Air interface bilateral AI model performance detection and switching method, device and system

Abstract

The application discloses an air interface double-sided AI model performance detection and switching method, device and system, and belongs to the technical field of wireless communication. In view of the problem of how to dynamically select the optimal model according to the time-varying channel when multiple matched double-sided AI model pairs are deployed between a base station and a terminal, the method provides that the terminal simultaneously reports multiple (including the currently activated) terminal-side model inference results and their pairing identifiers in uplink transmission; the base station performs parallel matching base station-side model inference, compares the comprehensive performance of multiple model pairs, and if a better model pair is found, issues a switching instruction carrying a target pairing identifier to the terminal; and the terminal switches the model accordingly. The method completes parallel detection and evaluation of multiple model performances through single uplink interaction, realizes low-overhead and adaptive dynamic switching of the model, can effectively utilize the AI model to improve the air interface transmission performance, and is especially suitable for 5G-A / 6G networks.

Description

Technical Field

[0001] This application relates to the field of wireless communication technology, and in particular to a method, device and system for air interface bilateral AI model performance detection and switching. Background Technology

[0002] In the development of fifth-generation mobile communication technology (5G) and its evolution (5G-Advanced), and even sixth-generation mobile communication technology (6G), the deep integration of artificial intelligence and machine learning technologies with mobile communication systems is one of the key directions. Utilizing artificial intelligence / machine learning (AI / ML) technologies to improve 5G network performance is a crucial component in realizing the integration of 5G and AI / ML technologies and building 5G-Advanced intelligent dimensions. The 3GPP standards organization initiated standard research for intelligent radio access networks (RANs) starting with Rel-16, and in Rel-18, it initiated a project on AI / ML-based 5G air interface enhancement, thus beginning international standardization work on the integration of 5G air interface and AI / ML. With the further deep integration of 5G-Advanced / 6G mobile communication technologies and AI / ML technologies, more and more air interface functions will apply AI / ML technologies to improve user experience and optimize network performance, resource management, and energy consumption.

[0003] Based on their deployment location within the network, AI / ML models can be categorized into one-sided and two-sided models. One-sided models are deployed only at one end of the communication link, such as on the terminal side or the base station side. Two-sided models, on the other hand, require deployment on both the base station and the terminal side, and they must work together to perform model inference to achieve full functionality. A typical application is AI-based channel state information compression and feedback.

[0004] For this type of bilateral AI model deployment, a crucial issue is ensuring that the models used by the base station and the terminal are mutually compatible, and how to dynamically select the most suitable base station-terminal model pair for the current transmission link in complex and rapidly changing wireless channel environments. To address the model matching problem, relevant standards have introduced the concept of pairing IDs, whose core idea is to associate a pair of co-trained models with a unified identifier. In real-world networks, base stations and terminals may pre-store multiple pairs of matched models, each optimized for a specific wireless channel environment. Therefore, in actual air interface transmission, the wireless channel environment is rapidly changing, requiring performance testing of the currently used model pairs and the ability to select and switch to a more suitable matched model for the current wireless channel environment, thereby ensuring user transmission rates and network performance.

[0005] Currently, for deployment scenarios involving bilateral AI models in 5G evolution or future 6G air traffic, multiple matched model pairs exist on both the base station and terminal sides. When the base station and terminal communicate, if an appropriate bilateral AI model cannot be selected and used based on the current wireless channel environment, it will cause a deterioration in terminal throughput, thus affecting overall network performance. A conservative solution is to revert to traditional non-AI transmission schemes on both the terminal and base station sides. However, this approach wastes the matched model pairs and fails to fully leverage AI technology to improve wireless air interface transmission performance. Therefore, a mechanism is urgently needed that can effectively detect model pair performance and achieve dynamic and smooth switching. Summary of the Invention

[0006] This application proposes a method, device, and system for air interface bilateral AI model performance detection and switching, which solves the problem that when there are multiple bilateral AI model pairs between the base station and the terminal, the optimal model pair cannot be dynamically selected according to the time-varying channel environment. It is particularly suitable for AI-based air interface performance optimization scenarios in 5G-A / 6G networks.

[0007] Firstly, this application proposes a method for performance detection and switching of air interface bilateral AI models. The method includes the following steps: when there are M ≥ 2 matched bilateral AI model pairs, and the currently active model pair is pairingID#K1, bilateral inference is performed. The inference outputs of N terminal-side models and their corresponding pairingIDs are exchanged via uplink signals, where 2 ≤ N ≤ M, and the N terminal-side models include the currently active pairingID#K1 model; based on the inference outputs and pairingIDs of the N terminal-side models, corresponding N network-side models are selected for inference, resulting in the outputs of N bilateral AI model pairs; the outputs of the N bilateral AI model pairs are compared to determine the target model pair to be selected from the N bilateral AI model pairs, and a switching indication carrying the pairingID of the target model pair is exchanged via downlink signals.

