WO2024046140A1 - 一种反馈处理方法、装置、存储介质及电子装置 - Google Patents

一种反馈处理方法、装置、存储介质及电子装置 Download PDF

Info

Publication number
WO2024046140A1
WO2024046140A1 PCT/CN2023/113645 CN2023113645W WO2024046140A1 WO 2024046140 A1 WO2024046140 A1 WO 2024046140A1 CN 2023113645 W CN2023113645 W CN 2023113645W WO 2024046140 A1 WO2024046140 A1 WO 2024046140A1
Authority
WO
WIPO (PCT)
Prior art keywords
target
csi data
csi
group
data
Prior art date
Application number
PCT/CN2023/113645
Other languages
English (en)
French (fr)
Inventor
卞青
于涵
陈玉
Original Assignee
深圳市中兴微电子技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳市中兴微电子技术有限公司 filed Critical 深圳市中兴微电子技术有限公司
Publication of WO2024046140A1 publication Critical patent/WO2024046140A1/zh

Links

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B7/00—Radio transmission systems, i.e. using radiation field
    • H04B7/02—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas
    • H04B7/04—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas
    • H04B7/06—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the transmitting station
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L5/00—Arrangements affording multiple use of the transmission path

Definitions

  • Embodiments of the present disclosure relate to, but are not limited to, the field of communications, and specifically, to a feedback processing method, device, storage medium, and electronic device.
  • the 5G system uses large-scale antenna arrays and beam forming technology to improve system performance.
  • Beam forming technology based on large-scale antenna arrays requires that the transmitter can accurately obtain channel state information (CSI), and select the optimal beam to transmit data based on this channel information.
  • Channel state information needs to be fed back to the sending end through the UE (User Equipment) at the receiving end.
  • This information includes channel state information - reference signals (CSI Reference Signals, referred to as CSI-RS), resource indication CRI (CSI-RS Resource Indicator) ), the rank indicator of the channel matrix (rank indicator, referred to as RI), the precoding codebook index (Precoding Matrix Index, referred to as PMI), and the channel quality indicator (Channel Quality Indicator, referred to as CQI).
  • the UE uses channel estimation to obtain the channel coefficient matrix H and noise coefficient No, and then calculates and reports the corresponding parameters according to the codebook type and feedback parameters configured by the base station.
  • Embodiments of the present disclosure provide a feedback processing method, device, storage medium and electronic device to at least solve the problem in the related art that short-term feedback interruption caused when the feedback process is abnormal will reduce the stability of the feedback-based communication system.
  • a feedback processing method is provided, which is applied to a base station.
  • the method includes: receiving CSI data reported by the UE in the target group according to the corresponding reporting configuration information, wherein the reporting configuration information is used to Instruct the UE in the target group to report the CSI data in the preset reporting time slot; if the CSI data reported by the target UE in the target group is not received in the preset reporting time slot, based on training
  • a good channel prediction model uses the CSI data reported by other UEs in the target group except the target UE and the CSI data reported by the target UE in a preset time period to perform channel prediction and obtain optimized CSI data; Downlink resources used for sending downlink data are determined according to the optimized CSI data.
  • a feedback processing method is also provided, which is applied to the target UE.
  • the method includes: determining channel state information CSI data, wherein the CSI data is calculated CSI data or the calculated CSI data. Optimize the CSI data based on the calculated CSI; report the CSI data to the base station according to the reporting configuration information corresponding to the target group, so that if the base station fails to report in the preset reporting time slot indicated by the reporting configuration information, After receiving the CSI data, based on the trained channel prediction model, use the CSI data reported by other UEs in the target group except the target UE and the CSI data reported in the preset time period to perform channel prediction, and obtain optimization the optimized CSI data; determine the downlink resources used for downlink data transmission according to the optimized CSI data, wherein the reporting configuration information is used to indicate that the UE in the target group is in the preset reporting time slot Report the CSI data.
  • a feedback processing device is also provided, which is applied to a base station.
  • the device includes: a receiving data module configured to receive CSI data reported by the UE in the target group according to the corresponding reporting configuration information, Wherein, the reporting configuration information is used to instruct the UE in the target group to report the CSI data in the preset reporting time slot; the first channel prediction module is configured to: After receiving the CSI data reported by the target UE in the target group, based on the trained channel prediction model, use the CSI data reported by other UEs in the target group except the target UE. Perform channel prediction on the CSI data and the CSI data reported by the target UE in a preset time period to obtain optimized CSI data; the resource determination module is configured to determine the downlink resources used for sending downlink data based on the optimized CSI data.
  • a UE packet processing device is also provided, which is applied to a target UE.
  • the device includes: a determining data module configured to determine channel state information CSI data, wherein the CSI data is The calculated CSI data or the optimized CSI data of the calculated CSI; the reporting module is configured to report the CSI data to the base station according to the reporting configuration information corresponding to the target group, so that if the base station is The CSI data is not received in the preset reporting time slot indicated by the reporting configuration information. Based on the trained channel prediction model, the CSI data reported by other UEs except the target UE in the target group is used.
  • a computer-readable storage medium is also provided, and a computer program is stored in the storage medium, wherein the computer program is configured to execute any of the above method embodiments when running. steps in.
  • an electronic device including a memory and a processor.
  • a computer program is stored in the memory, and the processor is configured to run the computer program to perform any of the above. Steps in method embodiments.
  • Figure 1 is a hardware structure block diagram of a mobile terminal of a feedback processing method according to an embodiment of the present disclosure
  • Figure 2 is a flow chart of a feedback processing method according to an embodiment of the present disclosure
  • Figure 3 is a flow chart of a feedback processing method according to another embodiment of the present disclosure.
