Csi reporting method, information processing method, device, equipment and medium

Abstract

The application provides a CSI reporting method, an information processing method, a device, equipment and a medium. The method of the application comprises: obtaining first CSI by compressing target CSI through a first model; grouping the first CSI according to a priority relationship of each output dimension of the first model; when the terminal needs to discard the reported CSI, reporting part of the first CSI according to the priority corresponding to each group of CSI; or, grouping the target CSI according to a preset grouping mode, obtaining second CSI by compressing each group of CSI obtained through the first model; and when the terminal needs to discard the reported CSI, reporting part of the second CSI according to a first priority rule. The application solves the problem that the priority grouping and UCI discard rule of the CSI reporting based on model compression are not clear, and guarantees the performance of the CSI reporting based on model compression under the premise that the terminal does not need to repeatedly calculate CSI.

Description

Technical Field

[0001] This application relates to the field of communication technology, and in particular to a CSI reporting method, information processing method, apparatus, device and medium. Background Technology

[0002] When a terminal reports Channel State Information (CSI), if the terminal needs to discard some of the reported CSI information, it needs to discard a portion of the CSI information to ensure the effective transmission of the remaining CSI.

[0003] CSI reporting based on Artificial Intelligence (AI) or Machine Learning (ML) models suffers from several drawbacks. The CSI generation part (the terminal-side model for CSI compression) outputs a low-dimensional vector or bit string obtained after compressing the target CSI, making it impossible to distinguish the priority of each bit within this low-dimensional vector or bit string. Furthermore, if the target CSI involves joint compression of multiple subbands, the CSI generation part cannot differentiate the correspondence between elements in the low-dimensional vector or bit string and the subbands. Therefore, in AI / ML model-based CSI reporting, priority grouping and Uplink Control Information (UCI) discarding rules cannot be defined as in existing codebook-based CSI reporting; in other words, existing priority and discarding rules are no longer applicable. Summary of the Invention

[0004] The purpose of this application is to provide a CSI reporting method, information processing method, apparatus, equipment and medium to solve the problem of unclear priority grouping and UCI discarding rules in CSI reporting based on model compression.

[0005] To achieve the above objectives, in a first aspect, embodiments of this application provide a CSI reporting method applied to a terminal, comprising:

[0006] The target channel state information (CSI) is compressed using a first model to obtain a first CSI; the first CSI is then grouped according to the priority relationships of each output dimension of the first model; when the terminal needs to discard reported CSIs, it reports a portion of the first CSIs according to the priority of each group of CSIs; or,

[0007] The target CSI is grouped according to a preset grouping mode. Each group of CSIs is compressed by a first model to obtain a second CSI. When the terminal needs to discard the reported CSI, it reports part of the second CSI according to a first priority rule. The preset grouping mode is a grouping mode with at least two resources as the granularity, and the resources under each granularity satisfy a preset relationship.

[0008] Secondly, embodiments of this application also provide an information processing method applied to a network device, comprising:

[0009] The system receives a portion of the first channel state information (CSI) reported by the receiving terminal; after decompressing this portion of the first CSI using a second model, it obtains the target CSI. The portion of the first CSI is related to the priority relationship of each output dimension of the first model, which is deployed on the terminal side. Alternatively,

[0010] The system receives a portion of the second CSI reported by the receiving terminal; decompresses the portion of the second CSI using a second model to obtain the model output result; and performs interpolation processing on the model output result to obtain the target CSI. The portion of the second CSI is related to a preset grouping mode and a first priority rule. The preset grouping mode is a grouping mode with at least two resources as granularity, and the resources at each granularity satisfy a preset relationship.

[0011] Thirdly, embodiments of this application also provide a CSI reporting method applied to a terminal, wherein the terminal has models for different feedback overheads, and the method includes:

[0012] Based on the second reference signal sent by the network device, obtain the rank indication and target channel state information (CSI);

[0013] Based on the rank indicator and the resource size or assumed resource size reported by CSI, select the matching fourth target model;

[0014] The target CSI is compressed using the fourth target model to obtain the third CSI;

[0015] Report the rank indication and the third CSI.

[0016] Fourthly, embodiments of this application also provide an information processing method applied to a network device, the network device having models for different feedback overheads, the method comprising:

[0017] Receive the rank indication and third CSI reported by the receiving terminal;

[0018] Based on the rank indication and the resource size or assumed resource size reported by CSI, select the matching fifth target model;

[0019] The target CSI is obtained by decompressing the third CSI using the fifth target model.

[0020] Fifthly, embodiments of this application also provide a terminal, including: a memory, a transceiver, and a processor: the memory for storing computer programs; the transceiver for sending and receiving data under the control of the processor, and performing the following operations:

[0021] The target channel state information (CSI) is compressed using a first model to obtain a first CSI; the first CSI is then grouped according to the priority relationships of each output dimension of the first model; when the terminal needs to discard reported CSIs, it reports a portion of the first CSIs according to the priority of each group of CSIs; or,

[0022] The target CSI is grouped according to a preset grouping mode. Each group of CSIs is compressed by a first model to obtain a second CSI. When the terminal needs to discard the reported CSI, it reports part of the second CSI according to a first priority rule. The preset grouping mode is a grouping mode with at least two resources as the granularity, and the resources under each granularity satisfy a preset relationship.

[0023] Sixthly, embodiments of this application also provide a CSI reporting device, comprising:

[0024] The first processing unit is configured to compress the target channel state information (CSI) using a first model to obtain a first CSI; group the first CSI according to the priority relationship of each output dimension of the first model; and when the terminal needs to discard reported CSIs, report a portion of the first CSIs according to the priority of each group of CSIs; or...

[0025] The second processing unit is used to group the target CSI according to a preset grouping mode, compress each group of CSIs using a first model, and obtain a second CSI. When the terminal needs to discard the reported CSIs, it reports part of the second CSIs according to a first priority rule. The preset grouping mode is a grouping mode with at least two resources as the granularity, and the resources at each granularity satisfy a preset relationship.

[0026] In a seventh aspect, embodiments of this application also provide a network device, including: a memory, a transceiver, and a processor: the memory for storing computer programs; the transceiver for sending and receiving data under the control of the processor, and performing the following operations:

[0027] The system receives a portion of the first channel state information (CSI) reported by the receiving terminal; after decompressing this portion of the first CSI using a second model, it obtains the target CSI. The portion of the first CSI is related to the priority relationship of each output dimension of the first model, which is deployed on the terminal side. Alternatively,

[0028] The system receives a portion of the second CSI reported by the receiving terminal; decompresses the portion of the second CSI using a second model to obtain the model output result; and performs interpolation processing on the model output result to obtain the target CSI. The portion of the second CSI is related to a preset grouping mode and a first priority rule. The preset grouping mode is a grouping mode with at least two resources as granularity, and the resources at each granularity satisfy a preset relationship.

[0029] Eighthly, embodiments of this application also provide an information processing apparatus, including:

[0030] The third processing unit is configured to receive a portion of the first channel state information (CSI) reported by the terminal; decompress the portion of the first CSI using a second model to obtain the target CSI, wherein the portion of the first CSI is related to the priority relationship of each output dimension of the first model, and the first model is deployed on the terminal side; or,

[0031] The fourth processing unit is used to receive a portion of the second CSI reported by the terminal; decompress the portion of the second CSI using the second model to obtain the model output result; and perform interpolation processing on the model output result to obtain the target CSI. The portion of the second CSI is related to a preset grouping mode and a first priority rule. The preset grouping mode is a grouping mode with at least two resources as the granularity, and the resources at each granularity satisfy a preset relationship.

[0032] Ninthly, embodiments of this application also provide a terminal, the terminal having models for different feedback overheads, including: a memory, a transceiver, and a processor: the memory for storing computer programs; the transceiver for sending and receiving data under the control of the processor, and performing the following operations:

[0033] Based on the second reference signal sent by the network device, obtain the rank indication and target channel state information (CSI);

[0034] Based on the rank indicator and the resource size or assumed resource size reported by CSI, select the matching fourth target model;

[0035] The target CSI is compressed using the fourth target model to obtain the third CSI;

[0036] Report the rank indication and the third CSI.

[0037] In a tenth aspect, embodiments of this application also provide a CSI reporting device applied to a terminal, the terminal having models for different feedback overheads, including:

[0038] The fifth processing unit is used to obtain the rank indication and target channel state information (CSI) based on the second reference signal sent by the network device.

[0039] The sixth processing unit is used to select a matching fourth target model based on the rank indication and the resource size or assumed resource size reported by CSI.

[0040] The seventh processing unit is used to compress the target CSI using the fourth target model to obtain the third CSI;

[0041] The first reporting unit is used to report the rank indication and the third CSI.

[0042] Eleventhly, embodiments of this application also provide a network device having models for different feedback overheads, including: a memory, a transceiver, and a processor: the memory for storing computer programs; the transceiver for sending and receiving data under the control of the processor, and performing the following operations:

[0043] Receive the rank indication and third CSI reported by the receiving terminal;

[0044] Based on the rank indication and the resource size or assumed resource size reported by CSI, select the matching fifth target model;

[0045] The target CSI is obtained by decompressing the third CSI using the fifth target model.

[0046] In a twelfth aspect, embodiments of this application also provide an information processing apparatus applied to a network device, the network device having models for different feedback overheads, including:

[0047] The first receiving unit is used to receive the rank indication and the third CSI reported by the terminal;

[0048] The eighth processing unit is used to select a matching fifth target model based on the rank indication and the resource size or assumed resource size reported by CSI.

[0049] The ninth processing unit is used to decompress the third CSI through the fifth target model to obtain the target CSI.

[0050] In a thirteenth aspect, embodiments of this application also provide a processor-readable storage medium storing a computer program for causing the processor to perform the steps of the CSI reporting method described in the first or third aspect, or to perform the steps of the information processing method described in the second or fourth aspect.

[0051] The above-mentioned technical solution of this application has at least the following beneficial effects:

[0052] In the above technical solution of this application embodiment, the target channel state information (CSI) is compressed using a first model to obtain a first CSI; the first CSI is grouped according to the priority relationship of each output dimension of the first model; when the terminal needs to discard the reported CSI, part of the first CSI is reported according to the priority of each group of CSI; or, the target CSI is grouped according to a preset grouping mode, and each group of CSI is compressed using a first model to obtain a second CSI; when the terminal needs to discard the reported CSI, part of the second CSI is reported according to a first priority rule. The preset grouping mode is a grouping mode with at least two resources as granularity, and each resource under each granularity satisfies a preset relationship. This solves the problem of unclear priority grouping and UCI discarding rules in CSI reporting based on model compression. It ensures the performance of CSI reporting based on model compression without the need for the terminal to repeatedly calculate CSI. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of spatial frequency domain CSI compressed feedback based on an AI / ML model;

[0054] Figure 2 This is one of the flowcharts illustrating the CSI reporting method according to an embodiment of this application;

[0055] Figure 3 This is one of the flowcharts illustrating the information processing method according to an embodiment of this application;

[0056] Figure 4 This is a training block diagram for a neural network with predefined output dimension priorities, as described in this application.

