求1至5任一所述的方法,其特征在于,所述AI模型用于所述网络设备进行上行CSI处理,且所述上行CSI处理包括上行波束信息恢复;
    所述AI模型的输入信息包括M个SRS资源或M个波束对应的接收信号测量值,所述AI模型的输出信息包括至少一个SRI,所述至少一个SRI中的每个SRI用于指示N个SRS资源中的一个SRS资源;其中,M为大于或等于1的整数,N为大于1的整数,M小于N。
  11. 根据权利要求10所述的方法,其特征在于,所述调度信息用于配置所述N个SRS资源,所述N个SRS资源包括所述M个SRS资源,所述上行传输包括所述M个SRS资源上的SRS传输。
  12. 根据权利要求6至11任一所述的方法,其特征在于,在所述AI模型的数量为多个的情况下,不同的AI模型对应不同的取值组合,所述取值组合为N和M的取值组合;或者,所述AI模型的输入信息还包括N和M的取值。
  13. 根据权利要求1至5任一所述的方法,其特征在于,所述AI模型用于所述网络设备进行上行信号处理,所述AI模型的输入信息包括所述网络设备检测到的调制信号,所述AI模型的输出信息包括经过非线性补偿后的调制信号。
  14. 根据权利要求1至13任一所述的方法,其特征在于,所述上行传输与所述一个或多个AI模型中的至少一个AI模型关联,体现在如下一个或多个方面:所述上行传输采用的天线端口,与所述至少一个AI模型的训练数据集中的输入信息所采用的天线端口相同;所述上行传输采用的发送波束,与所述至少一个AI模型的训练数据集中的输入信息所采用的发送波束相同;所述上行传输采用的功率放大器,与所述至少一个AI模型的训练数据集中的输入信息所采用的功率放大器相同。
  15. 根据权利要求1至14任一所述的方法,其特征在于,所述一个或多个AI模型中与所述上行传输关联的AI模型,基于所述调度信息中的第一指示信息确定。
  16. 根据权利要求1至15任一所述的方法,其特征在于,所述方法还包括:
    接收第二指示信息,所述第二指示信息用于指示所述第一终端设备更新所述一个或多个AI模型,或者,用于指示所述第一终端设备采用不基于AI模型的传输方式。
  17. 根据权利要求16所述的方法,其特征在于,所述方法还包括:
    上报所述一个或多个AI模型的更新模型信息。
  18. 一种基于AI模型的通信方法,其特征在于,所述方法由网络设备执行,所述方法包括:
    接收一个或多个AI模型的模型信息,所述AI模型用于所述网络设备进行上行CSI处理或上行信号处理;
    发送调度信息,所述调度信息用于调度第一终端设备进行上行传输,所述上行传输与所述一个或多个AI模型中的至少一个AI模型关联。
  19. 根据权利要求18所述的方法,其特征在于,所述模型信息包括如下一项或多项:模型数据集,模型结构,模型参数。
  20. 根据权利要求19所述的方法,其特征在于,所述模型信息还包括如下一项或多项:模型功能,模型标识ID,输入信息,输出信息,数据集ID。
  21. 根据权利要求20所述的方法,其特征在于,所述模型功能包括如下一种或多种:上行部分CSI恢复,上行波束信息恢复,上行非线性补偿。
  22. 根据权利要求18至21任一所述的方法,其特征在于,所述一个或多个AI模型由所述第一终端设备训练得到;或者,所述一个或多个AI模型由第二终端设备训练得到,所述第二终端设备与所述第一终端设备采用相同的硬件配置;或者,所述多个AI模型中的一部分AI模型由所述第一终端设备训练得到,所述多个AI模型中的另一部分AI模型由所述第二终端设备训练得到。
  23. 根据权利要求18至22任一所述的方法,其特征在于,所述AI模型用于所述网络设备进行上行CSI处理,且所述上行CSI处理包括上行部分CSI恢复;
    所述AI模型的输入信息包括M个天线端口对应的上行CSI,所述AI模型的输出信息包括N个天线端口对应的上行CSI;其中,M为大于或等于1的整数,N为大于1的整数,M小于N。
  24. 根据权利要求23所述的方法,其特征在于,所述AI模型的输入信息包括所述M个天线端口和K个子带对应的上行CSI,所述AI模型的输出信息包括所述N个天线端口和L个子带对应的上行CSI;