  • Figure 4 is a schematic diagram of the main application scenarios of the 5G system according to an embodiment of the present disclosure.
  • FIG. 5 is a block diagram of a WIFI and receiver system according to an embodiment of the present disclosure
  • Figure 6 is a flow chart of the process of grouping decision-making and grouping channel prediction according to an embodiment of the present disclosure
  • Figure 7 is a block diagram of a feedback processing device according to an embodiment of the present disclosure.
  • FIG. 8 is a block diagram of a feedback processing device according to another embodiment of the present disclosure.
  • FIG. 1 is a hardware structure block diagram of a mobile terminal of the feedback processing method according to an embodiment of the present disclosure.
  • the mobile terminal may include one or more (only one is shown in Figure 1 ) processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data, wherein the above-mentioned mobile terminal may also include a processor for communication functions.
  • Transmission device 106 and input and output device 108 may be executed in a mobile terminal, a computer terminal, or a similar computing device.
  • the structure shown in Figure 1 is only illustrative, and it does not limit the structure of the above-mentioned mobile terminal.
  • the mobile terminal may also include more or fewer components than shown in FIG. 1 , or have a different configuration than shown in FIG. 1 .
  • the memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as the computer program corresponding to the feedback processing method in the embodiment of the present disclosure.
  • the processor 102 executes the computer program stored in the memory 104 to execute various tasks. Functional applications and business chain address pool slicing processing implement the above method.
  • Memory 104 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory.
  • the memory 104 may further include memory located remotely relative to the processor 102, and these remote memories may be connected to the mobile terminal through a network. Examples of the above-mentioned networks include but are not limited to the Internet, intranets, local area networks, mobile communication networks and combinations thereof.
  • the transmission device 106 is used to receive or send data via a network.
  • Specific examples of the above-mentioned network may include a wireless network provided by a communication provider of the mobile terminal.
  • the transmission device 106 includes a network adapter (Network Interface Controller, NIC for short), which can be connected to other network devices through a base station to communicate with the Internet.
  • the transmission device 106 may be a radio frequency (Radio Frequency, RF for short) module, which is used to communicate with the Internet wirelessly.
  • NIC Network Interface Controller
  • FIG. 2 is a flow chart of a feedback processing method according to an embodiment of the present disclosure. As shown in Figure 2, this method is applied to a base station. Includes the following steps:
  • Step S202 Receive CSI data reported by the UE in the target group according to the corresponding reporting configuration information, where the reporting configuration information is used to instruct the UE in the target group to report the CSI data in the preset reporting time slot;
  • Step S204 If the CSI data reported by the target UE in the target group is not received in the preset reporting time slot, based on the trained channel prediction model, use the CSI data reported by other UEs in the target group except the target UE.
  • the CSI data and the CSI data reported by the target UE in the preset time period are used for channel prediction to obtain optimized CSI data;
  • Step S206 Determine downlink resources used for downlink data transmission based on the optimized CSI data.
  • the CSI data in the target group except the target UE is used. Perform channel prediction on the CSI data reported by other UEs and the CSI data of the target UE to obtain the optimized CSI data; and/or if the target in the target group is received in the preset reporting time slot Based on the CSI data reported by the UE, determine whether the target UE is retained in the target group; if the determination result is that the target UE is not retained in the target group, the target UE is Remove the target UE from the target group to update the target group; if the judgment result is that the target UE is retained in the target group, retain the target UE in the target group.
  • determining whether the target UE remains in the target group based on the CSI data may specifically include: predicting the target UE according to a trained channel prediction model corresponding to the target group.
  • the reported CSI data is subjected to channel prediction to obtain the optimized CSI data; further, the channel in the CSI data is decomposed by SVD to obtain the eigenvalues and eigenvectors of the target UE on the subband,
  • the subband channel quality indication and the UE scheduling subband weight in the CSI data reported by the target UE can also be used to modify the feature value and feature vector corresponding to the CSI data of the target UE;
  • the eigenvalues and eigenvectors corresponding to the CSI data of the target UE, as well as the eigenvalues and eigenvector values corresponding to the CSI data of other UEs in the target group are input into the trained channel prediction model to obtain the trained channel Predictive model output of the optimized CSI data, the optimized CSI data is the optimized
  • the above-mentioned determination of whether the target UE remains in the target group based on the CSI data may specifically include: determining the group of the target UE based on the CSI data, and further, determining the group of the target UE according to the CSI data.
  • the CSI data of the target UE and the CSI data of each UE except the target UE determine the weighted correlation value of the UE pair composed of the target UE and each UE except the target UE, and determine the The group composed of the UE pairs with the correlation value greater than the preset threshold is the group of the target UE; or the CSI data is input into the trained grouping decision-making model to obtain the target output by the trained grouping decision-making model.
  • Grouping of UEs judging whether the grouping of the target UE is the same as the target grouping; when the grouping of the target UE is the same as the target grouping, determine that the judgment result is that the target UE remains in the In the target group; when the group of the target UE is different from the target group, it is determined that the judgment result is that the target UE is not retained in the target group.
  • the method further includes: when the judgment result is that the target UE is not retained in the target group, setting the target UE as an ungrouped user; If the result is that the target UE remains in the target group, the trained grouping decision model is calculated based on the CSI data of the target UE and the CSI data of other UEs in the target group except the target UE. Training is performed to update the trained group decision model.
  • the method further includes: if the CSI data reported by the target UE is received in the preset reporting time slot, based on the CSI data of the target UE and the stored target packet
  • the trained channel prediction model is trained with CSI data of other UEs except the target UE to update the trained channel prediction model.