[0057] Figure 5 This is a second schematic flowchart of the CSI reporting method according to an embodiment of this application;

[0058] Figure 6 A schematic diagram of an AI / ML-based CSI compression model designed for scalable feedback overhead in this application;

[0059] Figure 7 This is a second schematic flowchart of the information processing method according to an embodiment of this application;

[0060] Figure 8 This is one of the structural block diagrams of the terminal in an embodiment of this application;

[0061] Figure 9 This is one of the module schematic diagrams of the CSI reporting device according to an embodiment of this application;

[0062] Figure 10 This is one of the structural block diagrams of a network device according to an embodiment of this application;

[0063] Figure 11 This is one of the schematic diagrams of the information processing device according to an embodiment of this application;

[0064] Figure 12 This is a second structural block diagram of the terminal according to an embodiment of this application;

[0065] Figure 13 This is a second schematic diagram of the CSI reporting device according to an embodiment of this application;

[0066] Figure 14 This is a second structural block diagram of a network device according to an embodiment of this application;

[0067] Figure 15 This is a second schematic diagram of the information processing device according to an embodiment of this application. Detailed Implementation

[0068] In the embodiments of this application, the term "and / or" describes the relationship between associated objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following associated objects have an "or" relationship.

[0069] In the embodiments of this application, the term "multiple" refers to two or more, and other quantifiers are similar.

[0070] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0071] To facilitate understanding of the solution proposed in this application, the relevant content will be introduced first.

[0072] AI / ML model-based CSI compression utilizes AI / ML models to compress and report the target CSI to be reported at the terminal side, and then decompresses and recovers the target CSI at the network device side. Its basic working mechanism is as follows:

[0073] After estimating the channel matrix H in the spatial frequency domain, the user terminal inputs the original channel or the preprocessed CSI (i.e., the original channel) to the CSI. Figure 1 X) uses the UE-side model, i.e., the encoder, to generate compressed codeword Y, and then feeds it back to the base station.

[0074] After receiving the compressed codeword Y, the base station uses the network equipment-side model, i.e., the decoder, to reconstruct / recover the original channel input CSI (i.e., Figure 1 (Z)

[0075] The above describes the basic principle of spatial-frequency domain CSI compressed feedback based on the AI / ML model. In other words, spatial-frequency domain CSI compressed feedback based on the AI / ML model is a two-sided model with dual-sided deployment. Its schematic diagram is shown below. Figure 1 As shown.

[0076] On the UE side: the original channel or the pre-processed channel is compressed using the AI / ML generation part (or CSI generation part, CSI generation part; AI / ML encoder, AI / MLencoder, encoder, etc.), and the compressed data is quantized into binary bits and fed back to the base station.

[0077] On the network side: The binary bits fed back by the UE are dequantized, and then the channel information is recovered using the AI / ML reconstruction part (or CSI reconstruction part; AI / ML decoder, decoder, etc.). Optionally, the base station side also performs some post-processing operations.

[0078] When a terminal reports CSI, if the CSI overhead reported by the terminal exceeds the size of the Physical Uplink Shared Channel (PUSCH) resource allocated by the base station, or if the base station has configured Physical Uplink Control Channel (PUCCH) resources and indicated the code rate, and if the terminal's reported CSI overhead still exceeds the terminal's assumed PUCCH resource size after adjusting the code rate to the maximum, the terminal needs to discard the reported CSI to ensure the effective transmission of the remaining CSI. However, in CSI reporting based on AI / ML model compression, priority grouping and UCI discarding rules cannot be defined as in existing codebook-based CSI reporting; that is, existing priority and discarding rules are no longer applicable.

[0079] To address the aforementioned technical problems, embodiments of this application provide a CSI reporting method, information processing method, apparatus, device, and medium. The method and apparatus are based on the same application concept. Since the methods and apparatus solve problems in similar principles, their implementations can be mutually referenced, and repeated details will not be elaborated further.

[0080] like Figure 2 The diagram shown is a flowchart of the CSI reporting method provided in this embodiment. This method is applied to a terminal, meaning it is executed by the terminal. The method may include:

[0081] Step 201: Compress the target channel state information (CSI) using the first model to obtain the first CSI; group the first CSI according to the priority relationship of each output dimension of the first model; when the terminal needs to discard the reported CSI, report part of the first CSI according to the priority of each group of CSI.

[0082] The terminal can receive reference signals sent by network devices (such as base stations), perform channel estimation on the reference signals, and obtain the channel matrix in the spatial frequency domain; then, the channel matrix in the spatial frequency domain is preprocessed to obtain the target CSI.

[0083] Here, the first model refers to the AI / ML model, which can also be called the AI / ML generation part; or the CSI generation part; or the AI / ML encoder; or the encoder.

[0084] The target CSI is compressed by the first model to obtain the first CSI, which is the binary bits of the target CSI after compression and quantization. It is a low-dimensional vector or bit string.

[0085] It should be noted that the first CSI is grouped according to the priority relationship of each output dimension of the first model, and different groups correspond to different priorities.

[0086] In the following situations, the terminal needs to discard the reported CSI:

[0087] The CSI overhead reported by the terminal exceeds the PUSCH resource size allocated by the base station;

[0088] The base station configures PUCCH resources and indicates the code rate. The terminal assumes that each CSI report is a CSI report with a rank of 1 to determine the PUCCH resources and the number of Physical Resource Blocks (PRBs) of the PUCCH resources, or to determine the number of second-part CSI reports. When the terminal reports back, even after the code rate is adjusted to the maximum, the CSI overhead reported by the terminal still exceeds the size of the PUCCH resources assumed by the terminal.

[0089] When a terminal needs to discard reported CSIs, it discards low-priority CSIs and reports high-priority CSIs (i.e., the first CSI in the reported subset) according to the priority of each group of CSIs. This solves the problem of unclear priority grouping and UCI discarding rules in model-compression-based CSI reporting, ensuring both the terminal avoids unnecessary recalculation of CSIs and the performance of model-compression-based CSI reporting.

[0090] Alternatively, in step 202, the target CSI is grouped according to a preset grouping mode, and each group of CSIs is compressed using a first model to obtain a second CSI. When the terminal needs to discard the reported CSI, part of the second CSI is reported according to a first priority rule. The preset grouping mode is a grouping mode with at least two resources as the granularity, and the resources at each granularity satisfy a preset relationship.

[0091] Here, the preset grouping mode is a grouping mode with at least two resources as the granularity, and the resources at each granularity satisfy a preset relationship. It can be understood that two or more resources form a resource group, and the resources in the resource group satisfy a preset relationship. The preset relationship can be that the resources are continuous, or it can be spaced at a preset number N, where N is greater than or equal to 1.

[0092] This can be understood as: dividing the target CSI into groups based on the CSI portion of the corresponding resource group, thus obtaining the CSIs in each group.

[0093] Optionally, the bearer channels reported by CSI include the Physical Uplink Shared Channel (PUSCH) or the Physical Uplink Control Channel (PUCCH).

[0094] Optionally, the first priority rule is obtained through one or more of the following methods:

[0095] Protocol predefined;

[0096] Network device configuration;

[0097] Reporting from the terminal side.

[0098] Optionally, the resource includes one of the following:

[0099] Frequency domain resources;

[0100] Time-domain resources;

[0101] Airspace resources;

[0102] Angle domain resources.

[0103] For example, the CSI reporting bandwidth is 12 subbands, and the preset grouping mode is 4 consecutive subbands. After grouping the target CSI according to the preset grouping mode, 4 consecutive subband Precoding Matrix Indicators (PMIs) can be obtained. Then, the 4 consecutive subband PMIs are input into the first model to obtain a low-dimensional vector after joint compression of the 4 consecutive subband PMIs in the spatial and frequency domains. When compressing and reporting CSIs from the 12 subbands of the CSI reporting bandwidth, the 12 subband PMIs need to be divided into 3 groups. Each group of 4 consecutive subband PMIs is subjected to model-based CSI compression and then reported. If the terminal needs to discard reported CSIs, according to the first priority rule, some subband CSIs are reported, and the remaining subband CSIs are discarded.

[0104] When a terminal needs to discard reported CSIs, it discards low-priority CSIs and reports high-priority CSIs (i.e., reports a portion of the second CSI) according to the first priority rule. This solves the problem of unclear priority grouping and UCI discarding rules in model-compression-based CSI reporting, ensuring both the terminal avoids unnecessary recalculation of CSIs and the performance of model-compression-based CSI reporting.

[0105] In some embodiments, the priority relationship of each output dimension of the first model is obtained after training the first target model according to predefined rules. The first target model includes the first model and a second model paired with the first model, and the second model is deployed on the network device side.

[0106] Here, the predefined rules refer to the predefined priority relationships among the output dimensions of the first model. The specific training process of the first target model will be detailed in the subsequent explanation on the network device side.

[0107] It should be noted that the priority relationship of each output dimension of the first model can be obtained after training the first target model according to predefined rules. Of course, other methods can also be used to obtain the priority relationship of each output dimension of the first model. For example, in some embodiments, the method of this application further includes:

[0108] 1) Based on the second target model, determine the weights of each output dimension of the first model. The second target model includes the first model and the third model, wherein the third model is a replacement model for the second model, and the second model is deployed on the network device side.

[0109] As an optional embodiment, the weights of each output dimension of the first model are determined according to the second target model, including:

[0110] For each output dimension of the first model, obtain the first performance index of the second target model before the output dimension is removed;

[0111] By removing the output dimension, a second performance metric of the second target model is obtained.

[0112] Removing the output dimension can be understood as setting that output dimension to zero and then observing the changes in the second target model, i.e., the second performance metric, to test the impact of the output dimension on the performance of the second target model. Here, the model's performance metric can be accuracy, recall, or other performance metrics.

[0113] The weight of the output dimension is determined by comparing the difference between the first performance metric and the second performance metric.

[0114] For example, if there is a large difference between the first and second performance metrics, it means that removing that output dimension will cause a significant drop in model performance, and therefore that output dimension has a larger weight; conversely, if the performance change is not significant, then that output dimension has a smaller weight.

[0115] 2) Based on the weights of each output dimension of the first model, obtain the priority relationship of each output dimension of the first model.

[0116] Here, based on the weights of each output dimension of the first model, the output dimensions of the first model can be divided into m priority parts. Specifically, the weights of all N output dimensions of the first model are sorted from largest to smallest, with G0 representing the dimensions of the output vectors corresponding to weights from 1 to N / m, G1 representing the dimensions of the output vectors corresponding to weights from N / m+1 to 2N / m, ... G m-1 The weights are sorted by the dimensions of the output vectors from (m-1)N / m+1 to N. The priority is G0 > G1 > ... G m-2 >G m-1 .

[0117] That is, when the terminal side has a replacement model of the second model (CSI reconstruction part) deployed on the network device side, the priority relationship of each output dimension of the first model is obtained by using ablation experiments to obtain the CSI reporting using the "first model and third model" pair of CSI compression-decompression models (i.e. the second target model).

[0118] The following example illustrates the detailed steps of an ablation experiment.

[0119] 1. Use the model parameters of the two-sided model, "Network-side CSI reconstruction part / CSI reconstruction part proxy model -- UE-side model", as the experimental initialization parameters.

[0120] 2. Based on the experimental requirements, select the dimension of the output vector to be ablated.

