Model switching method and apparatus, network device, terminal, storage medium, and computer program product

Abstract

The present disclosure provides a model switching method and apparatus, a network device, a terminal, a storage medium, and a computer program product. The method comprises: a network device receiving first information sent by a terminal, the first information at least indicating a first model to be switched; and sending second information to the terminal, the second information indicating a transmission configuration of the terminal or indicating a reason for declining model switching.

Description

Model switching methods, devices, network equipment, terminals, storage media, and computer program products

[0001] Cross-references to related applications

[0002] This disclosure claims priority to Chinese Patent Application No. 202411897425.7, filed in China on December 20, 2024, the entire contents of which are incorporated herein by reference. Technical Field

[0003] This disclosure relates to the field of wireless technology, and in particular to a model switching method, apparatus, network device, terminal, storage medium, and computer program product. Background Technology

[0004] In bilateral models where terminals interact with the network, changes in the input or performance metrics of the bilateral model—that is, changes in the usage environment—need model switching to prevent performance degradation caused by significant changes in the usage environment. However, the specific methods for model switching are currently unclear, making it impossible to perform model switching effectively. Summary of the Invention

[0005] To address the related technical issues, this disclosure provides a model switching method, apparatus, network device, terminal, storage medium, and computer program product.

[0006] The technical solution of this disclosure embodiment is implemented as follows:

[0007] This disclosure provides a model switching method applied to a network device, the method comprising:

[0008] The receiving terminal sends first information, which at least indicates the first model to be switched;

[0009] Send a second message to the terminal, the second message indicating the terminal's transmission configuration or indicating the reason for disagreeing with the switching model.

[0010] In the above scheme, the first information includes one or more of the following:

[0011] The identifier of the first model;

[0012] Channel similarity between two adjacent time points.

[0013] In the above scheme, the second information includes one or more of the following:

[0014] The transmission resources corresponding to the first model;

[0015] The transmission resources corresponding to the second model, where the second model represents the currently used model or the model before the switch;

[0016] The transmission resources corresponding to the time window are used for the terminal to feed back channel data. Within the time window, the first model and the second model are called simultaneously. The second model represents the currently used model or the model before the switch.

[0017] A first identifier indicates whether new data can be replaced by historical channel data or cannot be replaced by historical channel data.

[0018] The second identifier indicates whether uplink resources are insufficient or sufficient.

[0019] The third identifier indicates that the first model and the second model are invoked simultaneously, and the second model represents the currently used model or the model before the switch.

[0020] The method in the above scheme further includes:

[0021] The third information is determined based on the first information; the third information at least indicates the model switching type of the first model;

[0022] Based on the third information, the second information is determined.

[0023] In the above scheme, the third information includes one or more of the following:

[0024] The model switching type;

[0025] Available uplink resources;

[0026] Channel similarity between two adjacent time points;

[0027] The time window is pre-configured or determined based on the feedback period of the model invoked by the network device and the number of historical channel data that need to be input into the model.

[0028] The method in the above scheme further includes:

[0029] The model switching type is determined based at least on the feature information of the input data of the first model; wherein the feature information represents whether it is temporal or non-temporal.

[0030] In the above scheme, when the feature information representation is temporal and different input data are encoded by different encoders included in the first model, the model switching type is either a first switching type or a second switching type; or,

[0031] When the feature information representation is non-temporally sequential, the model switching type is the third switching type; or,

[0032] When the feature information representation is temporal and different input data are encoded by the same encoder included in the first model, the model switching type is the third switching type.

[0033] In the above scheme, the model switching type includes one or more of the following:

[0034] The first handover type is where the terminal and the network device simultaneously run the first model and the second model, and / or simultaneously transmit two different sets of channel data output by the two models on the uplink channel; the second model represents the currently used model or the model before the handover.

[0035] The second switching type is that after the network device receives the number of historical channel data that need to be input into the first model, it stops calling the second model and runs the first model; the second model represents the currently used model or the model before the switch.

[0036] The third switching type is to stop calling the second model and then run the first model; the second model represents the currently used model or the model before the switch.

[0037] The method in the above scheme further includes:

[0038] At least based on available uplink resources, determine whether uplink resources are insufficient or sufficient.

[0039] The method in the above scheme further includes:

[0040] Based on a time window and / or the channel similarity between two adjacent moments, it is determined whether new data can be replaced by historical channel data or not; the time window is pre-configured or determined based on the feedback period of the model invoked by the network device and the number of historical channel data that need to be input into the model.

[0041] The method in the above scheme further includes:

[0042] If the time window is less than the first threshold and the channel similarity between two adjacent times is greater than the second threshold, then it is determined that historical channel data can be used to replace the new data.

[0043] The method in the above scheme further includes:

[0044] If one or more of the following conditions are met, it is determined that historical channel data cannot be used to replace new data:

[0045] The time window is greater than or equal to the first threshold;

[0046] The channel similarity between two adjacent time points is less than or equal to the second threshold.

[0047] The method in the above scheme further includes:

[0048] Receive fourth information; the fourth information includes one or more of the following:

[0049] The model's encoder;

[0050] The encoder identifier of the model;

[0051] The model's decoder;

[0052] The identifier of the model's decoder;

[0053] The number of historical channel data points that the model needs to input;

[0054] Model switching type;

[0055] The amount of feedback data from the model of the terminal.

[0056] This disclosure also provides a model switching method applied to a terminal, the method comprising:

[0057] Send first information to the network device, wherein the first information at least indicates the first model to be switched;

[0058] The terminal receives second information sent by the network device, the second information indicating the transmission configuration of the terminal or indicating the reason for disagreeing on the switching model.

[0059] In the above scheme, the first information includes one or more of the following:

[0060] The identifier of the first model;

[0061] Channel similarity between two adjacent time points.

[0062] In the above scheme, the second information includes one or more of the following:

[0063] The transmission resources corresponding to the first model;

[0064] The transmission resources corresponding to the second model, where the second model represents the currently used model or the model before the switch;

[0065] The transmission resources corresponding to the time window are used for the terminal to feed back channel data. Within the time window, the first model and the second model are called simultaneously. The second model represents the currently used model or the model before the switch.

[0066] A first identifier indicates whether new data can be replaced by historical channel data or cannot be replaced by historical channel data.

[0067] The second identifier indicates whether uplink resources are insufficient or sufficient.

[0068] The third identifier indicates that the first model and the second model are invoked simultaneously, and the second model represents the currently used model or the model before the switch.

[0069] The method in the above scheme further includes:

[0070] Receive fifth information; the fifth information includes one or more of the following:

[0071] The model's encoder;

[0072] The encoder identifier of the model;

[0073] The number of historical channel data points required to input the model;

[0074] The inference complexity or computational complexity of the model.

[0075] The method in the above scheme further includes:

[0076] The first model is determined based on one or more of the following:

[0077] The processing power of the terminal;

[0078] The inference or computational cost of the model;

[0079] The accuracy of the model;

[0080] The amount of feedback data from the model;

[0081] The model's input data can be either time-series or non-time-series.

[0082] This disclosure also provides a model switching device, the device comprising:

[0083] The first receiving module is used to receive first information sent by the terminal, wherein the first information at least indicates the first model to be switched;

[0084] A first sending module is used to send second information to the terminal, the second information indicating the terminal's transmission configuration or indicating the reason for disagreeing on switching modes.

[0085] This disclosure also provides a model switching device, the device comprising:

[0086] The second sending module is used to send first information to the network device, wherein the first information at least indicates the first model to be switched;

[0087] The second receiving module is used to receive second information sent by the network device, the second information indicating the transmission configuration of the terminal or indicating the reason for disagreeing on the switching model.

[0088] This disclosure also provides a network device, including: a first processor and a first communication interface; wherein,

[0089] The first communication interface is used to receive first information sent by the terminal and send second information to the terminal; the first information at least indicates a first model to be switched; the second information indicates the transmission configuration of the terminal or indicates the reason for disagreeing on switching models.

[0090] This disclosure also provides a terminal, including: a second processor and a second communication interface; wherein,

[0091] The second communication interface is used to send first information to the network device and receive second information sent by the network device; the first information at least indicates a first model to be switched; the second information indicates the transmission configuration of the terminal or indicates the reason for disagreeing with the model switching.

[0092] This disclosure also provides a network device, including a first processor and a first memory for storing computer programs capable of running on the first processor.

[0093] Wherein, when the first processor is used to run the computer program, it executes the steps of any of the methods described above on the network device side.

[0094] This disclosure also provides a terminal, including a second processor and a second memory for storing computer programs capable of running on the second processor.

[0095] Wherein, when the second processor is running the computer program, it executes the steps of any of the methods described above on the terminal side.

[0096] This disclosure also provides a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of any of the methods described above on the network device side, or implements the steps of any of the methods described above on the terminal side.

[0097] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the methods described above on the network device side, or implements the steps of any of the methods described above on the terminal side.