    其中,K为大于或等于1的整数,L为大于1的整数,K小于L。
  25. 根据权利要求23或24所述的方法,其特征在于,所述第一终端设备的天线为部分相干天线,且所述第一终端设备包括相干的h个天线端口组,h为大于1的整数;
    所述M个天线端口包括所述h个天线端口组中每个天线端口组的至少一个天线端口。
  26. 根据权利要求23至25任一所述的方法,其特征在于,所述M个天线端口包括M个SRS端口,所述上行传输包括所述M个SRS端口上的SRS传输。
  27. 根据权利要求18至22任一所述的方法,其特征在于,所述AI模型用于所述网络设备进行上行CSI处理,且所述上行CSI处理包括上行波束信息恢复;
    所述AI模型的输入信息包括M个SRS资源或M个波束对应的接收信号测量值,所述AI模型的输出信息包括至少一个SRI,所述至少一个SRI中的每个SRI用于指示N个SRS资源中的一个SRS资源;
    其中,M为大于或等于1的整数,N为大于1的整数,M小于N。
  28. 根据权利要求27所述的方法,其特征在于,所述调度信息用于配置所述N个SRS资源,所述N个SRS资源包括所述M个SRS资源,所述上行传输包括所述M个SRS资源上的SRS传输。
  29. 根据权利要求23至28任一所述的方法,其特征在于,在所述AI模型的数量为多个的情况下,不同的AI模型对应不同的取值组合,所述取值组合为N和M的取值组合;或者,所述AI模型的输入信息还包括N和M的取值。
  30. 根据权利要求18至22任一所述的方法,其特征在于,所述AI模型用于所述网络设备进行上行信号处理,所述AI模型的输入信息包括所述网络设备检测到的调制信号,所述AI模型的输出信息包括经过非线性补偿后的调制信号。
  31. 根据权利要求18至30任一所述的方法,其特征在于,所述上行传输与所述一个或多个AI模型中的至少一个AI模型关联,体现在如下一个或多个方面:所述上行传输采用的天线端口,与所述至少一个AI模型的训练数据集中的输入信息所采用的天线端口相同;所述上行传输采用的发送波束,与所述至少一个AI模型的训练数据集中的输入信息所采用的发送波束相同;所述上行传输采用的功率放大器,与所述至少一个AI模型的训练数据集中的输入信息所采用的功率放大器相同。
  32. 根据权利要求18至31任一所述的方法,其特征在于,所述一个或多个AI模型中与所述上行传输关联的AI模型,基于所述调度信息中的第一指示信息确定。
  33. 根据权利要求18至32任一所述的方法,其特征在于,所述方法还包括:
    基于所述至少一个AI模型中的目标AI模型和所述上行传输,进行上行CSI处理和/或上行信号处理。
  34. 根据权利要求33所述的方法,其特征在于,所述基于所述至少一个AI模型中的目标AI模型和所述上行传输,进行上行CSI处理和/或上行信号处理,包括如下一种或多种处理:
    基于所述上行传输得到上行部分CSI,将所述上行部分CSI作为所述目标AI模型的输入信息,得到 上行完整CSI;
    基于所述上行传输得到上行部分波束对应的接收信号测量值,将所述上行部分波束对应的接收信号测量值作为所述目标AI模型的输入信息,得到上行完整波束中最优波束对应的SRI;
    将所述网络设备检测到的调制信号作为所述目标AI模型的输入信息,得到经过非线性补偿的调制信号。
  35. 根据权利要求18至34任一所述的方法,其特征在于,所述方法还包括:
    发送第二指示信息,所述第二指示信息用于指示所述第一终端设备更新所述一个或多个AI模型,或者,用于指示所述第一终端设备采用不基于AI模型的传输方式。
  36. 根据权利要求35所述的方法,其特征在于,所述发送第二指示信息,包括:
    基于所述一个或多个AI模型的性能监测结果,发送所述第二指示信息。
  37. 根据权利要求35或36所述的方法,其特征在于,所述方法还包括:
    接收所述一个或多个AI模型的更新模型信息。
  38. 一种基于AI模型的通信装置,其特征在于,所述装置包括:
    发送模块,用于上报一个或多个AI模型的模型信息,所述AI模型用于网络设备进行上行CSI处理或上行信号处理;