  • the method further includes: when the target UE does not belong to any group, determining the grouping result of the target UE according to the CSI data reported by the target UE, wherein the grouping
  • the results include: the target UE joins an existing group, or the target UE forms a new group with other ungrouped users.
  • the method further includes: obtaining CSI data reported by multiple UEs within the coverage, wherein the target UE is any UE among the multiple UEs;
  • the plurality of UEs are initially grouped according to the CSI data of the plurality of UEs to obtain the plurality of groups, wherein the target grouping is any one of the plurality of groups;
  • each of the plurality of UEs is Group selection pairs A correspondingly trained channel prediction model is generated, and the trained channel prediction model, grouping parameters and reported configuration information are fed back to the multiple UEs.
  • Figure 3 is a flow chart of a feedback processing method according to another embodiment of the present disclosure. As shown in Figure 3, it is applied to a target UE.
  • the method include:
  • Step S302 Determine channel state information CSI data, where the CSI data is calculated CSI data or CSI data optimized for the calculated CSI;
  • Step S304 Report CSI data to the base station according to the reporting configuration information corresponding to the target group, so that if the base station does not receive CSI data in the preset reporting time slot indicated by the reporting configuration information, it will use the target based on the trained channel prediction model.
  • Downlink resources, wherein the reporting configuration information is configured to instruct the UE in the target group to report CSI data in the preset reporting time slot.
  • step S302 to S304 it is possible to solve the problem in the related technology that short-term feedback interruption caused when the feedback process is abnormal, which will cause the stability of the feedback-based communication system to be reduced.
  • channel prediction is performed based on the CSI data reported within the preset time period and the CSI data reported by other UEs in the group, and downlink resources are configured based on the channel prediction results to avoid the base station reconfiguring feedback resources. Effectively reduce the feedback reporting load of the 5G system and improve the overall robustness of the system.
  • the method before step S302, the method further includes: receiving a CSI-RS detection signal sent by the base station; determining whether the reception quality of the CSI-RS detection information meets a preset condition; If the result is yes, calculate the CSI data to obtain the calculated CSI data, and perform the trained channel prediction model according to the CSI data and the CSI data of other UEs in the target group issued by the base station. Training to update the trained channel prediction model; when the judgment result is no, calculate CSI data to obtain the calculated CSI data, and calculate the calculated CSI data according to the trained channel prediction model Perform channel prediction to obtain the optimized CSI data.
  • predicting the calculated CSI data according to the trained channel prediction model, and obtaining the optimized CSI data may specifically include: performing SVD decomposition on the channels in the CSI data to obtain the subbands.
  • the eigenvalues and eigenvectors of obtain the optimal output of the trained channel prediction model transformed CSI data.
  • the eigenvalues and eigenvectors corresponding to the CSI data may also be modified using the subband channel quality indication and the UE scheduling subband weight in the CSI data.
  • the method further includes: receiving the trained channel prediction model and grouping parameters selected for the target group sent by the base station; and/or setting the CSI data reporting period.
  • This embodiment is based on the feedback method of adaptive learning and UE grouping technology.
  • the base station Based on the traditional AI codec feedback process, the base station performs UE channel correlation grouping based on adaptive learning based on the UE feedback reporting results; the base station further The group-DCI is issued for each UE group to control each UE group to report periodically; then the base station puts the feedback UE data reported normally in each group into the training data set, and the UE that reports abnormally or does not report this time Use the data reported by the remaining UEs in the group to perform reporting reconstruction based on adaptive learning; the corresponding UE can also perform the aforementioned model training and reporting reconstruction on the UE side according to the grouping determined by the base station, thereby further reducing the reporting load and improving system feedback robustness sex.
  • Figure 4 is a schematic diagram of the main application scenarios of the 5G system according to an embodiment of the present disclosure. As shown in Figure 4, there can be multiple UEs within the coverage of a gNB. The gNB and each UE perform physical layer data transmission, feedback, control and other processes based on the 3GPP protocol. The feedback and other related behaviors of each UE are controlled and scheduled by gNB.
  • the current mainstream 5G system feedback methods are codebook-based feedback reporting, AI codec-based compressed CSI reporting, etc.
  • the basic process of these methods is that the base station sends the corresponding NZP CSI-RS detection signal to each UE.
  • the UE uses different methods (codebooks, AI codecs, etc.) to calculate the CSI reporting information based on the channel estimation results of the signal and pass the uplink channel This information is transmitted to gNB. After obtaining these CSI report information, gNB performs adaptive scheduling (including beam, precoding, MCS scheduling, etc.) for each user.
  • CSI reporting is often a periodic or non-periodic process.
  • detection and reporting are often based on beamforming transmission. Short-term occlusion may cause interruption of the feedback process. Under the current protocol framework, such interruption often requires the reconstruction of the beam connection between gNB and UE. After the process, re-enter the CSI reporting process, which increases system overhead and may cause system instability after frequent interruptions.
  • FIG. 5 is a block diagram of a WIFI and receiver system according to an embodiment of the present disclosure. As shown in Figure 5, the main module functions of the system are:
  • the CSI data set/UE grouping set module 501 is mainly responsible for storing and updating the decompressed CSI information of each UE; pushing the corresponding UECSI set data to the modules 503-506 according to the grouping results and UECSI reception evaluation results.
  • the CSI decompression and encoding module 502 is mainly responsible for decompressing the received compressed CSI information.
  • the UE grouping decision-making module 503 calculates the grouping decision-making result of the current UE set based on the UE CSI set data pushed by the module 501, and outputs the UE grouping, intra-group CSI compression codec model, etc.