[0008] In some embodiments, the uplink signal can be triggered in one of the following ways: active triggering by downlink control signaling; triggering by downlink control signaling in response to a scheduling request; or triggering by periodic indication of configuration information.

[0009] In some embodiments, the value of parameter N can be configured by the network side for the terminal side via radio resource control signaling, or determined by the network side in response to capability information reported by the terminal side.

[0010] In a specific implementation, comparing the outputs of the N bilateral AI model pairs may include: evaluating the performance of each model pair based on the outputs of the N bilateral AI model pairs; and determining whether to switch to the target model pair so that its output performance is higher than that of the pairingID#K1 model pair.

[0011] In some embodiments, the target model carries a handover indication for the pairingID, which may be transmitted via downlink control signaling or radio resource control signaling.

[0012] The method described in any embodiment of the first aspect of this application can be specifically used in a network-side device. The method includes the following steps: receiving an uplink signal containing inference outputs of N terminal-side models and their corresponding pairingIDs, where N ≥ 2 and includes the terminal-side model corresponding to the currently active model pair; selecting the corresponding N network-side models for inference based on the inference outputs and pairingIDs to obtain the final outputs of N bilateral AI model pairs; comparing the final outputs of the N model pairs and determining a target model pair to be selected from the N model pairs based on the comparison result; and sending a switching instruction carrying the pairingID of the target model pair.

[0013] In some embodiments, the method may further include: sending a trigger instruction to trigger the terminal side to execute uplink signals for sending the inference outputs of N terminal-side models and their corresponding pairingIDs. The trigger instruction may be sent via downlink control signaling, or the trigger instruction may be generated in response to a scheduling request from the terminal side.

[0014] In some embodiments, the method may further include: configuring parameter N for the terminal side via radio resource control signaling; or determining parameter N based on capability information reported by the terminal side. The network-side device can evaluate the performance of each model pair based on the outputs of the N bilateral AI model pairs; and determine to switch to a target model pair whose output performance is higher than that of the pairingID#K1 model pair. Preferably, the switching indication is specifically sent via downlink control signaling or radio resource control signaling.

[0015] The method described in any embodiment of the first aspect of this application can also be specifically used in a terminal-side device. The method includes the following steps: simultaneously transmitting the inference outputs of N terminal-side models and their corresponding pairingIDs on uplink transmission resources, wherein N ≥ 2 and includes the terminal-side model corresponding to the currently active model pair; receiving a handover instruction from the network side, the handover instruction carrying the pairingID of the target model pair; and switching to the target model pair and performing the inference task according to the handover instruction.

[0016] In some embodiments, the step of simultaneously sending the inference outputs of N terminal-side models and their corresponding pairingIDs can be executed after receiving a trigger instruction from the network side. The trigger instruction can be received via downlink control signaling; alternatively, the method further includes sending a scheduling request to the network side before receiving the trigger instruction. The parameter N can be configured by the network side via radio resource control signaling, or determined by the capability information reported by the terminal side via capability signaling.

[0017] Secondly, this application also proposes a network-side device for implementing the method described in any one of the first aspects. The network-side device includes: a network receiving module for receiving inference outputs and pairingIDs of N terminal-side models sent by the terminal side; a network determining module for selecting a corresponding network-side model for inference based on the inference outputs and pairingIDs, obtaining model pair outputs, and determining whether to switch and select a target model pair based on the comparison results; and a network sending module for sending a switching instruction carrying the pairingID of the target model pair to the terminal side.

[0018] Thirdly, this application also proposes a terminal-side device for implementing the method described in any one of the first aspects. The terminal-side device includes: a terminal transmitting module for simultaneously transmitting the inference outputs and pairingIDs of N terminal-side models on uplink transmission resources; a terminal receiving module for receiving a handover instruction from the network side; and a terminal determining module for switching to the target model pair according to the handover instruction and performing the inference task.

[0019] Fourthly, this application also proposes a communication device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the method as described in any one of the first aspects of this application.

[0020] Fifthly, this application also proposes a computer-readable storage medium on which a computer program is stored, which, when executed by a processor, implements the steps of the method as described in any one of the first aspects of this application.

[0021] Sixthly, this application also proposes a mobile communication system comprising at least one network-side device as described above, and / or at least one terminal-side device as described above.

