Artificial intelligence or machine learning (ai / ML)-based wireless communication method and wireless communication device

Abstract

Provided in the present disclosure is an artificial intelligence or machine learning (AI / ML)-based wireless communication method, which is executed by a terminal device. The method comprises: receiving data and / or a data identifier (ID), wherein the data comprises a training data set and / or model parameters, and the data identifier (ID) is composed of a plurality of sub-data segment identifiers and / or is determined on the basis of a quasi-co-model identifier (QCM ID).

Description

A wireless communication method and wireless communication device based on artificial intelligence or machine learning (AI / ML) Technical Field

[0001] This disclosure relates to the field of wireless communication, and more particularly to a wireless communication method and wireless communication device based on artificial intelligence or machine learning (AI / ML). Background Technology

[0002] Within the framework of Massive-MIMO communication systems, the measurement and feedback of Channel State Information (CSI) is one of the core technologies of the physical layer. Accurate measurement and feedback of CSI information can help the base station determine appropriate data modulation and coding schemes to improve the spectral efficiency of the communication system.

[0003] Existing technologies have researched CSI compression in the spatial and frequency domains to increase CSI accuracy while reducing the air interface overhead of CSI reporting. The AI / ML-based CSI spatial-frequency domain compression process is based on an encoder-decoder model. The encoder is deployed on the UE side, using spatial-frequency domain channel information (such as precoding matrices and eigenvectors) obtained from channel measurements as model input, and outputting compressed channel information. The decoder is deployed on the base station side, using the compressed channel information as model input, and outputting complete channel information. A quantizer can be deployed after the encoder to further reduce the transmission overhead of CSI feedback, forming a complete CSI generation module with the encoder. Correspondingly, a dequantizer needs to be deployed before the decoder to output the unquantized compressed channel information, forming a complete CSI reconstruction module with the decoder. Furthermore, the quantizer and dequantizer can also be integrated into the encoder and decoder to reduce model complexity. However, in the use of AI / ML models, data transmission for the management functions of AI / ML-based CSI compression faces a series of problems. Therefore, there is a need to propose a wireless communication method and wireless communication device based on artificial intelligence or machine learning (AI / ML) to improve existing technologies. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a wireless communication method based on artificial intelligence or machine learning (AI / ML) to address the above-mentioned deficiencies in the prior art, thereby solving the problems existing in the prior art.

[0005] According to one aspect of this disclosure, a wireless communication method based on artificial intelligence or machine learning (AI / ML) is provided, executed in a terminal device, comprising:

[0006] Receive data and / or data identifier ID, wherein the data includes training dataset and / or model parameters, and the data identifier ID is composed of several sub-data segment identifiers and / or determined based on the Quasi-Common Model Identifier (QCM ID).

[0007] According to one aspect of this disclosure, a wireless communication method based on artificial intelligence or machine learning (AI / ML) is provided, executed on a terminal device, the method comprising: receiving data and / or a data identifier ID, wherein the data includes a training dataset and / or model parameters, and the data identifier ID is composed of several sub-data segment identifiers and / or determined based on a quasi-common model identifier (QCM ID).

[0008] According to one aspect of this disclosure, a wireless communication method based on artificial intelligence or machine learning (AI / ML) is provided, executed in an access network device, the method comprising:

[0009] Send data and / or data identifier ID, wherein the data includes training dataset and / or model parameters, and the data identifier ID is composed of several sub-data segment identifiers and / or determined based on the Quasi-Common Model Identifier (QCM ID).

[0010] According to one aspect of this disclosure, a wireless communication method based on artificial intelligence or machine learning (AI / ML) is provided, executed in an access network device, the method comprising: transmitting data and / or a data identifier ID, wherein the access network device data includes a training dataset and / or model parameters, and the data identifier ID is composed of several sub-data segment identifiers and / or determined based on a quasi-common model identifier (QCM ID).

[0011] According to one aspect of this disclosure, a wireless communication method based on artificial intelligence or machine learning (AI / ML) is provided, executed in a first device, the method comprising:

[0012] Receive a first part of data and / or a second part of data, wherein the first part of data includes at least one of the following: channel measurement data, quality indication of channel measurement data, and / or timestamp of channel measurement data; the second part of data includes at least one of the following: tag, quality indication of tag, and / or timestamp information of tag.

[0013] According to one aspect of this disclosure, a wireless communication method based on artificial intelligence or machine learning (AI / ML) is provided, executed in a second device, the method comprising:

[0014] Send a second part of the data; wherein the second part of the data includes at least one of the following: a tag, a tag quality indicator, and / or a tag timestamp information.

[0015] According to one aspect of this disclosure, a wireless communication device is provided, including a processor and a memory for storing a computer program, the processor for calling and running the computer program stored in the memory to perform steps in the data processing method as described in any of the preceding claims.

[0016] According to one aspect of this disclosure, a readable storage medium is provided for storing a computer program that is invoked and executed by a processor to perform any of the methods described above. Attached Figure Description

[0017] To more clearly illustrate the embodiments of this disclosure or related technologies, the following figures will be briefly described in the embodiments. Obviously, the figures are merely some embodiments of this disclosure, and those skilled in the art can obtain other figures based on these figures without creative effort.

[0018] Figure 1 illustrates a schematic diagram of the wireless communication system architecture provided in this disclosure.

[0019] Figure 2 illustrates one of the flowcharts of the wireless communication method based on artificial intelligence or machine learning (AI / ML) provided in this disclosure.

[0020] Figure 3 illustrates the second flowchart of the wireless communication method based on artificial intelligence or machine learning (AI / ML) provided in this disclosure.

[0021] Figure 4 illustrates the third flowchart of the wireless communication method based on artificial intelligence or machine learning (AI / ML) provided in this disclosure.

[0022] Figure 5 illustrates an exemplary block diagram of a wireless communication system provided in this disclosure. Detailed Implementation

[0023] The embodiments of this disclosure have been described in detail with reference to the accompanying drawings, outlining technical aspects, structural features, objectives, and effects, as described below. Specifically, the terminology used in the embodiments of this disclosure is for the purpose of describing particular embodiments only and is not intended to limit the disclosure.

[0024] In this disclosure, “A or B” may mean “A only”, “B only”, or “both A and B”.

[0025] In other words, in this disclosure, “A or B” can be interpreted as “A and / or B”. For example, in this disclosure, “A, B or C” can mean “A only”, “B only”, “C only” or “any combination of A, B, and C”.

[0026] The forward slash ( / ) or comma used in this disclosure can mean "and / or". For example, "A / B" can mean "A and / or B". Therefore, "A / B" can mean "A only", "B only", or "both A and B". For example, "A, B, C" can mean "A, B, or C".

[0027] In this disclosure, "at least one of A and B" may mean "only A", "only B" or "both A and B". Furthermore, in this disclosure, the expression "at least one of A or B" or "at least one of A and / or B" may be interpreted as "at least one of A and B".

[0028] Additionally, in this disclosure, "at least one of A, B, and C" may mean "A only", "B only", "C only" or "any combination of A, B, and C". Furthermore, "at least one of A, B, or C" or "at least one of A, B, and / or C" may mean "at least one of A, B, and C".

[0029] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0030] Those skilled in the art will recognize and understand that the details of the described examples are merely illustrative of some embodiments, and that the teachings set forth herein are applicable to various alternative settings.

[0031] The technical solutions disclosed herein can be applied to various wireless communication systems, such as: Long Term Evolution (LTE) systems, LTE Frequency Division Duplex (FDD) systems, LTE Time Division Duplex (TDD) systems, 5G communication systems, or future wireless communication systems, etc.

[0032] For example, the wireless communication system 100 of this disclosure is shown in FIG1. ​​The wireless communication system 100 may include a base station 110, which may be a device communicating with user equipment (UE) 120. The base station 110 can provide communication coverage for a specific geographical area and can communicate with user equipment located within that coverage area. Optionally, the base station 110 may be an evolved Node B (eNB or eNodeB) in an LTE system, or it may be a mobile switching center, relay station, access point, vehicle-mounted equipment, wearable device, hub, switch, bridge, router, access network equipment in a 5G network, or a base station in a future communication system, etc.

[0033] The wireless communication system 100 also includes at least one user equipment 120 located within the coverage area of ​​the base station 110. "User equipment" as used herein includes, but is not limited to, devices configured to receive / transmit communication signals via wired connections, such as via Public Switched Telephone Networks (PSTN), Digital Subscriber Line (DSL), digital cable, direct cable connection; and / or another data connection / network; and / or via a wireless interface, such as for cellular networks, Wireless Local Area Networks (WLAN), digital television networks such as DVB-H networks, satellite networks, AM-FM broadcast transmitters; and / or other user equipment. User equipment configured to communicate via a wireless interface may be referred to as a "wireless communication terminal," "wireless terminal," or "mobile terminal." Examples of mobile terminals include, but are not limited to, satellite or cellular phones; personal communications system (PCS) terminals that can combine cellular radiotelephone with data processing, fax, and data communication capabilities; PDAs that may include radiotelephones, pagers, Internet / intranet access, web browsers, notebooks, calendars, and / or Global Positioning System (GPS) receivers; and conventional laptop and / or handheld receivers or other electronic devices that include radiotelephone transceivers. User equipment can refer to access terminals, user units, user stations, mobile stations, mobile stations, remote stations, remote user equipment, mobile devices, wireless communication equipment, or user agents. Access terminals can be cellular phones, cordless phones, Session Initiation Protocol (SIP) phones, Wireless Local Loop (WLL) stations, personal digital assistants (PDAs), handheld devices with wireless communication capabilities, computing devices or other processing devices connected to a wireless modem, in-vehicle equipment, wearable devices, user equipment in 5G networks, or user equipment in future PLMN evolutions, etc.

[0034] Optionally, user equipment 120 can perform device-to-device (D2D) communication with each other.

[0035] Alternatively, 5G communication systems or 5G networks may also be referred to as New Radio (NR) systems or NR networks.

[0036] The wireless communication system 100 also includes a core network 130. The core network 130 may be an IP mobile communication network operated by a mobile communication operator. For example, the core network 130 may be the core network used by the mobile communication operator that operates and manages the wireless communication system 100, or it may be the core network used by a virtual mobile communication operator such as an MVNO (Mobile Virtual Network Operator).

[0037] The core network 130 can connect to the base station 110, serving as a relay device for transmitting user data. User equipment 120 transmits and receives user data via the core network 130. It should be noted that user data communication is not limited to IP communication; it can also be non-IP communication.

[0038] Figure 1 illustrates an exemplary base station 110, two user equipment 120, and a core network 130. Optionally, the wireless communication system 100 may include multiple base stations, and each base station may include other numbers of user equipment within its coverage area. This disclosure does not limit this.

[0039] Optionally, the wireless communication system 100 may also include other network entities such as a network controller, a mobility management entity, and network elements, and this disclosure does not limit this. For example, the core network 130 may include other network entities such as a network controller, a mobility management entity, and network elements, and this disclosure does not limit this.

[0040] It should be understood that devices with wireless communication capabilities in the network / system described in this disclosure may be referred to as wireless communication devices. Taking the wireless communication system 100 shown in Figure 1 as an example, the wireless communication devices may include a base station 110, a user equipment 120, and a core network 130 with communication capabilities. The base station 110 and the user equipment 120 may be the specific devices described above, which will not be repeated here. The wireless communication devices may also include other devices in the wireless communication system 100 (core network 130). For example, the core network 130 may include other network entities such as network controllers and mobility management entities, which are not limited in this disclosure.

[0041] In existing technologies, the IDs currently discussed include model ID, associated ID, and ID-X. ID-X is carried when transmitting datasets and is used for model identification, pairing UE-side models and base station-side models, and indicating activated UE-side models. ID-X can be used in inter-vendor collaborations. In this approach, the UE needs to train, retrain, or fine-tune its UE-side model based on the received dataset. Considering that the UE may have multiple built-in model structures, if ID-X is only used for model identification and has no actual physical meaning, then problem 1 arises: it wastes signaling resources and hinders the UE from selecting a suitable model structure for training its UE-side model. Updating all UE-side models also results in energy waste.