  • the UE grouping decision model training module 504 trains the UE grouping decision model based on the data pushed by the module 501, and pushes the trained upgraded model to the module 503.
  • the intra-group channel prediction model training module 505 trains the channel prediction model of the group based on the data and grouping results pushed by the module 501, and pushes the trained upgraded model to the module 506.
  • the intra-group channel prediction calculation module 506 calculates the predicted CSI of the current UE based on the CSI set data of the group pushed by the module 501, and outputs the CSI prediction result.
  • the channel prediction result saving module 507 saves the prediction results for subsequent downlink transmission precoding calculations, downlink CSI compression transmission, etc.
  • the grouping and model selection saving module 508 saves the output of the module 503 and transmits the corresponding grouping and/or model selection results to the required modules according to the requirements of each module.
  • the CSI compression encoding module 509 compresses the channel data provided by the module 507 according to the compression model selected by the module 508 and transmits it to the module 515.
  • the downlink data sending module 510 sends data or reference signals of the corresponding UE according to the protocol.
  • the CSI compression encoding module 511 is mainly responsible for compressing the received CSI information.
  • the intra-group channel prediction calculation module 512 calculates the predicted CSI of the current UE based on the CSI set data of the group pushed by the module 513, and outputs the CSI prediction result.
  • the intra-group CSI data collection module 513 saves the UECSI estimation results and the intra-group CSI information issued by the gNB, and at the same time controls the data output to the corresponding module.
  • the intra-group channel prediction model training module 514 trains the UE CSI prediction model based on the CSI set data of this group pushed by the module 513, and outputs the model results.
  • the downlink data receiving/processing module 515 receives and decodes the data sent by the gNB and outputs the results, receives the CSI reference signal sent by the gNB and outputs the CSI estimation results.
  • the uplink data sending module 516 sends the downlink data decoding result/uplink data.
  • the uplink data receiving/processing module 517 receives and processes uplink data.
  • the link quality evaluation module 518 evaluates the link quality of the UE according to the decoding results reported by the UE.
  • Figure 6 is a flow chart of the process of grouping decision-making and grouping channel prediction according to an embodiment of the present disclosure. As shown in Figure 6, the process includes the following steps:
  • Step 601 short-cycle reporting.
  • the gNB schedules UEs within the coverage area to report CSI, and the reporting results are used as initial data for UE group selection.
  • Step 602 gNB grouping decision and grouping configuration. Specifically, all UEs are initially grouped based on adaptive learning based on the initial data, and the group feedback learning model of each group is selected, and the reporting period and reporting offset of each UE in each group are selected.
  • Step 603 long-term reporting.
  • gNB notifies each UE of its grouping feedback learning model, grouping parameters such as reporting period and reporting offset, and the UE uses the corresponding learning model to perform periodic reporting at a specific time based on the above parameters.
  • gNB After gNB receives a feedback report based on adaptive learning from a certain UE, it uses the reported information as training data for the channel prediction model in its group to determine whether the group continues to be established. If true, the data will also be used to upgrade the intra-group channel prediction model and be used as training data for the full UE grouping model to upgrade the grouping model; if not true, the data will be used as input data for the full UE grouping model, and gNB will use this data to The grouped UEs are re-grouped.
  • Step 604 UE side packet channel prediction. After receiving the grouping parameters issued by gNB and the CSI information of other UEs in the group, the UE can use the CSI information to enhance the UE's CSI estimate or train the UE-side prediction model through the UE-side intra-group channel prediction model.
  • Step 605 gNB side packet channel prediction. Specifically, if the gNB does not receive the reporting information of a certain UE in the CSI reporting time slot of the UE, it uses other UEs in the group where the UE is located and the latest CSI reporting information of the UE as the input of the intra-group channel prediction model, and outputs the The predicted channel result of the UE is used to continue the subsequent beam and codebook selection process. If the predicted channel result still does not maintain normal data communication for the UE, the beam failure process specified in the protocol will be entered to reselect the beam, and the UE will be deleted from the group. After the beam is rebuilt, the new CSI report will be used for regrouping.
  • the UE does not receive the corresponding CSI-RS signal in the time slot where CSI feedback should be performed.
  • the latest intra-group CSI information and the UE CSI information are used to input the UE-side intra-group channel prediction model, and the corresponding CSI information is output as the current CSI-RS signal.
  • CSI reported times are used to input the UE-side intra-group channel prediction model, and the corresponding CSI information is output as the current CSI-RS signal.
  • the UE grouping and intra-group prediction method in this embodiment will be described by taking an example in which there are 6 UEs within a gNB as shown in Figure 4.
  • gNB configures UE0-5 to report CSI.
  • UE0-5 uses the compression model at a specific moment according to the time, compression model, etc. configured by gNB (the codebook specified by the protocol can be used, AI-based adaptive compression coding, etc. ) compresses the CSI and reports it to the gNB.
  • the gNB After receiving the corresponding CSI information, the gNB performs decompression processing based on the compression model and stores the decompressed CSI information in the module 501.
  • gNB After gNB collects the CSI information of UE0-5, it inputs the CSI information of all UEs into the UE grouping decision-making module. This module processes the input CSI information based on the trained UE grouping decision-making model, outputs the UE grouping decision-making results, and groups the UE UEs whose channel correlation is higher than a certain threshold are grouped into a group, and a channel prediction model is selected for each UE group.
  • gNB After gNB completes the UE grouping, it passes the grouping results to module 501, and sends the grouping results and prediction model selection result module 505. At the same time, it notifies the UE module 512 of the corresponding grouping results, prediction models, CSI reporting configuration and other information through downlink DCI.