[0022] The above-mentioned at least one technical solution adopted in the embodiments of this application can achieve the following beneficial effects: the method can flexibly and dynamically report the model results of multiple matched models, complete the model performance detection, and thus determine whether it is necessary to switch model pairs according to the current wireless channel transmission environment, and select a suitable target model pair to optimize terminal transmission efficiency and improve network performance. Attached Figure Description

[0023] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram of an AI-based compressed feedback process, used to illustrate the basic workflow of a two-sided AI model; Figure 2 This is a schematic diagram of bilateral AI model matching based on pairingID, used to illustrate the model matching mechanism; Figure 3 This is an overall method flowchart of one embodiment of this application; Figure 4 This is a schematic diagram of the process interaction in which the base station actively triggers performance detection and handover in the embodiment. Figure 5 This is a flowchart illustrating an embodiment of the method of this application used in a network-side device; Figure 6 This is a flowchart illustrating an embodiment of the method of this application used in a terminal-side device; Figure 7 This is a schematic diagram of an embodiment of a network-side device; Figure 8 This is a schematic diagram of an embodiment of the terminal-side device; Figure 9 This is a schematic diagram of the structure of a network-side device according to another embodiment of the present invention; Figure 10 This is a block diagram of a terminal-side device according to another embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] Existing technology Figure 1The diagram illustrates the workflow of a bilateral AI model in a typical application—AI-based Channel State Information (CSI) compression and feedback. As shown, this process involves collaboration between the terminal and network (base station) sides. First, the terminal compresses and encodes the measured or sensed channel state information based on its local AI model; this process can be viewed as the inference output of the terminal-side model. Then, the terminal sends the compressed data to the base station via the uplink air interface (e.g., PUSCH). Upon receiving the data, the base station uses its paired base station-side AI model to decode and decompress the data, recovering the channel information; this process represents the inference of the base station-side model. The collaborative work of both models constitutes a complete bilateral AI inference task. A crucial issue for deploying this type of bilateral AI model is ensuring that the models used by the base station and the terminal are mutually compatible, and how to dynamically select the most suitable base station-terminal model pair for the current transmission link in complex and rapidly changing wireless channel environments.

[0026] To address the model matching problem, the concept of pairing ID was introduced in relevant standard discussions. Figure 2 The role of pairing IDs in model matching is explained in detail. As shown in the figure, the core idea is that model matching is guaranteed by the "shared origin" of the models. During the offline or preparation phase, for a specific task (such as CSI compression in a specific scenario), a pair of co-optimized models are generated simultaneously, one deployed on the terminal and the other on the base station, and each pair of models is assigned a unique pairing ID. Therefore, in subsequent deployment and communication, the terminal and the base station only need to confirm that they are using models with the same pairing ID to ensure model matching. They can then jointly deploy for inference without transmitting complex model parameters. A base station and a terminal can store multiple model pairs with different pairing IDs to cope with diverse channel environments. Specifically, each model pair (corresponding to a pairing ID) may be specifically designed for or optimized for a particular wireless channel environment, such as data acquisition and model training under certain typical channel conditions. Therefore, in actual air interface transmission, the wireless channel environment is rapidly changing. It is necessary to test the performance of the model being used and to be able to select and switch to a more suitable matched model for the current wireless channel environment, so as to ensure user transmission rate and network performance.

[0027] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0028] Example 1 Figure 3This is a flowchart illustrating an embodiment of an air-interface bilateral AI model performance detection and switching method. The method is applicable to communication systems comprising network-side devices (such as base stations, access points, distributed units (DUs), or network servers with computing capabilities connected to base stations) and terminal-side devices (such as user equipment (UEs), mobile phones, IoT terminals, or vehicle-to-everything (V2X) terminals).

[0029] Step 110: If there are M ≥ 2 matched bilateral AI model pairs, and the currently active model pair is pairingID#K1, perform bilateral inference.

[0030] In the communication system, the terminal side and the network side pre-store or configure M (M≥2) matched bilateral AI model pairs via signaling, and each model pair is identified by a unique pairing ID.

[0031] Currently, both communicating parties have negotiated and activated the model pair with pairing ID K1 (pairingID#K1) to perform bilateral inference tasks, such as channel prediction, beam management, or data compression. The execution entity can be the terminal's application processor, baseband processor, or dedicated AI accelerator, as well as the baseband processing unit of the base station or the network-side AI server.

[0032] Step 120: Exchange the inference outputs of N terminal-side models and their corresponding pairingIDs through uplink signals, where 2≤N≤M, and the N terminal-side models must include the currently active pairingID#K1 model.