[0098] In the model switching method, apparatus, network device, terminal, storage medium, and computer program product provided in this disclosure, the terminal selects a first model to be switched to and sends first information to the network device to at least indicate the first model to be switched to; the network device receives the first information sent by the terminal and decides whether to agree to switch to the first model. When agreeing to switch to the first model, the network device indicates the terminal's transmission configuration through second information to realize the model switching. When disagreeing to switch to the first model, the network device indicates the reason for disagreeing to switch the model through second information. This not only clarifies the way the terminal and the network device perform model switching, but also avoids resource waste caused by unnecessary model switching. Attached Figure Description

[0099] Figure 1 is a schematic flowchart of a model switching method according to an embodiment of this disclosure;

[0100] Figure 2 is a schematic diagram of the first switching type according to an embodiment of this disclosure;

[0101] Figure 3 is a schematic flowchart of another model switching method according to an embodiment of this disclosure;

[0102] Figure 4 is a schematic diagram of the channel compression feedback model according to an embodiment of this disclosure;

[0103] Figure 5 is a schematic diagram of the model switching process in an embodiment of this disclosure;

[0104] Figure 6 is a schematic diagram of the interaction between the terminal and the network device according to an embodiment of this disclosure;

[0105] Figure 7 is a schematic diagram of a model switching device according to an embodiment of the present disclosure;

[0106] Figure 8 is a schematic diagram of another model switching device according to an embodiment of the present disclosure;

[0107] Figure 9 is a schematic diagram of the network device structure according to an embodiment of this disclosure;

[0108] Figure 10 is a schematic diagram of the terminal structure according to an embodiment of this disclosure. Detailed Implementation

[0109] Two-sided models based on Artificial Intelligence (AI) include time-based channel compression feedback models and time-independent channel compression feedback models. A two-sided model refers to a model group in which the network-side and terminal models work together to complete one or more functions. Examples of time-based channel compression feedback models include Long Short-Term Memory (LSTM) networks and Transformer models. Examples of time-independent channel compression feedback models include Multilayer Perceptron (MLP) models. The advantages of time-based channel compression feedback models are: they can adapt to the time-varying characteristics of the channel, improve the feedback efficiency of channel state information, and reduce feedback overhead while ensuring feedback accuracy.

[0110] AI-based two-sided models, due to their limited generalization capabilities, are often not applicable to all scenarios, and model switching may occur when the scenario changes. Furthermore, changes in terminal processing power (such as processing capacity, uplink bandwidth, RANK, and condition number) can also lead to model switching. For example, limited terminal processing power often restricts the model's inference computation, and with the same feedback quantity, the accuracy of a two-sided model is often positively correlated with the inference computation quantity. In this case, model switching will tend to choose a two-sided model with lower accuracy. Conversely, limited uplink bandwidth restricts the model's feedback quantity, such as the number of bits. In this case, model switching will tend to choose a model with a smaller feedback quantity, but smaller feedback quantities often result in lower model accuracy. When terminal processing power and / or uplink bandwidth change from limited to unlimited, a switch to a model with a larger inference computation and / or feedback quantity can be made.

[0111] Currently, the main approach is to monitor the inputs or performance metrics of the bilateral model to detect changes in its operating environment (e.g., channel environment). When significant changes in the operating environment lead to a performance degradation in the bilateral model, a model switch is deemed necessary. However, the specific methods for model switching are currently unclear, preventing effective model switching.

[0112] For example, existing solutions do not consider that time-based channel compression feedback models require historical channel data to function. Specifically, the channel data transmission process in a time-based channel compression feedback model involves compressing the channel data at the terminal using an encoder, transmitting the compressed channel data to the network side, and then decompressing the compressed channel data at the network side using a decoder. When a model switch occurs, because the encoder used to compress the channel data at the terminal is different, even if the network side stores historical channel data, it cannot meet the input requirements of the new network-side model, thus preventing a successful model switch. Historical channel data can be understood as channel data from historical moments.

[0113] Based on this, in various embodiments of this disclosure, the terminal selects a first model to be switched to and sends first information to the network device to at least indicate the first model to be switched to; the network device receives the first information sent by the terminal and decides whether to agree or disagree to switch to the first model. When agreeing to switch to the first model, the network device indicates the terminal's transmission configuration through second information to realize the model switch. When disagreeing to switch to the first model, the network device indicates the reason for disagreeing to switch the model through second information. This not only clarifies the way the terminal and the network device perform model switching, but also avoids resource waste caused by unnecessary model switching.

[0114] The present disclosure will now be described in further detail with reference to the accompanying drawings and embodiments.

[0115] This disclosure provides a model switching method applied to a network device located on the network side, including a base station. As shown in Figure 1, the method includes:

[0116] Step 101: Receive the first information sent by the terminal.

[0117] The first information indicates at least the first model to be switched.

[0118] Here, the network device receives the first message sent by the terminal, indicating that the terminal has initiated a model switch, and a model switch is required between the network device and the terminal. For example, the terminal initiates a model switch and sends the first message to the network device when its processing power and / or uplink resources change. The first model to be switched to can be determined or selected by the terminal; for example, if the terminal is currently using a second model, the terminal determines or selects the first model to be switched to based on the terminal's processing power, and / or the model's inference or computational load, and / or the model's accuracy, and / or the amount of feedback data from the model, and / or whether the model's input data is sequential or non-sequential, in order to request a switch from the second model to the first model. The first and second models are bilateral models, meaning that the first and second models need to be deployed on both the terminal and the network device.

[0119] The network device performs model switching or determines or decides the reason for disagreeing with the model switching based at least on the first information; if the network device agrees to perform model switching, the network device sends the terminal's transmission configuration to the terminal to realize the model switching.

[0120] In one embodiment, the first information includes one or more of the following:

[0121] The identifier of the first model;

[0122] Channel similarity between two adjacent time points.

[0123] Here, the identifier of the first model is used to indicate the first model to be switched. The form of the identifier of the first model is not limited. For example, the identifier of the first model can be the model number of the first model. Two adjacent times can be the time when the terminal initiates the model switch and the previous time, for example, time t-1 and time t. The channel similarity between two adjacent times represents the channel similarity between two adjacent times of the currently used model or the model before the switch.

[0124] The channel similarity between two adjacent time points can be calculated using cosine similarity. For example, the channel similarity between two adjacent time points can be expressed as:

[0125] Where t represents the moment when the terminal initiates model switching, t-1 represents the moment before the moment when the terminal initiates model switching, S(t-1,t) represents the channel similarity between two adjacent moments, H(t) represents the channel characteristics, channel data, or channel status between the network device and the terminal at moment t, H(t-1) represents the channel characteristics, channel data, or channel status between the network device and the terminal at moment t-1, and ||| represents the 2-norm. H(t) and H(t-1) are two channel matrices stretched into vectors with columns as units. Channel stationarity is indicated when the channel similarity between two adjacent moments is greater than the second threshold.

[0126] When a terminal initiates a model switch, the first information includes the identifier of the first model, or it includes the identifier of the first model and the channel similarity between two adjacent time points.

[0127] When the terminal cannot run two models simultaneously and / or the network device indicates insufficient uplink resources to the terminal, the first information includes the channel similarity between two adjacent time points, or, it includes the identifier of the first model and the channel similarity between two adjacent time points. Insufficient uplink resources can be understood as the inability to simultaneously transmit two different sets of channel data output by two models on the uplink channel.

[0128] In this embodiment, the content of the first information is clearly defined so that the network device can decide whether to agree to the model switch, thereby improving the efficiency or accuracy of the model switch.

[0129] Step 102: Send the second information to the terminal.

[0130] The second information indicates the terminal's transmission configuration or the reason for disagreeing on the switching model.

[0131] Here, the second information is determined at least based on the first information. The terminal's transmission configuration is configured when the network device determines or agrees to perform a model switch. For example, if the network device cannot run two models simultaneously and / or the network device determines that it cannot simultaneously transmit two different sets of channel data output by the two models on the uplink channel, and the channel similarity at two adjacent times is less than or equal to a second threshold, the second information indicates the reason for disagreeing with the model switch. In this case, the reason for disagreeing with the model switch may include: the network device cannot run two models simultaneously and the first model on the network device side cannot replace new data with historical channel data, and / or, the network device does not support the terminal simultaneously transmitting two different sets of channel data output by the two models on the uplink channel and the first model on the network device side cannot replace new data with historical channel data.

[0132] In one embodiment, the second information includes one or more of the following:

[0133] The transmission resources corresponding to the first model;

[0134] The transmission resources corresponding to the second model, where the second model represents the currently used model or the model before the switch;

[0135] The transmission resources corresponding to the time window are used for the terminal to feed back channel data. Within the time window, the first model and the second model are called simultaneously. The second model represents the currently used model or the model before the switch.

[0136] A first identifier indicates whether new data can be replaced by historical channel data or cannot be replaced by historical channel data.

[0137] The second identifier indicates whether uplink resources are insufficient or sufficient.

[0138] The third identifier indicates that the first model and the second model are invoked simultaneously, and the second model represents the currently used model or the model before the switch.

[0139] Here, the transmission resources corresponding to the time window include the transmission resources corresponding to the first model and the second model within the time window.

[0140] When the network device determines or agrees to perform a model switch, the second information indicates the terminal's transmission configuration, including one or more of the following: transmission resources corresponding to the first model, transmission resources corresponding to the second model, transmission resources corresponding to the time window, a first identifier, a second identifier, and a third identifier. It should be noted that when both the first and second models are invoked simultaneously, the second information must include at least the transmission resources corresponding to the time window and / or the third identifier.

[0141] For example, if the input data of the first model is sequential, and different input data are encoded by different encoders contained in the first model, and the uplink resources are sufficient and the network device determines to switch models, the second information includes at least one or more of the following: the transmission resources corresponding to the first model, the transmission resources corresponding to the second model, the transmission resources corresponding to the time window, the first identifier, the second identifier indicating sufficient uplink resources, and the third identifier; the sequential nature of the input data indicates that the input data has a time order, that is, different input data need to be input into the model in the order in which the input data is generated or obtained.

[0142] When the input data of the first model is sequential and different input data are encoded by different encoders contained in the first model, and when uplink resources are insufficient and the network device determines to switch models, the second information includes at least one or more of the following: transmission resources corresponding to the first model, transmission resources corresponding to the second model, a first identifier indicating that historical channel data can be used to replace new data, and a second identifier indicating that uplink resources are insufficient.