[0042] Furthermore, the CSI compression framework can also include quantizers and dequantizers to further reduce the overhead of CSI feedback. The generalization performance of the quantizer and dequantizer also affects the overall performance of CSI compression. Current root cause analysis focuses on the CSI generation and reconstruction modules, without considering the presence or absence of quantization functionality (when quantization is present, quantization and encoding / decoding are treated as a whole). Therefore, it's impossible to distinguish between encoding / decoding and quantization issues. When conducting root cause analysis including quantization functionality, the following two issues need to be considered:

[0043] Question 2: If the decision is made on the network side, the UE can assist in monitoring the encoder's training performance and report the performance monitoring metrics or results to the network side to avoid the significant overhead of directly reporting the encoder output. In this case, it is necessary to consider how to define the performance monitoring metrics and thresholds for UE-side auxiliary monitoring. If the encoder-decoder's training performance is not problematic, it can be determined that data drift is causing the performance degradation of CSI compression. The UE needs to report UE-side monitoring data to the network side, such as the input of model inference (i.e., the target CSI), the encoder output, the quantizer output, and to determine whether there are problems with the generalization performance of the encoder-decoder and quantization on the network side. In this case, it is necessary to consider how to design the reporting configuration of the UE-side monitoring data.

[0044] Question 3: If the decision is made on the UE side, the network side can similarly assist in monitoring the decoder's training performance to avoid the significant overhead of directly sending the decoder output (i.e., reconstructing CSI). Considering that the network-side monitoring method is transparent to the UE, the network side can monitor the decoder's training performance based on its internal algorithms. If the decoder's training performance is fine, it can be determined that data drift is causing a performance degradation in CSI compression. The UE needs to report UE-side monitoring data to the network side, such as the encoder's output and the quantizer's output, to further determine if there are problems with the generalization performance of the codec and quantization. How should the triggering mechanism for reporting UE-side monitoring data be designed? The judgment of generalization performance includes the network-side-assisted judgment of quantization generalization performance and the UE-side judgment of the codec's generalization performance based on the precoding reference signal. At this point, it is necessary to consider how to indicate the network-side-assisted quantization generalization performance judgment results to the UE.

[0045] Question 4: When decision-making is performed on the NW side, if the training performance of the encoder currently running on the UE side is detected to be insufficient, the network side can take model control measures, including model activation, deactivation, and switching, to improve the performance of CSI compression. The proposed Model ID is used for model identification. If the Model ID is used to indicate the model to be controlled, considering the current discussion of a one-to-one correspondence between Model ID and model information (such as model architecture and model parameters), the following two drawbacks exist: 1. The UE needs to report the Model ID to complete the identification of the updated UE-side model, which introduces additional signaling overhead; 2. To reduce the reconfiguration of the correspondence between Model ID and model information, Model ID will occupy a large amount of signaling overhead to support the description of more models, especially when model control targets multiple UE-side models, the indication of multiple Model IDs will further increase the signaling overhead.

[0046] The information sending method provided in this application will be described in detail below with reference to the accompanying drawings and through some embodiments and application scenarios.

[0047] Figure 2 illustrates one of the flowcharts of a wireless communication method based on artificial intelligence or machine learning (AI / ML) provided in this disclosure, the method including at least one of the following steps:

[0048] Step S100: Receive data and / or data identifier ID, wherein the data includes training dataset and / or model parameters, and the data identifier ID is composed of several sub-data segment identifiers and / or determined based on the Quasi-Common Model Identifier (QCM ID).

[0049] Specifically, after receiving the data from the access network device, the user equipment (e.g., UE) can train the UE-side model. The data can be from the network side (or the access network device or the core network device), or data from other devices, without limitation. The training dataset in the data may include the target CSI and / or the labels output by the encoder. The user equipment can also receive a data identifier ID from the access network device. The data ID can be represented in a segmented manner, meaning it can consist of several sub-data segment identifiers, used to indicate at least one of the following information about the CSI compressed bilateral model trained by the access network device: encoder architecture, decoder architecture, and / or quantization method. The UE can autonomously select a suitable internal encoder architecture to train the UE-side model based on the data ID.

[0050] In some embodiments of this disclosure, the sub-data segment may include at least one of the following: model type, number of layers, number of neurons, linear quantization type, linear quantization scaling factor, linear quantization offset, number of quantization bits corresponding to each dimension of the input quantizer, number of data units (tokens), output dimension of each data unit, number of blocks within the Transformer model, and / or number of self-attention heads.

[0051] The model type may be indicated in at least one of the following ways:

[0052] 1. An index is used, with each index corresponding to a model type. At least one bit can be used to represent the index, and the number of bits can be a default integer value and / or indicated by the base station. For example, model types include, but are not limited to, at least one of the following: Multilayer Perceptron (MLP), Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and / or Transformer. Indices 00, 01, 10, and 11 each correspond to one of these model types. It is worth noting that there are no restrictions on the correspondence between indices and model types. Index 00 can correspond to any one of Multilayer Perceptron (MLP), Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and / or Transformer. Other indices are similar and will not be elaborated further here.

[0053] 2. A bitmap is used, with the number of bits representing the number of model types. This number can be a default value and / or indicated by the base station. Each bit corresponds to one model type. If a bit is 1, it means that the access network device can use the model type corresponding to that bit for training the CSI compressed dual-side model.

[0054] The number of layers refers to the total number of layers. For the specified number of layers: if the model type is one of the following: MLP, CNN, or RNN, then the layer refers to a layer within that neural network model; if the model type is Transformer, then the layer refers to a layer within the MLP module of that neural network model.

[0055] When the number of layers is the total number of layers, the total number of layers includes at least one of the following forms:

[0056] 1) Includes an input layer and an output layer;

[0057] 2) Includes the input layer, but excludes the output layer;

[0058] 3) Includes the output layer, but excludes the input layer; and / or

[0059] 4) It does not include the input layer and the output layer.

[0060] At this time, the number of layers can be indicated in the following ways: binary representation and / or index indication, with the number of bits being a default integer value, and each index corresponding to an optional number of layers.

[0061] When the number of layers refers to the number of layers of different types, the number of layers of each type can be a default integer value and / or indicated by the base station. In this case, the number of layers of each type can be indicated using at least one of the following methods:

[0062] 1) The number of layers of various types can be represented in binary and / or indicated by index. The number of bits can be a default integer value. When using index indication, each index corresponds to an optional number of layers. The indication order of different layer types can be predefined and / or indicated by the base station. For example, if the number of layers of each type corresponds to 4 bits, and the number of layer types is N, where the layer types include, but are not limited to, at least one of the following: Fully Connected Layer, Convolutional Layer, Pooling Layer, Activation Layer, Normalization Layer, Recurrent Layer, Transformer Layer, and / or Dropout Layer, then the number of layers is represented by 4N bits.

[0063] 2) The number of layers of each type consists of two parts: the first part indicates the layer type, and the second part indicates the number of layers of that type. The first part can be represented by an index, with each index corresponding to a layer type. At least one bit can be used to represent the index, and the number of bits can be a default integer value and / or indicated by the base station. The second part can be represented by a binary representation and / or by an index, with the number of bits also being a default integer value. When indicating the number of layers by an index, each index corresponds to an optional number of layers. For example, layer types include, but are not limited to, at least one of the following: fully connected layer, convolutional layer, pooling layer, activation layer, normalized layer, recurrent layer, Transformer layer, and / or dropout layer. The first part can use one of the indices 000, 001, 010, 011, 100, 101, 110, 111 to correspond to one of the aforementioned layer types. The second part has 4 bits, thus the number of layers of each type is represented by 7 bits.

[0064] The number of neurons includes the number of neurons in all layers and the number of neurons in different types of layers.

[0065] 1. Statistics on the number of neurons in all layers, including but not limited to at least one of the following: sum, average, minimum, and / or maximum. The statistics on the number of neurons in all layers can be represented in binary and / or indicated by an index, with the number of bits being a default integer value. When indicated by an index, each index corresponds to an optional number of neurons;

[0066] 2. Statistical measures for the number of neurons in different layer types, wherein the number of different layer types can be a default integer value and / or indicated by the base station. The statistical measures for the number of neurons in different layer types include, but are not limited to, at least one of the following: sum, average, minimum, and / or maximum. The statistical measures for the number of neurons in different layer types can be indicated using at least one of the following methods:

[0067] 1) The statistics on the number of neurons in each type of layer can be represented in binary and / or indicated by indexes, with the number of bits being a default integer value. When using indexes, each index corresponds to an optional number of neurons, and the order of indication for different layer types can be predefined and / or indicated by base stations;

[0068] 2) The statistics on the number of neurons in each type of layer consist of two parts: the first part is an indication of the layer type, and the second part is the statistics on the number of neurons in that type of layer. The first part can be represented using an index, with each index corresponding to a layer type. At least one bit can be used to represent the index, and the number of bits can be a default integer value and / or indicated by the base station. The second part can be represented using binary representation and / or an index indication, with the number of bits also being a default integer value. When using an index indication, each index corresponds to an optional number of neurons.

[0069] 3. Regarding the number of neurons in each layer, the number of layers is indicated by the components of the data ID and / or by the base station. The number of neurons in each layer can be represented in binary and / or indicated by an index. The number of bits can be a default integer value. When an index is used, each index corresponds to an optional number of neurons. When the order of the layers is consistent with the encoder or decoder architecture that can be used when the access network device trains the CSI compressed two-sided model, the layer order is consistent with the encoder or decoder architecture that can be used.

[0070] The linear quantization type can be indicated in at least one of the following ways:

[0071] 1. An index is used, with each index corresponding to a linear quantization type. At least one bit can be used to represent the index, and the number of bits can be a default integer value and / or indicated by the base station. For example, linear quantization types include, but are not limited to, symmetric quantization and / or asymmetric quantization, with indices 0 and 1 corresponding to one of these two linear quantization types, respectively.

[0072] 2. A bitmap is used, with the number of bits being the number of linear quantization types. This number can be a default value and / or indicated by the base station. Each bit corresponds to a linear quantization type. If a bit is 1, it means that the access network device can use the linear quantization type corresponding to that bit for training the CSI compression two-sided model.

[0073] Among them, the linear quantization scaling factor indicates the scaling scale of the linear quantization for floating-point numbers used during the training of the access network equipment. It can be represented in binary, and the number of bits is a default integer value. For example, when this component can be represented in single-precision floating-point numbers, the number of bits is 32 bits.

[0074] The linear quantization offset indicates the quantization value of the floating-point zero point using linear quantization during access network equipment training. It is represented in binary, with the number of bits being a default integer value. For example, when this component is represented as an 8-bit integer, the number of bits is 8 bits. This component is omitted when the linear quantization type is symmetric quantization, where the linear quantization type is indicated by the components of the data ID and / or by the base station.

[0075] The number of quantized bits for each dimension of the quantizer input indicates the number of quantized bits for each dimension of the quantizer input used during the training of the access network device. This number of quantized bits can be represented in binary and / or indicated by an index. The number of bits is a default integer value. When an index is used, each index corresponds to an optional number of quantized bits.

[0076] The number of data units (Tokens) indicates the number of data units input to the encoder. It can be represented in binary and / or indicated by an index. The number of bits is a default integer value. When an index is used, each index corresponds to an optional number of data units.

[0077] The output dimension of each data unit indicates the output dimension corresponding to each data unit output by the encoder. It is represented in binary and / or indicated by an index. The number of bits is a default integer value. When indicated by an index, each index corresponds to an optional output dimension.

[0078] The number of blocks within the Transformer model indicates the number of Transformer blocks in the encoder, represented in binary and / or indicated by index. The number of bits is a default integer value, and when indicated by index, each index corresponds to an optional number of Transformer blocks.