  • the DCI is transmitted using the CSI DCI format specified by the UE.
  • Table 1 The fields included in the CSI DCI format are shown in Table 1 below.
  • the CSI information of other UEs in the UE group except the own UE is compressed through module 509 and then transmitted to the UE.
  • the gNB transmits the latest updated compressed CSI information of the UE in the group to all UEs in the group through multicast.
  • the UE After receiving the grouping results, prediction model, CSI reporting configuration and other information, the UE parses the corresponding information and sets Set the prediction model of module 512, and set the reporting period, etc. at the same time. After receiving the CSI information of other UEs in the group, the UE first decompresses the corresponding CSI, and then stores the corresponding data in the module 513.
  • gNB and UE have completed the initial CSI information storage, initial UE grouping and signaling exchange between gNB and UE.
  • UE0 is an independent UE
  • UE1-3 is user group
  • UE4-5 is UE group 2.
  • UE0 will not obtain grouping related information, and its CSI reporting process uses the traditional process.
  • User group 0 contains 3 UEs (UE1-3).
  • gNB notifies each UE in group 0 of its prediction model selection results, other UE IDs in the group, and the CSI information of the corresponding UE. For example, gNB will notify UE1 Its prediction model informs that there are UE2 and UE3 in the group, and sends the CSI of UE2 and UE3 to UE1.
  • User group 1 contains 2 UEs (UE4-5).
  • gNB notifies each UE in group 1 of its prediction model selection results, other UE IDs in the group, and the CSI information of the corresponding UE. For example, gNB will notify UE4 of its prediction The model informs that there is UE5 in the group and sends the CSI of UE5 to UE4.
  • UE0 that has not joined the group receives the CSI-RS detection signal sent by gNB and the CSI-RS reception quality is good this time (the SNR of the detection signal can be used as a reference, a threshold is set, and the quality is considered to be better if it is higher than a certain threshold), it is calculated
  • the corresponding CSI information is stored in module 513.
  • Module 513 inputs the calculated CSI and the previously saved CSI information of UE0 into module 514 for channel prediction model training.
  • Module 514 outputs the upgraded model to module 512 for subsequent CSI compression reporting. Use the CSI information calculated this time directly.
  • UE1 in user group 0 When UE1 in user group 0 receives the CSI-RS sounding signal sent by gNB and the CSI-RS reception quality is good this time, it calculates the corresponding CSI information and stores it in module 513.
  • Module 513 stores the calculated UE1 CSI information this time and the previous
  • the saved CSI information of UE1 and other users in user group 0 (UE2/3, sent by gNB to UE1) is input to module 514 for channel prediction model training.
  • Module 514 outputs the upgraded model to module 512, and subsequent CSI compression reports are also used directly. The CSI information calculated this time.
  • the module 513 inputs the calculated UE1 CSI, the previously saved UE1 and the CSI information of other users in user group 0 (UE2/3, gNB delivers to UE1) into the module 512 Perform channel prediction and correction, and module 512 outputs the calculated CSI information to module 511 for compression. Report.
  • UE1 in user group 0 receives the CSI compression information of other UEs (such as UE2) in the group issued by gNB, it first decompresses the CSI and stores the decompressed CSI information in module 513.
  • Module 513 stores the previously saved CSI information.
  • the UE1CSI and UE3CSI input modules 514 perform channel prediction model training, and the module 514 outputs the upgraded model to the module 512.
  • the post-processing method is the same as that of UE1.
  • UE4 in user group 1 When UE4 in user group 1 receives the CSI-RS sounding signal sent by gNB and the CSI-RS reception quality is good this time, it calculates the corresponding CSI information and stores it in the module 513.
  • the module 513 stores the calculated UE4 CSI information this time and the previous
  • the saved CSI information of UE4 and other users in user group 1 (UE5, sent by gNB to UE4) is input to module 514 for channel prediction model training.
  • Module 514 outputs the upgraded model to module 512. Subsequent CSI compression reports are also directly used this time. Calculated CSI information.
  • UE4 in user group 1 After UE4 in user group 1 receives the CSI compression information of other UEs (such as UE5) in the group issued by gNB, it first performs CSI decompression and stores the decompressed CSI information in module 513.
  • Module 513 inputs the previously saved UE4 CSI into module 514 for channel prediction model training, and module 514 outputs the upgraded model to module 512.
  • gNB After receiving the CSI reported by UE0, gNB first decompresses the CSI. Since UE0 is not in any user group, it can directly store the received CSI information into module 501. The precoding matrix, number of layers, MCS, etc. used for subsequent downlink data transmission by UE0 are also directly calculated using its latest reported results. Since UE0 does not belong to any user group, module 501 will input the current CSI information of UE0 and the current grouping policy into the process grouping decision-making process of module 503 to decide whether UE0 should join a current user group or form a new user with other ungrouped users. Group.
  • gNB After receiving the CSI reported by UE1, gNB first decompresses the CSI. Since UE1 is in user group 0, gNB first stores the received CSI information in module 501. Module 101 determines whether to use the updated data for training or testing. If you choose to test, the updated UE1 CSI information and the stored UE2/3 CSI information will be input into the module 506 at the same time. Based on the model prediction results, it will be judged whether the current UE0 can continue to be retained in user group 0. If it cannot be retained, the module 501 will be notified to refresh the group.
  • gNB After gNB receives the CSI reported by UE2/3, the subsequent steps are the same as those received by UE1.