[0143] When the input data of the first model is non-temporally ordered, or when the input data of the first model is temporally ordered and different input data are encoded by the same encoder included in the first model, and when the network device determines to switch models, the second information includes at least one or more of the following: the transmission resources corresponding to the first model, and the second identifier indicating that the uplink resources are sufficient; the input data being non-temporally ordered indicates that the input data does not have a time sequence.

[0144] Here, if the network device does not agree to switch models, the second information indicates the reason for not agreeing to switch models, including at least one or more of the following: a first identifier and a second identifier.

[0145] For example, if the input data of the first model is non-sequential, or if the input data of the first model is sequential and different input data are encoded by the same encoder included in the first model, and if the network device determines that it does not agree to switch models, the second information shall at least include a second identifier indicating insufficient uplink resources.

[0146] When the input data of the first model is time-series, different input data are encoded by different encoders contained in the first model, and the network device determines that it does not agree to switch models, the second information includes at least one or more of the following: the first identifier indicates that the new data cannot be replaced by historical channel data, and the second identifier indicates that the uplink resources are insufficient.

[0147] In this embodiment, the content of the second information is clearly defined to further facilitate model switching. When the model switching is not agreed upon, the second information indicates the reason for disagreeing with the model switching, thus avoiding unnecessary waste of resources caused by model switching.

[0148] Upon receiving the first information from the terminal, the network device needs to determine the second information. In one embodiment, the model switching method further includes:

[0149] The third information is determined based on the first information; the third information at least indicates the model switching type of the first model;

[0150] Based on the third information, the second information is determined.

[0151] Here, the third information is determined based on the first information. This can be based on the identifier of the first model included in the first information, such as the model switching type of the first model included in the third information; or it can be determined after receiving the first information, such as the third information also including available uplink resources.

[0152] For example, based on the identifier of the first model, it is determined that the input data of the first model has temporal sequence, and different input data are encoded by different encoders contained in the first model, the model switching type of the first model is either the first switching type or the second switching type.

[0153] Based on the identifier of the first model, if it is determined that the input data of the first model is non-temporally sequential, or if it is determined that the input data of the first model is temporally sequential and different input data are encoded by the same encoder contained in the first model, the model switching type of the first model is the third switching type.

[0154] The second piece of information can be different or the same depending on the model switching type of the first model. The model switching type can also be described as the switching type.

[0155] When the model switching type of the first model is either the first switching type or the second switching type, the second information includes one or more of the following: the transmission resources corresponding to the first model, the transmission resources corresponding to the second model, the transmission resources corresponding to the time window, the first identifier, the second identifier, and the third identifier;

[0156] When the model switching type of the first model is the third switching type, the second information includes at least one or more of the following: the transmission resources corresponding to the first model and the second identifier.

[0157] In this embodiment, the third information, namely the model switching type of the first model, is determined based on the first information. Based on the model switching type of the first model, the second information is determined. This clarifies that the model switching types of different first models may be different, and the second information may also be different depending on the model switching type of the first model, which can help improve the accuracy of the second information.

[0158] In one embodiment, the third information includes one or more of the following:

[0159] The model switching type;

[0160] Available uplink resources;

[0161] Channel similarity between two adjacent time points;

[0162] The time window is pre-configured or determined based on the feedback period of the model invoked by the network device and the number of historical channel data that need to be input into the model.

[0163] Here, the time window is determined based on the feedback period of the second model invoked by the network device and the number of historical channel data points required to input into the first model. This can be achieved by multiplying the feedback period of the second model invoked by the network device and the number of historical channel data points required to input into the first model; or by multiplying the feedback period of the second model invoked by the network device, the number of historical channel data points required to input into the first model, and a correction coefficient. The correction coefficient can be set according to actual needs. It is worth noting that, with sufficient computing resources on the terminal, the model deployed on the terminal can complete one invocation of the model within one feedback period; when the number of historical channel data points required to input into the model is L... i In this case, the sequence length of the model's input data is L. i + 1 The model's feedback cycle can be specified by the protocol or set based on actual needs.

[0164] The third information determined based on the information included in the first information includes one or more of the following: model switching type, channel similarity between two adjacent time points;

[0165] After receiving the first information, the third information determined by the network device includes one or more of the following: available uplink resources and time windows.

[0166] In this embodiment, the content of the third information is clearly defined in order to more accurately determine the second information.

[0167] In one embodiment, the model switching method further includes:

[0168] The model switching type is determined based at least on the feature information of the input data of the first model; wherein the feature information represents whether it is temporal or non-temporal.

[0169] Here, when determining the third information based on the first information, the model switching type can be determined at least based on the feature information of the input data of the first model. For example, if the feature information of the input data of the first model represents a temporal sequence, the model switching type of the first model can be determined as a first switching type, a second switching type, or a third switching type.

[0170] When the feature information representation of the input data of the first model is non-temporally sequential, the model switching type of the first model is determined to be the third switching type.

[0171] In this embodiment, determining the model switching type based on the feature information of the input data of the first model can improve the efficiency and accuracy of determining the model switching type of the first model.

[0172] In one embodiment, when the feature information representation is temporal and different input data are encoded by different encoders included in the first model, the model switching type is a first switching type or a second switching type; or,

[0173] When the feature information representation is non-temporally sequential, the model switching type is the third switching type; or,

[0174] When the feature information representation is temporal and different input data are encoded by the same encoder included in the first model, the model switching type is the third switching type.

[0175] Here, different input data are encoded by different encoders included in the first model. This can be understood as one input data point corresponding to one encoder, and the output data of a predetermined number of adjacent encoders with temporal sequence are decoded by the same decoder included in the first model.

[0176] It is worth noting that the first, second, and third switching types are applicable to all bilateral models, i.e., the first model.

[0177] In this embodiment, the characteristics of the first model corresponding to the three model switching types are clearly defined, which further improves the efficiency of determining the model switching type of the first model.

[0178] In one embodiment, the model switching type includes one or more of the following:

[0179] The first handover type is where the terminal and the network device simultaneously run the first model and the second model, and / or simultaneously transmit two different sets of channel data output by the two models on the uplink channel; the second model represents the currently used model or the model before the handover.

[0180] The second switching type is that after the network device receives the number of historical channel data that need to be input into the first model, it stops calling the second model and runs the first model; the second model represents the currently used model or the model before the switch.

[0181] The third switching type is to stop calling the second model and then run the first model; the second model represents the currently used model or the model before the switch.

[0182] Here, a schematic diagram of the first switching type is shown in Figure 2. In Figure 2, the network side corresponds to the network device, E1 represents the encoder of the second model, D1 represents the decoder of the second model, E2 represents the encoder of the first model, and D2 represents the decoder of the first model. It is easy to see that Figure 2 shows the model before switching, during switching, and after switching. It is worth noting that the model switching is performed within a time window.

[0183] It should be noted that the first handover type can be understood as soft handover, and the third handover type can be understood as hard handover. The second handover type, as a supplement to the first handover type, is determined for the first model where the characteristic information representation of the input data has temporal sequence, and different input data are encoded by different encoders contained in the first model. When the terminal and network equipment cannot run the first model and the second model simultaneously, and / or uplink resources are insufficient, and historical channel data can be used to replace new data, the model handover type of the first model is determined to be the second handover type.

[0184] In this embodiment, the model switching method corresponding to each model switching type is specified to help achieve model switching.

[0185] In one embodiment, the model switching method further includes:

[0186] At least based on available uplink resources, determine whether uplink resources are insufficient or sufficient.

[0187] Here, network devices can determine whether uplink resources are insufficient or sufficient based on available uplink resources; alternatively, they can determine whether uplink resources are insufficient or sufficient based on the identifier of the first model and the available uplink resources.

[0188] Based on available uplink resources, determine whether uplink resources are insufficient or sufficient, including:

[0189] If the available uplink resources are greater than or equal to the third threshold, then the uplink resources are deemed sufficient.

[0190] If the available uplink resources are less than the third threshold, it is determined that the uplink resources are insufficient; the third threshold can be set according to actual needs.

[0191] Based on the identifier of the first model and the available uplink resources, determine whether the uplink resources are insufficient or sufficient, including one or more of the following:

[0192] If the available uplink resources support the terminal to simultaneously transmit two different sets of channel data output by the first model and the second model in the uplink channel, then the uplink resources are determined to be sufficient; or, if the available uplink resources meet the resource requirements for simultaneously transmitting two different sets of channel data output by the first model and the second model, then the uplink resources are determined to be sufficient.

[0193] If the available uplink resources support the terminal in transmitting the channel data output by the first model in the uplink channel, it is determined that the uplink resources are sufficient.

[0194] If the available uplink resources do not support the terminal in simultaneously transmitting two different sets of channel data output by the first and second models on the uplink channel, it is determined that the uplink resources are insufficient.

[0195] If the available uplink resources do not support the terminal in transmitting the channel data output by the first model in the uplink channel, it is determined that the uplink resources are insufficient.

[0196] In this embodiment, a method is defined to determine whether uplink resources are insufficient or sufficient (i.e., the second identifier in the second information) based at least on available uplink resources (i.e., the third information), thereby helping to achieve model switching.

[0197] In one embodiment, the model switching method further includes:

[0198] Based on a time window and / or the channel similarity between two adjacent moments, it is determined whether new data can be replaced by historical channel data or not; the time window is pre-configured or determined based on the feedback period of the model invoked by the network device and the number of historical channel data that need to be input into the model.

[0199] Here, based on the channel similarity between two adjacent moments transmitted by the terminal, if both the time window and the channel similarity between the two adjacent moments meet the conditions, it is determined that historical channel data can be used to replace new data; if the time window and / or the channel similarity between the two adjacent moments meet the conditions, it is determined that historical channel data cannot be used to replace new data.