    所述发送模块,还用于根据所述网络设备的调度信息进行上行传输,所述上行传输与所述一个或多个AI模型中的至少一个AI模型关联。
  39. 根据权利要求38所述的装置,其特征在于,所述模型信息包括如下一项或多项:模型数据集,模型结构,模型参数。
  40. 根据权利要求39所述的装置,其特征在于,所述模型信息还包括如下一项或多项:模型功能,模型标识ID,输入信息,输出信息,数据集ID。
  41. 根据权利要求40所述的装置,其特征在于,所述模型功能包括如下一种或多种:上行部分CSI恢复,上行波束信息恢复,上行非线性补偿。
  42. 根据权利要求38至41任一所述的装置,其特征在于,所述一个或多个AI模型由所述装置训练得到;或者,所述一个或多个AI模型由第二终端设备训练得到,所述第二终端设备与所述装置采用相同的硬件配置;或者,所述多个AI模型中的一部分AI模型由所述装置训练得到,所述多个AI模型中的另一部分AI模型由所述第二终端设备训练得到。
  43. 根据权利要求38至42任一所述的装置,其特征在于,所述AI模型用于所述网络设备进行上行CSI处理,且所述上行CSI处理包括上行部分CSI恢复;
    所述AI模型的输入信息包括M个天线端口对应的上行CSI,所述AI模型的输出信息包括N个天线端口对应的上行CSI;其中,M为大于或等于1的整数,N为大于1的整数,M小于N。
  44. 根据权利要求43所述的装置,其特征在于,所述AI模型的输入信息包括所述M个天线端口和K个子带对应的上行CSI,所述AI模型的输出信息包括所述N个天线端口和L个子带对应的上行CSI;
    其中,K为大于或等于1的整数,L为大于1的整数,K小于L。
  45. 根据权利要求43或44所述的装置,其特征在于,所述装置的天线为部分相干天线,且所述装置包括相干的h个天线端口组,h为大于1的整数;
    所述M个天线端口包括所述h个天线端口组中每个天线端口组的至少一个天线端口。
  46. 根据权利要求43至45任一所述的装置,其特征在于,所述M个天线端口包括M个SRS端口,所述上行传输包括所述M个SRS端口上的SRS传输。
  47. 根据权利要求38至42任一所述的装置,其特征在于,所述AI模型用于所述网络设备进行上行CSI处理,且所述上行CSI处理包括上行波束信息恢复;
    所述AI模型的输入信息包括M个SRS资源或M个波束对应的接收信号测量值,所述AI模型的输出信息包括至少一个SRI,所述至少一个SRI中的每个SRI用于指示N个SRS资源中的一个SRS资源;
    其中,M为大于或等于1的整数,N为大于1的整数,M小于N。
  48. 根据权利要求47所述的装置,其特征在于,所述调度信息用于配置所述N个SRS资源,所述N个SRS资源包括所述M个SRS资源,所述上行传输包括所述M个SRS资源上的SRS传输。
  49. 根据权利要求43至48任一所述的装置,其特征在于,在所述AI模型的数量为多个的情况下,不同的AI模型对应不同的取值组合,所述取值组合为N和M的取值组合;或者,所述AI模型的输入信息还包括N和M的取值。
  50. 根据权利要求38至42任一所述的装置,其特征在于,所述AI模型用于所述网络设备进行上行信号处理,所述AI模型的输入信息包括所述网络设备检测到的调制信号,所述AI模型的输出信息包括经过非线性补偿后的调制信号。
  51. 根据权利要求38至50任一所述的装置,其特征在于,所述上行传输与所述一个或多个AI模型中的至少一个AI模型关联,体现在如下一个或多个方面:所述上行传输采用的天线端口,与所述至少一个AI模型的训练数据集中的输入信息所采用的天线端口相同;所述上行传输采用的发送波束,与所述至少一个AI模型的训练数据集中的输入信息所采用的发送波束相同;所述上行传输采用的功率放大器,与所述至少一个AI模型的训练数据集中的输入信息所采用的功率放大器相同。
  52. 根据权利要求38至51任一所述的装置,其特征在于,所述一个或多个AI模型中与所述上行传输关联的AI模型,基于所述调度信息中的第一指示信息确定。
  53. 根据权利要求38至52任一所述的装置,其特征在于,所述装置还包括:
    接收模块,用于接收第二指示信息,所述第二指示信息用于指示所述装置更新所述一个或多个AI模型,或者,用于指示所述装置采用不基于AI模型的传输方式。