  • gNB After receiving the CSI reported by UE4, gNB first decompresses the CSI. Since UE4 is in user group 1, gNB first stores the received CSI information in module 501. Module 101 determines whether to use the updated data for training or testing. If you choose to test, the updated UE4 CSI information and the stored UE5 CSI information will be input into the module 506 at the same time. Based on the model prediction results, it will be judged whether the current UE4 can continue to be retained in user group 1. If it cannot be retained, the module 501 will be notified to refresh the grouping strategy.
  • module 101 Since user group 1 only has two users, UE4/5, module 101 will disband user group 1 at this time and put UE4/5 into the ungrouped user set; if retained, module 501 will be notified of the result, and module 501 will add UE4
  • the latest CSI data is used as the training data of module 504 to upgrade the grouping decision model. If training is selected, the updated UE4 CSI information and the stored UE5 CSI information are input into the module 505 at the same time, and the trained model is output.
  • the precoding matrix, number of layers, MCS, etc. used for UE4's downlink data transmission are also directly calculated using its latest reported results.
  • Module 513 will input the previously saved CSI information of UE0 into module 512 for channel prediction and correction.
  • Module 512 outputs the calculated CSI information to module 511 for compression and reporting.
  • module 513 will deliver the previously saved UE1 and other users in user group 0 (UE2/3, gNB) to UE1 )'s CSI information input module 512 performs channel prediction and correction, and the module 512 outputs the calculated CSI information to the module 511 for compression and reporting.
  • the post-processing method is the same as that of UE1.
  • module 513 will deliver the previously saved data of UE4 and other users in user group 1 (UE5, gNB) to UE4.
  • the CSI information input module 512 performs channel prediction and correction, and the module 512 outputs the calculated CSI information to the module 511 for compression and reporting.
  • the post-processing method is the same as that of UE4.
  • the module 501 inputs the CSI information previously saved by UE0 into the module 506 to predict the UE0 CSI information and outputs the predicted CSI result.
  • the precoding matrix, number of layers, MCS, etc. used for downlink data transmission of UE0 are also Calculated directly using its latest forecast results.
  • module 501 will update the CSI information previously saved by UE1 with other users (UE2/3) in user group 0.
  • the reported CSI information is simultaneously input into the module 506 to predict the UE1 CSI information and output the predicted CSI results.
  • the precoding matrix, number of layers, MCS, etc. used for UE1's downlink data transmission are also directly calculated using its latest prediction results.
  • gNB does not receive the reported CSI of UE2/3 in the specified CSI information reporting time slot, the subsequent steps are the same as those of UE0.
  • module 501 will combine the CSI information previously saved by UE4 with the latest reported by other users (UE5) in user group 1.
  • the CSI information simultaneous input module 506 predicts UE4 CSI information and outputs predicted CSI results.
  • the precoding matrix, number of layers, MCS, etc. used for UE4's downlink data transmission are also directly calculated using its latest prediction results.
  • gNB does not receive the reported CSI of UE5 in the specified CSI information reporting time slot, the subsequent steps are the same as those of UE4.
  • the user grouping decision on the gNB side in this embodiment can use correlation decision-making based on the UE channel feature vector, including:
  • gNB uses 16port NZP CSI-RS to detect UR0-5, then UE i decompresses H on subband s. for:
  • Rx,i represents the number of receiving antennas of UEi
  • Tx,gNB represents the number of sounding signal ports sent by gNB.
  • UE pairs whose correlation value is greater than the threshold (the threshold can be obtained through simulation), it is considered that they can be divided into Group, if there is overlap between groups, for example, the correlation value between UE0 and UE1 is greater than the threshold, and the correlation value between UE1 and UE2 is also greater than the threshold, then UE0, 1, and 2 are divided into one group.
  • Modified correlation decisions based on UE channel feature vectors can also be used, including:
  • gNB uses 16port NZP CSI-RS to detect UR0-5, then UE i decompresses H on subband s. for:
  • Rx,i represents the number of receiving antennas of UE i
  • Tx,gNB represents the number of sounding signal ports sent by gNB.
  • q i,s represents the sub-band CQI equivalent weight, which can be obtained using the following formula.
  • the modulation coefficient corresponding to the current sub-band CQI is x, where the value of x is as shown in Table 2 below.
  • w i,s represents the average scheduling rate of UEi in subband s within T seconds before this calculation, in the interval (0,1] Inside.
  • the threshold can be obtained through simulation
  • UE pairs whose correlation value is greater than the threshold it is considered that they can be divided into one group. If there is overlap between the groups, for example, the correlation value between UE0 and UE1 is greater than the threshold, and the correlation value between UE1 and UE2 is also greater than the threshold, then UE0,1,2 are divided into one group.
  • Adaptive learning-based user grouping decisions can also be used, including:
  • gNB uses 16port NZP CSI-RS to detect UR0-5, then the decompressed H of UE i on subband s is: After SVD decomposition, the eigenvalues and eigenvectors of UE i on subband s are obtained:
  • Rx,i represents the number of receiving antennas of UE i
  • Tx,gNB represents the number of sounding signal ports sent by gNB.
  • the model After preprocessing the channel information of each UE, the preprocessed feature value S and feature vector V (which may also include other weight parameters, such as subband quality parameters, scheduling parameters, etc.) are input into the trained Grouping decision-making model, the model calculates the final grouping result.
  • the preprocessed feature value S and feature vector V which may also include other weight parameters, such as subband quality parameters, scheduling parameters, etc.