[0200] It is worth noting that the time window can be pre-configured, representing a certain stability of the channel within the time window. Based on the feedback period and time window of the model invoked by the network device, the maximum number of historical channel data that can be input into the model can be determined. The maximum number of historical channel data that can be input into the model is compared with the number of historical channel data that needs to be input into the model. The comparison result is used to determine whether historical channel data can replace new data or not. The number can also be described as the number of records.

[0201] Based on the comparison results and / or the channel similarity between two adjacent time points, it can be determined whether historical channel data can replace new data or not. Specifically, if the maximum number of historical channel data that can be input into the model is greater than or equal to the number of historical channel data that needs to be input into the model, and the channel similarity between two adjacent time points is greater than a second threshold, then historical channel data can be used to replace new data. If the maximum number of historical channel data that can be input into the model is less than the number of historical channel data that needs to be input into the model, and / or the channel similarity between two adjacent time points is less than or equal to the second threshold, then historical channel data cannot be used to replace new data. The second threshold can be set according to actual needs.

[0202] In this embodiment, the third information used to determine whether new data can be replaced by historical channel data or cannot be replaced by historical channel data is clearly defined, thereby improving the efficiency of determining the first identifier in the second information.

[0203] In one embodiment, the model switching method further includes:

[0204] If the time window is less than the first threshold and the channel similarity between two adjacent times is greater than the second threshold, then it is determined that historical channel data can be used to replace the new data.

[0205] Here, the first and second thresholds can be set according to actual needs. The first threshold can be understood as a time threshold, and the second threshold can be understood as a channel stability threshold. The role of the first threshold is to ensure that even if the channel is stable, it can only maintain stability within a limited time period, thus enabling the channel to be reused. Setting the first threshold can limit the accumulation of errors that may be caused by a stable channel.

[0206] In this embodiment, clearly defining the conditions under which historical channel data can be used to replace new data can help improve the accuracy of the second information.

[0207] In one embodiment, the model switching method further includes:

[0208] If one or more of the following conditions are met, it is determined that historical channel data cannot be used to replace new data:

[0209] The time window is greater than or equal to the first threshold;

[0210] The channel similarity between two adjacent time points is less than or equal to the second threshold.

[0211] In this embodiment, the conditions under which historical channel data cannot be used to replace new data are clearly defined, thereby helping to improve the accuracy of the second information.

[0212] To ensure the normal operation of the bilateral model on network devices, corresponding information needs to be configured for the network devices. In one embodiment, the model switching method further includes:

[0213] Receive fourth information; the fourth information includes one or more of the following:

[0214] The model's encoder;

[0215] The encoder identifier of the model;

[0216] The model's decoder;

[0217] The identifier of the model's decoder;

[0218] The number of historical channel data points that the model needs to input;

[0219] Model switching type;

[0220] The amount of feedback data from the model of the terminal.

[0221] Here, the network device receives the fourth message sent by the Radio Access Network (RAN) AI model training system. The RAN AI model training system can be deployed on the network device's Central Unit (CU) and / or Distributed Unit (DU) and / or logical entities across CUs. The network device may include one CU and one or more DUs.

[0222] The encoder identifier of the model is used to indicate the encoder of the model. The form of the encoder identifier is not limited. For example, the encoder identifier could be the model number. The decoder identifier of the model is used to indicate the decoder of the model. The form of the decoder identifier is not limited. For example, the decoder identifier could be the model number. The encoder of the model can be understood as a channel compression model, and the decoder of the model can be understood as a channel decompression model.

[0223] When the input data of the bilateral model running on the network device is time-series, the fourth information includes at least the number of historical channel data that the model needs to input. The number of historical channel data that the model needs to input can be understood as the number of historical channel data that the model on the network device side needs to input.

[0224] The model switching type of a model can be understood as the model switching type to which the model is applicable.

[0225] It should be understood that the model in the fourth piece of information can be the currently used model or the model before the switch, or it can include both the currently used model and the model after the switch.

[0226] In this embodiment, the network device receives the fourth piece of information to ensure that the model can operate normally.

[0227] Correspondingly, this disclosure also provides a model switching method applied to a terminal, as shown in Figure 3, the method including:

[0228] Step 301: Send the first message to the network device.

[0229] The first information indicates at least the first model to be switched.

[0230] Step 302: Receive the second information sent by the network device.

[0231] The second information indicates the terminal's transmission configuration or the reason for disagreeing on the switching model.

[0232] Here, the terminal side corresponds to the network device side, and the repeated content will not be repeated here.

[0233] In this embodiment, the terminal selects a first model to be switched to and sends first information to the network device to at least indicate the first model to be switched to; the network device receives the first information sent by the terminal and decides whether to agree or disagree to switch to the first model. When agreeing to switch to the first model, the network device indicates the terminal's transmission configuration through second information to realize the model switch. When disagreeing to switch to the first model, the network device indicates the reason for disagreeing to switch the model through second information. This not only clarifies the way the terminal and the network device switch models, but also avoids resource waste caused by unnecessary model switching.

[0234] In one embodiment, the first information includes one or more of the following:

[0235] The identifier of the first model;

[0236] Channel similarity between two adjacent time points.

[0237] In this embodiment, the content of the first information is clearly defined so that the network device can decide whether to agree to the model switch, thereby improving the efficiency or accuracy of the model switch.

[0238] In one embodiment, the second information includes one or more of the following:

[0239] The transmission resources corresponding to the first model;

[0240] The transmission resources corresponding to the second model, where the second model represents the currently used model or the model before the switch;

[0241] The transmission resources corresponding to the time window are used for the terminal to feed back channel data. Within the time window, the first model and the second model are called simultaneously. The second model represents the currently used model or the model before the switch.

[0242] A first identifier indicates whether new data can be replaced by historical channel data or cannot be replaced by historical channel data.

[0243] The second identifier indicates whether uplink resources are insufficient or sufficient.

[0244] The third identifier indicates that the first model and the second model are invoked simultaneously, and the second model represents the currently used model or the model before the switch.

[0245] In this embodiment, the content of the second information is clearly defined to further facilitate model switching. When the model switching is not agreed upon, the second information indicates the reason for disagreeing with the model switching, thus avoiding unnecessary waste of resources caused by model switching.

[0246] To ensure the bilateral model functions correctly on the terminal, corresponding information needs to be configured for the terminal. In one embodiment, the model switching method further includes:

[0247] Receive fifth information; the fifth information includes one or more of the following:

[0248] The model's encoder;

[0249] The encoder identifier of the model;

[0250] The number of historical channel data points required to input the model;

[0251] The inference complexity or computational complexity of the model.

[0252] Here, the terminal receives the fifth message sent by the RAN AI model training system. The RAN AI model training system can be deployed on the CU and / or DU and / or logical entities across CUs of a network device. A network device may include one CU and one or more DUs.

[0253] The number of historical channel data that needs to be input into the model can be understood as the number of historical channel data that the terminal side needs to input into the network device side.

[0254] The inference complexity or computational complexity of a model is proportional to the amount of feedback data it generates.

[0255] It is worth noting that the fifth piece of information may also include the model's decoder and / or the identifier of the model's decoder.

[0256] In this embodiment, the terminal receives the fifth piece of information to ensure that the model can operate normally.

[0257] When the terminal's processing power and / or uplink resources change, the terminal initiates a model switch, requiring the determination of the first model to be switched to. Based on this, in one embodiment, the model switching method further includes:

[0258] The first model is determined based on one or more of the following:

[0259] The processing power of the terminal;

[0260] The inference or computational cost of the model;

[0261] The accuracy of the model;

[0262] The amount of feedback data from the model;

[0263] The model's input data can be either time-series or non-time-series.

[0264] Here, the inference or computational load of the model needs to be adapted to the processing capabilities of the terminal.

[0265] It is worth noting that the greater the inference or computational cost of the model, and / or the larger the amount of feedback data, and / or the more time-series the input data, the higher the accuracy of the corresponding model usually is.

[0266] In this embodiment, the reference information of the first model is clearly determined so that when switching between different models, the triangular balance of evaluation indicators can be achieved, which takes into account the capabilities (selecting models with high accuracy), costs (considering terminal performance, i.e., the terminal's processing power and limited uplink resources), and quality (ensuring the integrity and robustness of model functions through model switching).

[0267] The present disclosure will now be described in further detail with reference to application examples.

[0268] The two-sided model for network equipment and terminal operation includes the channel compression feedback model, which can be divided into three types: a, b, and c, as shown in Figure 4. In Figure 4, type a represents a time-series-based channel compression feedback model, where the encoder receives historical channel data and the decoder does not uniformly feed back. For example, the input data is time-series, and different input data are encoded by different encoders included in the first model. Type b represents a time-series-based channel compression feedback model, where the encoder receives historical channel data and uniformly feeds back to the decoder. For example, the input data is time-series, and different input data are encoded by the same encoder included in the first model. Type c represents a non-time-series-based channel compression feedback model, such as a first model where the input data is not time-series. In Figure 4, the network equipment is a base station, E. i Characterizing the encoder, D i Characterization decoder.

[0269] The terminal-side data for type a model includes time-series input data V. t-2 V t-1 V t and V t+1 The input data is encoded by the encoder on the terminal side model, and the encoder's output data is transmitted to the base station via the air interface. The base station's decoder then decodes the encoder's output data. The base station's output data includes V' t and V' t+1 Each input data corresponds to an encoder, and the output data of every three adjacent encoders corresponds to a decoder.

[0270] The terminal-side data for the type b model includes time-series input data V. t-2 V t-1 V t and V t+1 The input data is encoded by the encoder on the terminal side model, and the encoder's output data is transmitted to the base station via the air interface. The base station's decoder then decodes the encoder's output data. The base station's output data includes V' t and V' t+1 In this system, every three adjacent input data points correspond to one encoder, and the output data of one encoder corresponds to one decoder.