[0033] Uplink transmission resources include, but are not limited to, Uplink Control Channel (PUCCH) transmission resources or Uplink Shared Channel (PUSCH) transmission resources. The N inference outputs of the interaction are generated based on the same input data (e.g., the raw channel measurements at the current moment) to ensure fair comparison. The inference outputs of the aforementioned N terminal-side models are obtained through the same model input. Here, "uplink signal interaction" refers to the process by which the terminal-side device transmits on the uplink physical channel and is received by the network-side device. The transmission behavior can be actively triggered by downlink control signaling (such as DCI) from the network-side device; it can also be initiated by the terminal-side device sending a Scheduling Request (SR), to which the network-side device responds with a DCI containing a trigger instruction; or it can be pre-configured by the network-side device as periodic triggering via Radio Resource Control (RRC) signaling.

[0034] The value of parameter N can be directly configured by the network-side device to the terminal-side device via RRC signaling; alternatively, the network-side device can determine the value of N based on the supported capabilities (e.g., maximum number of parallel models) reported by the terminal-side device via UE capability signaling. The value of parameter N can be confirmed through the following methods, including but not limited to: UE reporting via UE capability signaling (optional step); base station configuration via RRC signaling; and when the configured value is greater than or equal to M. ;otherwise Configuration value. The above N terminal-side models refer to the parts of the N paired bilateral AI model pairs that need to perform inference on the terminal side, and the N terminal-side models include the currently active pairingID#K1 model.

[0035] Step 130: Based on the inference outputs and pairingIDs of the N terminal-side models, select the corresponding N network-side models for inference to obtain the final outputs of the N bilateral AI model pairs.

[0036] Specifically, after receiving the uplink signal, the network-side equipment (such as the receiver and demodulator of the base station) retrieves the matching network-side model from the model library based on the pairingID attached to each output, and uses the model to further process the received inference output, thereby obtaining the complete functional output of the model (terminal-side model + network-side model).

[0037] Step 140: Compare the final outputs of the N bilateral AI model pairs, determine a target model pair from the N model pairs based on the comparison results, and carry the switching indication of the pairingID of the target model pair through downlink signal interaction.

[0038] The base station compares the final outputs of N bilateral AI model pairs to determine whether it needs to switch the currently active model with pairingID#K1. If a switch is required, the base station instructs the UE to do so in one of the following ways, including but not limited to: 1) The base station sends a DCI to the terminal, indicating the pairingID of the target model to the UE through the DCI; 2) Or the base station sends an RRC signaling to the terminal, configuring the pairingID of the target model through the RRC signaling.

[0039] Performance comparison can be performed by the processor of the network-side device. The comparison metrics can include the signal-to-noise ratio (SNR) of the output data, bit error rate (BER), the degree of matching between prediction and actual conditions, or the spectral efficiency gain estimated from these metrics. Specifically, the network-side device evaluates the performance of each of the N bilateral AI model pairs based on their outputs. When the evaluation finds that the performance (e.g., prediction accuracy) of a model pair (whose pairingID is not K1) is higher than that of the currently active pairingID#K1, it is determined that a switch to the superior model pair is needed, and this superior model pair is selected as the target model pair. The handover indication can be transmitted quickly via downlink control signaling (DCI) at the physical layer or MAC layer to achieve low-latency handover; alternatively, it can be configured more stably via higher-layer Radio Resource Control (RRC) signaling.

[0040] It should be noted that the above steps are used for network entities in a wireless communication system, including terminal-side devices, network-side devices, or other intermediate devices; the above steps can also be used for service devices that provide information processing for the network entity devices; the above steps can also be used for any device, system, subsystem, circuit, chip, or software entity that provides information reception, transmission, identification, and processing for terminal-side devices or network-side devices.

[0041] Figure 4 This document illustrates an implementation example of a performance detection and handover process initiated proactively by the network side (base station). The process begins with the base station sending a trigger command to the terminal via Downlink Control Information (DCI) and allocating specific uplink resources. Upon receiving the command, the terminal runs multiple (e.g., N) terminal-side models in parallel on its computing unit (such as a processor or dedicated AI chip) using the same input data, generating inference outputs. Subsequently, the terminal simultaneously sends these N inference outputs and their respective pairing IDs on the uplink resources. Upon receiving these, the base station, on its computing unit, calls the corresponding base station-side model for inference based on each pairing ID, obtaining N complete model pairs as outputs. The base station's decision unit (which may be an algorithm running on the processor) compares the performance (e.g., prediction accuracy, signal quality gain) of these N outputs. If it finds that the performance is better than the currently active model pairing, it instructs the terminal to switch to the model corresponding to the target pairing ID via DCI or RRC signaling. Upon receiving the instruction, the terminal, under the control of its model management module, loads and activates the new model, completing the handover.

[0042] Example 2 Figure 5 This is a flowchart of one embodiment of the method of this application used in a network-side device.

[0043] Step 210: The network-side device receives an uplink signal, which contains the inference outputs of N terminal-side models and their corresponding pairingIDs, where N≥2 and includes the terminal-side model corresponding to the currently active model pair.