[0079] The number of self-attention heads indicates the number of self-attention heads in the multi-head attention module of the Transformer block. It can be represented in binary and / or indicated by index. The number of bits is a default integer value. When indicated by index, each index corresponds to an optional number of self-attention heads.

[0080] It is worth noting that the order of the IDs of the aforementioned sub-data segments within the data ID is not restricted.

[0081] In one implementation, the IDs of at least two sub-data segments in the aforementioned data ID can form a new part to reduce overhead. This new part is indicated by an index, each index corresponding to a combination of IDs of at least two sub-data segments. The index can be represented by at least one bit, and the number of bits can be determined in at least one of the following ways: a default integer value, indicated by the constituent parts of the data ID, and / or indicated by the base station.

[0082] For example, the number of Transformer blocks and the number of self-attention heads in the data ID form a new part. If the number of optional Transformer blocks is 2, 4, 6, 8, or 10, and the number of optional self-attention heads is 4, 8, or 16, then the number of bits required before reassembly is 5 bits, and only 4 bits are needed after reassembly. The index table for the value of this new component is shown in Table 1.

[0083] Table 1 Index of values ​​for the new components

[0084] The segmented data ID (i.e., data ID composed of the IDs of sub-data segments) is sent via at least one of the following signaling methods: DCI, MAC CE, and / or RRC.

[0085] In some embodiments of this disclosure, the data ID is determined based on the Quasi-Common Model Identifier (QCM ID). The method for determining the QCM ID includes: determining the QCM ID based on independent variables and a preset function; wherein, the independent variables include at least one of the following IDs: data ID, model ID, functionality ID, feature ID, resource configuration ID, resource set ID, resource ID, and / or associated ID; and the function includes at least one of the following forms: composed of at least one subfield of the QCM ID, an encryption / decryption algorithm, and / or a neural network model employing an encoder-decoder architecture.

[0086] Specifically, after receiving the data from the access network equipment, the user equipment (UE) can train the UE-side model. The data can be from the network side (access network equipment or core network equipment) or other devices, without limitation. The training dataset in the data may include the target CSI and / or the labels output by the encoder. The UE can also receive a data identifier ID from the access network equipment, which indicates at least one of the following information about the CSI compression two-sided model trained by the access network equipment: encoder architecture, decoder architecture, and / or quantization method. The UE can autonomously select a suitable internal encoder architecture for training the UE-side model based on this information. This data ID can be obtained by solving for the QCM ID (i.e., after obtaining the QCM ID, the UE can reverse-parse the QCM ID to obtain the data ID through a function), which reduces the overhead of signaling transmission.

[0087] In one implementation, the QCM ID is obtained by calculating the independent variable (at least one ID).

[0088] In one implementation, the QCM ID is calculated by inputting the independent variable (at least one ID) into a function (f(·) function), that is, the QCM ID can be expressed as QCM ID = f(ID1, ID2, ..., ID2). S (where S is the number of IDs. The independent variable contains information as described above and will not be repeated here. The function f(·) can take at least one of the following forms:

[0089] 1. A QCM ID consists of at least one subfield of the QCM ID. Each QCM ID subfield corresponds to a set of at least one subfield of the independent variable ID. Assume that the subfield of the independent variable ID used for calculation of this QCM ID subfield is x. n If n = 1, 2, ..., N, then the QCM ID subfield can be calculated using the following formula:

[0090] In the above formula, X i ,i=1,2,…,N-1 represents the maximum value of the i-th independent variable ID subfield. The ordinal number and maximum value of each independent variable ID subfield are determined by the base station configuration and / or by a predefined method;

[0091] 2. A QCM ID consists of at least one subfield of a QCM ID, and each QCM ID subfield corresponds to at least one independent variable ID. This mapping table is determined by the base station configuration and / or by a predefined method.

[0092] 3. Employs encryption / decryption algorithms, including at least one of the following:

[0093] 1) Symmetric encryption algorithms, such as Advanced Encryption Standard (AES); and / or

[0094] 2) Asymmetric encryption algorithm, RSA (Rivest-Shamir-Adleman) algorithm;

[0095] and / or

[0096] 4. A neural network model employing an encoder-decoder architecture.

[0097] Optionally, the QCM ID is calculated only from the data ID, assuming the data ID is represented as ID. A Then QCM ID can be represented as QCM ID = f(ID) A ).

[0098] Alternatively, if the QCM ID is calculated using more than one type of ID, including the data ID, then the QCM ID can be expressed as QCM ID = f(ID). A ID1, ID2, ..., ID S-1 ), where ID A For data ID, ID s ,s=1,…,S-1 are other IDs. Optionally, if the QCM ID is composed of at least one QCM ID subfield, then each QCM ID subfield corresponds to the set of all subfields of one of the independent variable IDs or corresponds to one of the independent variable IDs.

[0099] In one implementation, the QCM ID is sent via at least one of the following signaling methods: DCI, MAC CE, and / or RRC.

[0100] Step S200: The UE updates the UE-side model based on the data (or data from the access network device) and the data ID, including training, retraining, or fine-tuning.

[0101] Step S300: Receive at least one auxiliary monitoring data first reporting configuration, wherein the auxiliary monitoring data includes monitoring indicator values ​​and / or monitoring results, and the first reporting configuration includes reporting quantity, monitoring indicator, granularity of monitoring indicator values ​​and / or monitoring indicator threshold group.

[0102] Specifically, the auxiliary monitoring data can be UE-side auxiliary monitoring data. There can be at least one first reporting configuration; one reporting configuration can correspond to one auxiliary monitoring data point, or one reporting configuration can correspond to multiple auxiliary monitoring data points. When the auxiliary monitoring data is UE-side (i.e., UE) auxiliary monitoring data, the base station can configure at least one UE auxiliary monitoring data point as the first reporting configuration for the UE, supporting access network devices in monitoring and making decisions regarding CSI compressed training performance with minimal signaling overhead.

[0103] The report quantity in the first reporting configuration is indicated in at least one of the following ways:

[0104] 1. Direct indication: Optional reporting quantities include UE-side monitoring indicator values ​​and / or UE-side monitoring results. The descriptions of each reporting quantity are as follows:

[0105] 1) If the indicated reporting quantity is a UE-side monitoring indicator value, the UE reports the monitoring indicator value calculated by the actual encoder output and the encoder output label. The actual encoder output represents the output after the target CSI is input into the UE-side model, and the encoder output label represents the encoder output provided by the access network device or the encoder output after the target CSI is input into the access network device. The monitoring indicator is configured by the base station, and the base station can use an internal evaluation algorithm to judge the performance of the UE-side model based on the monitoring indicator value.

[0106] 2) If the indicated reported quantity is the UE-side monitoring result, the UE compares the monitoring indicator value with the monitoring indicator threshold group indicated by the base station, judges the UE-side model performance, obtains the judgment result, and reports the judgment result, i.e., the monitoring result. The base station can directly know the UE-side model performance and make subsequent decisions through this monitoring result. For example, when there is only one threshold in the monitoring indicator threshold group, the reported monitoring result can use 0 to indicate abnormal UE-side model performance and 1 to indicate normal UE-side model performance.

[0107] 2. A bitmap is used, and the number of bits can be the number of reporting types. This number uses a default value and / or is determined by the UE capability. Each bit corresponds to a reporting type. If a bit is 1, it instructs the UE to report the reporting type corresponding to that bit; otherwise, it does not report.

[0108] 3. Use an index, with each index corresponding to a reporting quantity. The index can be represented by at least one bit, and the number of bits can be a default integer value and / or determined by the UE capability.

[0109] Regarding the monitoring indicator in the first reporting configuration, this parameter exists when the base station indicates that the reported quantity is a UE-side monitoring indicator value or a UE-side monitoring result. The monitoring indicator can be indicated in at least one of the following ways:

[0110] 1. Direct indication; optional monitoring indicators include, but are not limited to, at least one of the following: Squared Generalized Cosine Similarity (SGCS), Normalized Mean Squared Error (NMSE), and / or Maximum Error, wherein the Maximum Error can be expressed as... In this formula, x i Let x' represent the i-th element of the actual vector. i This represents the i-th element of the label vector;

[0111] 2. A bitmap is used, with the number of bits equal to the number of monitoring indicators. This number uses a default value and / or is determined by the UE's capabilities. Each bit corresponds to one monitoring indicator. If a bit is 1, it instructs the UE to use the monitoring indicator corresponding to that bit to calculate the monitoring indicator value; otherwise, it does not use it.

[0112] 3. An index is used, with each index corresponding to a monitoring indicator. At least one bit can be used to represent the index, with the number of bits using a default integer value and / or determined by the UE's capabilities. For example, the monitoring indicators include, but are not limited to, at least one of the following: SGCS, Normalized Mean Squared Error (NMSE), and / or maximum error. Indices 00, 01, and 10 each correspond to one of these monitoring indicators.

[0113] The monitoring metric granularity parameter indicates whether the monitoring metric is calculated for the model output of each layer corresponding to the target CSI (i.e., the precoding matrix) or for all layers when the UE-side model performs independent inference on each layer of the target CSI. It uses a 1-bit indicator; for example, 0 indicates that the monitoring metric is calculated for the model output of all layers of the target CSI, and 1 indicates that the monitoring metric is calculated for the model output corresponding to each layer of the target CSI. This parameter can be left as a default when the UE-side model performs inference on all layers of the target CSI.

[0114] The monitoring indicator threshold group is used by the UE to compare the calculated monitoring indicator values ​​with this threshold group to determine the UE-side model performance. This parameter exists when the base station indicates that the reported quantity is the UE-side monitoring result. The base station can indicate at least one set of monitoring indicator thresholds based on the granularity of the monitoring indicator values. For example, when the monitoring indicator values ​​are calculated for the model output of all layers of the target CSI, the base station indicates a set of monitoring indicator thresholds. When the monitoring indicator values ​​are calculated for the model output corresponding to each layer of the target CSI, the number of monitoring indicator threshold groups indicated by the base station is equal to the number of layers. The monitoring indicator threshold group includes at least one monitoring indicator threshold, wherein the monitoring indicator threshold is indicated in at least one of the following forms:

[0115] 1. Direct indication, for example, optional thresholds include {0, 0.1, 0.2, ..., 0.8, 0.9, 1};

[0116] 2. Use an index, where each index corresponds to a threshold. At least one bit can be used to represent the index, with the number of bits set to a default integer value. For example, with 4 bits, index values ​​0 through 11 represent {0, 0.1, 0.2, ..., 0.8, 0.9, 1} respectively.

[0117] Optionally, the first reporting configuration of the aforementioned UE auxiliary monitoring data can be configured using RRC signaling, and the CSI-ReportConfig IE can be reused or carried on a new IE during configuration.

[0118] Step S400: The UE monitors the training performance of the encoder;

[0119] Step S500: Report monitoring indicator values ​​and / or monitoring results. When the monitoring indicator threshold group contains more than one monitoring indicator threshold, the reporting method of the monitoring results includes at least one of the following: bitmap, index, and / or candidate value, to increase the accuracy and reliability of the monitoring results.

[0120] When the monitoring results are reported using bitmap, assuming the number of thresholds in the monitoring indicator threshold group is M, the number of bits is M+1. This number uses the default value and / or is configured by the base station. If the m-th bit is 1, it means that the monitoring indicator value is within the range of the (m-1)-th threshold and the m-th threshold. When the first bit is 1, it means that the monitoring indicator value is less than the first threshold. When the (M+1)-th bit is 1, it means that the monitoring indicator value is greater than the M-th threshold.

[0121] When the monitoring results are reported using an index, each index corresponds to a threshold range. At least one bit can be used to represent the index, and the number of bits can be a default integer value and / or configured by the base station. For example, the first index indicates that the monitoring index value is less than the first threshold, the second index indicates that the monitoring index value is within the range of the first and second thresholds, the third index indicates that the monitoring index value is within the range of the second and third thresholds, and so on.