[0271] The terminal-side data for the C-type model includes V tThe input data is encoded by the encoder of the terminal-side model, and the encoder output data is transmitted to the air interface of the base station. The encoder output data is decoded by the decoder of the base station. The base station output data includes V't.

[0272] It should be understood that Figure 4 is merely an example and does not constitute a limitation on the number of input data for the bilateral model.

[0273] When the first model is Class B or Class C, the terminal sends first information to the network device. This first information includes the identifier of the first model, such as its model number. If the network device determines that uplink resources are sufficient, it sends second information to the terminal. This second information includes the transmission resources corresponding to the first model. If the network device determines that uplink resources are insufficient, it sends second information to the terminal, including a second identifier indicating insufficient uplink resources. The terminal then re-determines the first model. It should be understood that when the first model is Class B, the model on the network device side still requires historical channel data, but the network device does not need to allocate additional uplink transmission resources to the terminal. When the first model is Class A, since model data transmission is at the Transmission Time Interval (TTI) level, to ensure normal model operation, the network device needs to allocate additional uplink transmission resources and design time windows for the terminal to receive the number of historical channel data required for input to the first model. It is worth noting that the number of time windows is the same as the number of models that need to switch to the Class A first model; the number can also be described as a count.

[0274] For the first model where the input data is time-series, a storage queue can be designed. The queue length is equal to the length of the longest input sequence among all time-series-based channel compression feedback models, denoted as max{L}. i}, to pre-store historical channel data. The terminal's storage queue is used to store the most recent max{L} data from the current time. i The historical channel data refers to the data encoded by the encoder; the storage queue designed for network devices is used to store the most recent max{L} data. i The historical channel data refers to the data after being decoded by the decoder.

[0275] Taking the first model to be switched as class a as an example, the model switching process is shown in Figure 5, including:

[0276] Step 1: The terminal determines that the first model to be switched is class a.

[0277] Step 2: The terminal determines whether it can run the first model and the second model simultaneously. If the terminal can run the first model and the second model simultaneously, proceed to Step 3; if the terminal cannot run the first model and the second model simultaneously, proceed to Step 7.

[0278] The second model represents the model currently in use or the model before the switch.

[0279] Step 3: The terminal sends the first information to the network device, which includes the model number of the first model.

[0280] Step 4: The network device determines whether there are sufficient uplink resources, i.e., whether the uplink resources are sufficient or insufficient. If the uplink resources are sufficient, proceed to Step 5; if the uplink resources are insufficient, proceed to Step 6.

[0281] Step 5: The network device sends a second message to the terminal to enable the network device and the terminal to switch models.

[0282] Specifically, the second information includes the transmission resources corresponding to the first model, the transmission resources corresponding to the second model, the transmission resources corresponding to the time window, and the third identifier.

[0283] Step 6: The network device sends a second message to the terminal, which includes a second identifier indicating insufficient uplink resources.

[0284] Step 7: The terminal calculates the channel similarity between two adjacent time points and sends the first information to the network device, which includes the channel similarity between the two adjacent time points; and the network device determines the time window.

[0285] Step 8: The network device determines whether the time window is less than the first threshold and whether the channel similarity between two adjacent times is greater than the second threshold. If the time window is less than the first threshold and the channel similarity between two adjacent times is greater than the second threshold, proceed to step 9. If the time window is greater than or equal to the first threshold and / or the channel similarity between two adjacent times is less than or equal to the second threshold, proceed to step 10.

[0286] Step 9: The network device determines that historical channel data can be used to replace new data. After receiving the number of historical channel data that need to be input into the first model, it stops calling the second model and sends second information to the terminal. The second information includes the transmission resources corresponding to the first model, so that the network device and the terminal can run the first model.

[0287] Step 10: The network device determines that historical channel data cannot be used to replace new data, and sends second information to the terminal. The second information includes a first identifier indicating that historical channel data cannot be used to replace new data. The terminal re-determines the first model, or if no first model meets the requirements, the terminal determines to switch to a traditional method, such as the etypeII method, and sends the first information or the method number of the traditional method to the network device. The first information includes the model number of the re-determined first model.

[0288] Step 11: The network device determines whether the model to be switched to is Class A, First Model. If the model to be switched to is Class A, return to Step 1; if the model to be switched to is not the First Model or the First Model is not Class A, for example, if the First Model is Class B or Class C, proceed to Step 12.

[0289] Step 12: Switch the model to class b or class c using the traditional method or the first model.

[0290] To further illustrate the model switching method provided in this disclosure, Figure 6 shows a schematic diagram of the interaction between the terminal and the network device, including the following steps:

[0291] Step 1: The RAN AI model training system pre-collects downlink channel data with different channel characteristics required for air interface AI model training. The encoder for offline training of the AI-based bilateral model can be denoted as E. i The decoder can be denoted as D. i The system sends a fifth message to the terminal and a fourth message to the network device. The RAN AI model training system is deployed on the CU and / or DU and / or logical entities across CUs of the network device. The RAN AI model training system can set the identifier of the time-frequency resources of the original model (i.e., the transmission resources corresponding to the second model) to 0, and the identifier of the time-frequency resources of the new model (i.e., the transmission resources corresponding to the first model) to 1. It is worth noting that this identifier (the third identifier) ​​can be set to only appear when both the original and new models are called simultaneously. It can also set the identifier of sufficient uplink resources (the second identifier) ​​to 0, and the identifier of insufficient uplink resources (the second identifier) ​​to 1. It is worth noting that this identifier can be set to only be fed back when uplink resources are insufficient. Furthermore, it can set the identifier of available historical channel data to replace new data (the first identifier) ​​to 0, and the identifier of inability to replace new data with historical channel data (the first identifier) ​​to 1. All identifier bits can be set using 1 bit data. The RAN AI model training system also sets a first threshold δ2 and a second threshold δ1.

[0292] It should be noted that the first model can be understood as the new model, and the second model can be understood as the original model.

[0293] Step 2: The terminal starts the encoder of the second model according to the initial configuration, connects to the network, and enters the Radio Resource Control Connected (RRC-CONNECTED) state.

[0294] Step 3: The network device starts the decoder of the second model according to the initial configuration, allocates time and frequency resources to the terminal for feedback channel data, and sends the Channel State Information-Reference Signal (CSI-RS).

[0295] Step 4: The terminal compresses and feeds back the corresponding channel information based on CSI-RS.

[0296] Step 5: The network device receives and decompresses the channel information.

[0297] Step 6: When the terminal's processing power changes, a model switch is required. The terminal determines the model number of the first model to be switched to based on its processing power, and / or the model's inference or computational load, and / or the model's accuracy, and / or the amount of feedback data from the model, and / or whether the model's input data is sequential or non-sequential. The terminal sends the first information to the network device.

[0298] Step 7: The network device determines the third information based on the first information, and then determines the second information based on the third information. If the network device determines to perform a model switch, it sends one or more of the first and second identifiers from the second information to the terminal via Media Access Control (MAC) signaling. It also sends one or more of the transmission resources corresponding to the first model, the transmission resources corresponding to the second model, the transmission resources corresponding to the time window, and the third identifier from the second information to the terminal via Downlink Control Information (DCI) signaling. If the network device does not agree to switch models, it sends one or more of the first identifier indicating that new data cannot be replaced by historical channel data and the second identifier indicating insufficient uplink resources to the terminal.

[0299] Steps 6 and 7 can be divided into four cases, including:

[0300] Scenario 1: Uplink resources are not limited. The terminal's processing capacity changes from limited to unlimited, requiring a model switch.

[0301] Specifically, in the case of switching to the first model of type b and type c, the terminal sends the model number of the first model to the network device. The network device determines that the model switching type of the first model is the third switching type, sends the second information to the terminal, and allocates the corresponding transmission resources to the terminal so that the terminal stops calling the second model and switches to the first model.

[0302] For the case of switching to the first model of type A, both the terminal and the network device need to simultaneously invoke both the first and second models within the time window, meaning the models need to be started in advance so that the network device can receive the number of historical channel data required to input the first model. The terminal sends the model number of the first model to the network device. The network device determines the model switching type of the first model as the first switching type, determines the time window based on the feedback period of the model invoked by the network device and the number of historical channel data required to input the model, allocates the corresponding time-frequency resources to the terminal according to the time window, and uses an identifier to indicate which part of the time-frequency resources corresponds to the second model and which part corresponds to the first model. The second information is sent to the terminal. At this time, both the first and second models are in the invocation state, and the interaction data between the first and second models is transmitted through the air interface. After the terminal sends the historical channel data required by the network device to input the number of historical channel data for the first model, both the terminal and the network device stop invoking the second model and switch to running the first model. At this time, the network device reallocates the time-frequency resources corresponding to the terminal, only needing to allocate the transmission resources corresponding to the first model for the terminal.

[0303] It is worth noting that, when the terminal's processing power is not limited, the terminal can complete the invocation of the first and second models within one feedback cycle.

[0304] Scenario 2: Uplink resources are unrestricted, but the terminal's processing capacity changes from unrestricted to restricted, requiring a model switch.

[0305] Specifically, the case of switching to the first model of type b and type c is the same as case 1, and will not be repeated here.