  • the group decision-making model algorithm can use adaptive machine learning algorithms, adaptive reinforcement learning algorithms, adaptive deep learning algorithms, policy optimization reinforcement learning algorithms and other adaptive learning algorithms; the group decision-making model can use the partially observed Markov decision process (POMDP). ), artificial neural networks (including deep belief network (DBN), deep convolutional network (DCN), recurrent neural network (RNN), multi-layer perceptual neural network (MLPCN), convolutional neural network (CNN), etc.).
  • DBN deep belief network
  • DCN deep convolutional network
  • RNN recurrent neural network
  • CNN multi-layer perceptual neural network
  • CNN convolutional neural network
  • the calculation of the group decision-making model can be generated through offline training in the simulation environment, or online training and upgrade through the CSI information received by the system in real time.
  • the channel H obtained after channel estimation of this UE is preprocessed (usually SVD decomposition to extract feature values).
  • gNB uses 16port NZP CSI-RS for detection, then UE i H after channel estimation in time slot t subband s is:
  • the eigenvalues and eigenvectors of UE i on subband s are obtained:
  • Rx,i represents the number of receiving antennas of UE i
  • Tx,gNB represents the number of sounding signal ports sent by gNB
  • t represents the time slot number
  • the characteristic values saved in time slots t+n, t, t-n,.... of UE i are compared with Feature vector, the latest saved feature value and feature vector of other UEs (UE j) in the group (for example, the latest saved CSI of UE j in time slot t+x), the above data:
  • the grouping channel prediction algorithm can use adaptive machine learning algorithms, adaptive reinforcement learning algorithms, adaptive deep learning algorithms, policy optimization reinforcement learning algorithms and other adaptive learning algorithms; the grouping channel prediction can use the partially observed Markov decision process (POMDP). ), artificial neural networks (including deep belief network (DBN), deep convolutional network (DCN), recurrent neural network (RNN), multi-layer perceptual neural network (MLPCN), convolutional neural network (CNN), etc.).
  • POMDP partially observed Markov decision process
  • artificial neural networks including deep belief network (DBN), deep convolutional network (DCN), recurrent neural network (RNN), multi-layer perceptual neural network (MLPCN), convolutional neural network (CNN), etc.
  • the calculation of packet channel prediction can be generated through offline training in the simulation environment, or online training and upgrade through the CSI information received by the system in real time.
  • the gNB side user group channel prediction method described in this method is described as follows:
  • the decompressed channel H of all UEs in the group is preprocessed (usually SVD decomposition to extract feature values).
  • gNB uses 16port NZP CSI-RS for detection, then UE i is detected in time slot t H after decompressing CSI on band s is:
  • Rx,i represents the number of receiving antennas of UE i
  • Tx,gNB represents the number of sounding signal ports sent by gNB
  • t represents the time slot number
  • the feature value and feature vector saved in time slot t,t-n,.... of UE i will be updated by other UEs (UE j) in the group.
  • the saved eigenvalues and eigenvectors (for example, the latest CSI of UE j is saved in time slot t+x), the above data:
  • the grouping channel prediction algorithm can use adaptive machine learning algorithms, adaptive reinforcement learning algorithms, adaptive deep learning algorithms, policy optimization reinforcement learning algorithms and other adaptive learning algorithms; the grouping channel prediction can use the partially observed Markov decision process (POMDP). ), artificial neural networks (including deep belief network (DBN), deep convolutional network (DCN), recurrent neural network (RNN), multi-layer perceptual neural network (MLPCN), convolutional neural network (CNN), etc.).
  • POMDP partially observed Markov decision process
  • artificial neural networks including deep belief network (DBN), deep convolutional network (DCN), recurrent neural network (RNN), multi-layer perceptual neural network (MLPCN), convolutional neural network (CNN), etc.
  • the calculation of packet channel prediction on the gNB side can be generated through offline training in the simulation environment, or online training and upgrade through the CSI information received by the system in real time.
  • the gNB side packet channel prediction also supports testing using the received CSI information of the UE. Assume that H after UE i decompresses CSI on time slot t+n subband s is:
  • the eigenvalues and eigenvectors of the corresponding subbands of UEi are obtained.
  • the eigenvalues and eigenvectors saved in time slots t, tn,... of UE i will be used, and the latest eigenvalues and eigenvectors saved by other UEs (UE j) in the group (such as the latest in time slot t+ x saves the CSI of UE j), and the above data:
  • the packet channel prediction on the UE side and the gNB side can be implemented in an optional manner (only one side is reserved) or both sides are reserved, and can be flexibly configured according to the capabilities of the gNB and UE.
  • FIG. 7 is a block diagram of a feedback processing device according to an embodiment of the present disclosure. As shown in Figure 7, it is applied to a base station.
  • the device includes:
  • the receiving data module 72 is configured to receive the CSI data reported by the UE in the target group according to the corresponding reporting configuration information, wherein the reporting configuration information is used to instruct the UE in the target group to report the CSI data in the preset reporting time slot. Describe CSI data;
  • the first channel prediction module 74 is configured to use the target group based on the trained channel prediction model if the CSI data reported by the target UE in the target group is not received in the preset reporting time slot. Perform channel prediction on CSI data reported by other UEs other than the target UE and CSI data reported by the target UE in a preset time period to obtain optimized CSI data;
  • the resource determination module 76 is configured to determine the downlink resources used for downlink data transmission according to the optimized CSI data.
  • the device further includes:
  • the second channel prediction module is configured to, if the CSI data reported by the target UE in the target group is received in the preset reporting time slot, based on the trained channel prediction model, use the target group except the Perform channel prediction on the CSI data reported by other UEs other than the target UE and the CSI data of the target UE to obtain the optimized CSI data; and/or
  • a determination module configured to determine whether the target UE remains in the target group according to the CSI data if the CSI data reported by the target UE in the target group is received in the preset reporting time slot; in When the judgment result is that the target UE is not retained in the target group, the target UE is removed from the target group to update the target group; when the judgment result is that the target UE is retained in the target group If the target UE is in the target group, the target UE is retained in the target group.