[0122] When candidate values ​​are used to report monitoring results, the candidate values ​​include values ​​within the threshold range of the monitoring indicator value and / or values ​​ranging from 0 to 1. The values ​​within the threshold range can be the minimum and / or maximum values ​​of the threshold range. Candidate values ​​ranging from 0 to 1 can be obtained based on the number of thresholds. Assuming the number of thresholds is M, the selectable values ​​are...

[0123] Step S600: The access network equipment monitors and makes decisions regarding the training performance of CSI compression;

[0124] In some embodiments of this disclosure, after the method completes step S600, it executes step S700a;

[0125] Step S700a: If the monitoring result indicates that the training performance of the encoder currently running on the UE side is insufficient (or if the access network device determines that the monitoring result indicates that the training performance of the encoder currently running on the UE side is insufficient), the method further includes the access network device (e.g., a base station) sending model control information to the UE, wherein the model control information indicates at least one of the following: model activation, model deactivation, and / or model switching.

[0126] Specifically, insufficient encoder training performance refers to the monitoring index value reported by the UE not being within the range of the index value corresponding to normal training performance, or the monitoring result reported by the UE indicating that the monitoring index value is not within the range of the index value corresponding to normal training performance.

[0127] In some embodiments of this disclosure, the indication method of the model control information includes at least one of the following: consisting of an indication of activating the model and / or an additional ID, a QCM ID, and / or an updated indication of activating the model, so as to reduce the signaling overhead of the model indication.

[0128] When a base station provides instructions for a single model to manage that model, such as model activation, model deactivation, and / or model switching, the instructions may be provided in at least one of the following ways:

[0129] 1. The model indication consists of an activation model indication and / or an additional ID. The activation model indication uses a data ID or a CSI generation module ID. The forms of the data ID and CSI generation module ID can be found in step S100. A subfield of the CSI generation module ID is associated with the CSI generation module trained on the activated UE side and is indicated to the base station by the UE after model activation. The additional ID indication is one of the UE-side models trained based on the data ID corresponding to the activated CSI generation module, excluding the currently activated UE-side model. The UE-side model corresponding to the additional ID is determined through at least one of the following methods:

[0130] 1) The additional IDs corresponding to each UE-side model obtained from base station configuration training, excluding the currently activated UE-side model;

[0131] 2) The UE indicates to the base station the additional IDs corresponding to each UE-side model obtained from training, excluding the currently activated UE-side model. Thus, the corresponding UE-side model can be determined through the additional IDs indicated by the base station.

[0132] 3) The order of reporting data related to each UE-side model (excluding the currently activated UE-side model) is determined based on the reporting order of UE-side auxiliary monitoring data or UE-side monitoring data. Each additional ID corresponds to the reporting sequence number of a UE-side model's related data. The additional ID is represented in binary. For example, the earlier the UE-side model's related data is reported, the smaller the additional ID corresponding to that UE-side model. For instance, the additional ID corresponding to the first reported UE-side model is 1, the additional ID corresponding to the second reported UE-side model is 2, and so on.

[0133] 2. The model indicator is represented by a QCM ID, which is calculated from the indicator that activates the model and / or an additional ID. Let's assume the indicator that activates the model is represented by ID. B The appended ID is represented as ID. C Then QCM ID can be represented as QCM ID = f(ID) B ID C ), where the form of the f(·) function can be found in the previous text;

[0134] 3. The model indicator is the indicator of the updated active model, that is, the model is indicated by changing at least one subfield in the active model indicator, and it takes at least one of the following forms:

[0135] 1) Set the increment of the subfield. For example, if the updated subfield is the number of layers for all layers, assuming that the number of bits in this part is 4, indicating a maximum of 8 layers, then the number of bits other than the number of bits indicating the number of layers is used to indicate one of the UE-side models other than the currently activated UE-side model. The number of bits corresponding to each UE-side model is determined by at least one of the following methods: base station configuration, indication by the UE to the base station, and / or based on the reporting order of the relevant data of each UE-side model other than the currently activated UE-side model when the UE auxiliary monitoring data is reported or the UE-side monitoring data is reported. For example, the earlier the relevant data of the UE-side model is reported, the smaller the binary number of this part corresponding to that UE-side model. For example, the binary number of this part corresponding to the first reported UE-side model is 1001, the binary number of this part corresponding to the second reported UE-side model is 1010, and so on.

[0136] 2) Increase the number of bits in the subfield. For example, if the updated subfield is the number of layers for all layers, and assuming the number of bits in this part is 4, then at least 5 bits can be used to indicate one of the UE-side models other than the currently activated UE-side model. The number of bits corresponding to each UE-side model is determined by at least one of the following methods: base station configuration, UE indicating to the base station, and / or based on the reporting order of relevant data of each UE-side model other than the currently activated UE-side model when UE auxiliary monitoring data is reported or UE-side monitoring data is reported. For example, the earlier the relevant data of the UE-side model is reported, the smaller the binary number of this part corresponding to that UE-side model. For example, the binary number of this part corresponding to the first reported UE-side model is 00000, the binary number of this part corresponding to the second reported UE-side model is 00001, and so on.

[0137] Optionally, the updated model indication described above can be further represented by a QCM ID, assuming the updated model indication is represented as ID'. B Then QCM ID can be represented as QCM ID = f(ID' B ), where the form of the f(·) function can be found in the previous text;

[0138] When a base station provides instructions for a set of models to manage that set of models, such as model activation, model deactivation, and / or model switching, the instructions may be provided in at least one of the following ways:

[0139] 1. The model indication consists of an activation model indication and / or an additional ID. The activation model indication uses a data ID or a CSI generation module ID. The forms of the data ID and CSI generation module ID can be found in step S100. The subfield of the CSI generation module ID is associated with the CSI generation module trained on the activated UE side and is indicated to the base station by the UE after model activation. The additional ID indicates a set of UE-side models trained based on the data ID corresponding to the activated CSI generation module, excluding the currently activated UE-side model. This additional ID uses at least one of the following forms:

[0140] 1) It consists of at least one subfield with an additional ID, and each subfield with an additional ID corresponds to a UE-side model. The UE-side model corresponding to the additional ID subfield is determined by at least one of the following methods:

[0141] ① The additional ID subfield corresponding to each UE-side model obtained from base station configuration training, excluding the currently activated UE-side model;

[0142] ②The UE indicates to the base station the additional ID subfield corresponding to each UE-side model obtained from training, excluding the currently activated UE-side model. Thus, a set of UE-side models can be determined through the additional ID subfield in the additional ID indicated by the base station.

[0143] ③ The order of reporting data related to each UE-side model (excluding the currently activated UE-side model) is determined based on the reporting order of UE-side auxiliary monitoring data or UE-side monitoring data. Each additional ID subfield corresponds to the reporting sequence number of a UE-side model-related data. The additional ID subfield is represented in binary. For example, the earlier the UE-side model-related data is reported, the smaller the additional ID subfield corresponding to that UE-side model. For example, the additional ID subfield corresponding to the first reported UE-side model is 1, the additional ID subfield corresponding to the second reported UE-side model is 2, and so on.

[0144] 2) Each additional ID corresponds to a set of UE-side models, and this correspondence is configured by the base station and / or indicated by the UE to the base station;

[0145] 2. The model indicator is represented by a QCM ID, which is calculated from the indicator that activates the model and / or an additional ID. Let's assume the indicator that activates the model is represented by ID. B The appended ID is represented as ID. D Then QCM ID can be represented as QCM ID = f(ID) B ID D ), where the form of the f(·) function can be referred to the form of the f(·) function described in step S100;

[0146] 3. The model indicator is the indicator of the updated active model, that is, the model is indicated by changing at least one subfield in the active model indicator, and it takes at least one of the following forms:

[0147] 1) Set the increment of the subfield. For example, if the updated subfield is the number of layers for all layers, assuming that the number of bits in this part is 4, indicating a maximum of 8 layers, then the number of bits other than the number of layers is used to indicate a group of UE-side models other than the currently activated UE-side models. The number of bits corresponding to the combination of each UE-side model is indicated to the base station through base station configuration and / or by the UE to the base station;

[0148] 2) Increase the number of bits in the subfield. One example is that the updated subfield is the number of layers for all layers. Assuming that the number of bits in this part is 4, then at least 5 bits can be used to indicate a set of UE-side models other than the currently activated UE-side model. The number of bits corresponding to the combination of each UE-side model is indicated to the base station through base station configuration and / or by the UE to the base station.

[0149] Optionally, the updated model indication described above can be further represented by a QCM ID, assuming the updated model indication is represented as ID'. B Then QCM ID can be represented as QCM ID = f(ID' B), where the form of the function f(·) can be found in the previous text.

[0150] In some embodiments of this disclosure, after the method completes step S600, steps S700b to S1000 are executed.

[0151] Step S700b: If the monitoring result indicates that the CSI compression training performance is normal (or if the access network device determines that the monitoring result indicates that the CSI compression training performance is normal), the method further includes the access network device sending a second reporting configuration of the monitoring data to the UE, wherein the monitoring data includes at least one of the following: target CSI, UE-side model output, and / or UE-side quantizer output, and the second reporting configuration includes at least one of the following: reporting amount, compressed precoding matrix group size, quantized compressed precoding matrix group size, and / or time-frequency allocation of at least one physical resource carrying the monitoring data, to support the access network device in monitoring and making decisions on the generalization performance of the decoder and quantization.

[0152] The reporting volume in the second reporting configuration is indicated in at least one of the following ways:

[0153] 1. Direct indication: Optional reporting quantities include uncompressed CSI, UE-side model output, and UE-side quantizer output. The descriptions of each reporting quantity are as follows:

[0154] 1) If the reported quantity is uncompressed CSI, the UE reports at least one of the following parameters obtained from measuring CSI-RS: CSI-RS Resource Indicator (CRI), Rank Indicator (RI), Layer Indicator (LI), Precoding Matrix Indicator (PMI), and / or Channel Quality Indicator (CQI), wherein the PMI is reported according to the codebook configuration indicated by the base station;

[0155] 2) If the indicated reported quantity is the output of the UE-side model, then the UE reports at least one of the following parameters: CRI, RI, LI, compressed precoding matrix, and / or CQI, wherein the compressed precoding matrix is ​​the data output after the precoding matrix is ​​input into the UE-side model, and the base station can input the compressed precoding matrix into the base station-side model to obtain the reconstructed precoding matrix.

[0156] 3) If the indicated reported quantity is the output of the UE-side quantizer, then the UE reports at least one of the following parameters: CRI, RI, LI, quantized compressed precoding matrix, and / or CQI, wherein the quantized compressed precoding matrix is ​​the data output after the compressed precoding matrix is ​​input into the UE-side quantizer;

[0157] 2. A bitmap is used, with the number of bits representing the number of reporting types. This number uses a default value and / or is determined by the UE's capabilities. Each bit corresponds to a reporting type. If a bit is 1, it instructs the UE to report the reporting type corresponding to that bit; otherwise, it does not report.

[0158] 3. Use an index, with each index corresponding to a reporting quantity. The index can be represented by at least one bit, and the number of bits can be a default integer value and / or determined by the UE capability.