[0306] For the case of switching to the first model of type a, if the terminal's processing power can support the terminal and network device to run the first and second models simultaneously, the situation is the same as in case 1, and will not be repeated here. If the terminal's processing power cannot support the simultaneous operation of the first and second models by the terminal and network device, the terminal calculates the channel similarity between two adjacent time points and sends first information to the network device. The first information includes the model number of the first model and the channel similarity between the two adjacent time points. If the network device's time window is less than a first threshold and the channel similarity between the two adjacent time points is greater than a second threshold, it determines that historical channel data can be used to replace new data, determines the model switching type of the first model to be the second switching type, and sends at least the first identifier in the second information to the terminal indicating that historical channel data can be used to replace new data. The network device uses historical channel data to replace new data. After receiving the number of historical channel data points required to be input for the first model, the network device stops calling the second model and runs the first model. If the network device's time window is greater than or equal to the first threshold and / or the channel similarity between the two adjacent time points is less than or equal to the second threshold, it determines that historical channel data cannot be used to replace new data. It sends second information to the terminal indicating the reason for disagreeing with the model switching, namely the first identifier indicating that historical channel data cannot be used to replace new data. The terminal re-determines the first model. If no first model meets the requirements, the terminal determines to switch to the traditional method and sends this information to the network device.

[0307] Scenario 3: Uplink resources are limited, changing from limited terminal processing capacity to unlimited terminal processing capacity, requiring model switching.

[0308] Specifically, for the case of switching to the first model of type b and type c, the terminal sends the model number of the first model to the network device. The network device determines whether the uplink resources support the amount of feedback data added by the first model compared to the second model. If the amount of feedback data of the first model is less than or equal to the amount of feedback data of the second model, the network device does not make this determination. If the uplink resources are sufficient and support the amount of feedback data of the first model, the model switching type of the first model is determined to be the third switching type. The network device sends the second information to the terminal and allocates the corresponding transmission resources to the terminal so that the terminal stops calling the second model and switches to running the first model. If the uplink resources are insufficient, the network device sends the second information to the terminal indicating the reason for not agreeing to switch models, that is, the second identifier indicates insufficient uplink resources. The terminal re-determines the first model and sends the re-determined model number of the first model to the network device.

[0309] For the case of switching to the first model of type a, if the uplink resources can support the simultaneous transmission of two different sets of channel data output by the two models on the uplink channel, and the uplink resources support the amount of feedback data of the first model, the process is the same as in case 1, and will not be repeated here. If the uplink resources cannot support the simultaneous transmission of two different sets of channel data output by the two models on the uplink channel, and / or the uplink resources cannot support the amount of feedback data of the first model, the network device receives the model number of the first model sent by the terminal, determines the model switching type of the first model as either the first switching type or the second switching type, and sends the second identifier in the second information to the terminal to indicate insufficient uplink resources. The terminal calculates the channel similarity between two adjacent time points and sends the channel similarity between the two adjacent time points to the network device. If the time window is less than a first threshold and the channel similarity between the two adjacent time points is greater than a second threshold, the network device determines that historical channel data can replace the new data, determines the model switching type of the first model to be the second switching type, and at least sends the first identifier in the second information to the terminal indicating that historical channel data can replace the new data. The network device uses historical channel data to replace the new data. After receiving the number of historical channel data points required for the first model, the network device stops calling the second model and runs the first model. If the time window is greater than or equal to the first threshold, and / or the channel similarity between the two adjacent time points is less than or equal to the second threshold, the network device determines that historical channel data cannot replace the new data, and sends the second information to the terminal indicating the reason for disagreeing with the model switching, i.e., the first identifier indicating that historical channel data cannot replace the new data. The terminal re-determines the first model. If no first model meets the requirements, the terminal determines to switch to the traditional method and sends this information to the network device.

[0310] Scenario 4: Uplink resources are limited, changing from unlimited processing power to limited processing power of the terminal, requiring model switching.

[0311] Specifically, the case of switching to the first model of type b and type c is the same as case 3, and will not be repeated here.

[0312] For the case of switching to the first model of type a, if the uplink resources can support the simultaneous transmission of two different sets of channel data from the two models via the uplink channel, the situation is the same as in case 2, and will not be repeated here. If the uplink resources cannot support the simultaneous transmission of two different sets of channel data from the two models via the uplink channel, the situation is the same as in case 3, and will not be repeated here.

[0313] The model switching method disclosed herein determines the model switching type based on the input characteristics of the model, such as whether the input data is sequential or non-sequential. By setting a first threshold and a second threshold, a second switching type is added to supplement cases where the first switching type is not applicable. This ensures the integrity and robustness of the function when switching between different models, and ensures that the model with higher accuracy is selected when resources are sufficient, thereby improving system performance and user experience, achieving the goal of widespread application. Furthermore, it achieves a balance between capability (selecting a model with higher accuracy), cost (considering terminal performance, i.e., the terminal's processing power and limited uplink resources), and quality (ensuring the integrity and robustness of model function through model switching) in the evaluation metrics when switching between different models.

[0314] To implement the method on the network device side of this disclosure embodiment, this disclosure embodiment also provides a model switching device, which is installed on the network device, as shown in FIG7. The device includes:

[0315] The first receiving module 701 is used to receive first information sent by the terminal, wherein the first information at least indicates the first model to be switched.

[0316] The first sending module 702 is used to send second information to the terminal, the second information indicating the terminal's transmission configuration or indicating the reason for disagreeing with the switching model.

[0317] In one embodiment, the first information includes one or more of the following:

[0318] The identifier of the first model;

[0319] Channel similarity between two adjacent time points.

[0320] In one embodiment, the second information includes one or more of the following:

[0321] The transmission resources corresponding to the first model;

[0322] The transmission resources corresponding to the second model, where the second model represents the currently used model or the model before the switch;

[0323] The transmission resources corresponding to the time window are used for the terminal to feed back channel data. Within the time window, the first model and the second model are called simultaneously. The second model represents the currently used model or the model before the switch.

[0324] A first identifier indicates whether new data can be replaced by historical channel data or cannot be replaced by historical channel data.

[0325] The second identifier indicates whether uplink resources are insufficient or sufficient.

[0326] The third identifier indicates that the first model and the second model are invoked simultaneously, and the second model represents the currently used model or the model before the switch.

[0327] In one embodiment, the device further includes:

[0328] A first determining module is configured to determine third information based on the first information; the third information at least indicates the model switching type of the first model;

[0329] The second determining module is used to determine the second information based on the third information.

[0330] In one embodiment, the third information includes one or more of the following:

[0331] The model switching type;

[0332] Available uplink resources;

[0333] Channel similarity between two adjacent time points;

[0334] The time window is pre-configured or determined based on the feedback period of the model invoked by the network device and the number of historical channel data that need to be input into the model.

[0335] In one embodiment, the first determining module is specifically used to determine the model switching type based at least on the feature information of the input data of the first model; wherein the feature information represents whether it is temporal or non-temporal.

[0336] In one embodiment, when the feature information representation is temporal and different input data are encoded by different encoders included in the first model, the model switching type is a first switching type or a second switching type; or,

[0337] When the feature information representation is non-temporally sequential, the model switching type is the third switching type; or,

[0338] When the feature information representation is temporal and different input data are encoded by the same encoder included in the first model, the model switching type is the third switching type.

[0339] In one embodiment, the model switching type includes one or more of the following:

[0340] The first handover type is where the terminal and the network device simultaneously run the first model and the second model, and / or simultaneously transmit two different sets of channel data output by the two models on the uplink channel; the second model represents the currently used model or the model before the handover.

[0341] The second switching type is that after the network device receives the number of historical channel data that need to be input into the first model, it stops calling the second model and runs the first model; the second model represents the currently used model or the model before the switch.

[0342] The third switching type is to stop calling the second model and then run the first model; the second model represents the currently used model or the model before the switch.

[0343] In one embodiment, the second determining module is specifically used to determine, at least based on the available uplink resources, whether the uplink resources are insufficient or sufficient.

[0344] In one embodiment, the second determining module is specifically used to determine whether new data can be replaced by historical channel data or cannot be replaced by historical channel data based on a time window and / or the channel similarity between two adjacent time points; the time window is pre-configured or determined based on the feedback period of the model invoked by the network device and the number of historical channel data that need to be input into the model.

[0345] In one embodiment, the second determining module is specifically used to determine that historical channel data can be used to replace new data when the time window is less than a first threshold and the channel similarity between two adjacent times is greater than a second threshold.

[0346] In one embodiment, the second determining module is specifically configured to determine that new data cannot be replaced by historical channel data if one or more of the following conditions are met:

[0347] The time window is greater than or equal to the first threshold;

[0348] The channel similarity between two adjacent time points is less than or equal to the second threshold.

[0349] In one embodiment, the device further includes:

[0350] The third receiving module is used to receive fourth information; the fourth information includes one or more of the following:

[0351] The model's encoder;

[0352] The encoder identifier of the model;

[0353] The model's decoder;

[0354] The identifier of the model's decoder;

[0355] The number of historical channel data points that the model needs to input;

[0356] Model switching type;

[0357] The amount of feedback data from the model of the terminal.

[0358] In practical applications, the first receiving module 701, the first sending module 702, the first determining module, and the second determining module can be implemented by the processor in the model switching device in conjunction with the communication interface.

[0359] To implement the terminal-side method of this disclosure embodiment, this disclosure embodiment also provides a model switching device, disposed on the terminal, as shown in FIG8, the device comprising:

[0360] The second sending module 801 is used to send first information to the network device, wherein the first information at least indicates the first model to be switched;

[0361] The second receiving module 802 is used to receive second information sent by the network device, the second information indicating the transmission configuration of the terminal or indicating the reason for disagreeing on the switching model.

[0362] In one embodiment, the first information includes one or more of the following:

[0363] The identifier of the first model;

[0364] Channel similarity between two adjacent time points.

[0365] In one embodiment, the second information includes one or more of the following:

[0366] The transmission resources corresponding to the first model;

[0367] The transmission resources corresponding to the second model, where the second model represents the currently used model or the model before the switch;

[0368] The transmission resources corresponding to the time window are used for the terminal to feed back channel data. Within the time window, the first model and the second model are called simultaneously. The second model represents the currently used model or the model before the switch.