  54. 根据权利要求53所述的装置,其特征在于,
    所述发送模块,还用于上报所述一个或多个AI模型的更新模型信息。
  55. 一种基于AI模型的通信装置,其特征在于,所述装置包括:
    接收模块,用于接收一个或多个AI模型的模型信息,所述AI模型用于所述装置进行上行CSI处理或上行信号处理;
    发送模块,用于发送调度信息,所述调度信息用于调度第一终端设备进行上行传输,所述上行传输与所述一个或多个AI模型中的至少一个AI模型关联。
  56. 根据权利要求55所述的装置,其特征在于,所述模型信息包括如下一项或多项:模型数据集,模型结构,模型参数。
  57. 根据权利要求56所述的装置,其特征在于,所述模型信息还包括如下一项或多项:模型功能,模型标识ID,输入信息,输出信息,数据集ID。
  58. 根据权利要求57所述的装置,其特征在于,所述模型功能包括如下一种或多种:上行部分CSI恢复,上行波束信息恢复,上行非线性补偿。
  59. 根据权利要求55至58任一所述的装置,其特征在于,所述一个或多个AI模型由所述第一终端设备训练得到;或者,所述一个或多个AI模型由第二终端设备训练得到,所述第二终端设备与所述第一终端设备采用相同的硬件配置;或者,所述多个AI模型中的一部分AI模型由所述第一终端设备训练得到,所述多个AI模型中的另一部分AI模型由所述第二终端设备训练得到。
  60. 根据权利要求55至59任一所述的装置,其特征在于,所述AI模型用于所述装置进行上行CSI处理,且所述上行CSI处理包括上行部分CSI恢复;
    所述AI模型的输入信息包括M个天线端口对应的上行CSI,所述AI模型的输出信息包括N个天线端口对应的上行CSI;其中,M为大于或等于1的整数,N为大于1的整数,M小于N。
  61. 根据权利要求60所述的装置,其特征在于,所述AI模型的输入信息包括所述M个天线端口和K个子带对应的上行CSI,所述AI模型的输出信息包括所述N个天线端口和L个子带对应的上行CSI;
    其中,K为大于或等于1的整数,L为大于1的整数,K小于L。
  62. 根据权利要求60或61所述的装置,其特征在于,所述第一终端设备的天线为部分相干天线,且所述第一终端设备包括相干的h个天线端口组,h为大于1的整数;
    所述M个天线端口包括所述h个天线端口组中每个天线端口组的至少一个天线端口。
  63. 根据权利要求60至62任一所述的装置,其特征在于,所述M个天线端口包括M个SRS端口,所述上行传输包括所述M个SRS端口上的SRS传输。
  64. 根据权利要求55至59任一所述的装置,其特征在于,所述AI模型用于所述装置进行上行CSI处理,且所述上行CSI处理包括上行波束信息恢复;
    所述AI模型的输入信息包括M个SRS资源或M个波束对应的接收信号测量值,所述AI模型的输出信息包括至少一个SRI,所述至少一个SRI中的每个SRI用于指示N个SRS资源中的一个SRS资源;
    其中,M为大于或等于1的整数,N为大于1的整数,M小于N。
  65. 根据权利要求64所述的装置,其特征在于,所述调度信息用于配置所述N个SRS资源,所述N个SRS资源包括所述M个SRS资源,所述上行传输包括所述M个SRS资源上的SRS传输。
  66. 根据权利要求60至65任一所述的装置,其特征在于,在所述AI模型的数量为多个的情况下,
    不同的AI模型对应不同的取值组合,所述取值组合为N和M的取值组合;或者,
    所述AI模型的输入信息还包括N和M的取值。
  67. 根据权利要求55至59任一所述的装置,其特征在于,所述AI模型用于所述装置进行上行信号处理,所述AI模型的输入信息包括所述装置检测到的调制信号,所述AI模型的输出信息包括经过非线性补偿后的调制信号。
  68. 根据权利要求55至67任一所述的装置,其特征在于,所述上行传输与所述一个或多个AI模型中的至少一个AI模型关联,体现在如下至少一个方面:所述上行传输采用的天线端口,与所述至少一个AI模型的训练数据集中的输入信息所采用的天线端口相同;所述上行传输采用的发送波束,与所述至少一个AI模型的训练数据集中的输入信息所采用的发送波束相同;所述上行传输采用的功率放大器,与所述至少一个AI模型的训练数据集中的输入信息所采用的功率放大器相同。