[0121] 3. Test the impact of the target dimension on model performance by individually removing (setting) the selected dimension to zero and observing the changes in model performance. For example, the importance of a dimension can be evaluated by comparing the changes in accuracy, recall, or other performance metrics before and after removing a feature from that dimension.

[0122] 4. The results of ablation experiments can quantify the contribution of each dimension to the model performance and explain the weight of each dimension (i.e., their contribution to the model output).

[0123] If removing a dimension causes a significant drop in model performance, then that dimension has a larger weight; conversely, if the performance change is not significant, then that dimension has a smaller weight.

[0124] Statistical analysis was performed on the results of the ablation experiments. By analyzing the model performance data, we can understand the specific impact of different dimensions on the model performance.

[0125] 5. Based on the weight of each dimension, divide the output vector into three priority parts. Sort the weights of all N dimensions from largest to smallest. G0 represents the dimensions of the output vectors with weights from 1 to N / 3, G1 represents the dimensions of the output vectors with weights from N / 3+1 to 2N / 3, and G2 represents the dimensions of the output vectors with weights from 2N / 3+1 to N. The priority is G0 > G1 > G2.

[0126] Based on the above ablation experiment method, the priority relationship of each output dimension of the UE-side first model can be obtained. When the UE-side first model is used for CSI compression reporting, if the terminal needs to discard part of the CSI information when reporting 2-part CSI, it is grouped according to priority G0 > G1 > G2. The lower priority CSI information group is discarded first until the part of the CSI information to be transmitted can be transmitted on the allocated PUSCH resources.

[0127] In some embodiments, before grouping the target CSI according to a preset grouping module, the method of this application further includes:

[0128] Receive the first reference signal sent by the network device;

[0129] Before receiving the first reference signal sent by the network device, the terminal can receive configuration information sent by the network device via higher-layer signaling. This higher-layer signaling can be Radio Resource Control (RRC) signaling. The configuration information includes, but is not limited to: the CSI reporting bandwidth, subband size, CSI reporting subbands, and reference signal configuration configured by the network device for the terminal. The CSI reporting subband (csi-ReportingBand) uses a bitmap to indicate whether the subbands within the CSI reporting bandwidth are continuous or non-contiguous; CSI from these subbands will be reported.

[0130] Channel estimation is performed based on the first reference signal to obtain the first downlink channel matrix;

[0131] It should be noted that the first downlink channel matrix can be the downlink channel matrix within the CSI-reported resources in the configuration information sent by the network device. Specifically, the first downlink channel matrix can be the downlink channel matrix in the spatial frequency domain (such as the downlink channel matrix within the CSI-reported subband), or it can be the downlink channel matrix in other relevant domains.

[0132] When the first downlink channel matrix is ​​a downlink channel matrix of other related domains, this step, performing channel estimation based on the first reference signal to obtain the first downlink channel matrix, may include:

[0133] Channel estimation is performed based on the first reference signal to obtain the second downlink channel matrix;

[0134] Here, the second downlink channel matrix is ​​the downlink channel matrix in the spatial frequency domain.

[0135] The second downlink channel matrix is ​​preprocessed to obtain the first downlink channel matrix;

[0136] Here, the first downlink channel matrix obtained after preprocessing is the downlink channel matrix transformed from the spatial frequency domain to other related domains.

[0137] The preprocessing includes at least one of the following:

[0138] Angular domain transformation;

[0139] Time delay domain transformation;

[0140] Doppler domain transform.

[0141] The first correlation matrix is ​​calculated based on the first downlink channel matrix;

[0142] In one example, the terminal calculates the spatial subband correlation matrix (i.e., the first correlation matrix mentioned above) based on the downlink channel matrix within the CSI-reported subband.

[0143] Record CSI reporting sub-band S l The channel matrix of subcarrier i is H i Dimension N r ×(2N1N2), where N r 2N1N2 is the number of receiving antennas at the terminal, 2N1N2 is the number of antenna ports for CSI-RS transmitted by the base station, l = 1, ..., L is the reporting subband index, and the reporting subband set {S1, S2, ... S} is the number of antenna ports for CSI-RS transmitted by the base station. L This comprises all subbands configured on the network device side and indicated by the csi-ReportingBand that will perform CSI reporting. The spatial subband correlation matrix is ​​calculated as follows:

[0144]

[0145] The first correlation matrix is ​​subjected to eigenvalue decomposition to obtain an eigenvector, and the target CSI includes the eigenvector.

[0146] Continuing with the example above, the terminal performs eigenvalue decomposition on the spatial sub-band correlation matrix (the first correlation matrix mentioned above) and takes the largest rank indicator (RI) principal eigenvectors. remember

[0147] The specific implementation process of step 202 described above will be illustrated by the following embodiments.

[0148] Example 1

[0149] For CSI compression based on AI / ML models, the preset grouping pattern is 3 consecutive sub-bands. After grouping the target CSI according to this preset grouping pattern, the v-th layer feature vectors of the 3 consecutive sub-bands can be obtained. Then, the v-th layer feature vectors of the 3 consecutive sub-bands are input into the terminal-side model for joint compression, that is:

[0150]

[0151] The output of the terminal-side model: Y = encoder(X) is the output result of joint compression of the feature vectors of the vth layer of three consecutive sub-bands.

[0152] Suppose the set of reporting subbands indicated by the csi-ReportingBand is {S1, S2, ... S}. 12 Assuming the input to the terminal-side model is the feature vector of the v-th layer of the three consecutive sub-bands mentioned above, joint compression is performed, assuming that in {S1, S2, ... S... 12 If a transmission with RI=1 is performed on {S1,S2,…S}, then the transmission in {S1,S2,…S} will be... 12 The CSI report based on the AI / ML model can be denoted as: {Y1,Y2,Y3,Y4}, where Y1 corresponds to the joint compression of the first layer feature vectors of S1 to S3, Y2 corresponds to the joint compression of the first layer feature vectors of S4 to S6, and so on.

[0153] It is necessary to consider the feature vectors of the vth layer based on the three consecutive sub-bands (i.e. The total reporting bandwidth obtained (i.e., {S1, S2, ... S) 12 The CSI reporting (i.e., {Y1,Y2,Y3,Y4}) on the network is defined by a first priority rule. Specifically, one possible first priority rule is: {Y1,Y3} have the same priority, {Y2,Y4} have the same priority, and {Y1,Y3} has a higher priority than {Y2,Y4}. According to the above first priority rule, when reporting 2-part CSI information based on AI / ML model-compressed CSI, if it is necessary to discard some CSI information, then {Y2,Y4} is discarded first, and {Y1,Y3} is reported, i.e., the reporting bandwidth {S1,S2,…S} is... 12 The compressed CSI corresponding to {S1, S2, S3, S7, S8, S9} in the code will be reported. Here, the first priority rule can be predefined by the protocol, configured on the network device side, or reported by the terminal side.

[0154] In the aforementioned CSI compression based on the AI / ML model for partial subband CSI reporting, considering that using the PMI of adjacent subbands to calculate the Channel Quality Indicator (CQI) of the current subband would increase the number of CQI calculations, it is preferable to use the PMI of the current subband to calculate its own CQI. If the reported CQI lacks a corresponding PMI due to the aforementioned CSI discarding, the base station can approximate the PMI of the current subband by interpolating the PMI of adjacent subbands. That is, the base station can obtain {S4, S5, S6} and {S...} from the recovered PMI of {S1, S2, S3, S7, S8, S9} through interpolation or extrapolation. 10 ,S 11 ,S 12 The PMI of}

[0155] Example 2

[0156] For CSI compression based on AI / ML models, the preset grouping pattern is 3 sub-bands, with each pair of sub-bands separated by one sub-band. After grouping the target CSI according to this preset grouping pattern, three layer v feature vectors with each pair separated by one sub-band can be obtained. Then, these three layer v feature vectors with each pair separated by one sub-band are input into the terminal-side model for joint compression, i.e.:

[0157]

[0158] The output of the terminal-side model: Y = encoder(X) is the output result of joint compression of three feature vectors of the vth layer that are spaced one subband apart.

[0159] Suppose the set of reporting subbands indicated by the csi-ReportingBand is {S1, S2, ... S}. 12 Assuming the input to the terminal-side model is the joint compression of the three layer v feature vectors spaced one subband apart, and assuming that in {S1, S2, ... S... 12 If a transmission with RI=1 is performed on {S1,S2,…S}, then the transmission in {S1,S2,…S} will be... 12 The CSI report based on the AI / ML model can be denoted as: {Y1,Y2,Y3,Y4}, where Y1 corresponds to the first layer feature vector of {S1,S3,S5} for joint compression, and Y2 corresponds to {S7,S9,S... 11 The first layer feature vectors of {S2, S4, S6} are jointly compressed, and Y3 corresponds to the first layer feature vectors of {S8, S6}. The first layer feature vectors of {S2, S4, S6} are also jointly compressed, and Y4 corresponds to the first layer feature vectors of {S8, S6}. 10 ,S 12 The first layer of feature vectors is jointly compressed.

[0160] Specifically, one possible first priority rule is: {Y1,Y2} have the same priority, {Y3,Y4} have the same priority, and {Y1,Y2} has a higher priority than {Y3,Y4}. According to this first priority rule, when reporting 2-part CSI information based on AI / ML model-compressed CSI, if some CSI information needs to be discarded, then {Y3,Y4} should be discarded first, and {Y1,Y2} should be reported as CSI, i.e., the reporting bandwidth {S1,S2,…S}... 12 In}, {S1,S3,S5,S7,S9,S} 11 The corresponding compressed CSI will be reported. Here, the first priority rule can be predefined by the protocol, configured on the network device side, or reported by the terminal side.

[0161] In the aforementioned CSI compression based on the AI / ML model for partial subband CSI reporting, considering that using the PMI of adjacent subbands to calculate the CQI of the current subband would increase the number of CQI calculations, the method of calculating the CQI of the current subband using the PMI of the current subband can be adopted. If the reported CQI has no corresponding PMI due to the aforementioned CSI discarding, the base station can approximate the PMI of the current subband by interpolating the PMI of adjacent subbands. That is, the base station can use the recovered {S1, S3, S5, S7, S9, S... 11 The PMI of {S2, S4, S6, S8, S} is obtained through interpolation or extrapolation. 10 ,S 12 The PMI of}

[0162] like Figure 3 The diagram shown is a flowchart illustrating an information processing method provided in an embodiment of this application. This method is applied to a network device, meaning it is executed by the network device. The method may include:

[0163] Step 301: Receive a portion of the first channel state information (CSI) reported by the terminal; decompress the portion of the first CSI using a second model to obtain the target CSI, wherein the portion of the first CSI is related to the priority relationship of each output dimension of the first model, and the first model is deployed on the terminal side;

[0164] Here, the method executed on the network device side is the same as... Figure 2 The methods executed on the terminal side are shown below. The meanings and explanations of related terms or steps can be found in the description of the terminal side methods, and will not be repeated here.

[0165] The second model on the network device side is paired with the first model on the terminal side. The first model is used to compress CSI, and the second model is used to decompress CSI. Here, the second model refers to the AI / ML model, which can also be called the AI / ML reconstruction part; or the CSI reconstruction part; or the AI / ML decoder; or simply the decoder.