[0044] For example, a base station receives uplink transmissions from a terminal via its receiving antenna and receiver. These uplink transmissions contain the inference outputs of N (N≥2) terminal-side AI models simultaneously reported by the terminal, along with a pairingID corresponding to each output. These N models include the terminal-side models in the active model pairs (e.g., pairingID#K1) currently used for communication between the base station and the terminal.

[0045] Step 220: The network-side device selects the corresponding N network-side models for inference based on the inference output and pairingID, and obtains the final output of N bilateral AI model pairs.

[0046] For example, the base station's processing unit (such as a baseband processor or a dedicated AI inference chip) parses the received information. For each reported inference output and its pairingID, the base station's model selection unit, based on the pairingID, calls the precisely matching base station-side AI model from the base station's local model library or a connected model server. Subsequently, using these called base station-side models, the reported inference outputs are further processed, thereby obtaining the final output of N complete bilateral AI model pairs (terminal model + base station model) in parallel. For example, in a CSI feedback scenario, this would be N reconstructed channel matrices.

[0047] Step 230: The network-side device compares the final outputs of the N model pairs and determines the target model pair to be selected from the N model pairs based on the comparison results.

[0048] For example, the base station's performance evaluation and decision-making unit (which may be a software module running on a processor) compares and evaluates the N final output results obtained in step 220. The purpose of the evaluation is to determine whether the currently active model pair is still the best performing among the N candidate model pairs. The evaluation can be based on certain quality metrics of the output results. For example, in a channel information reconstruction scenario, the error between the channel information reconstructed by different model pairs and the reference channel estimated by other means (such as pilots) can be compared. Specifically, the network-side device evaluates the performance of each model pair based on the output of the N bilateral AI model pairs; when it is determined that switching to a target model pair can make the system's output performance (such as throughput, energy efficiency) higher than continuing to use the current pairingID#K1 model pair, then that model pair is determined as the target model pair. If the evaluation finds that the best performing one is not the currently active pairingID#K1, the decision-making unit generates a switching decision and marks the pairingID of the best model pair as the target ID.

[0049] Step 240: The network-side device sends a switching instruction carrying the target model's pairingID.

[0050] For example, the base station's transmitting units (such as the scheduler and transmitter) generate a handover indication message based on the decision in step 230. The core content of this indication message is the pairingID of the target model pair. The base station sends this indication to the terminal via the downlink. Specifically, the pairingID can be encoded in a downlink control message (DCI) and quickly transmitted via the physical downlink control channel (PDCCH); or it can be included in a radio resource control (RRC) reconfiguration message and transmitted via the downlink shared channel (PDSCH) to achieve a more stable configuration update.

[0051] Furthermore, in this embodiment, the base station can also proactively initiate a performance testing process. Specifically, the base station can send a specific trigger command (e.g., a newly defined DCI format) to the terminal through its scheduling and transmission module to command the terminal to perform the multi-model result reporting behavior in step 120 of embodiment one. The trigger command can be sent directly via DCI; or, the generation and transmission of the trigger command can be a response made by the scheduler on the base station side after receiving a scheduling request (SR) from the terminal. The base station can also pre-configure the number N of reporting models for the terminal through its RRC signaling configuration module; or, the base station's capability management module can determine and notify the terminal of the N value to be reported this time based on the UE capability information previously reported by the terminal (e.g., the maximum number of concurrent AI model inferences supported by the terminal).

[0052] Example 3 Figure 6This is a flowchart of one embodiment of the method of this application used in a terminal-side device.

[0053] Step 310: The terminal-side device simultaneously sends the inference outputs of N terminal-side models and their corresponding pairingIDs on the uplink transmission resources.

[0054] Wherein, N≥2 and includes the terminal-side model corresponding to the currently active model pair.

[0055] Under the control of its uplink transmission module, the terminal simultaneously transmits the inference outputs of N (N≥2) terminal-side AI models and their corresponding pairingIDs on designated resources of the uplink control channel (PUCCH) or uplink shared channel (PUSCH). These N models are selected by the terminal from its local model library and must include the terminal-side models in the model pairs currently active in communication with the base station (such as the model corresponding to pairingID#K1). The number of models N reported by the terminal can be pre-configured to the terminal by the base station's RRC signaling, and the terminal's behavior control module reads and executes this configuration; alternatively, the value of N can also be indirectly determined by the capability information previously reported to the base station by the terminal's capability reporting module, and the terminal executes according to the agreement with the base station.

[0056] Step 320: The terminal device receives a handover instruction from the network side, the handover instruction carrying the pairingID of the target model pair.