  69. 根据权利要求55至68任一所述的装置,其特征在于,所述一个或多个AI模型中与所述上行传输关联的AI模型,基于所述调度信息中的第一指示信息确定。
  70. 根据权利要求55至69任一所述的装置,其特征在于,所述装置还包括:
    处理模块,用于基于所述至少一个AI模型中的目标AI模型和所述上行传输,进行上行CSI处理或上行信号处理。
  71. 根据权利要求70所述的装置,其特征在于,所述处理模块用于执行一种或多种处理:
    基于所述上行传输得到上行部分CSI,将所述上行部分CSI作为所述目标AI模型的输入信息,得到上行完整CSI;
    基于所述上行传输得到上行部分波束对应的接收信号测量值,将所述上行部分波束对应的接收信号测量值作为所述目标AI模型的输入信息,得到上行完整波束中最优波束对应的SRI;
    将所述装置检测到的调制信号作为所述目标AI模型的输入信息,得到经过非线性补偿的调制信号。
  72. 根据权利要求55至71任一所述的装置,其特征在于,
    所述发送模块,还用于发送第二指示信息,所述第二指示信息用于指示所述第一终端设备更新所述一个或多个AI模型,或者,用于指示所述第一终端设备采用不基于AI模型的传输方式。
  73. 根据权利要求72所述的装置,其特征在于,
    所述发送模块,还用于基于所述一个或多个AI模型的性能监测结果,发送所述第二指示信息。
  74. 根据权利要求72或73所述的装置,其特征在于,
    所述接收模块,还用于接收所述一个或多个AI模型的更新模型信息。
  75. 一种基于AI模型的通信设备,其特征在于,所述通信设备包括:处理器;与所述处理器相连的收发器;用于存储所述处理器的可执行指令的存储器;其中,所述处理器被配置为加载并执行所述可执行指令以实现如权利要求1至17任一所述的基于AI模型的通信方法。
  76. 一种基于AI模型的通信设备,其特征在于,所述通信设备包括:处理器;与所述处理器相连的收发器;用于存储所述处理器的可执行指令的存储器;其中,所述处理器被配置为加载并执行所述可执行指令以实现如权利要求18至37任一所述的基于AI模型的通信方法。
  77. 一种计算机可读存储介质,其特征在于,所述计算机可读存储介质中存储有至少一段程序,所述至少一段程序由处理器加载并执行以实现如权利要求1至17任一所述的基于AI模型的通信方法,或如权利要求18至37任一所述的基于AI模型的通信方法。
  78. 一种计算机程序产品,其特征在于,所述计算机程序产品包括计算机指令,所述计算机指令存储在计算机可读存储介质中,处理器从所述计算机可读存储介质中获取所述计算机指令,所述处理器执行所述计算机指令以实现如权利要求1至17任一所述的基于AI模型的通信方法,或如权利要求18至37任一所述的基于AI模型的通信方法。
  79. 一种计算机程序,其特征在于,所述计算机程序包括计算机指令,所述计算机指令存储在计算机可读存储介质中,处理器从所述计算机可读存储介质中获取所述计算机指令,所述处理器执行所述计算机指令以实现如权利要求1至17任一所述的基于AI模型的通信方法,或如权利要求18至37任一所述的基于AI模型的通信方法。
  80. 一种芯片,其特征在于,所述芯片包括可编程逻辑电路和/或至少一段程序,所述芯片用于基于所述可编程逻辑电路和/或所述至少一段程序,实现如权利要求1至17任一所述的基于AI模型的通信方法,或如权利要求18至37任一所述的基于AI模型的通信方法。
PCT/CN2024/098789 2024-06-12 2024-06-12 基于ai模型的通信方法、装置、设备、介质和程序产品 Pending WO2025255752A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2024/098789 WO2025255752A1 (zh) 2024-06-12 2024-06-12 基于ai模型的通信方法、装置、设备、介质和程序产品