[0166] Alternatively, in step 302, a portion of the second CSI reported by the terminal is received; the portion of the second CSI is decompressed by the second model to obtain the model output result; the model output result is interpolated to obtain the target CSI, wherein the portion of the second CSI is related to a preset grouping mode and a first priority rule, wherein the preset grouping mode is a grouping mode with at least two resources as granularity, and the resources at each granularity satisfy a preset relationship.

[0167] Optionally, the bearer channels reported by CSI include the Physical Uplink Shared Channel (PUSCH) or the Physical Uplink Control Channel (PUCCH).

[0168] Optionally, the first priority rule is obtained through one or more of the following methods:

[0169] Protocol predefined;

[0170] Network device configuration;

[0171] Reporting from the terminal side.

[0172] Optionally, the resource includes one of the following:

[0173] Frequency domain resources;

[0174] Time-domain resources;

[0175] Airspace resources;

[0176] Angle domain resources.

[0177] The information processing method of this application decompresses a portion of the CSI reported by the terminal using a second model paired with the first model of the terminal, or further interpolates the model output after decompression to recover the target CSI. In this way, through cooperation with the terminal side, the problem of unclear priority grouping and UCI discarding rules in CSI reporting based on model compression can be solved. While ensuring that the terminal does not need to repeatedly calculate CSI, the performance of CSI reporting based on model compression is also guaranteed.

[0178] In some embodiments, the method of this application further includes:

[0179] The first target model is trained according to predefined rules to obtain the trained first target model, which includes a first model that satisfies the predefined priority relationship of each output dimension and a second model that is paired with the first model.

[0180] Here, the predefined rules refer to the predefined priority relationships of each output dimension of the first model.

[0181] Specifically, training the first target model according to predefined rules to obtain the trained first target model may include:

[0182] During model training, a random mask is applied to the first output vector of the first model, and the resulting second output vector is input into the second model to obtain the output result. The random mask is determined by the predefined rules.

[0183] It should be noted that during model training, the priority relationship of each output dimension of the first model is predefined. For example, assuming that the output of the first model on the terminal side is a vector of dimension N, the output vector is divided into three priority parts: G0 is the first to N / 3 dimensions of the output vector, G1 is the N / 3+1 to 2N / 3 dimensions of the output vector, and G2 is the 2N / 3+1 to N dimensions of the output vector, with the priority being G0 > G1 > G2.

[0184] During model training, a random mask module is added after the output layer of the first model on the terminal side, such as... Figure 4 As shown. By adding a random mask module, the neural network (a neural network consisting of the first model and the second model) updates the trained neural network by calculating the gradient of the loss function with respect to the neural network weights. The design of the random mask is determined by predefined rules.

[0185] The effect of a random mask operation on vector Y is denoted as: in, Both Y and M are Dimensions (for example, if Y is an image with a length of 64 and a width of 64, then L...) Y =64×64×3, where 3 represents the R of the three primary color channels (red, green, and blue). Y 1st order vector It is the i-th dimension of the input vector, and the operation... This represents a pointwise product of vectors, where M is the mask vector. This is the masking result. The mask vector M is a random vector, and it is independently and identically distributed for each training sample and each training epoch. R M The mask vector M of order M has dimension M. The design of the mask vector M can be, but is not limited to:

[0186] 1. Hard-Masking

[0187] This indicates the order of the mask to be applied (e.g., for a 6×6×3 dimensional image, a 6×6 dimension mask is applied, but no mask is applied to the 3 dimensions of the red, green, and blue primary color channels, then I = {1, 2}). Vector Index is The element values ​​are:

[0188]

[0189] Where, integer Following a given probability distribution (e.g., a uniform distribution), for each I i t are independent and identically distributed.

[0190] It should be understood that the design of a hard mask refers to setting the first t part of the mask vector M to 1 and the last t part to 0.

[0191] 2. Exponentially decaying soft-masking

[0192] vector Index is The element values ​​are:

[0193]

[0194] Where, real number λ∈[0,λ max [I] follows a certain distribution (e.g., uniform distribution), for each I i λ is independently and identically distributed.

[0195] 3. In the mask vector M, part of it uses hard mask and part of it uses soft mask, that is, a combination of hard mask and soft mask.

[0196] When the mask vector described above is applied to a certain vector, it does not change the dimension of the vector itself; this is equivalent to performing a weighted operation on the vector. Vector values ​​at smaller indices are assigned greater weight. For example, for a first-order vector, a hard mask... It is a case where the first t bits are 1, and the last t bits are 1. A vector with 0 bits. Equivalent to the latter of Y The bit values ​​are set to zero, while the vector dimensions remain unchanged. During training, the optimizer calculates gradients through backpropagation based on the loss and updates the neural network weight parameters. Furthermore, in Figure 4 middle:

[0197] The neural network can be, but is not limited to, convolutional neural networks, feedforward fully connected neural networks, recurrent neural networks, and combinations thereof. The loss function can be, but is not limited to, mean squared error, L1 norm, classification cross-entropy, and cosine similarity. The random mask module avoids gradient vanishing or non-differentiability problems by using custom gradients and treating the mask vector as a constant vector.

[0198] Adjust the model parameters based on the output results, and train the model based on the first target model with adjusted parameters until the first target model is obtained.

[0199] Specifically, the first model in the first target model after training is the model whose output dimensions satisfy a predefined priority relationship. In other words, after obtaining the first target model after training, the priority relationship of each output dimension of the first model can be obtained.

[0200] It should be noted that before model training, the priority relationship of each output dimension of the first model is predefined. For example, the total length of the output vector of the first model on the terminal side is divided into three equal parts G0, G1, and G2. Among them, the first to N / 3 dimensions of the output vector Y (G0) have the highest priority, the N / 3+1 to 2N / 3 dimensions of the output vector Y (G1) have the second highest priority, and the 2N / 3+1 to N dimensions of the output vector Y (G2) have the lowest priority. For the three priority parts of the output vector Y, namely the random masks corresponding to G0 (the first to N / 3 dimensions of the output vector Y), G1 (the N / 3+1 to 2N / 3 dimensions of the output vector Y), and G2 (the 2N / 3+1 to N dimensions of the output vector Y), the random masks in the following embodiment can be used for vector weighting.

[0201] Example 1

[0202] During the training phase, the parameters t∈{N / 3,2N / 3,N} of the hard mask module follow an equal probability distribution. It is applied to the encoder (UE-side first model) output.

[0203] It should be understood that during training, each training iteration uses one of the following three mask vectors M:

[0204] M = [1,1,...,1,0,0,...,0], where the number of 1s in M ​​is N / 3.

[0205] Alternatively, M = [1,1,...,1,0,0,...,0], where the number of 1s in M ​​is 2N / 3.

[0206] Alternatively, M = [1,1,...,1], where there are N 1s in M.

[0207] The probability of each of the three M's occurring is 1 / 3.

[0208] Example 2

[0209] During the training phase, a soft-value mask module is used for a first-order mask vector M of length N, with parameters λ∈[0,20] uniformly distributed, and applied to the encoder output.

[0210] It should be understood that in the i-th training iteration, a random variable λ is first generated according to a uniform distribution λ∈[0,20]: λ i During this training session:

[0211]

[0212] That is, the values ​​of the elements in M ​​above are applied to the output vector Y of the UE-side model.

[0213] Example 3

[0214] During the training phase, for a first-order mask vector M of length N, a mask module parameter combining hard and soft values ​​is used: t∈{N / 3,2N / 3,N} follows an equal probability distribution. The parameter λ∈[0,1] is uniformly distributed and acts on the encoder output.

[0215] It should be understood that in the i-th training iteration, a random variable λ is first generated according to a uniform distribution λ∈[0,1]: λ i In this training, one of the following three mask vectors M is used:

[0216]

[0217] The probability of each of the three M's occurring is 1 / 3.

[0218] After training the AI / ML model using the above model training method, the generated UE-side AI / ML model (first model) has predefined priority relationships for each output dimension. When using this UE-side first model for CSI compression reporting, the first model output is grouped according to the priority relationship of each output dimension. Different groups correspond to different priorities. When the terminal performs 2-part CSI reporting, if it needs to discard some CSI information, it is grouped according to priority G0 > G1 > G2. The lower priority CSI information groups are discarded first until the CSI information part to be transmitted can be transmitted on the allocated PUSCH resources.

[0219] It should be noted that the priority relationship of each output dimension of the first model can be obtained after training the first target model according to predefined rules. Of course, other methods can also be used to obtain the priority relationship of each output dimension of the first model. For example, in some embodiments, the method of this application further includes:

[0220] 1) Based on the third target model, determine the weights of each output dimension of the first model. The third target model includes the second model and the fourth model. The fourth model is a replacement model for the first model. The first model is deployed on the terminal side.

[0221] As an optional embodiment, the weights of each output dimension of the first model are determined according to the third target model, including:

[0222] For each output dimension of the first model, obtain the third performance index of the third target model before the output dimension is removed;

[0223] The output dimension is removed to obtain the fourth performance metric of the third objective model;

[0224] Removing the output dimension can be understood as setting that output dimension to zero, and then observing the changes in the third objective model, i.e., the fourth performance metric, to test the impact of the output dimension on the performance of the third objective model. Here, the model's performance metric can be accuracy, recall, or other performance metrics.

[0225] The weights of the output dimensions are determined by comparing the differences between the third and fourth performance metrics.

[0226] For example, if there is a large difference between the third and fourth performance metrics, it means that removing that output dimension will cause a significant drop in model performance, and therefore that output dimension has a larger weight; conversely, if the performance change is not significant, then that output dimension has a smaller weight.

[0227] 2) Based on the weights of each output dimension of the first model, obtain the priority relationship of each output dimension of the first model.

[0228] Here, based on the weights of each output dimension of the first model, the output dimensions of the first model can be divided into m priority parts. Specifically, the weights of all N output dimensions of the first model are sorted from largest to smallest, with G0 representing the dimensions of the output vectors corresponding to weights from 1 to N / m, G1 representing the dimensions of the output vectors corresponding to weights from N / m+1 to 2N / m, ... G m-1 The weights are sorted by the dimensions of the output vectors from (m-1)N / m+1 to N. The priority is G0 > G1 > ... G m-2 >G m-1 .

[0229] That is, when the network device has a replacement model (fourth model) for the first model (CSI generation part) deployed on the terminal side, the priority relationship of each output dimension of the first model is obtained by using ablation experiments to obtain the CSI reporting using the "fourth model and second model" pair of CSI compression-decompression models (i.e., the third target model).

[0230] In some embodiments, the method of this application further includes:

[0231] A first reference signal is sent to the terminal, and the first reference signal is used to generate a first downlink channel matrix.

[0232] Before sending the first reference signal to the terminal, the network device can send configuration information to the terminal via higher-layer signaling. This higher-layer signaling can be RRC signaling. The configuration message includes, but is not limited to: the CSI reporting bandwidth, subband size, CSI reporting subband, and reference signal configuration configured by the network device for the terminal. The CSI reporting subband (csi-ReportingBand) uses a bitmap to indicate continuous or non-contiguous subbands within the CSI reporting bandwidth; the CSI of these subbands will be reported. The configuration information sent by the network device assists the terminal in generating the first downlink channel matrix based on the first reference signal and in performing CSI reporting.