[0159] The compressed precoding matrix group size in the second reporting configuration indicates the maximum number of compressed precoding matrices that can be carried within a physical resource carrying UE-side monitoring data. This parameter can be left as a default when the UE-side model infers all layers of the target CSI. If the UE-side model infers each layer of the target CSI independently, and the number of layers is not greater than this parameter, the compressed precoding matrices corresponding to each layer are reported within a single physical resource carrying UE-side monitoring data. If the UE-side model infers each layer of the target CSI independently, and the number of layers is greater than this parameter, the compressed precoding matrices corresponding to each layer are reported within multiple physical resources carrying UE-side monitoring data. Assuming the number of layers is N... layer If the size of the compressed precoding matrix group is K, then among them Each physical resource carrying UE-side monitoring data carries K compressed precoding matrices. If N layer mod K>0, another physical resource carrying UE-side monitoring data carries N layer mod K compressed precoding matrices. The size of the compressed precoding matrix group is indicated in at least one of the following ways:

[0160] 1) Directly indicate a positive integer;

[0161] 2) A bitmap is used, with the number of bits equal to the number of candidate compressed precoding matrix group sizes. This number is determined in at least one of the following ways: predefined, determined by UE capabilities, and / or configured by the base station. Each bit corresponds to one compressed precoding matrix group size. If a bit is 1, it indicates that the UE should use the compressed precoding matrix group size corresponding to that bit; otherwise, it should not be used.

[0162] 3) An index is used, with each index corresponding to a compressed precoding matrix group size. The index can be represented by at least one bit, and the number of bits is determined in at least one of the following ways: predefined, determined by UE capabilities, and / or configured by the base station. An example of an index table for a compressed precoding matrix group size is shown in Table 2.

[0163] Table 2 Index of compressed precoding matrix group sizes

[0164] Specifically, regarding the quantized compressed precoding matrix group size in the second reporting configuration: this indicates the maximum number of quantized compressed precoding matrices that can be carried within a single physical resource carrying UE-side monitoring data. This parameter can be left as a default when the UE-side model infers all layers of the target CSI. If the UE-side model performs independent inference for each layer of the target CSI and the number of layers is not greater than this parameter, then the quantized compressed precoding matrices corresponding to each layer are reported within a single physical resource carrying UE-side monitoring data. If the UE-side model performs independent inference for each layer of the target CSI and the number of layers is greater than this parameter, then the quantized compressed precoding matrices corresponding to each layer are reported within multiple physical resources carrying UE-side monitoring data. Assuming the number of layers is N... layer If the size of the quantized compression precoding matrix group is K, then where Each physical resource carrying UE-side monitoring data carries K quantized compressed precoding matrices. If N layer mod K>0, another physical resource carrying UE-side monitoring data carries N layer The compressed precoding matrices are quantized by mod K. The size of the quantized compressed precoding matrix group is indicated in at least one of the following ways:

[0165] 1. Directly indicate a positive integer;

[0166] 2. A bitmap is used, the number of bits being the number of candidate quantized compressed precoding matrix group sizes. This number is determined in at least one of the following ways: predefined, determined by UE capabilities, and / or configured by the base station. Each bit corresponds to one quantized compressed precoding matrix group size. If a bit is 1, it instructs the UE to use the quantized compressed precoding matrix group size corresponding to that bit; otherwise, it does not use it.

[0167] 3. An index is used, with each index corresponding to a quantized compressed precoding matrix group size. The index can be represented by at least one bit, and the number of bits is determined in at least one of the following ways: predefined, determined by UE capabilities, and / or configured by the base station. An example of an index table for a quantized compressed precoding matrix group size is shown in Table 3.

[0168] Table 3 Index of Quantized Compressed Precoding Matrix Group Sizes

[0169] Specifically, for the time-frequency allocation of at least one physical resource carrying UE-side monitoring data in the second reporting configuration, the indication of the time-frequency allocation of each physical resource adopts at least one of the following methods:

[0170] 1. Direct instruction, wherein the parameters of the instruction include at least one of the following: the time slot in which the physical resource is located, the starting symbol of the physical resource in its time slot, the starting resource block of the physical resource, the number of symbols occupied by the physical resource, and / or the number of resource blocks occupied by the physical resource;

[0171] 2. Indication is based on reference physical resources, wherein the reference physical resources include at least one of the following: physical resources carrying DCI, physical resources carrying MAC CE, and / or physical resources carrying UE auxiliary monitoring data, wherein the time-frequency allocation of the physical resources carrying UE-side monitoring data includes at least one of the following parameters:

[0172] 1) The offset of the time slot where the physical resource is located relative to the time slot where the reference physical resource is located, in units of time slots. This offset value can be a predefined integer value and / or configured by at least one of the following signaling: DCI, MAC CE, and / or RRC;

[0173] 2) The starting symbol of the physical resource in its time slot, which is configured by a predefined integer value and / or by at least one of the following signaling: DCI, MAC CE, and / or RRC;

[0174] 3) The offset of the physical resource start symbol relative to the reference physical resource start symbol or the last symbol, in symbols. This offset value can be a predefined integer value and / or configured by at least one of the following signaling: DCI, MAC CE, and / or RRC;

[0175] 4) The offset of the starting resource block of the physical resource relative to the starting resource block of the reference physical resource, in units of RB. This offset value can be a predefined integer value and / or configured by at least one of the following signaling: DCI, MAC CE, and / or RRC;

[0176] 5) The number of symbols occupied by physical resources, which uses a predefined integer value and / or is configured by at least one of the following signaling: DCI, MAC CE, and / or RRC;

[0177] 6) The number of resource blocks occupied by physical resources, which is a predefined integer value and / or configured by at least one of the following signaling: DCI, MAC CE, and / or RRC.

[0178] Optionally, the second reporting configuration of the UE-side monitoring data is configured using RRC signaling, and the CSI-ReportConfig IE is reused or carried on a new IE during configuration.

[0179] Step S800: The UE runs the CSI generation module to generate the UE-side model output and / or quantizer output;

[0180] Step S900: UE reports the monitoring data to the base station based on the second reporting configuration. When the monitoring data includes two different reporting quantities, the reporting method includes at least one of the following: completely carried on different physical resource reports, completely reused on the same physical resource report, and / or part of the information carried on different physical resource reports and part of the information reused on the same physical resource report, so as to support the access network device to monitor and make decisions on the generalization performance of the decoder and quantization.

[0181] Exemplarily, assume the number of layers is N layer , the group size of the compressed precoding matrix is K1, and the group size of the quantized compressed precoding matrix is K2, where N layer mod K1=0, N layer mod K2=0, K1<K2. If two different reporting quantities are completely carried on different physical resource reports, it is required to configure physical resources for carrying the UE-side monitoring data. If two different reporting quantities are completely reused on the same physical resource report, it is required to configure physical resources for carrying the UE-side monitoring data. If for two different reporting quantities, part of the information is carried on different physical resource reports and part of the information is reused on the same physical resource report, the number of physical resources for carrying the UE-side monitoring data that needs to be configured is greater than and less than

[0182] Optionally, if the physical resource cannot carry the complete monitoring data, report the monitoring data (or UE-side monitoring data) based on a priority rule, where the priority rule is related to the layer ordinal, so as to ensure the stability of model monitoring.

[0183] Specifically, if the physical resource allocated by the base station cannot carry the complete UE-side monitoring data, at least one of the following priority rules is used to report the UE-side monitoring data:

[0184] 1. The smaller the layer ordinal, the higher the reporting priority. An example is that the UE reports the reporting quantities according to the priority order shown in Table 4

[0185] , where N layer represents the number of layers;

[0186] Table 4 UE-side monitoring data reporting priority mode 1

[0187] 2. The priority of odd-numbered layers is higher than that of even-numbered layers. An example is that the UE reports the reporting quantities according to the priority order shown in Table 5 or Table 6, where N layer represents the number of layers;

[0188] Table 5 UE-side monitoring data reporting priority method 2 (N) layer (Odd number)

[0189] Table 6 UE-side monitoring data reporting priority method 2 (N) layer (even number)

[0190] 3. Even-numbered layers have higher priority than odd-numbered layers. An example is that the UE reports the reported data according to the priority order shown in Table 7 or Table 8, where N... layer Indicates the number of layers;

[0191] Table 7 UE-side monitoring data reporting priority method 3 (N) layer (Odd number)

[0192] Table 8 UE-side monitoring data reporting priority method 3 (N) layer (even number)

[0193] 4. The UE reports the reported data according to the priority order shown in Table 9, where N layer The number of layers is indicated by D, which represents the layer sequence interval. D can be indicated in at least one of the following ways: predefined, base station configured, and / or indicated by the UE.

[0194] Table 9 UE-side monitoring data reporting priority method 4

[0195] Step S1000: The base station performs generalization performance monitoring and decision-making for CSI compression based on the received UE-side monitoring data.

[0196] It is worth noting that, in practice, there are no restrictions on the execution order of steps S100 to S700a. That is, any number of the described steps can be skipped or combined in any order to implement the method or an alternative method. Similarly, there are no restrictions on the execution order of steps S100 to S700b and steps S700b to S1000. That is, any number of the described steps can be skipped or combined in any order to implement the method or an alternative method.

[0197] Figure 3 illustrates a second flowchart of a wireless communication method based on artificial intelligence or machine learning (AI / ML) provided in this disclosure, the method comprising at least one of the following steps:

[0198] Step H100: The UE receives data and / or data identifier ID sent by the base station, wherein the data includes training dataset and / or model parameters, and the data identifier ID is composed of several sub-data segment identifiers and / or determined based on the Quasi-Common Model Identifier (QCM ID).

[0199] For details of step H100, please refer to the specific content under step S100, which will not be repeated here.

[0200] Step H200: The UE updates the UE-side model based on the access network device data and data ID, including training, retraining, or fine-tuning.

[0201] In step H300, the UE monitors the encoder's training performance.

[0202] Step H400: If the UE-side monitoring result is that the encoder training performance is normal, where normal encoder training performance means that the monitoring index value related to the encoder is within the range of index values ​​corresponding to normal training performance, the UE requests the base station to monitor the training performance of the decoder.

[0203] Step H500: The base station monitors the training performance of the decoder;

[0204] In some embodiments of this disclosure, after step H500 is completed, steps H600a, H700 to H1000 are executed.

[0205] Step H600a: If the monitoring result indicates that the decoder training performance is normal (or if the base station determines that the monitoring result indicates that the decoder training performance is normal), the method further includes a third reporting configuration for sending monitoring data, wherein the monitoring data includes at least one of the following: target CSI, UE-side model output, and / or UE-side quantizer output, and the third reporting configuration includes at least one of the following parameters: compressed precoding matrix group size, quantized compressed precoding matrix group size, and / or time-frequency allocation of at least one physical resource carrying the monitoring data.

[0206] Specifically, "normal decoder training performance" means that the monitoring metrics related to the decoder are within the range of values ​​corresponding to normal training performance. The monitoring data can be UE-side monitoring data. The specific content of the third reporting configuration is the same as that of the second reporting configuration, and will not be repeated here.

[0207] Optionally, if the monitoring result indicates that the decoder training performance is normal (or if the base station determines that the monitoring result indicates that the decoder training performance is normal), the base station sends downlink control information (DCI) for scheduling the Physical Uplink Shared Channel (PUSCH), media access control element (MAC CE), and / or radio resource control (RRC); triggering the UE to report monitoring data to the base station to support the monitoring of the generalization performance of CSI compression by the access network equipment.

[0208] When a base station sends downlink control information (DCI) to schedule the Physical Uplink Shared Channel (PUSCH), including but not limited to DCI formats 0_0, 0_1, 0_2, 0_3, and 2_X (where X is a non-negative integer), the UE-side monitoring data reporting can be triggered using at least one of the following methods:

[0209] 1. The DCI includes a field for UE-side monitoring data reporting indicator. This field indicates whether the UE needs to report monitoring data. It uses 1 bit to indicate, for example, 1 indicates that the UE reports monitoring data, and 0 indicates that the UE does not report monitoring data, which means that the training performance of the decoder on the base station side is poor.

[0210] 2. The DCI is scrambled by an RNTI used for CSI compression performance monitoring (e.g., defined as CSI Compression Performance Monitoring Radio Network Temporary Identity, CSIC-PM-RNTI), which can be configured in at least one of the following ways:

[0211] 1) Configured by RRC signaling, with a value range of 0001-FFF2; and / or

[0212] 2) Predefine a hexadecimal number.