[0057] For example, the terminal's downlink receiving module continuously listens to the downlink control channel from the base station. After sending a multi-model report, the terminal receives a handover indication from the base station. The terminal parses this indication and extracts the pairingID of the target model pair carried within it. This indication may be a DCI, parsed by the terminal's physical layer control module; or it may be an RRC signaling, parsed by the terminal's RRC layer processing module.

[0058] Step 330: According to the switching instruction, switch to the target model pair and perform the inference task.

[0059] For example, upon receiving the switching instruction in step 320, the terminal's model management module immediately executes a model switching operation. Based on the target pairingID in the instruction, this module loads the corresponding terminal-side AI model from its model storage area (such as memory or a dedicated model cache) and sets it to an active state, replacing the original pairingID#K1 model. Subsequently, all subsequent bilateral AI inference tasks on the terminal (such as new CSI compression) will be executed using the newly activated model.

[0060] After receiving the handover instruction as described in step 320 (i.e., DCI or RRC signaling from the network side carrying the pairingID of the target model), the terminal switches to the target model and performs the inference task.

[0061] In this embodiment, the timing of the terminal executing step 310 (multi-model reporting) can be controlled. Specifically, the behavior of simultaneously sending multiple model inference outputs can be initiated by the terminal's transmission behavior control module only after receiving an explicit trigger instruction (such as a specific DCI) from the base station. The trigger instruction can be received and decoded from the PDCCH by the terminal's physical layer receiving module; or, in another mode, the terminal's scheduling request module can first send an SR to the base station to request uplink resources, and then wait for and receive the DCI containing the trigger instruction from the base station.

[0062] Example 4 Figure 7 This is a schematic diagram of a module of one embodiment of a network-side device. This device is used to implement the method described in Embodiment 2 above.

[0063] To implement the above technical solution, this application proposes a network-side device 400, which includes a network transmitting module 401, a network determining module 402, and a network receiving module 403 that are interconnected.

[0064] The network receiving module 403 is used to implement the function of step 210. Specifically, it is responsible for receiving wireless signals from the terminal-side device through the wireless interface, and performing demodulation, decoding and other processing on the signals to extract the inference outputs of N terminal-side models reported by the terminal at the same time and the corresponding pairingID information.

[0065] The network determination module 402 implements the functions of steps S220 and S230. It further includes a model selection submodule and a performance evaluation decision submodule. The model selection submodule selects N corresponding network-side models from the local model library based on the pairingID provided by the network receiving module 403. The performance evaluation decision submodule is responsible for driving these selected models to perform inference, obtaining the outputs of N model pairs, comparing the performance of these outputs, and ultimately determining whether switching is necessary and which model pair to select as the target.

[0066] The network transmission module 401, which implements the function of step 240, is responsible for generating and sending a handover indication carrying the target model's pairingID. This module encodes the decision result into appropriate signaling (such as DCI or RRC messages) and sends it to the terminal via the downlink of the radio interface.

[0067] The specific methods for implementing the functions of the network sending module, network determining module, and network receiving module are as described in the various method embodiments of this application, and will not be repeated here.

[0068] The network-side equipment described in this application may refer to base station facilities, network-side equipment or servers connected to base stations, systems that provide services for the aforementioned equipment, or any system, subsystem, module, circuit, chip or software operating device that provides information reception, transmission, identification and processing for the aforementioned equipment.

[0069] Example 5 Figure 8 This is a schematic diagram of a module of one embodiment of a terminal-side device. This device is used to implement the method described in Embodiment 3 above. To implement the above technical solution, this application proposes a terminal-side device 500, which includes a terminal transmitting module 501, a terminal determining module 502, and a terminal receiving module 503 connected to each other.

[0070] The terminal transmission module 501 is used to implement the function of step 310. Driven by the behavior control logic, this module multiplexes and encodes the inference outputs of multiple terminal-side models and their pairingIDs, and transmits them through a power amplifier and other radio frequency front-ends on the specified uplink time-frequency resources.

[0071] The terminal receiving module 503 is used to implement the function of step 320. It is responsible for receiving downlink wireless signals through the antenna and receiver, processing the signals to decode the handover indication sent by the base station, and extracting the target pairingID.

[0072] The terminal determination module 502 is used to implement the function of step 330. This module is essentially a model management and execution unit. It receives the target pairingID from the terminal receiving module 503, then controls the terminal to load the corresponding terminal-side AI model from the storage unit, deploys it to the execution unit (such as the NPU), and updates the system configuration so that subsequent AI inference tasks use the new model, thus completing the switch. The specific methods for implementing the functions of the terminal sending module, terminal determination module, and terminal receiving module are as described in the various method embodiments of this application, and will not be repeated here.