[0233] The information processing method of this application decompresses a portion of the CSI reported by the terminal using a second model paired with the first model of the terminal, or further interpolates the model output after decompression to recover the target CSI. In this way, through cooperation with the terminal side, the problem of unclear priority grouping and UCI discarding rules in CSI reporting based on model compression can be solved. While ensuring that the terminal does not need to repeatedly calculate CSI, the performance of CSI reporting based on model compression is also guaranteed.

[0234] like Figure 5 The diagram shown is a flowchart of the CSI reporting method provided in an embodiment of this application. This method is applied to a terminal, meaning it is executed by the terminal. The terminal has models for different feedback overheads.

[0235] Before explaining the process of this application, the design background and principles of the terminal's models for different feedback overheads will be explained below.

[0236] In traditional codebook-based CSI reporting, the overhead of CSI varies greatly across different ranks. In the following situations, some CSIs will need to be discarded.

[0237] Scenario 1: The PUSCH resources scheduled by the base station are greater than the feedback overhead reported by CSI.

[0238] Scenario 2: The base station has configured PUCCH resources and indicated the code rate. The terminal assumes that each CSI report is a rank 1 CSI report to determine the PUCCH resources and the number of Physical Resource Blocks (PRBs) of the PUCCH resources or to determine the number of part 2 CSI reports. The actual code rate reported by the CSI is greater than or equal to the maximum code rate configured by the RRC. That is, when the terminal responds, even after the code rate is adjusted to the maximum, the CSI overhead reported by the terminal still exceeds the size of the PUCCH resources assumed by the terminal.

[0239] For CSI reporting based on AI / ML models with CSI compression, if the AI / ML model has scalability for different feedback overheads (i.e., if the AI / ML-based CSI compression model has model branches for different feedback overheads (compression rates), then when the bearer channel for CSI reporting is PUSCH, the terminal can select the model branch of the CSI compression model based on the size of the PUSCH resources allocated by the base station, ensuring that the output overhead of the terminal-side model is always less than or equal to the size of the PUSCH resources allocated by the base station. When the bearer channel for CSI reporting is PUCCH, the terminal and the base station need to reach an agreement on the total feedback overhead assumed for each CSI report by the UE (i.e., they need to reach an agreement on the assumed PUCCH resource size for CSI reporting). The terminal can select the appropriate model branch of the CSI compression model based on the assumed total feedback overhead for each CSI report and the rank indicator (rank number) of the actual CSI reporting, ensuring that the total feedback overhead of CSI compression across all layers is less than or equal to the assumed total overhead of each CSI report.

[0240] See Figure 6 This diagram illustrates an AI / ML-based CSI compression model designed for scalable feedback overhead in this application. In the diagram, EN block represents the coding block, DE block represents the decoding block, DS represents downsampling, and US represents upsampling.

[0241] Assuming the base station allocates PUSCH resources or the terminal and base station assume the total feedback overhead for each CSI report is 40 bits, if the terminal actually determines to report compressed CSI feedback at rank=2, meaning it needs to report compressed PMI information from both layer 1 and layer 2, then the terminal will choose... Figure 6 The UE-side model branch of the DS-20 block (a downsampled block with a sampling depth of 20 bits) has a reporting overhead of 20 bits for each layer, and the total CSI reporting overhead for the two layers is 40 bits of the PUSCH resources allocated by the base station.

[0242] Thus, if the AI / ML CSI compression model has scalability for different feedback overheads, the CSI reporting problem based on AI / ML model CSI compression will be transformed into the problem of selecting the branch model that matches the feedback overhead on the network device side and the terminal side to perform CSI compression feedback of the AI / ML model.

[0243] Based on the above model, the method of this application may include:

[0244] Step 501: Obtain the rank indication and target channel state information (CSI) based on the second reference signal sent by the network device;

[0245] Step 502: Select a matching fourth target model based on the rank indicator and the resource size or assumed resource size reported by CSI;

[0246] It should be understood that the fourth target model is a branch of the AI / ML-based CSI compression model and matches the rank indicator and the resource size or assumed resource size reported by CSI.

[0247] Step 503: After compressing the target CSI using the fourth target model, the third CSI is obtained;

[0248] Step 504: Report the rank indication and the third CSI.

[0249] Optionally, for 2-part CSI reporting based on AI / ML models, the rank indicator RI is in CSI part 1 of the CSI reporting, and the third CSI is in CSI part 1 of the CSI reporting.

[0250] Optionally, the bearer channels reported by CSI include the Physical Uplink Shared Channel (PUSCH) or the Physical Uplink Control Channel (PUCCH).

[0251] Optionally, when the bearer channel reported by CSI is PUCCH, the assumed PUCCH resource size reported by CSI is obtained through one or more of the following methods:

[0252] Protocol predefined;

[0253] Network device indication;

[0254] Reporting from the terminal side.

[0255] The CSI reporting method of this application embodiment designs an AI / ML-based CSI compression model with scalable feedback overhead. It can select a matching model based on the resource size or assumed resource size of the reported CSI and the rank indication obtained from the second reference signal sent by the network device. After the target CSI is compressed by the model, the compressed CSI is obtained and the rank indication and the compressed CSI are reported. This makes the total feedback overhead of the actual reported CSI compression less than or equal to the PUSCH resource size allocated by the base station or the assumed total overhead of each CSI report. This solves the problem of unclear priority grouping and UCI discarding rules in model-based CSI reporting. It ensures that the terminal does not need to repeatedly calculate CSI and also guarantees the performance of model-based CSI reporting.

[0256] like Figure 7 The diagram shown is a flowchart illustrating an information processing method provided in an embodiment of this application. This method is applied to a network device, meaning it is executed by the network device. The network device has models for different feedback overheads. The method may include:

[0257] Step 701: Receive the rank indication and third CSI reported by the terminal;

[0258] Here, the method executed on the network device side is the same as... Figure 5 The methods executed on the terminal side are shown below. The meanings and explanations of related terms or steps can be found in the description of the terminal side methods, and will not be repeated here.

[0259] For a schematic diagram of the network device for different feedback overheads, please refer to [link to relevant documentation]. Figure 6 For related explanations, please refer to the description in the terminal-side method section above, which will not be repeated here.

[0260] For 2-part CSI reporting based on AI / ML models, the rank indicator can be obtained from the reported CSI part 1, and the third CSI can be obtained from the reported CSI part 2.

[0261] Step 702: Select the matching fifth target model based on the rank indication and the resource size or assumed resource size reported by CSI;

[0262] Step 703: After decompressing the third CSI through the fifth target model, the target CSI is obtained.

[0263] Optionally, the bearer channels reported by CSI include the Physical Uplink Shared Channel (PUSCH) or the Physical Uplink Control Channel (PUCCH).

[0264] Optionally, when the bearer channel reported by CSI is PUCCH, the assumed PUCCH resource size reported by CSI is obtained through one or more of the following methods:

[0265] Protocol predefined;

[0266] Network device indication;

[0267] Reporting from the terminal side.

[0268] The information processing method of this application embodiment designs an AI / ML-based CSI compression model with scalable feedback overhead. It can select a matching model based on the resource size or assumed resource size reported by the CSI and the received rank indication. After the third CSI is decompressed by the model, the target CSI is obtained. In this way, by cooperating with the terminal side, the problem of unclear priority grouping and UCI discarding rules in model-compressed CSI reporting is solved. It ensures the performance of model-compressed CSI reporting while ensuring that the terminal does not need to repeatedly calculate the CSI.

[0269] like Figure 8 As shown, this application embodiment also provides a terminal, including: a transceiver 800, a memory 820, a processor 810, and a computer program stored in the memory 820 and executable on the processor 810; the transceiver 800 is used to receive and send data under the control of the processor 810; the processor 810 is used to read the program in the memory 820 and execute the following processes:

[0270] The target channel state information (CSI) is compressed using a first model to obtain a first CSI; the first CSI is then grouped according to the priority relationships of each output dimension of the first model; when the terminal needs to discard reported CSIs, it reports a portion of the first CSIs according to the priority of each group of CSIs; or,

[0271] The target CSI is grouped according to a preset grouping mode. Each group of CSIs is compressed by a first model to obtain a second CSI. When the terminal needs to discard the reported CSI, it reports part of the second CSI according to a first priority rule. The preset grouping mode is a grouping mode with at least two resources as the granularity, and the resources under each granularity satisfy a preset relationship.

[0272] Among them, Figure 8In this context, the bus architecture can include any number of interconnected buses and bridges, specifically linking various circuits of one or more processors represented by processor 810 and memory represented by memory 820 together. The bus architecture can also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. The bus interface provides an interface. The transceiver 800 can be multiple components, including transmitters and receivers, providing a unit for communicating with various other devices over a transmission medium, such as wireless channels, wired channels, optical fibers, etc. For different user equipment, the user interface 830 can also be an interface capable of connecting external or internal devices, including but not limited to keypads, displays, speakers, microphones, joysticks, etc.

[0273] The processor 810 is responsible for managing the bus architecture and general processing, while the memory 820 can store the data used by the processor 810 during operation.

[0274] Optionally, the processor 810 can be a central processing unit (CPU), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a complex programmable logic device (CPLD). The processor can also adopt a multi-core architecture.

[0275] The processor 810 executes any of the methods provided in the embodiments of this application according to the obtained executable instructions by calling program instructions stored in the memory. The processor 810 and the memory 820 may also be physically separated.

[0276] In some embodiments, the priority relationship of each output dimension of the first model is obtained after training the first target model according to predefined rules. The first target model includes the first model and a second model paired with the first model, and the second model is deployed on the network device side.

[0277] In some embodiments, the processor 810 is further configured to:

[0278] Based on the second target model, the weights of each output dimension of the first model are determined. The second target model includes the first model and the third model, wherein the third model is a replacement model for the second model, and the second model is deployed on the network device side.

[0279] Based on the weights of each output dimension of the first model, the priority relationship of each output dimension of the first model is obtained.

[0280] In some embodiments, the processor 810 is further configured to:

[0281] For each output dimension of the first model, obtain the first performance index of the second target model before the output dimension is removed;

[0282] By removing the output dimension, a second performance metric of the second target model is obtained.

[0283] The weight of the output dimension is determined by comparing the difference between the first performance metric and the second performance metric.

[0284] In some embodiments, the first priority rule is obtained through one or more of the following methods:

[0285] Protocol predefined;

[0286] Network device configuration;

[0287] Reporting from the terminal side.

[0288] In some embodiments, the resource includes one of the following:

[0289] Frequency domain resources;

[0290] Time-domain resources;

[0291] Airspace resources;

[0292] Angle domain resources.

[0293] In some embodiments, the transceiver 800 is further configured to receive a first reference signal transmitted by a network device; the processor 810 is further configured to:

[0294] Channel estimation is performed based on the first reference signal to obtain the first downlink channel matrix;

[0295] The first correlation matrix is ​​calculated based on the first downlink channel matrix;

[0296] The first correlation matrix is ​​subjected to eigenvalue decomposition to obtain an eigenvector, and the target CSI includes the eigenvector.