[0213] If the UE detects a DCI scrambled by the RNTI used for CSI compression performance monitoring, it needs to report auxiliary data for CSI compression performance monitoring. This DCI includes a field indicating the auxiliary data type for CSI compression performance monitoring. This field indicates the data type reported by the UE for CSI performance monitoring and can be indicated by a 1-bit field; for example, 0 indicates the reported data type is UE auxiliary monitoring data, and 1 indicates the reported data type is UE-side monitoring data; and / or

[0214] 3. The DCI is scrambled by the RNTI used for UE-side monitoring data reporting (e.g., defined as Monitoring Data Report Radio Network Temporary Identity, MDR-RNTI), which is configured in at least one of the following ways:

[0215] 1) Configured by RRC signaling, with a value range of 0001-FFF2; and / or

[0216] 2) Predefine a hexadecimal number.

[0217] If the UE detects DCI scrambled by RNTI used for UE-side monitoring data reporting, it needs to report UE-side monitoring data.

[0218] In some embodiments of this disclosure, after the method completes step H500, steps H600b, H700 to H1000 are executed;

[0219] When the base station sends downlink control information (MAC CE) to schedule the Physical Uplink Shared Channel (PUSCH), it triggers the reporting of monitoring data on the UE side using at least one of the following methods:

[0220] 1. The base station sends at least one of the following MAC CEs: aperiodic CSI trigger state subselection MAC CE, semi-persistent CSI reporting on PUCCH activation / deactivation MAC CE, and / or enhanced SP CSI reporting on PUCCH activation / deactivation MAC CE. The MAC CE includes a field for a UE-side monitoring data reporting indicator, which indicates whether the UE needs to report monitoring data. It uses 1 bit to indicate, for example, 1 indicates that the UE reports monitoring data, and 0 indicates that the UE does not report monitoring data, which means that the training performance of the decoder on the base station side is poor.

[0221] 2. The base station sends a MAC CE for CSI compression performance monitoring. This MAC CE is identified by a MAC subheader carrying a Logical Channel Identity (LCID) or an Extended Logical Channel Identity (eLCID). Optionally, the LCID can be an integer from 35 to 46, and the eLCID can be an integer from 0 to 215. Upon receiving the MAC CE for CSI compression performance monitoring, the UE needs to report auxiliary data for CSI compression performance monitoring. This MAC CE includes a field indicating the auxiliary data type for CSI compression performance monitoring. This field indicates the data type reported by the UE for CSI performance monitoring, using a 1-bit indication; for example, 0 indicates the data type is UE auxiliary monitoring data, and 1 indicates the data type is UE-side monitoring data.

[0222] 3. The base station sends a MAC CE for UE-side monitoring data reporting. This MAC CE is identified by a MAC sub-header carrying LCID or eLCID. Optionally, LCID can be an integer from 35 to 46, and eLCID can be an integer from 0 to 215. This MAC CE uses a fixed size of 0 bits. When the UE receives this MAC CE, it needs to report the UE-side monitoring data.

[0223] When a base station sends a Line Resource Control (RRC) message to schedule the Physical Uplink Shared Channel (PUSCH), the RRC signaling includes at least one of the following Information Elements (IEs):

[0224] 1. UE-side monitoring data reporting indicator: This IE indicates whether the UE needs to report monitoring data, and can take at least one of the following forms:

[0225] 1) Boolean data type, for example, 1 indicates that the UE reports monitoring data, and 0 indicates that the UE does not report monitoring data, which means that the training performance of the decoder on the base station side is poor;

[0226] 2) Enumerated data types, such as 'enabled' indicating that the UE should report monitoring data;

[0227] 2. CSI Compression Performance Monitoring Indicator, which indicates that the DCI contains at least one of the following fields to indicate the auxiliary data type reported by the UE for CSI performance monitoring and / or whether auxiliary data needs to be reported:

[0228] 1) Fields of the UE-side monitoring data reporting indicator;

[0229] 2) Fields of CSI compression performance monitoring auxiliary data types;

[0230] For an explanation of the above fields, see the method of DCI triggering UE-side monitoring data reporting when the base station sends scheduling PUSCH to the UE;

[0231] In some embodiments of this disclosure, after the method completes step H500, steps H600b, H700 to H1000 are executed;

[0232] Step H600b: Configure the timer for the UE; when the UE receives the indication information of the base station reporting monitoring data, stop the timer counting to reduce the overhead of signaling indication.

[0233] Specifically, the UE configures a timer, which starts running on at least one of the following events:

[0234] 1. The UE begins monitoring the training performance of the UE-side model (encoder); and / or

[0235] 2. The UE requests the access network equipment to monitor the training performance of the base station-side model (decoder);

[0236] The timer stops running when the UE receives an instruction from the base station to report UE-side monitoring data. This instruction can be carried by at least one of the following signaling methods: DCI, MAC CE, and / or RRC. If the timer expires, the UE will not report UE-side monitoring data, indicating poor training performance of CSI compression.

[0237] Step H700: Based on the third reporting configuration, report monitoring data. When the monitoring data contains two different reporting quantities, the reporting method includes at least one of the following: reporting entirely on different physical resources, reporting entirely on the same physical resource, and / or reporting some information on different physical resources and reusing some information on the same physical resource.

[0238] Specifically, the specific content of the third reported configuration refers to the specific content of the second reported configuration, while the specific content of step H700 refers to the specific content of step S900.

[0239] Optionally, if the physical resources cannot support the complete monitoring data, the monitoring data (or UE-side monitoring data) is reported based on priority rules, wherein the priority rules are related to the layer ordinal number to ensure the stability of model monitoring. Details of this part are as described above and will not be repeated here.

[0240] Step H800: The base station performs generalization performance monitoring of CSI compression based on the received monitoring data from the UE side;

[0241] Step H900, the method further includes: the UE receiving a precoding reference signal and / or quantization generalization performance monitoring result sent by the base station, wherein the indication method of the quantization generalization performance monitoring result includes at least one of the following: implicit indication via RRC indication, via MAC CE indication, via DCI indication, and / or via change of time-frequency parameters of the precoding reference signal, to associate the quantization generalization performance monitoring result and the precoding reference signal, the precoding reference signal being used for monitoring the generalization performance of the codec, wherein the quantization generalization performance monitoring result is associated with the precoding reference signal.

[0242] Specifically, when a base station sends quantization generalization performance monitoring results via RRC signaling, it can use at least one of the following methods:

[0243] 1. Indicate the quantization generalization performance monitoring result in at least one of the following IEs: CSI-ResourceConfig IE, NZP-CSI-RS-ResourceSet IE, and / or NZP-CSI-RS-Resource IE; and / or

[0244] 2. Indicate the quantization generalization performance monitoring results in the RRC signaling, and associate the corresponding precoded CSI-RS resource through at least one of the following IEs: CSI-ResourceConfigId IE, NZP-CSI-RS-ResourceSetId IE, and / or NZP-CSI-RS-ResourceId IE;

[0245] When the base station sends SP CSI-RS / CSI-IM Resource Set Activation / Deactivation MAC CE to the UE, the MAC CE contains a field that carries the quantization generalization performance monitoring result to indicate the quantization generalization performance monitoring result. At the same time, the SP CSI-RS resource set ID field indicates the precoded CSI-RS resource associated with the monitoring result.

[0246] When the base station sends DCI to the UE, it indicates the quantization generalization performance monitoring results using at least one of the following methods:

[0247] 1. The transmitted DCI is the DCI for scheduling PUSCH, including but not limited to DCI format 0_0, DCI format 0_1, DCI format 0_2, DCI format 0_3, and DCI format 2_X (X is a non-negative integer). This DCI includes a field carrying the quantization generalization performance monitoring result to indicate the quantization generalization performance monitoring result, and indicates the precoded CSI-RS resource associated with the monitoring result through the CSI request field; and / or

[0248] 2. The transmitted DCI is the DCI for scheduling PDSCH, including but not limited to DCI format 1_0, DCI format 1_1, DCI format 1_2, DCI format 1_3, and DCI format 2_X (X is a non-negative integer). This DCI includes a field carrying the quantization generalization performance monitoring result to indicate the quantization generalization performance monitoring result, and associates the quantization generalization performance monitoring result with the precoded reference signal using at least one of the following methods:

[0249] 1) The monitoring result is correlated with the precoded DMRS corresponding to the PDSCH scheduled by the DCI; and / or

[0250] 2) The CSI-RS resource index indicates the precoded CSI-RS resource associated with the monitoring result through a field that may indicate at least one of the following parameters: CSI resource configuration ID, CSI-RS resource set ID, and / or CSI-RS resource ID;

[0251] 3. The DCI sent is the DCI for scheduling PDSCH, including but not limited to DCI format 0_0, DCI format 0_1, DCI format 0_2, DCI format 0_3, and DCI format 2_X (X is a non-negative integer). The DCI schedules at least two physical resources carrying PDSCH, one of which carries the quantization generalization performance monitoring result, and associates the monitoring result with the precoding DMRS corresponding to the other scheduled PDSCH.

[0252] When the UE determines the quantization generalization performance judgment result indicated by the network side through blind detection of the precoding reference signal, the overhead of the indication is reduced. If the UE detects the precoding reference signal in the time slot, symbol, or resource element (RE) configured by the base station, it indicates that the network side indicates the first judgment result, which optionally indicates that the quantization generalization performance is normal. If the time-frequency parameters corresponding to the precoding reference signal detected by the UE are offset from the time-frequency parameters configured by the base station, wherein the time-frequency parameters include at least one of the following: time slot, start symbol, start resource block, and / or start resource element, and the offset value is predefined or configured by the base station, it indicates that the network side indicates the second judgment result, which optionally indicates that the quantization generalization performance is abnormal.

[0253] In step H1000, the UE performs CSI compression generalization performance monitoring and decision-making based on the received precoded reference signal and / or quantization generalization performance monitoring results.

[0254] It is worth noting that there are no restrictions on the execution order of steps H100 to H600a and steps H600a to H1000, nor are there any restrictions on the execution order of steps H100 to H600b and steps H600b to H1000. In other words, any number of the described steps can be skipped or combined in any order to implement the method or alternative methods.

[0255] The above method can achieve at least one of the following technical effects: segmented data IDs and QCM-based data IDs support the UE in selecting an appropriate model architecture for UE-side model training, while ensuring quantization matching between the UE and network sides. The QCM-based approach... The data ID of the ID can further reduce signaling overhead; the UE-assisted monitoring data reporting configuration and reporting method solve the problem of high overhead when directly reporting encoder output during network-side decision-making, supporting network-side monitoring and decision-making for CSI compression training performance with less signaling overhead; the triggering mechanism and reporting configuration and method of UE-side monitoring data reporting can support network-side monitoring of CSI compression generalization performance, where the timer-based triggering mechanism further reduces signaling indication overhead, and the reporting configuration and method of UE-side monitoring data ensure the integrity of monitoring data and the stability of model monitoring; the proposed indication method for network-side assisted quantization generalization performance monitoring results can associate quantization generalization performance monitoring results with precoding reference signals, supporting UE-side monitoring and decision-making for CSI compression generalization performance, while improving the accuracy of encoder-decoder generalization performance monitoring, where the UE blind detection of precoding reference signals further reduces indication overhead; the indication method for model control solves the problem of high signaling overhead when using model ID indication, supporting model ID-based LCM with less overhead.