[0369] A first identifier indicates whether new data can be replaced by historical channel data or cannot be replaced by historical channel data.

[0370] The second identifier indicates whether uplink resources are insufficient or sufficient.

[0371] The third identifier indicates that the first model and the second model are invoked simultaneously, and the second model represents the currently used model or the model before the switch.

[0372] In one embodiment, the device further includes:

[0373] The fourth receiving module is used to receive the fifth information; the fifth information includes one or more of the following:

[0374] The model's encoder;

[0375] The encoder identifier of the model;

[0376] The number of historical channel data points required to input the model;

[0377] The inference complexity or computational complexity of the model.

[0378] In one embodiment, the device further includes:

[0379] The third determining module is used to determine the first model based on one or more of the following:

[0380] The processing power of the terminal;

[0381] The inference or computational cost of the model;

[0382] The accuracy of the model;

[0383] The amount of feedback data from the model;

[0384] The model's input data can be either time-series or non-time-series.

[0385] In practical applications, the second sending module 801, the second receiving module 802, the fourth receiving module, and the third determining module can be implemented by the processor in the model switching device in conjunction with the communication interface.

[0386] It should be noted that the model switching device provided in the above embodiments is only illustrated by the division of the above program modules when performing model switching. In actual applications, the above processing can be assigned to different program modules as needed, that is, the internal structure of the device can be divided into different program modules to complete all or part of the processing described above. In addition, the model switching device and the model switching method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0387] Based on the hardware implementation of the above program modules, and in order to implement the method on the network device side of this disclosure embodiment, this disclosure embodiment also provides a network device, as shown in FIG9, the network device 900 includes:

[0388] The first communication interface 901 is capable of exchanging information with other network nodes;

[0389] The first processor 902 is connected to the first communication interface 901 to enable information interaction with other network nodes. When running a computer program, it executes the methods provided by one or more technical solutions on the network device side. The computer program is stored in the first memory 903.

[0390] Specifically, the first communication interface 901 is used to receive first information sent by the terminal and send second information to the terminal; the first information at least indicates a first model to be switched; the second information indicates the transmission configuration of the terminal or indicates the reason for disagreeing with the model switching.

[0391] In one embodiment, the first information includes one or more of the following:

[0392] The identifier of the first model;

[0393] Channel similarity between two adjacent time points.

[0394] In one embodiment, the second information includes one or more of the following:

[0395] The transmission resources corresponding to the first model;

[0396] The transmission resources corresponding to the second model, where the second model represents the currently used model or the model before the switch;

[0397] The transmission resources corresponding to the time window are used for the terminal to feed back channel data. Within the time window, the first model and the second model are called simultaneously. The second model represents the currently used model or the model before the switch.

[0398] A first identifier indicates whether new data can be replaced by historical channel data or cannot be replaced by historical channel data.

[0399] The second identifier indicates whether uplink resources are insufficient or sufficient.

[0400] The third identifier indicates that the first model and the second model are invoked simultaneously, and the second model represents the currently used model or the model before the switch.

[0401] In one embodiment, the first processor 902 is further configured to determine third information based on the first information; and determine second information based on the third information; wherein the third information at least indicates the model switching type of the first model.

[0402] In one embodiment, the third information includes one or more of the following:

[0403] The model switching type;

[0404] Available uplink resources;

[0405] Channel similarity between two adjacent time points;

[0406] The time window is pre-configured or determined based on the feedback period of the model invoked by the network device and the number of historical channel data that need to be input into the model.

[0407] In one embodiment, the first processor 902 is specifically configured to determine the model switching type based at least on feature information of the input data of the first model; wherein the feature information represents whether it is temporal or non-temporal.

[0408] In one embodiment, when the feature information representation is temporal and different input data are encoded by different encoders included in the first model, the model switching type is a first switching type or a second switching type; or,

[0409] When the feature information representation is non-temporally sequential, the model switching type is the third switching type; or,

[0410] When the feature information representation is temporal and different input data are encoded by the same encoder included in the first model, the model switching type is the third switching type.

[0411] In one embodiment, the model switching type includes one or more of the following:

[0412] The first handover type is where the terminal and the network device simultaneously run the first model and the second model, and / or simultaneously transmit two different sets of channel data output by the two models on the uplink channel; the second model represents the currently used model or the model before the handover.

[0413] The second switching type is that after the network device receives the number of historical channel data that need to be input into the first model, it stops calling the second model and runs the first model; the second model represents the currently used model or the model before the switch.

[0414] The third switching type is to stop calling the second model and then run the first model; the second model represents the currently used model or the model before the switch.

[0415] In one embodiment, the first processor 902 is specifically configured to determine, at least based on the available uplink resources, whether the uplink resources are insufficient or sufficient.

[0416] In one embodiment, the first processor 902 is specifically used to determine, based on a time window and / or the channel similarity between two adjacent time points, whether new data can be replaced by historical channel data or not; the time window is pre-configured or determined based on the feedback period of the model invoked by the network device and the number of historical channel data that need to be input into the model.

[0417] In one embodiment, the first processor 902 is specifically configured to determine that historical channel data can be used to replace new data when the time window is less than a first threshold and the channel similarity between two adjacent times is greater than a second threshold.

[0418] In one embodiment, the first processor 902 is specifically configured to determine that new data cannot be replaced by historical channel data if one or more of the following conditions are met:

[0419] The time window is greater than or equal to the first threshold;

[0420] The channel similarity between two adjacent time points is less than or equal to the second threshold.

[0421] In one embodiment, the first communication interface 901 is further configured to receive fourth information; the fourth information includes one or more of the following:

[0422] The model's encoder;

[0423] The encoder identifier of the model;

[0424] The model's decoder;

[0425] The identifier of the model's decoder;

[0426] The number of historical channel data points that the model needs to input;

[0427] Model switching type;

[0428] The amount of feedback data from the model of the terminal.

[0429] It should be noted that the specific processing procedures of the first processor 902 and the first communication interface 901 can be understood by referring to the above method.

[0430] Of course, in practical applications, the various components in network device 900 are coupled together through bus system 904. It can be understood that bus system 904 is used to implement communication between these components. In addition to the data bus, bus system 904 also includes a power bus, a control bus, and a status signal bus. However, for clarity, all buses are labeled as bus system 904 in Figure 9.

[0431] The first memory 903 in this embodiment is used to store various types of data to support the operation of the network device 900. Examples of such data include any computer program used to operate on the network device 900.

[0432] The methods disclosed in the above embodiments of this disclosure can be applied to the first processor 902, or implemented by the first processor 902. The first processor 902 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware or by instructions in the form of software in the first processor 902. The first processor 902 may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The first processor 902 can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this disclosure. The general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the methods disclosed in the embodiments of this disclosure can be directly manifested as being executed by a hardware decoding processor, or being executed by a combination of hardware and software modules in the decoding processor. The software modules may be located in a storage medium, which is located in the first memory 903. The first processor 902 reads the information in the first memory 903 and completes the steps of the aforementioned method in combination with its hardware.

[0433] In an exemplary embodiment, the network device 900 may be implemented by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers (MCUs), microprocessors, or other electronic components to perform the aforementioned method.

[0434] Based on the hardware implementation of the above program modules, and in order to implement the terminal-side method of this disclosure embodiment, this disclosure embodiment also provides a terminal, as shown in FIG10, the terminal 1000 includes:

[0435] The second communication interface 1001 is capable of exchanging information with other network nodes;

[0436] The second processor 1002 is connected to the second communication interface 1001 to enable information interaction with other network nodes. When running a computer program, it executes the methods provided by one or more of the aforementioned terminal-side technical solutions. The computer program is stored in the second memory 1003.

[0437] Specifically, the second communication interface 1001 is used to send first information to the network device and receive second information sent by the network device; the first information at least indicates a first model to be switched; the second information indicates the transmission configuration of the terminal or indicates the reason for disagreeing with the model switching.

[0438] In one embodiment, the first information includes one or more of the following:

[0439] The identifier of the first model;

[0440] Channel similarity between two adjacent time points.

[0441] In one embodiment, the second information includes one or more of the following:

[0442] The transmission resources corresponding to the first model;

[0443] The transmission resources corresponding to the second model, where the second model represents the currently used model or the model before the switch;

[0444] The transmission resources corresponding to the time window are used for the terminal to feed back channel data. Within the time window, the first model and the second model are called simultaneously. The second model represents the currently used model or the model before the switch.

[0445] A first identifier indicates whether new data can be replaced by historical channel data or cannot be replaced by historical channel data.

[0446] The second identifier indicates whether uplink resources are insufficient or sufficient.

[0447] The third identifier indicates that the first model and the second model are invoked simultaneously, and the second model represents the currently used model or the model before the switch.

[0448] In one embodiment, the second communication interface 1001 is further configured to receive fifth information; the fifth information includes one or more of the following:

[0449] The model's encoder;

[0450] The encoder identifier of the model;

[0451] The number of historical channel data points required to input the model;

[0452] The inference complexity or computational complexity of the model.

[0453] In one embodiment, the second processor 1002 is further configured to determine the first model based on one or more of the following:

[0454] The processing power of the terminal;

[0455] The inference or computational cost of the model;

[0456] The accuracy of the model;

[0457] The amount of feedback data from the model;

[0458] The model's input data can be either time-series or non-time-series.

[0459] It should be noted that the specific processing procedures of the second processor 1002 and the second communication interface 1001 can be understood by referring to the above method.

[0460] Of course, in practical applications, the various components in terminal 1000 are coupled together through bus system 1004. It can be understood that bus system 1004 is used to realize the connection and communication between these components. In addition to the data bus, bus system 1004 also includes a power bus, a control bus, and a status signal bus. However, for clarity, all buses are labeled as bus system 1004 in Figure 10.