[0297] In some embodiments, the processor 810 is further configured to:

[0298] Channel estimation is performed based on the first reference signal to obtain the second downlink channel matrix;

[0299] The second downlink channel matrix is ​​preprocessed to obtain the first downlink channel matrix;

[0300] The preprocessing includes at least one of the following:

[0301] Angular domain transformation;

[0302] Time delay domain transformation;

[0303] Doppler domain transform.

[0304] It should be noted that the apparatus provided in this application embodiment can implement all the method steps implemented in the above method embodiment and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.

[0305] like Figure 9 As shown in the illustration, this application also provides a CSI reporting device 900, comprising:

[0306] The first processing unit 901 is configured to compress the target channel state information (CSI) using a first model to obtain a first CSI; group the first CSI according to the priority relationship of each output dimension of the first model; and when the terminal needs to discard reported CSIs, report a portion of the first CSIs according to the priority of each group of CSIs; or...

[0307] The second processing unit 902 is used to group the target CSI according to a preset grouping mode, and compress each group of CSIs using a first model to obtain a second CSI; when the terminal needs to discard the reported CSI, it reports part of the second CSI according to a first priority rule, wherein the preset grouping mode is a grouping mode with at least two resources as granularity, and the resources under each granularity satisfy a preset relationship.

[0308] In some embodiments, the priority relationship of each output dimension of the first model is obtained after training the first target model according to predefined rules. The first target model includes the first model and a second model paired with the first model, and the second model is deployed on the network device side.

[0309] In some embodiments, the apparatus of this application further includes:

[0310] The tenth processing unit is used to determine the weights of each output dimension of the first model according to the second target model. The second target model includes the first model and the third model, wherein the third model is a replacement model for the second model, and the second model is deployed on the network device side.

[0311] The eleventh processing unit is used to obtain the priority relationship of each output dimension of the first model based on the weights of each output dimension of the first model.

[0312] In some embodiments, the tenth processing unit is specifically used for:

[0313] For each output dimension of the first model, obtain the first performance index of the second target model before the output dimension is removed;

[0314] By removing the output dimension, a second performance metric of the second target model is obtained.

[0315] The weight of the output dimension is determined by comparing the difference between the first performance metric and the second performance metric.

[0316] In some embodiments, the first priority rule is obtained through one or more of the following methods:

[0317] Protocol predefined;

[0318] Network device configuration;

[0319] Reporting from the terminal side.

[0320] In some embodiments, the resource includes one of the following:

[0321] Frequency domain resources;

[0322] Time-domain resources;

[0323] Airspace resources;

[0324] Angle domain resources.

[0325] In some embodiments, the apparatus of this application further includes:

[0326] The second receiving unit is used to receive the first reference signal sent by the network device;

[0327] A channel estimation unit is used to perform channel estimation based on the first reference signal to obtain a first downlink channel matrix;

[0328] The first calculation unit is used to calculate the first correlation matrix based on the first downlink channel matrix;

[0329] The feature decomposition unit is used to perform feature decomposition on the first correlation matrix to obtain a feature vector, and the target CSI includes the feature vector.

[0330] In some embodiments, the channel estimation unit is specifically used for:

[0331] Channel estimation is performed based on the first reference signal to obtain the second downlink channel matrix;

[0332] The second downlink channel matrix is ​​preprocessed to obtain the first downlink channel matrix;

[0333] The preprocessing includes at least one of the following:

[0334] Angular domain transformation;

[0335] Time delay domain transformation;

[0336] Doppler domain transform.

[0337] It should be noted that the division of units in the embodiments of this application is illustrative and only represents one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units.

[0338] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a processor-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0339] It should be noted that the apparatus provided in this application embodiment can implement all the method steps implemented in the above method embodiment and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.

[0340] like Figure 10 As shown in the illustration, this application also provides a network device, including: a transceiver 1000, a memory 1020, a processor 1010, and a computer program stored in the memory 1020 and executable on the processor 1010; the transceiver 1000 is configured to receive and transmit data under the control of the processor 1010, and the transceiver 1000 is further configured to:

[0341] Receive partial Channel Status Information (CSI) reported by the receiving terminal;

[0342] The processor 1010 is further configured to: decompress a portion of the first CSI using a second model to obtain a target CSI, wherein a portion of the first CSI is related to the priority relationship of each output dimension of the first model, and the first model is deployed on the terminal side; or,

[0343] The transceiver 1000 is also used to: receive a portion of the second CSI reported by the terminal;

[0344] The processor 1010 is further configured to: decompress a portion of the second CSI using a second model to obtain a model output result; and perform interpolation processing on the model output result to obtain a target CSI, wherein a portion of the second CSI is related to a preset grouping mode and a first priority rule, wherein the preset grouping mode is a grouping mode with at least two resources as granularity, and each resource satisfies a preset relationship at each granularity.

[0345] Among them, Figure 10 In this context, the bus architecture can include any number of interconnected buses and bridges, specifically linking various circuits together, represented by one or more processors (processor 1010) and memory (memory 1020). The bus architecture can also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. The bus interface provides an interface. The transceiver 1000 can be multiple elements, including transmitters and receivers, providing units for communicating with various other devices over transmission media, including wireless channels, wired channels, optical fibers, etc. The processor 1010 is responsible for managing the bus architecture and general processing, and the memory 1020 can store data used by the processor 1010 during operation.

[0346] The processor 1010 can be a central processing unit (CPU), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a complex programmable logic device (CPLD). The processor can also adopt a multi-core architecture.

[0347] The processor 1010 executes any of the methods described in the embodiments of this application by calling a computer program stored in the memory 1020, according to the obtained executable instructions. The processor 1010 and the memory 1020 may also be physically separated.

[0348] In some embodiments, the processor 1010 is further configured to:

[0349] The first target model is trained according to predefined rules to obtain the trained first target model, which includes a first model that satisfies the predefined priority relationship of each output dimension and a second model that is paired with the first model.

[0350] In some embodiments, the processor 1010 is further configured to:

[0351] During model training, a random mask is applied to the first output vector of the first model, and the resulting second output vector is input into the second model to obtain the output result. The random mask is determined by the predefined rules.

[0352] Adjust the model parameters based on the output results, and train the model based on the first target model with adjusted parameters until the first target model is obtained.

[0353] In some embodiments, the processor 1010 is further configured to:

[0354] Based on the third target model, the weights of each output dimension of the first model are determined. The third target model includes the second model and the fourth model. The fourth model is a replacement model for the first model. The first model is deployed on the terminal side.

[0355] Based on the weights of each output dimension of the first model, the priority relationship of each output dimension of the first model is obtained.

[0356] In some embodiments, the processor 1010 is further configured to:

[0357] For each output dimension of the first model, obtain the third performance index of the third target model before the output dimension is removed;

[0358] The output dimension is removed to obtain the fourth performance metric of the third objective model;

[0359] The weights of the output dimensions are determined by comparing the differences between the third and fourth performance metrics.

[0360] In some embodiments, the first priority rule is obtained through one or more of the following methods:

[0361] Protocol predefined;

[0362] Network device configuration;

[0363] Reporting from the terminal side.

[0364] In some embodiments, the resource includes one of the following:

[0365] Frequency domain resources;

[0366] Time-domain resources;

[0367] Airspace resources;

[0368] Angle domain resources.

[0369] In some embodiments, the transceiver 1000 is further configured to:

[0370] A first reference signal is sent to the terminal, and the first reference signal is used to generate a first downlink channel matrix.

[0371] It should be noted that the apparatus provided in this application embodiment can implement all the method steps implemented in the above method embodiment and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.

[0372] like Figure 11 As shown in the figure, this application embodiment also provides an information processing device 1100, including:

[0373] The third processing unit 1101 is configured to receive a portion of the first channel state information (CSI) reported by the terminal; decompress the portion of the first CSI using a second model to obtain the target CSI, wherein the portion of the first CSI is related to the priority relationship of each output dimension of the first model, and the first model is deployed on the terminal side; or,

[0374] The fourth processing unit 1102 is used to receive a portion of the second CSI reported by the terminal; decompress the portion of the second CSI using a second model to obtain the model output result; and perform interpolation processing on the model output result to obtain the target CSI. The portion of the second CSI is related to a preset grouping mode and a first priority rule. The preset grouping mode is a grouping mode with at least two resources as granularity, and the resources at each granularity satisfy a preset relationship.

[0375] In some embodiments, the apparatus of this application further includes:

[0376] The model training unit is used to train a first target model according to predefined rules to obtain a trained first target model. The trained first target model includes a first model that satisfies the predefined priority relationship of each output dimension and a second model that is paired with the first model.

[0377] In some embodiments, the model training unit is specifically used for:

[0378] During model training, a random mask is applied to the first output vector of the first model, and the resulting second output vector is input into the second model to obtain the output result. The random mask is determined by the predefined rules.

[0379] Adjust the model parameters based on the output results, and train the model based on the first target model with adjusted parameters until the first target model is obtained.

[0380] In some embodiments, the apparatus of this application further includes:

[0381] The twelfth processing unit is used to determine the weights of each output dimension of the first model according to the third target model. The third target model includes the second model and the fourth model. The fourth model is a replacement model for the first model. The first model is deployed on the terminal side.

[0382] The thirteenth processing unit is used to obtain the priority relationship of each output dimension of the first model based on the weights of each output dimension of the first model.

[0383] In some embodiments, the twelfth processing unit is specifically used for:

[0384] For each output dimension of the first model, obtain the third performance index of the third target model before the output dimension is removed;

[0385] The output dimension is removed to obtain the fourth performance metric of the third objective model;

[0386] The weights of the output dimensions are determined by comparing the differences between the third and fourth performance metrics.

[0387] In some embodiments, the first priority rule is obtained through one or more of the following methods:

[0388] Protocol predefined;

[0389] Network device configuration;

[0390] Reporting from the terminal side.

[0391] In some embodiments, the resource includes one of the following:

[0392] Frequency domain resources;

[0393] Time-domain resources;

[0394] Airspace resources;

[0395] Angle domain resources.

[0396] In some embodiments, the method of this application further includes:

[0397] The first transmitting unit is used to transmit a first reference signal to the terminal, the first reference signal being used to generate a first downlink channel matrix.

[0398] It should be noted that the division of units in the embodiments of this application is illustrative and only represents one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units.

[0399] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a processor-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0400] It should be noted that the apparatus provided in this application embodiment can implement all the method steps implemented in the above method embodiment and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.

[0401] In some embodiments of this application, a non-transitory readable storage medium is also provided, which stores a program for executing the above-described... Figure 2 The CSI reporting method described above, or the method described above, can be used to report CSI information. Figure 3 The information processing method described herein.

[0402] The non-transiently readable storage medium can be any available medium or data storage device that the processor can access, including but not limited to magnetic memory (e.g., floppy disk, hard disk, magnetic tape, magneto-optical disk (MO)), optical memory (e.g., compact disc (CD), digital video disc (DVD), Blu-ray disc (BD), high-definition versatile disc (HVD)), and semiconductor memory (e.g., ROM, erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), non-volatile memory (NAND (Non-volatile Memory Device) FLASH), solid-state drives (SSD), etc.).

[0403] When executed by the processor, this program can achieve the above-mentioned applications, such as... Figure 2 The terminal-side method shown or as... Figure 3 To avoid repetition, all implementation methods shown in the network device-side embodiments will not be described again here.