[0073] The terminal-side equipment described in this application may refer to user equipment (UE), personal mobile terminal, smart terminal, mobile phone, computer with communication function, system that provides services for the above-mentioned equipment, or any system, subsystem, module, circuit, chip or software running device that provides information reception, transmission, identification and processing for the above-mentioned equipment.

[0074] Example 6 Figure 9A schematic diagram of a network-side device according to another embodiment of the present invention is shown. As shown, the network-side device 600 includes a processor 601, a wireless interface 602, and a memory 603. The wireless interface may consist of multiple components, including a transmitter and a receiver, providing a unit for communication with various other devices over a transmission medium. The wireless interface implements communication functions with the terminal-side device, processes wireless signals through receiving and transmitting devices, and the data carried by the signals is communicated with the memory or processor via an internal bus structure. The memory 603 contains a computer program that executes any embodiment of this application, and the computer program runs or modifies the processor 601. When the memory, processor, and wireless interface circuit are connected through a bus system, the bus system includes a data bus, a power bus, a control bus, and a status signal bus, which will not be described in detail here.

[0075] Example 7 Figure 10 This is a block diagram of a terminal-side device according to another embodiment of the present invention. The terminal-side device 700 includes at least one processor 701, a memory 702, a user interface 703, and at least one wireless network interface 704. The various components in the terminal-side device 700 are coupled together via a bus system. The bus system is used to enable communication between these components. The bus system includes a data bus, a power bus, a control bus, and a status signal bus.

[0076] User interface 703 may include a display, keyboard, or clicking device, such as a mouse, trackball, touchpad, or touchscreen.

[0077] The memory 702 stores executable modules or data structures. The memory may store an operating system and application programs. The operating system includes various system programs, such as a framework layer, core library layer, and driver layer, used to implement various basic business functions and handle hardware-based tasks. The application programs include various applications, such as media players and browsers, used to implement various application functions.

[0078] In an embodiment of the present invention, the memory 702 contains a computer program that executes any embodiment of the present application, the computer program being run on or modified by the processor 701.

[0079] The memory 702 includes a computer-readable storage medium. The processor 701 reads the information in the memory 702 and, in conjunction with its hardware, completes the steps of the above-described method. Specifically, the computer-readable storage medium stores a computer program, which, when executed by the processor 701, implements the steps of the method embodiments described in any of the above embodiments.

[0080] Processors 601 and 701 may be integrated circuit chips with signal processing capabilities. In implementation, each step of the method in this application can be completed by integrated logic circuits in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, digital signal processor, application-specific integrated circuit, off-the-shelf programmable gate array or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of this invention can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor.

[0081] Other embodiments Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. In a typical configuration, the device of this application includes one or more processors (CPUs), an input / output user interface, a network interface, and memory.

[0082] Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0083] Therefore, this application also proposes a computer-readable medium storing a computer program that, when executed by a processor, implements the steps of the method described in any embodiment of this application. For example, the memory 603, 702 of the present invention may include non-permanent memory in the form of computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM.

[0084] Based on the embodiments of the above-described apparatus in this application, this application also proposes a mobile communication system, including at least one embodiment of any terminal-side device in this application and / or at least one embodiment of any network-side device in this application.

[0085] It should be noted that the specific mobile communication technology described in this invention is not limited, and can be WCDMA, CDMA2000, TD-SCDMA, WiMAX, LTE / LTE-A, LAA, MuLTEfire, 5G NR, and the sixth-generation and Nth-generation mobile communication technologies that may appear in the future.

[0086] The terminal described in this invention refers to a terminal-side product that can support the communication protocols of terrestrial mobile communication systems, and a specially designed wireless modem module that can be integrated into various types of terminal forms such as mobile phones, tablets, and data cards to complete communication functions.

[0087] For ease of description, a fifth-generation mobile communication system is used as an example, where the mobile communication terminal can be represented as UE (User Equipment), and the network-side access equipment can be represented as a base station or access point.

[0088] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0089] Those skilled in the art will understand that, unless otherwise stated, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be understood that when a device or component is “connected” to another device or component, it may be directly connected to the other device or component, or there may be an intermediary device or component. Furthermore, the term “connection” as used herein may include partially wireless connections as well as partially wired connections.

[0090] In the description of this application, it should be understood that the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances. Furthermore, in the description of this application, unless otherwise stated, "multiple" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship.

[0091] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for performance testing and switching of an air-to-ground bilateral AI model, characterized in that, Includes the following steps: If there are M ≥ 2 matched bilateral AI model pairs, and the currently active model pair is pairingID#K1, perform bilateral inference; The inference outputs of N terminal-side models and their corresponding pairingIDs are exchanged through uplink signals, where 2≤N≤M, and the N terminal-side models include the currently active pairingID#K1 model. Based on the inference outputs and pairingIDs of the N terminal-side models, select the corresponding N network-side models for inference to obtain the outputs of N bilateral AI model pairs; The outputs of the N bilateral AI model pairs are compared to determine the target model pair to be selected from the N bilateral AI model pairs. The switching indication of the pairingID of the target model pair is carried through downlink signal interaction.