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2024/098789 WO2025255752A1 (zh) 2024-06-12 2024-06-12 基于ai模型的通信方法、装置、设备、介质和程序产品

Publications (1)

Publication Number Publication Date
WO2025255752A1 true WO2025255752A1 (zh) 2025-12-18

Family

ID=98050043

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2024/098789 Pending WO2025255752A1 (zh) 2024-06-12 2024-06-12 基于ai模型的通信方法、装置、设备、介质和程序产品

Country Status (1)

Country Link
WO (1) WO2025255752A1 (zh)

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116074813A (zh) * 2021-10-29 2023-05-05 中国电信股份有限公司 无线通信方法及相关设备
WO2024031692A1 (zh) * 2022-08-12 2024-02-15 富士通株式会社 Ai/ml模型的监测方法和装置
CN117676544A (zh) * 2022-08-30 2024-03-08 维沃移动通信有限公司 发送方法、用户设备、网络侧设备及可读存储介质
CN117715106A (zh) * 2022-09-07 2024-03-15 维沃移动通信有限公司 信息传输方法、ai网络模型训练方法、装置和通信设备
CN117858251A (zh) * 2022-09-30 2024-04-09 大唐移动通信设备有限公司 下行波束预测方法、设备、装置及存储介质
CN117858117A (zh) * 2022-09-30 2024-04-09 大唐移动通信设备有限公司 模型信息上报方法、设备、装置及存储介质
CN117998443A (zh) * 2022-11-03 2024-05-07 展讯通信(上海)有限公司 报告传输方法及通信装置

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116074813A (zh) * 2021-10-29 2023-05-05 中国电信股份有限公司 无线通信方法及相关设备
WO2024031692A1 (zh) * 2022-08-12 2024-02-15 富士通株式会社 Ai/ml模型的监测方法和装置
CN117676544A (zh) * 2022-08-30 2024-03-08 维沃移动通信有限公司 发送方法、用户设备、网络侧设备及可读存储介质
CN117715106A (zh) * 2022-09-07 2024-03-15 维沃移动通信有限公司 信息传输方法、ai网络模型训练方法、装置和通信设备
CN117858251A (zh) * 2022-09-30 2024-04-09 大唐移动通信设备有限公司 下行波束预测方法、设备、装置及存储介质
CN117858117A (zh) * 2022-09-30 2024-04-09 大唐移动通信设备有限公司 模型信息上报方法、设备、装置及存储介质
CN117998443A (zh) * 2022-11-03 2024-05-07 展讯通信(上海)有限公司 报告传输方法及通信装置

Similar Documents

Publication Publication Date Title
EP4386620A1 (en) Model test method and apparatus
CN112368950B (zh) 使用深度学习的多用户多输入多输出波束选择和用户配对的方法和装置
US20210345134A1 (en) Handling of machine learning to improve performance of a wireless communications network
US20260025688A1 (en) Channel state information omissions from channel state information reports
US12495324B2 (en) Apparatus and method for reporting CSI in wireless communication system
US20240162957A1 (en) Method and device for reporting csi based on ai model in wireless communication system
WO2025035776A1 (zh) 信道状态信息的发送和接收方法、通信装置、及存储介质
EP4661479A1 (en) Communication method, apparatus, and system
WO2025148327A1 (zh) 信道状态信息的发送和接收方法、通信装置及存储介质
CN118509014A (zh) 一种通信的方法和通信装置
US20250337467A1 (en) Method and device for detecting channel variation, based on ai model in wireless communication system
CN117639852A (zh) 波束参数信息的反馈、接收方法,通信节点及存储介质
WO2025035777A1 (zh) 性能指示的发送和接收方法、通信装置、及存储介质
WO2025255752A1 (zh) 基于ai模型的通信方法、装置、设备、介质和程序产品
US20260058706A1 (en) Antenna location configurations for predictive beam management
WO2024152600A1 (zh) 信息发送方法、接收方法、通信设备和存储介质
JP2026503552A (ja) データセットの送信及び受信方法、装置並びに記憶媒体
JP2026507525A (ja) 端末装置及び方法
US20250016065A1 (en) Server and agent for reporting of computational results during an iterative learning process
US12381789B2 (en) Techniques for reporting correlation metrics for machine learning reproducibility
WO2024255037A1 (en) Communication method and communication apparatus
US20260128961A1 (en) Communication method and communication apparatus
US20260095267A1 (en) Interference prediction events
WO2024255035A1 (en) Communication method and communication apparatus
US20250183950A1 (en) Doppler channel state information (csi) based on a modified slepian basis

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 24942929

Country of ref document: EP

Kind code of ref document: A1