[0404] This application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the above-described... Figure 1 or Figure 3 The various processes of the method embodiments shown can achieve the same technical effect, and will not be described again here to avoid repetition.

[0405] like Figure 12 As shown, this application embodiment also provides a terminal, which has models for different feedback overheads, including: a transceiver 1200, a memory 1220, a processor 1210, and a computer program stored in the memory 1220 and executable on the processor 1210; the transceiver 1200 is used to receive and send data under the control of the processor 1210; the processor 1210 is used to read the program in the memory 1220 and execute the following processes:

[0406] Based on the second reference signal sent by the network device, obtain the rank indication and target channel state information (CSI);

[0407] Based on the rank indicator and the resource size or assumed resource size reported by CSI, select the matching fourth target model;

[0408] The target CSI is compressed using the fourth target model to obtain the third CSI;

[0409] Report the rank indication and the third CSI.

[0410] Among them, Figure 12 In this context, the bus architecture can include any number of interconnected buses and bridges, specifically linking various circuits of one or more processors represented by processor 1210 and memory represented by memory 1220 together. The bus architecture can also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. The bus interface provides an interface. The transceiver 1200 can be multiple components, including transmitters and receivers, providing a unit for communicating with various other devices over a transmission medium, such as wireless channels, wired channels, optical fibers, etc. For different user equipment, the user interface 1230 can also be an interface capable of connecting external or internal devices, including but not limited to keypads, displays, speakers, microphones, joysticks, etc.

[0411] The processor 1210 is responsible for managing the bus architecture and general processing, and the memory 1220 can store the data used by the processor 1210 when performing operations.

[0412] Optionally, the processor 1210 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a complex programmable logic device (CPLD), and the processor may also adopt a multi-core architecture.

[0413] The processor 1210 executes any of the methods provided in the embodiments of this application according to the obtained executable instructions by calling program instructions stored in the memory. The processor 1210 and the memory 1220 may also be physically separated.

[0414] In some embodiments, when the bearer channel reported by CSI is PUCCH, the assumed PUCCH resource size reported by CSI is obtained through one or more of the following methods:

[0415] Protocol predefined;

[0416] Network device indication;

[0417] Reporting from the terminal side.

[0418] It should be noted that the apparatus provided in this application embodiment can implement all the method steps implemented in the above method embodiment and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.

[0419] like Figure 13 As shown in the illustration, this application also provides a CSI reporting device, which is applied to a terminal. The terminal has models for different feedback overheads, and the device includes:

[0420] The fifth processing unit 1301 is used to obtain the rank indication and target channel state information (CSI) based on the second reference signal sent by the network device.

[0421] The sixth processing unit 1302 is used to select a matching fourth target model based on the rank indication and the resource size or assumed resource size reported by CSI.

[0422] The seventh processing unit 1303 is used to compress the target CSI using the fourth target model to obtain the third CSI;

[0423] The first reporting unit 1304 is used to report the rank indication and the third CSI.

[0424] In some embodiments, when the bearer channel reported by CSI is PUCCH, the assumed PUCCH resource size reported by CSI is obtained through one or more of the following methods:

[0425] Protocol predefined;

[0426] Network device indication;

[0427] Reporting from the terminal side.

[0428] It should be noted that the division of units in the embodiments of this application is illustrative and only represents one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units.

[0429] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a processor-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0430] It should be noted that the apparatus provided in this application embodiment can implement all the method steps implemented in the above method embodiment and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.

[0431] like Figure 14 As shown in the illustration, this application also provides a network device having models for different feedback overheads, including: a transceiver 1400, a memory 1420, a processor 1410, and a computer program stored in the memory 1420 and executable on the processor 1410; the transceiver 1400 is configured to receive and transmit data under the control of the processor 1410, and is further configured to: receive rank indications and third CSIs reported by terminals; the processor 1410 is further configured to:

[0432] Based on the rank indication and the resource size or assumed resource size reported by the CSI, a matching fifth target model is selected; the third CSI is decompressed by the fifth target model to obtain the target CSI.

[0433] Among them, Figure 14In this context, the bus architecture can include any number of interconnected buses and bridges, specifically linking various circuits together, represented by one or more processors (processor 1410) and memory (memory 1420). The bus architecture can also link together various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and therefore will not be described further herein. The bus interface provides an interface. The transceiver 1400 can be multiple elements, including transmitters and receivers, providing units for communicating with various other devices over transmission media, including wireless channels, wired channels, optical fibers, etc. The processor 1410 is responsible for managing the bus architecture and general processing, and the memory 1420 can store data used by the processor 1410 during operation.

[0434] The processor 1410 can be a central processing unit (CPU), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a complex programmable logic device (CPLD). The processor can also adopt a multi-core architecture.

[0435] The processor 1410 executes any of the methods described in the embodiments of this application by calling a computer program stored in the memory 1420, according to the obtained executable instructions. The processor 1410 and the memory 1420 may also be physically separated.

[0436] In some embodiments, when the bearer channel reported by CSI is PUCCH, the assumed PUCCH resource size reported by CSI is obtained through one or more of the following methods:

[0437] Protocol predefined;

[0438] Network device indication;

[0439] Reporting from the terminal side.

[0440] It should be noted that the apparatus provided in this application embodiment can implement all the method steps implemented in the above method embodiment and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.

[0441] like Figure 15As shown in the illustration, this application also provides an information processing apparatus applied to a network device, the network device having models for different feedback overheads, the apparatus comprising:

[0442] The first receiving unit 1501 is used to receive the rank indication and the third CSI reported by the terminal;

[0443] The eighth processing unit 1502 is used to select a matching fifth target model based on the rank indication and the resource size or assumed resource size reported by CSI.

[0444] The ninth processing unit 1503 is used to decompress the third CSI through the fifth target model to obtain the target CSI.

[0445] In some embodiments, when the bearer channel reported by CSI is PUCCH, the assumed PUCCH resource size reported by CSI is obtained through one or more of the following methods:

[0446] Protocol predefined;

[0447] Network device indication;

[0448] Reporting from the terminal side.

[0449] It should be noted that the division of units in the embodiments of this application is illustrative and only represents one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units.

[0450] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a processor-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0451] It should be noted that the apparatus provided in this application embodiment can implement all the method steps implemented in the above method embodiment and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.

[0452] In some embodiments of this application, a non-transitory readable storage medium is also provided, which stores a program for executing the above-described... Figure 5 The CSI reporting method described above, or the method described above, can be used to report CSI information. Figure 7 The information processing method described herein.

[0453] The non-transiently readable storage medium can be any available medium or data storage device that the processor can access, including but not limited to magnetic memory (e.g., floppy disk, hard disk, magnetic tape, magneto-optical disk (MO)), optical memory (e.g., compact disc (CD), digital video disc (DVD), Blu-ray disc (BD), high-definition versatile disc (HVD)), and semiconductor memory (e.g., ROM, erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), non-volatile memory (NAND (Non-volatile Memory Device) FLASH), solid-state drives (SSD), etc.).

[0454] When executed by the processor, this program can achieve the above-mentioned applications, such as... Figure 5 The terminal-side method shown or as... Figure 7 To avoid repetition, all implementation methods shown in the network device-side embodiments will not be described again here.

[0455] This application also provides a computer program product, including computer instructions, which, when executed by a processor, implement the above-described... Figure 5 or Figure 7 The various processes of the method embodiments shown can achieve the same technical effect, and will not be described again here to avoid repetition.

[0456] The technical solutions provided in this application can be applied to a variety of systems. For example, applicable systems may include Global System for Mobile Communication (GSM), Code Division Multiple Access (CDMA), Wideband Code Division Multiple Access (WCDMA) General Packet Radio Service (GPRS), Long Term Evolution (LTE), LTE Frequency Division Duplex (FDD), LTE Time Division Duplex (TDD), Long Term Evolution Advanced (LTE-A), Universal Mobile Telecommunication System (UMTS), Worldwide Interoperability for Microwave Access (WiMAX), 5G New Radio (NR) and its evolutionary communication systems, and 6G (sixth generation mobile communication technology) systems. All of these systems include terminal equipment and network equipment. The system may also include a core network component, such as the Evolved Packet System (EPS) or the 5G system (5GS).

[0457] The terminal devices involved in the embodiments of this application can be devices that provide voice and / or data connectivity to users, handheld devices with wireless connectivity, or other processing devices connected to a wireless modem. The names of the terminal devices may differ in different systems; for example, in 5G or 6G systems, the terminal device may be called User Equipment (UE). Wireless terminal devices can be USB storage devices, other personal computer memory devices, and dongles. They can also communicate with one or more core networks (CNs) via a Radio Access Network (RAN). Wireless terminal devices can be mobile terminal devices, such as mobile phones (or "cellular" phones) and computers with mobile terminal devices. For example, they can be portable, pocket-sized, handheld, computer-embedded, or vehicle-mounted mobile devices that exchange voice and / or data with the radio access network. Examples of such devices include Personal Communication Service (PCS) phones, cordless phones, Session Initiated Protocol (SIP) phones, Wireless Local Loop (WLL) stations, Personal Digital Assistants (PDAs), personal computers, tablets, and Machine-type Communication (MTC) terminal devices. Wireless terminal devices can also be referred to as systems, subscriber units, subscriber stations, mobile stations, mobile devices, remote stations, access points, remote terminals, access terminals, user terminals, user agents, user devices, and wireless access devices and routers / modems that meet the limitations of this definition; however, this application does not limit the scope of the embodiments.

[0458] The network device involved in this application embodiment can be a base station, which may include multiple cells providing services to terminals. Depending on the specific application, a base station may also be called an access point, or a device in the access network that communicates with wireless terminal devices through one or more sectors on the air interface, or other names. The network device can be used to exchange received air frames with Internet Protocol (IP) packets, acting as a router between the wireless terminal device and the rest of the access network, where the rest of the access network may include an Internet Protocol (IP) communication network. The network device can also coordinate the attribute management of the air interface. For example, the network equipment involved in the embodiments of this application can be a base transceiver station (BTS) in a Global System for Mobile communications (GSM) or Code Division Multiple Access (CDMA), a NodeB in a Wide-band Code Division Multiple Access (WCDMA) system, an evolved Node B (eNB or e-NodeB) in a long term evolution (LTE) system, a 5G base station (gNB) in a next generation system, a Home evolved Node B (HeNB), a relay node, a femto, a pico, network testing equipment, etc., and is not limited in the embodiments of this application. In some network structures, the network equipment may include centralized unit (CU) nodes and distributed unit (DU) nodes, and the centralized unit and distributed unit may also be geographically separated.

[0459] Network devices and terminal devices can each use one or more antennas for multiple-input multiple-output (MIMO) transmission. MIMO transmission can be single-user MIMO (SU-MIMO) or multiple-user MIMO (MU-MIMO). Depending on the configuration and number of antenna combinations, MIMO transmission can be 2D-MIMO, 3D-MIMO, FD-MIMO, or massive-MIMO, and can also be diversity transmission, precoding transmission, or beamforming transmission, etc.