[0256] Figure 4 illustrates a third flowchart of a wireless communication method based on artificial intelligence or machine learning (AI / ML) provided in this disclosure, executed in a first device, the method comprising at least one of the following steps:

[0257] Step A100: The first device receives a first portion of data sent by the third device and / or a second portion of data sent by the second device, wherein the first portion of data includes at least one of the following: channel measurement data, quality indication of channel measurement data, and / or timestamp of channel measurement data; the second portion of data includes at least one of the following: tag, quality indication of tag, and / or timestamp information of tag;

[0258] Specifically, the first device can be a base station, a UE, a core network device, or a device outside the communication system; there are no limitations on this. The second part of the data can be sent by the second device, and the first part of the data can be sent by the third device. The second and third devices can be the same or different; there are no restrictions on this. The second or third device can be at least one of the following: a positioning reference unit (PRU), a user equipment that is not a positioning reference unit (PRU), a transmission and reception point (TRP) / base station, and / or a location management function (LMF). For example, if the second and third devices are different, the second device can be a TRP / gNB, and the second device can be a PRU, a non-PRU UE, or an LMF. The first and second parts of the data are used for subsequent data matching.

[0259] Step A200: Determine the matching degree between the first part of the data and the second part of the data based on the timestamp of the channel measurement data and the timestamp information of the tag.

[0260] Specifically, based on the timestamp of the channel measurement data of the first part of the data and the timestamp of the label of the second part of the data, the matching degree between the first part of the data and the second part of the data can be determined. The matching degree can be either a match or a mismatch. Then, the matched first part of the data and the second part of the data can be used as samples for model training or model monitoring to improve the reliability of model training or model monitoring.

[0261] In some embodiments of this disclosure, the matching degree between the first part of the data and the second part of the data is determined based on the timestamp of the channel measurement data of the first part of the data and the timestamp of the tag of the second part of the data. The first part of the data and the second part of the data are considered to match when the timestamp corresponding to the first part of the data is within the valid time period of the tag corresponding to the second part of the data.

[0262] In some embodiments of this disclosure, the timestamp information of the tag includes at least one of the following: timestamp, timestamp distribution, tag valid start timestamp, tag valid end timestamp, and / or tag valid duration.

[0263] Optionally, the tag validity period is used to calculate the tag validity time period.

[0264] Optionally, the timestamp information of the tag is used to calculate the effective time period of the tag.

[0265] The timestamp indicates the timestamp when the tag was generated;

[0266] The timestamp distribution is used to indicate the position of the sent timestamp within the tag's valid time period. Optional timestamp distributions include at least one of the following:

[0267] 1) The timestamp sent is the maximum value of the tag's valid time period, that is, the entity that generates the second part of the data records or sends the tag before the change and the corresponding timestamp when the tag changes;

[0268] 2) The sent timestamp is the minimum value of the tag's valid time period, that is, the entity that generates the second part of the data records or sends the changed tag and the corresponding timestamp when the tag changes;

[0269] and / or

[0270] 3) The timestamp sent is within the valid time period of the tag (including the maximum and minimum values). That is, the entity that generates the second part of the data may not necessarily record or send the tag and the corresponding timestamp when the tag changes.

[0271] The effective start timestamp of the tag indicates the timestamp when the tag was first generated;

[0272] Among them, the tag's valid end timestamp indicates the timestamp when the tag was last generated;

[0273] The validity period of the label shall take at least one of the following forms:

[0274] 1) When sending a tag's validity period, if a timestamp distribution exists, and the sent timestamp represents the maximum value of the tag's valid time period, then the entire validity period before the sent timestamp is the tag's valid time period. If the timestamp distribution represents the minimum value of the tag's valid time period, then the entire validity period after the sent timestamp is the tag's valid time period. If a tag's valid start timestamp exists, the entire validity period after that timestamp is the tag's valid time period. If a tag's valid end timestamp exists, the entire validity period before that timestamp is the tag's valid time period.

[0275] 2) Send the first tag validity duration and the second tag validity duration. When the timestamp distribution exists, if the timestamp distribution is such that the sent timestamp is within the tag validity period (including the maximum and minimum values), then the range of the first tag validity duration before the sent timestamp is the tag validity period, and the range of the second tag validity duration after the sent timestamp is also the tag validity period.

[0276] The validity period of the above tags can be a non-negative integer or a non-negative real number.

[0277] Optionally, the unit is at least one of the following: symbol, time slot, frame, nanosecond, microsecond, millisecond, second, minute, and / or hour, represented in binary and / or indicated by index;

[0278] Optionally, each index may have an optional validity period.

[0279] Optionally, the above-mentioned effective start timestamp and effective end timestamp of the label can together form the effective time period of the label.

[0280] This article describes a wireless communication method based on artificial intelligence or machine learning (AI / ML), applicable to wireless communication such as communication between a user equipment (UE) and a base station. However, these inventive concepts, methods, apparatuses, devices, computer-readable storage media, chips, and computer program products are not limited to wireless communication and can be extended to other communication scenarios to achieve the same technical benefits and effects.

[0281] In these scalable communication scenarios, the primary device can be a User Equipment (UE), a base station (such as a gNB, eNodeB, Transmitter Receiving Point (TRP), a NodeB for next-generation communication, or a Wi-Fi access point), or a network element. A User Equipment (UE) is a device used for communication at the user end, such as a mobile phone; it can also be called a terminal, mobile station, or mobile terminal. A UE can be a variety of devices, including but not limited to mobile phones, tablets, virtual reality (VR) devices, augmented reality (AR) devices, wireless terminals for industrial control, wireless terminals for autonomous driving, wireless terminals for remote medical surgery, wireless terminals for smart grids, wireless terminals for environmental monitoring, wireless terminals for smart cities, and wireless terminals for smart homes, etc.

[0282] Furthermore, UEs and base stations can be deployed in various environments, including but not limited to indoor, outdoor, handheld devices, vehicle-mounted devices, or even on water, in the air, on airplanes, drones, or on satellites.

[0283] Therefore, although this document describes methods and devices for wireless communication, the inventive concepts and techniques contained herein can be extended to other communication scenarios and are expected to achieve the same technical benefits and effects. It is readily apparent that these inventive concepts have broad applicability and scalability, whether for communication between different types of base stations and user equipment, or for communication in different deployment environments.

[0284] It should be noted that the above steps are merely examples and do not limit the scope of the invention. Various modifications and variations can be made to the steps without departing from the spirit and scope of the invention.

[0285] The order of the described steps (signaling / boxes) is not intended to be construed as a limitation, and any number of the described steps (signaling / boxes) can be skipped or combined in any order to implement the method or an alternative method.

[0286] This disclosure describes examples of communication between terminals and network element components in the network architecture described in the above embodiments, which are primarily for illustrative purposes and not for limitation.

[0287] The order of the described steps (signaling / blocks) is not intended to be construed as limiting, and any number of the described steps (signaling / blocks) can be skipped or combined in any order to implement the method or alternative methods. Generally, any of the components, modules, methods, and operations described herein can be implemented using software, firmware, hardware (e.g., fixed logic circuitry), manual processing, or any combination thereof. Some operations of the example methods can be described in the general context of executable instructions stored on computer-readable storage located locally and / or remotely on a computer processing system, and implementations can include software applications, programs, functions, etc. Alternatively or additionally, any functionality described herein can be performed at least in part by one or more hardware logic components, such as, but not limited to, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), etc.

[0288] Furthermore, the signaling transmission described in the embodiments of this disclosure can be implemented in any manner known in the art. For example, signaling transmission can be explicit and / or implicit. Moreover, the illustrated steps (signaling / blocks) are for illustrative purposes only and are not intended to limit this application.

[0289] Figure 5 is a schematic structural diagram of a wireless communication device 900 provided in this disclosure. The wireless communication device includes a processor and a memory, the memory for storing computer programs, and the processor for calling and running the computer programs stored in the memory, executing instructions for at least one of the operations described above.

[0290] The wireless communication device can be a user equipment, a base station, or a network element. The wireless communication device 900 shown in Figure 5 includes a processor 910, which can call and run computer programs from memory to implement the methods in the embodiments of this application.

[0291] Optionally, as shown in FIG5, the wireless communication device 900 may further include a memory 920. The processor 910 can call and run computer programs from the memory 920 to implement the methods in the embodiments of this application. The memory 920 may be a separate device independent of the processor 910, or it may be integrated into the processor 910.

[0292] Optionally, as shown in Figure 5, the wireless communication device 900 may further include a transceiver 930. The processor 910 can control the transceiver 930 to communicate with other devices. Specifically, it can send information or data to other devices or receive information or data sent by other devices. The transceiver 930 may include a transmitter and a receiver. The transceiver 930 may further include an antenna, and the number of antennas may be one or more.

[0293] Optionally, the wireless communication device 900 may specifically be a base station in the embodiments of this application, and the wireless communication device 900 may implement the corresponding processes implemented by the base station in the various methods of the embodiments of this application. For the sake of brevity, it will not be described in detail here.

[0294] Optionally, the wireless communication device 900 may specifically be a mobile user equipment / user equipment in the embodiments of this application, and the wireless communication device 900 may implement the corresponding processes implemented by the mobile user equipment / user equipment in the various methods of the embodiments of this application. For the sake of brevity, it will not be described in detail here.

[0295] Optionally, the wireless communication device 900 may specifically be a network element in the embodiments of this application, and the wireless communication device 900 may implement the corresponding processes implemented by the network element in the various methods of the embodiments of this application. For the sake of brevity, it will not be described in detail here.

[0296] According to an example embodiment, a chip is provided, the chip including: a processor for calling and running a computer program from a memory, causing a device on which the chip is installed to perform the method according to any one of the above embodiments, examples, or example embodiments.

[0297] According to an example embodiment, a computer-readable storage medium is provided for storing a computer program that causes a computer to perform a method according to any one of the above embodiments, examples, or example embodiments.

[0298] According to an example embodiment, a computer program product is provided, including a computer program / instructions that, when executed by a processor (e.g., by the processor or an apparatus, device, computer, or machine including the processor), implement the method according to any one of the above embodiments, examples, or example embodiments.

[0299] Embodiments of this disclosure are combinations of technologies / processes that can be employed in 3GPP specifications to create a final product.

[0300] While this disclosure has been described in conjunction with what are considered to be the most practical and preferred embodiments, it should be understood that this disclosure is not limited to the disclosed embodiments, but is intended to cover various arrangements made without departing from the broadest interpretation of the appended claims.