[0461] The second memory 1003 in this embodiment of the present disclosure is used to store various types of data to support the operation of the terminal 1000. Examples of such data include any computer program used to operate on the terminal 1000.

[0462] The methods disclosed in the above embodiments of this disclosure can be applied to, or implemented by, the second processor 1002. The second processor 1002 may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by the integrated logic circuitry of the hardware or by instructions in the form of software within the second processor 1002. The second processor 1002 may be a general-purpose processor, a DSP, or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The second processor 1002 can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this disclosure. The general-purpose processor may be a microprocessor or any conventional processor, etc. The steps of the methods disclosed in the embodiments of this disclosure 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. The software modules may be located in a storage medium, specifically a second memory 1003. The second processor 1002 reads information from the second memory 1003 and, in conjunction with its hardware, completes the steps of the aforementioned method.

[0463] In an exemplary embodiment, terminal 1000 may be implemented by one or more ASICs, DSPs, PLDs, CPLDs, FPGAs, general-purpose processors, controllers, MCUs, microprocessors, or other electronic components to perform the aforementioned method.

[0464] It is understood that the memories (first memory 903, second memory 1003) in the embodiments of this disclosure can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), ferromagnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM); magnetic surface memory can be disk storage or magnetic tape storage. Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Synchronous Static Random Access Memory (SSRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), SyncLink Dynamic Random Access Memory (SLDRAM), and Direct Rambus Random Access Memory (DRRAM).The memories described in the embodiments of this disclosure are intended to include, but are not limited to, these and any other suitable types of memories.

[0465] In exemplary embodiments, this disclosure also provides a storage medium, namely a computer storage medium, specifically a computer-readable storage medium, such as a first memory 903 storing a computer program, which can be executed by a first processor 902 of a network device 900 to complete the steps described in the aforementioned network device-side method. Another example is a second memory 1003 storing a computer program, which can be executed by a second processor 1002 of a terminal 1000 to complete the steps described in the aforementioned terminal-side method. The computer-readable storage medium can be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, Flash Memory, magnetic surface memory, optical disc, or CD-ROM.

[0466] By way of example, this disclosure also provides a computer program product, including a computer program that can be executed by a first processor 902 of a network device 900 to perform the steps described in the aforementioned network device-side method. The computer program can also be executed by a second processor 1002 of a terminal 1000 to perform the steps described in the aforementioned terminal-side method.

[0467] It should be noted that terms such as "first" and "second" are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. The term "and / or" in this article merely describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, the term "one or more" in this article refers to any combination of at least two of any one or more items from a set of A, B, and C. For example, "one or more of A, B, and C" can represent any one or at least two or more elements selected from the set of A, B, and C.

[0468] Furthermore, the technical solutions described in the embodiments of this disclosure can be combined arbitrarily without conflict.

[0469] The above description is merely a preferred embodiment of this disclosure and is not intended to limit the scope of protection of this disclosure.

Claims

1. A model switching method applied to a network device, the method comprising: The receiving terminal sends first information, which at least indicates the first model to be switched; Send a second message to the terminal, the second message indicating the terminal's transmission configuration or indicating the reason for disagreeing with the switching model.

2. The method according to claim 1, wherein, The first information includes one or more of the following: The identifier of the first model; Channel similarity between two adjacent time points.

3. The method according to claim 1, wherein, The second information includes one or more of the following: The transmission resources corresponding to the first model; The transmission resources corresponding to the second model, where the second model represents the currently used model or the model before the switch; The transmission resources corresponding to the time window are used for the terminal to feed back channel data. Within the time window, the first model and the second model are called simultaneously. The second model represents the currently used model or the model before the switch. A first identifier indicates whether new data can be replaced by historical channel data or cannot be replaced by historical channel data. The second identifier indicates whether uplink resources are insufficient or sufficient. The third identifier indicates that the first model and the second model are invoked simultaneously, and the second model represents the currently used model or the model before the switch.

4. The method according to any one of claims 1 to 3, further comprising: The third information is determined based on the first information; The third information at least indicates the model switching type of the first model; Based on the third information, the second information is determined.

5. The method according to claim 4, wherein, The third information includes one or more of the following: The model switching type; Available uplink resources; Channel similarity between two adjacent time points; The time window is pre-configured or determined based on the feedback period of the model invoked by the network device and the number of historical channel data that need to be input into the model.

6. The method according to claim 4, further comprising: The model switching type is determined based at least on the feature information of the input data of the first model; wherein the feature information represents whether it is temporal or non-temporal.

7. The method according to claim 6, wherein, When the feature information representation is temporal and different input data are encoded by different encoders included in the first model, the model switching type is either a first switching type or a second switching type; or, When the feature information representation is non-temporally sequential, the model switching type is the third switching type; or, When the feature information representation is temporal and different input data are encoded by the same encoder included in the first model, the model switching type is the third switching type.

8. The method according to any one of claims 5 to 7, wherein, The model switching type includes one or more of the following: The first handover type is where the terminal and the network device simultaneously run the first model and the second model, and / or simultaneously transmit two different sets of channel data output by the two models on the uplink channel; the second model represents the currently used model or the model before the handover. The second switching type is that after the network device receives the number of historical channel data that need to be input into the first model, it stops calling the second model and runs the first model; the second model represents the currently used model or the model before the switch. The third switching type is to stop calling the second model and then run the first model; the second model represents the currently used model or the model before the switch.

9. The method according to any one of claims 1 to 3, 5 to 7, further comprising: At least based on available uplink resources, determine whether uplink resources are insufficient or sufficient.

10. The method according to any one of claims 1 to 3, 5 to 7, further comprising: Based on the time window and / or the channel similarity between two adjacent moments, determine whether new data can be replaced by historical channel data or not. The time window is pre-configured or determined based on the feedback period of the model invoked by the network device and the number of historical channel data that need to be input into the model.

11. The method according to claim 10, further comprising: If the time window is less than the first threshold and the channel similarity between two adjacent times is greater than the second threshold, then it is determined that historical channel data can be used to replace the new data.

12. The method according to claim 10, further comprising: If one or more of the following conditions are met, it is determined that historical channel data cannot be used to replace new data: The time window is greater than or equal to the first threshold; The channel similarity between two adjacent time points is less than or equal to the second threshold.

13. The method according to any one of claims 9 to 11, 15 to 16, further comprising: Receive fourth information; the fourth information includes one or more of the following: The model's encoder; The encoder identifier of the model; The model's decoder; The identifier of the model's decoder; The number of historical channel data points that the model needs to input; Model switching type; The amount of feedback data from the model of the terminal.

14. A model switching method applied to a terminal, the method comprising: Send first information to the network device, wherein the first information at least indicates the first model to be switched; The terminal receives second information sent by the network device, the second information indicating the transmission configuration of the terminal or indicating the reason for disagreeing on the switching model.

15. The method according to claim 14, wherein, The first information includes one or more of the following: The identifier of the first model; Channel similarity between two adjacent time points.

16. The method of claim 14, wherein, The second information includes one or more of the following: The transmission resources corresponding to the first model; The transmission resources corresponding to the second model, where the second model represents the currently used model or the model before the switch; The transmission resources corresponding to the time window are used for the terminal to feed back channel data. Within the time window, the first model and the second model are called simultaneously. The second model represents the currently used model or the model before the switch. A first identifier indicates whether new data can be replaced by historical channel data or cannot be replaced by historical channel data. The second identifier indicates whether uplink resources are insufficient or sufficient. The third identifier indicates that the first model and the second model are invoked simultaneously, and the second model represents the currently used model or the model before the switch.

17. The method according to any one of claims 14 to 16, wherein the method further comprises: Receive the fifth message; The fifth piece of information includes one or more of the following: The model's encoder; The encoder identifier of the model; The number of historical channel data points required to input the model; The inference complexity or computational complexity of the model.

18. The method according to any one of claims 14 to 16, wherein the method further comprises: The first model is determined based on one or more of the following: The processing power of the terminal; The inference or computational cost of the model; The accuracy of the model; The amount of feedback data from the model; The model's input data can be either time-series or non-time-series.

19. A model switching device, the device comprising: The first receiving module is used to receive first information sent by the terminal, wherein the first information at least indicates the first model to be switched; A first sending module is used to send second information to the terminal, the second information indicating the terminal's transmission configuration or indicating the reason for disagreeing on switching modes.

20. A model switching device, the device comprising: The second sending module is used to send first information to the network device, wherein the first information at least indicates the first model to be switched; The second receiving module is used to receive second information sent by the network device, the second information indicating the transmission configuration of the terminal or indicating the reason for disagreeing on the switching model.

21. A network device, comprising: A first processor and a first communication interface; wherein... The first communication interface is used to receive first information sent by the terminal and send second information to the terminal; the first information at least indicates a first model to be switched; the second information indicates the transmission configuration of the terminal or indicates the reason for disagreeing on switching models.

22. A terminal, comprising: A second processor and a second communication interface; wherein... The second communication interface is used to send first information to the network device and receive second information sent by the network device; the first information at least indicates a first model to be switched; the second information indicates the transmission configuration of the terminal or indicates the reason for disagreeing with the model switching.

23. A network device, comprising a first processor and a first memory for storing a computer program capable of running on the first processor. in, When the first processor is used to run the computer program, it performs the steps of the method according to any one of claims 1 to 13.

24. A terminal, comprising a second processor and a second memory for storing a computer program capable of running on the second processor. in, When the second processor is used to run the computer program, it performs the steps of the method according to any one of claims 14 to 18.

25. A storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 13, or implements the steps of the method according to any one of claims 14 to 18.

26. A computer program product comprising a computer program that, when executed by a processor, implements the steps of the method according to any one of claims 1 to 13, or implements the steps of the method according to any one of claims 14 to 18.

Images