[0460] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0461] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-executable instructions. These computer-executable instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0462] These processor-executable instructions may also be stored in a processor-readable memory that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the processor-readable memory produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0463] These processors can execute instructions that can also be loaded onto a computer or other programmable data processing device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable device for implementing the process. Figure 1One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0464] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A CSI reporting method, applied to a terminal, characterized in that, include: The target channel state information (CSI) is compressed using the first model to obtain the first CSI; The first CSI is grouped according to the priority relationship of each output dimension of the first model; When the terminal needs to discard the reported CSI, it reports part of the first CSI according to the priority of each group of CSIs; or, The target CSI is grouped according to a preset grouping pattern. Each group of CSIs is then compressed using a first model to obtain a second CSI. When the terminal needs to discard the reported CSI, it reports part of the second CSI according to the first priority rule. The preset grouping mode is a grouping mode with at least two resources as the granularity, and the resources under each granularity satisfy the preset relationship.

2. The method according to claim 1, characterized in that, The priority relationship of each output dimension of the first model is obtained after training the first target model according to predefined rules. The first target model includes the first model and a second model paired with the first model. The second model is deployed on the network device side.

3. The method according to claim 1, characterized in that, The method further includes: Based on the second target model, the weights of each output dimension of the first model are determined. The second target model includes the first model and the third model, wherein the third model is a replacement model for the second model, and the second model is deployed on the network device side. Based on the weights of each output dimension of the first model, the priority relationship of each output dimension of the first model is obtained.

4. The method according to claim 3, characterized in that, The step of determining the weights of each output dimension of the first model based on the second target model includes: For each output dimension of the first model, obtain the first performance index of the second target model before the output dimension is removed; By removing the output dimension, a second performance metric of the second target model is obtained. The weight of the output dimension is determined by comparing the difference between the first performance metric and the second performance metric.

5. The method according to claim 1, characterized in that, The first priority rule is obtained through one or more of the following methods: Protocol predefined; Network device configuration; Reporting from the terminal side.

6. The method according to claim 1, characterized in that, The resources include one of the following: Frequency domain resources; Time-domain resources; Airspace resources; Angle domain resources.

7. The method according to claim 1, characterized in that, Before grouping the target CSIs according to the preset grouping pattern, the method further includes: Receive the first reference signal sent by the network device; Channel estimation is performed based on the first reference signal to obtain the first downlink channel matrix; The first correlation matrix is ​​calculated based on the first downlink channel matrix; The first correlation matrix is ​​subjected to eigenvalue decomposition to obtain an eigenvector, and the target CSI includes the eigenvector.

8. The method according to claim 7, characterized in that, The step of performing channel estimation based on the first reference signal to obtain the first downlink channel matrix includes: Channel estimation is performed based on the first reference signal to obtain the second downlink channel matrix; The second downlink channel matrix is ​​preprocessed to obtain the first downlink channel matrix; The preprocessing includes at least one of the following: Angular domain transformation; Time delay domain transformation; Doppler domain transform.

9. An information processing method applied to a network device, characterized in that, include: The system receives a portion of the first channel state information (CSI) reported by the receiving terminal; after decompressing the portion of the first CSI using a second model, it obtains the target CSI. The portion of the first CSI is related to the priority relationship of each output dimension of the first model, which is deployed on the terminal side. Alternatively, The system receives a portion of the second CSI reported by the receiving terminal; decompresses the portion of the second CSI using a second model to obtain the model output result; and performs interpolation processing on the model output result to obtain the target CSI. The portion of the second CSI is related to a preset grouping mode and a first priority rule. The preset grouping mode is a grouping mode with at least two resources as granularity, and the resources at each granularity satisfy a preset relationship.

10. The method according to claim 9, characterized in that, The method further includes: The first target model is trained according to predefined rules to obtain the trained first target model, which includes a first model that satisfies the predefined priority relationship of each output dimension and a second model that is paired with the first model.

11. The method according to claim 10, characterized in that, The step of training the first target model according to predefined rules to obtain the trained first target model includes: During model training, a random mask is applied to the first output vector of the first model, and the resulting second output vector is input into the second model to obtain the output result. The random mask is determined by the predefined rules. Adjust the model parameters based on the output results, and train the model based on the first target model with adjusted parameters until the first target model is obtained.

12. The method according to claim 9, characterized in that, The method further includes: Based on the third target model, the weights of each output dimension of the first model are determined. The third target model includes the second model and the fourth model. The fourth model is a replacement model for the first model. The first model is deployed on the terminal side. Based on the weights of each output dimension of the first model, the priority relationship of each output dimension of the first model is obtained.

13. The method according to claim 11, characterized in that, The step of determining the weights of each output dimension of the first model based on the third target model includes: For each output dimension of the first model, obtain the third performance index of the third target model before the output dimension is removed; The output dimension is removed to obtain the fourth performance metric of the third objective model; The weights of the output dimensions are determined by comparing the differences between the third and fourth performance metrics.

14. The method according to claim 9, characterized in that, The first priority rule is obtained through one or more of the following methods: Protocol predefined; Network device configuration; Reporting from the terminal side.

15. The method according to claim 9, characterized in that, The resources include one of the following: Frequency domain resources; Time-domain resources; Airspace resources; Angle domain resources.

16. The method according to claim 9, characterized in that, The method further includes: A first reference signal is sent to the terminal, and the first reference signal is used to generate a first downlink channel matrix.

17. A CSI reporting method, applied to a terminal, characterized in that, The terminal has models for different feedback overheads, and the method includes: Based on the second reference signal sent by the network device, obtain the rank indication and target channel state information (CSI); Based on the rank indicator and the resource size or assumed resource size reported by CSI, select the matching fourth target model; The target CSI is compressed using the fourth target model to obtain the third CSI; Report the rank indication and the third CSI.

18. The method according to claim 17, characterized in that, When the bearer channel reported by CSI is PUCCH, the assumed PUCCH resource size reported by CSI is obtained through one or more of the following methods: Protocol predefined; Network device indication; Reporting from the terminal side.

19. An information processing method applied to a network device, characterized in that, The network device has models for different feedback overheads, and the method includes: Receive the rank indication and third CSI reported by the receiving terminal; Based on the rank indication and the resource size or assumed resource size reported by CSI, select the matching fifth target model; The target CSI is obtained by decompressing the third CSI using the fifth target model.

20. The method according to claim 19, characterized in that, When the bearer channel reported by CSI is PUCCH, the assumed PUCCH resource size reported by CSI is obtained through one or more of the following methods: Protocol predefined; Network device indication; Reporting from the terminal side.

21. A terminal, characterized in that, include: Memory, transceiver, processor: Memory is used to store program instructions; A transceiver, used to send and receive data under the control of the processor, and to perform the following operations: The target channel state information (CSI) is compressed using a first model to obtain a first CSI; the first CSI is then grouped according to the priority relationship of each output dimension of the first model. When the terminal needs to discard the reported CSI, it reports part of the first CSI according to the priority of each group of CSIs; or, The target CSI is grouped according to a preset grouping pattern. Each group of CSIs is then compressed using a first model to obtain a second CSI. When the terminal needs to discard the reported CSI, it reports part of the second CSI according to the first priority rule. The preset grouping mode is a grouping mode with at least two resources as the granularity, and the resources under each granularity satisfy the preset relationship.

22. A CSI reporting device, characterized in that, include: The first processing unit is used to compress the target channel state information (CSI) using a first model to obtain the first CSI. The first CSI is grouped according to the priority relationship of each output dimension of the first model; When the terminal needs to discard the reported CSI, it reports part of the first CSI according to the priority of each group of CSIs; or, The second processing unit is used to group the target CSI according to a preset grouping mode, and compress each group of CSI using the first model to obtain the second CSI. When the terminal needs to discard the reported CSI, it reports part of the second CSI according to the first priority rule. The preset grouping mode is a grouping mode with at least two resources as the granularity, and the resources under each granularity satisfy the preset relationship.

23. A network device, characterized in that, include: Memory, transceiver, processor: Memory is used to store program instructions; A transceiver, used to send and receive data under the control of the processor, and to perform the following operations: The system receives a portion of the first channel state information (CSI) reported by the receiving terminal; after decompressing the portion of the first CSI using a second model, it obtains the target CSI. The portion of the first CSI is related to the priority relationship of each output dimension of the first model, which is deployed on the terminal side. Alternatively, The system receives a portion of the second CSI reported by the receiving terminal; decompresses the portion of the second CSI using a second model to obtain the model output result; and performs interpolation processing on the model output result to obtain the target CSI. The portion of the second CSI is related to a preset grouping mode and a first priority rule. The preset grouping mode is a grouping mode with at least two resources as granularity, and the resources at each granularity satisfy a preset relationship.

24. An information processing device, characterized in that, include: The third processing unit is configured to receive a portion of the first channel state information (CSI) reported by the terminal; decompress the portion of the first CSI using a second model to obtain the target CSI, wherein the portion of the first CSI is related to the priority relationship of each output dimension of the first model, and the first model is deployed on the terminal side; or, The fourth processing unit is used to receive a portion of the second CSI reported by the terminal; decompress the portion of the second CSI using the second model to obtain the model output result; and perform interpolation processing on the model output result to obtain the target CSI. The portion of the second CSI is related to a preset grouping mode and a first priority rule. The preset grouping mode is a grouping mode with at least two resources as the granularity, and the resources at each granularity satisfy a preset relationship.

25. A terminal, characterized in that, The terminal has models for different feedback overheads, including: a memory, a transceiver, and a processor: the memory stores program instructions; the transceiver, under the control of the processor, sends and receives data and performs the following operations: Based on the second reference signal sent by the network device, obtain the rank indication and target channel state information (CSI); Based on the rank indicator and the resource size or assumed resource size reported by CSI, select the matching fourth target model; The target CSI is compressed using the fourth target model to obtain the third CSI; Report the rank indication and the third CSI.

26. A CSI reporting device, applied to a terminal, characterized in that, The terminal has models for different feedback overheads, including: The fifth processing unit is used to obtain the rank indication and target channel state information (CSI) based on the second reference signal sent by the network device. The sixth processing unit is used to select a matching fourth target model based on the rank indication and the resource size or assumed resource size reported by CSI. The seventh processing unit is used to compress the target CSI using the fourth target model to obtain the third CSI; The first reporting unit is used to report the rank indication and the third CSI.

27. A network device, characterized in that, The network device has models for different feedback overheads, including: a memory, a transceiver, and a processor: the memory stores program instructions; the transceiver, under the control of the processor, sends and receives data and performs the following operations: Receive the rank indication and third CSI reported by the receiving terminal; Based on the rank indication and the resource size or assumed resource size reported by CSI, select the matching fifth target model; The target CSI is obtained by decompressing the third CSI using the fifth target model.

28. An information processing device, applied to network equipment, characterized in that, The network device has models for different feedback overheads, including: The first receiving unit is used to receive the rank indication and the third CSI reported by the terminal; The eighth processing unit is used to select a matching fifth target model based on the rank indication and the resource size or assumed resource size reported by CSI. The ninth processing unit is used to decompress the third CSI through the fifth target model to obtain the target CSI.

29. A non-transiently readable storage medium, characterized in that, The non-transiently readable storage medium stores a program for performing the steps of the CSI reporting method according to any one of claims 1 to 8, or the steps of the information processing method according to any one of claims 9 to 16, or the steps of the CSI reporting method according to any one of claims 17 to 18, or the steps of the information processing method according to any one of claims 19 to 20.

Images