Claims

A wireless communication method based on artificial intelligence or machine learning (AI / ML), executed in a terminal device, the method comprising: Receive data and / or data identifier ID, wherein the data includes training dataset and / or model parameters, and the data identifier ID is composed of several sub-data segment identifiers and / or determined based on the Quasi-Common Model Identifier (QCM ID). According to the method of claim 1, wherein, The sub-data segment includes at least one of the following: model type, number of layers, number of neurons, linear quantization type, linear quantization scaling factor, linear quantization offset, number of quantization bits corresponding to each dimension of the input quantizer, number of data units, output dimension of each data unit, number of blocks within the transformer model, and / or number of self-attention heads. According to the method of claim 1, wherein, The methods for determining the QCM ID include: The QCM ID is determined based on independent variables and a preset function; wherein the independent variables include at least one of the following IDs: data ID, model ID, function ID, feature ID, resource configuration ID, resource set ID, resource ID, and / or association ID, and the function includes at least one of the following forms: composed of at least one subfield of QCM ID, encryption / decryption algorithm, and / or a neural network model using an encoder-decoder architecture. According to the method of claim 1, wherein, The method further includes a first reporting configuration for receiving at least one auxiliary monitoring data, wherein the auxiliary monitoring data includes monitoring indicator values ​​and / or monitoring results, and the first reporting configuration includes reporting quantity, monitoring indicators, granularity of monitoring indicator values, and / or monitoring indicator threshold groups. The method according to claim 4, wherein, The method further includes reporting monitoring indicator values ​​and / or monitoring results, wherein when the monitoring indicator threshold group contains more than one threshold, the reporting method of the monitoring results includes at least one of the following: bitmap, index, and / or candidate value. According to the method of claim 1, wherein, The method further includes receiving model control information, wherein the model control information indicates at least one of the following: model activation, model deactivation, and / or model switching. The method according to claim 6, wherein, The indication method of the model control information includes at least one of the following: consisting of an indication of the activated model and / or an additional ID, a QCM ID, and / or an updated indication of the activated model. According to the method of claim 1, wherein, The method further includes a second reporting configuration for receiving monitoring data, wherein the monitoring data includes at least one of the following: target CSI, UE-side model output, and / or UE-side quantizer output, and the second reporting configuration includes at least one of the following: reporting amount, compressed precoding matrix group size, quantized compressed precoding matrix group size, and / or time-frequency allocation of at least one physical resource carrying the monitoring data. The method according to claim 8, wherein, The method further includes reporting monitoring data based on the second reporting configuration. When the monitoring data contains two different reporting volumes, the reporting method includes at least one of the following: reporting entirely on different physical resources, reporting entirely on the same physical resource, and / or reporting some information on different physical resources and reusing some information on the same physical resource. The method according to claim 8, wherein, If the physical resources cannot support the complete monitoring data, the monitoring data is reported based on priority rules, wherein the priority rules are related to the ordinal number of the layer. A wireless communication method based on artificial intelligence or machine learning (AI / ML), executed in a terminal device, the method comprising: Receive data and / or data identifier ID, wherein the data includes training dataset and / or model parameters, and the data identifier ID is composed of several sub-data segment identifiers and / or determined based on the Quasi-Common Model Identifier (QCM ID). The method according to claim 11, wherein, The sub-data segment includes at least one of the following: model type, number of layers, number of neurons, linear quantization type, linear quantization scaling factor, linear quantization offset, number of quantization bits corresponding to each dimension of the input quantizer, number of data units, output dimension of each data unit, number of blocks within the transformer model, and / or number of self-attention heads. The method according to claim 11, wherein, The methods for determining the QCM ID include: The QCM ID is determined based on independent variables and a preset function; wherein the independent variables include at least one of the following IDs: data ID, model ID, function ID, feature ID, resource configuration ID, resource set ID, resource ID, and / or association ID, and the function includes at least one of the following forms: composed of at least one subfield of QCM ID, encryption / decryption algorithm, and / or a neural network model using an encoder-decoder architecture. The method according to claim 11, wherein, The method further includes a third reporting configuration for receiving monitoring data, wherein the monitoring data includes at least one of the following: target CSI, UE-side model output, and / or UE-side quantizer output, and the third reporting configuration includes at least one of the following parameters: compressed precoding matrix group size, quantized compressed precoding matrix group size, and / or time-frequency allocation of at least one physical resource carrying the monitoring data. The method according to claim 11 or 14, wherein, The method further includes: Receive and schedule downlink control information (DCI) for the Physical Uplink Shared Channel (PUSCH), media access control element (MAC CE), and / or radio resource control (RRC); Report monitoring data. The method according to claim 11, wherein, The method further includes: Configure the timer; When an instruction to report monitoring data is received, the timer stops counting. The method according to claim 14, wherein, The method further includes: reporting monitoring data based on the third reporting configuration, wherein when the monitoring data contains two different reporting volumes, the reporting method includes at least one of the following: reporting entirely on different physical resources, reusing entirely on the same physical resource, and / or some information being reported on different physical resources and some information being reused on the same physical resource. The method according to claim 17, wherein, If the physical resources cannot support the complete monitoring data, the monitoring data is reported based on priority rules, wherein the priority rules are related to the ordinal number of the layer. The method according to claim 11, wherein, The method further includes: Receive a precoded reference signal and / or quantization generalization performance monitoring results, wherein the indication method of the quantization generalization performance monitoring results includes at least one of the following: indication via RRC, indication via MAC CE, indication via DCI, and / or implicit indication via changes in the time-frequency parameters of the precoded reference signal, wherein the quantization generalization performance monitoring results are associated with the precoded reference signal. A wireless communication method based on artificial intelligence or machine learning (AI / ML), executed in an access network device, the method comprising: Send data and / or data identifier ID, wherein the data includes training dataset and / or model parameters, and the data identifier ID is composed of several sub-data segment identifiers and / or determined based on the Quasi-Common Model Identifier (QCM ID). The method according to claim 20, wherein, The sub-data segment includes at least one of the following: model type, number of layers, number of neurons, linear quantization type, linear quantization scaling factor, linear quantization offset, number of quantization bits corresponding to each dimension of the input quantizer, number of data units, output dimension of each data unit, number of blocks within the transformer model, and / or number of self-attention heads. The method according to claim 20, wherein, The methods for determining the QCM ID include: The QCM ID is determined based on independent variables and a preset function; wherein the independent variables include at least one of the following IDs: data ID, model ID, function ID, feature ID, resource configuration ID, resource set ID, resource ID, and / or related ID; and the function includes at least one of the following forms: consisting of at least one subfield of the QCM ID, an encryption / decryption algorithm, and / or a neural network model using an encoder-decoder architecture. The method according to claim 20, wherein, The method further includes a first reporting configuration for sending at least one auxiliary monitoring data, wherein the auxiliary monitoring data includes monitoring indicator values ​​and / or monitoring results, and the first reporting configuration includes reporting quantity, monitoring indicators, granularity of monitoring indicator values, and / or monitoring indicator threshold groups. The method according to claim 23, wherein, The method further includes receiving monitoring index values ​​and / or monitoring results, wherein when the monitoring index threshold group contains more than one threshold, the reporting method of the monitoring results includes at least one of the following: bitmap, index, and / or candidate value. The method according to claim 24, wherein, If the monitoring result indicates that the training performance of the encoder currently running on the UE side is insufficient, the method further includes sending model control information, wherein the model control information indicates at least one of the following: model activation, model deactivation, and / or model switching. The method according to claim 25, wherein, The indication method of the model control information includes at least one of the following: consisting of an indication of the activated model and / or an additional ID, a QCM ID, and / or an updated indication of the activated model. The method according to claim 24, wherein, If the monitoring result indicates that the CSI compression training performance is normal, the method further includes a second reporting configuration for sending monitoring data, wherein the monitoring data includes at least one of the following: target CSI, UE-side model output, and / or UE-side quantizer output, and the second reporting configuration includes at least one of the following: reporting amount, compressed precoding matrix group size, quantized compressed precoding matrix group size, and / or time-frequency allocation of at least one physical resource carrying the monitoring data. The method according to claim 27, wherein, The method also includes receiving monitoring data. The method according to claim 27, wherein, The monitoring data is received, wherein the monitoring data is determined based on a priority rule, and the priority rule is related to the ordinal number of the layer. A wireless communication method based on artificial intelligence or machine learning (AI / ML), executed in an access network device, the method comprising: Send data and / or data identifier ID, wherein the access network device data includes training datasets and / or model parameters, and the data identifier ID is composed of several sub-data segment identifiers and / or determined based on the Quasi-Common Model Identifier (QCM ID). The method according to claim 30, wherein, The sub-data segment includes at least one of the following: model type, number of layers, number of neurons, linear quantization type, linear quantization scaling factor, linear quantization offset, number of quantization bits corresponding to each dimension of the input quantizer, number of data units, output dimension of each data unit, number of blocks within the transformer model, and / or number of self-attention heads. The method according to claim 30, wherein, The methods for determining the QCM ID include: The QCM ID is determined based on independent variables and a preset function; wherein the independent variables include at least one of the following IDs: data ID, model ID, function ID, feature ID, resource configuration ID, resource set ID, resource ID, and / or related ID; and the function includes at least one of the following forms: composed of at least one subfield of QCM ID, encryption / decryption algorithm, and / or a neural network model using an encoder-decoder architecture. The method according to claim 30, wherein, If the monitoring result indicates that the decoder training performance is normal, the method further includes a third reporting configuration for sending monitoring data, wherein the monitoring data includes at least one of the following: target CSI, UE-side model output, and / or UE-side quantizer output, and the third reporting configuration includes at least one of the following parameters: compressed precoding matrix group size, quantized compressed precoding matrix group size, and / or time-frequency allocation of at least one physical resource carrying the monitoring data. The method according to claim 30 or 33, wherein, The method further includes: Send downlink control information (DCI) for scheduling the Physical Uplink Shared Channel (PUSCH), media access control element (MAC CE), and / or radio resource control (RRC); Receive monitoring data. The method according to claim 30, wherein, The method further includes: Send instruction information to the terminal device to report monitoring data. The method according to claim 33, wherein, The method also includes receiving monitoring data. The method according to claim 36, wherein, The monitoring data is received, wherein the monitoring data is determined based on a priority rule, and the priority rule is related to the ordinal number of the layer. The method according to claim 30, wherein, The method further includes: Send a precoded reference signal and / or quantization generalization performance monitoring results, wherein the indication method of the quantization generalization performance monitoring results includes at least one of the following: indication via RRC, indication via MAC CE, indication via DCI, and / or implicit indication via changes in the time-frequency parameters of the precoded reference signal, wherein the quantization generalization performance monitoring results are associated with the precoded reference signal. A wireless communication method based on artificial intelligence or machine learning (AI / ML), executed in a first device, the method comprising: Receive a first part of data and / or a second part of data, wherein the first part of data includes at least one of the following: channel measurement data, quality indication of channel measurement data, and / or timestamp of channel measurement data; the second part of data includes at least one of the following: tag, quality indication of tag, and / or timestamp information of tag. The method according to claim 39, wherein, The method further includes: determining the matching degree between the first part of data and the second part of data based on the timestamp of the channel measurement data and the timestamp information of the tag. The method according to claim 40, wherein, Determining the matching degree between the first part of data and the second part of data based on the timestamp of the channel measurement data and the timestamp information of the tag includes: When the timestamp corresponding to the first part of the data is within the valid time period of the tag corresponding to the second part of the data, the first part of the data and the second part of the data are matched. The method according to claim 39, wherein, The timestamp information of the tag includes at least one of the following: timestamp, timestamp distribution, tag valid start timestamp, tag valid end timestamp, and / or tag valid duration. The method according to claim 42, wherein, The tag validity period is used to calculate the effective time period of the tag. The method according to claim 42, wherein, The timestamp information of the tag is used to calculate the effective time period of the tag. The method according to claim 42, wherein, The effective duration of the tag is a non-negative integer or a non-negative real number. The method according to claim 42, wherein, The unit of the effective duration of the tag includes at least one of the following: symbol, time slot, frame, nanosecond, microsecond, millisecond, second, minute, and / or hour. The method according to claim 42, wherein, The validity period of the tag is represented in binary and / or indicated by an index. The method according to claim 41, wherein, The effective time period of the tag includes the effective start timestamp of the tag and / or the effective end timestamp of the tag. A wireless communication method based on artificial intelligence or machine learning (AI / ML), executed in a second device, the method comprising: Send a second part of the data; wherein the second part of the data includes at least one of the following: a tag, a tag quality indicator, and / or a tag timestamp information. The method according to claim 49, wherein, The timestamp information of the tag includes at least one of the following: timestamp, timestamp distribution, tag valid start timestamp, tag valid end timestamp, and / or tag valid duration. The method according to claim 50, wherein, The tag validity period is used to calculate the effective time period of the tag. The method according to claim 50, wherein, The timestamp information of the tag is used to calculate the effective time period of the tag. The method according to claim 50, wherein, The effective duration of the tag is a non-negative integer or a non-negative real number. The method according to claim 50, wherein, The unit of the effective duration of the tag includes at least one of the following: symbol, time slot, frame, nanosecond, microsecond, millisecond, second, minute, and / or hour. The method according to claim 50, wherein, The validity period of the tag is represented in binary and / or indicated by an index. The method according to claim 51 or 52, wherein, The effective time period of the tag includes the effective start timestamp of the tag and / or the effective end timestamp of the tag. A wireless communication device, wherein, The wireless communication device includes a processor and a memory for storing computer programs, the processor for calling and running the computer programs stored in the memory to perform the method as described in any one of claims 1 to 56. A readable storage medium for storing a computer program that is invoked and executed by a processor to perform the method as described in any one of 1-56.

Images