2. The method as described in claim 1, characterized in that, The uplink signal is triggered in one of the following ways: Downlink control signaling is actively triggered; Downlink control signaling is triggered in response to a scheduling request; Periodic indicators of configuration information are triggered.

3. The method as described in claim 1 or 2, characterized in that, The value of parameter N is configured by the network side for the terminal side through radio resource control signaling, or determined by the network side in response to the capability information reported by the terminal side.

4. The method as described in claim 1, characterized in that, include: Evaluate the performance of each model pair based on the outputs of the N bilateral AI model pairs; Determine whether to switch to the target model pair so that the output performance is higher than that of the pairingID#K1 model pair.

5. The method as described in claim 1, characterized in that, The target model carries a handover indication for the pairingID, specifically transmitted via downlink control signaling or radio resource control signaling.

6. A method for performance testing and switching of an air-interface bilateral AI model, used in network-side equipment, characterized in that, Includes the following steps: Receive uplink signals, which contain the inference outputs of N terminal-side models and their corresponding pairingIDs, where N≥2 and includes the terminal-side model corresponding to the currently active model pair; Based on the inference output and pairingID, select the corresponding N network-side models for inference to obtain the final output of N bilateral AI model pairs; Compare the final outputs of the N model pairs, and determine the target model pair to select from the N model pairs based on the comparison results; Send a switching instruction carrying the target model's pairingID.

7. The method as described in claim 6, characterized in that, Also includes: Send a trigger command to trigger the terminal side to execute the uplink signal of the inference output of N terminal side models and their corresponding pairingIDs.

8. The method as described in claim 7, characterized in that, The triggering instruction is sent via downlink control signaling, or the triggering instruction is generated in response to a scheduling request on the terminal side.

9. The method as described in claim 6, characterized in that, The method further includes: Parameter N is configured on the terminal side via radio resource control signaling; or, parameter N is determined based on the capability information reported by the terminal side.

10. The method as described in claim 6, characterized in that, The network testing device evaluates the performance of each model pair based on the output of the N bilateral AI model pairs; and determines to switch to the target model pair so that its output performance is higher than that of the pairingID#K1 model pair.

11. The method as described in claim 6, characterized in that, The handover instruction is specifically sent via downlink control signaling or radio resource control signaling.

12. A method for performance detection and switching of an air-to-ground bilateral AI model, used in a terminal-side device, characterized in that, Includes the following steps: On the uplink transmission resources, the inference outputs of N terminal-side models and their corresponding pairingIDs are sent simultaneously, where N≥2 and include the terminal-side models corresponding to the currently active model pair. Receive a handover instruction from the network side, the handover instruction carrying the pairingID of the target model pair; According to the switching instruction, switch to the target model pair and perform the inference task.

13. The method as described in claim 12, characterized in that, The step of simultaneously sending the inference outputs of N terminal-side models and their corresponding pairingIDs is executed after receiving a trigger command from the network side.

14. The method as described in claim 13, characterized in that, The triggering instruction is received via downlink control signaling; or, the method further includes sending a scheduling request to the network side before receiving the triggering instruction.

15. The method as described in claim 12, characterized in that, The parameter N is determined by or from the capability information reported by the terminal side via capability signaling.

16. A network-side device for implementing the method according to any one of claims 1-5 or 6-11, characterized in that, The network-side device includes: The network receiving module is used to receive the inference outputs and pairingIDs of N terminal-side models sent by the terminal side; The network determination module is used to select the corresponding network-side model for inference based on the inference output and pairingID, obtain the model pair output, and determine whether to switch and select the target model pair based on the comparison results. The network sending module is used to send a switching instruction carrying the target model's pairingID to the terminal side.

17. A terminal-side device for implementing the method according to any one of claims 1-5 or 12-15, characterized in that, The terminal-side device includes: The terminal sending module is used to simultaneously send the inference outputs and pairingIDs of N terminal-side models on uplink transmission resources; The terminal receiving module is used to receive handover instructions from the network side; The terminal determination module is used to switch to the target model pair and perform inference tasks according to the switching instruction.

18. A communication device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the method as described in any one of claims 1 to 15.

19. A computer-readable storage medium storing a computer program thereon, the computer program, when executed by a processor, implementing the steps of the method as claimed in any one of claims 1 to 15.

20. A mobile communication system comprising at least one network-side device as claimed in claim 16, and / or at least one terminal-side device as claimed in claim 17.

Images