  • the judgment module includes:
  • the channel prediction submodule is configured to perform channel prediction on the CSI data reported by the target UE according to the trained channel prediction model corresponding to the target group, and obtain the optimized CSI data;
  • the first determination sub-module is configured to determine whether the target UE remains in the target group according to the optimized CSI data.
  • the channel prediction sub-module is further configured to perform SVD decomposition on the channel in the CSI data to obtain the feature value and feature vector of the target UE on the subband;
  • the eigenvalues and eigenvectors corresponding to the CSI data, as well as the eigenvalues and eigenvector values corresponding to the CSI data of other UEs in the target group are input into the trained channel prediction model to obtain the trained channel prediction model.
  • the optimized CSI data output is the optimized feature value and feature vector of the target UE on the subband.
  • the device further includes:
  • the correction submodule is configured to use the subband channel quality indication and the UE scheduling subband weight in the CSI data reported by the target UE to correct the feature value and feature vector corresponding to the CSI data of the target UE.
  • the judgment module includes:
  • a first determination submodule configured to determine the grouping of the target UE according to the CSI data
  • the second judgment sub-module is configured to judge whether the group of the target UE is the same as the target group
  • the second determination submodule is configured to determine that the judgment result is that the target UE remains in the target group when the group of the target UE is the same as the target group;
  • the third determination sub-module is configured to determine that the determination result is that the target UE does not remain in the target group when the group of the target UE is different from the target group.
  • the first determining sub-module is further configured to determine the target UE and each UE except the target UE according to the CSI data of the target UE and the CSI data of each UE except the target UE.
  • the weighted correlation value of the UE pair composed of each UE other than the target UE is determined, and the group composed of the UE pair whose correlation value is greater than the preset threshold is determined to be the group of the target UE; or the CSI data is input and trained In the grouping decision-making model, the grouping of the target UE output by the trained grouping decision-making model is obtained.
  • the device further includes:
  • a setting submodule configured to set the target UE as an ungrouped user if the judgment result is that the target UE is not retained in the target group;
  • the training submodule is configured to, when the judgment result is that the target UE remains in the target group, based on the CSI data of the target UE and the data of other UEs in the target group except the target UE.
  • the CSI data trains the trained grouping decision model to update the trained grouping decision model.
  • the device further includes:
  • the training module is configured to: if the CSI data reported by the target UE is received in the preset reporting time slot, according to the CSI data of the target UE and the stored target group, in addition to the target UE
  • the trained channel prediction model is trained with CSI data of other UEs to update the trained channel prediction model.
  • the device further includes:
  • a grouping determination module configured to determine a grouping result of the target UE according to the CSI data reported by the target UE when the target UE does not belong to any group, wherein the grouping result includes: the target The UE joins an existing group, or the target UE forms a new group with other ungrouped users.
  • the device further includes:
  • the acquisition module is configured to acquire CSI data reported by multiple UEs within the coverage area, wherein the target UE is any UE among the multiple UEs;
  • a grouping module configured to perform initial grouping on the plurality of UEs according to the CSI data of the plurality of UEs to obtain the plurality of groups, wherein the target grouping is any group among the plurality of groups;
  • the feedback module is configured to select corresponding trained channel prediction models for the plurality of groups, and feed back the trained channel prediction models, grouping parameters and the reported configuration information to the plurality of UEs.
  • a UE packet processing device is also provided.
  • Figure 8 is a block diagram of a feedback processing device according to another embodiment of the present disclosure. As shown in Figure 8, it is applied to a target UE.
  • the device include:
  • the determining data module 82 is configured to determine channel state information CSI data, where the CSI data is calculated CSI data or CSI data optimized for the calculated CSI;
  • the reporting module 84 is configured to report the CSI data to the base station according to the reporting configuration information corresponding to the target group, so that if the base station does not receive the CSI in the preset reporting time slot indicated by the reporting configuration information, Data, based on the trained channel prediction model, use the CSI data reported by other UEs in the target group except the target UE and the CSI data reported in the preset time period to perform channel prediction to obtain optimized CSI data; Downlink resources used for downlink data transmission are determined according to the optimized CSI data, where the reporting configuration information is used to instruct the UE in the target group to report the CSI data in a preset reporting time slot.
  • the device further includes:
  • a receiving signal module configured to receive the CSI-RS detection signal sent by the base station
  • the second judgment module is configured to judge whether the reception quality of the CSI-RS sounding information meets the preset conditions
  • the third training module is configured to calculate CSI data when the judgment result is yes, and obtain the calculated CSI data. According to the CSI data and the data of other UEs in the target group issued by the base station, The CSI data trains the trained channel prediction model to update the trained channel prediction model;
  • the fourth training module is configured to calculate CSI data to obtain the calculated CSI data when the judgment result is negative, and perform channel prediction on the calculated CSI data according to the trained channel prediction model to obtain The optimized CSI data.
  • the fourth training module is further configured to perform SVD decomposition on the channels in the CSI data to obtain eigenvalues and eigenvectors on subbands; and eigenvector values, as well as eigenvalues and eigenvector values corresponding to CSI data of other UEs in the target group are input into the trained channel prediction model, and the optimized output of the trained channel prediction model is obtained.
  • CSI data is further configured to perform SVD decomposition on the channels in the CSI data to obtain eigenvalues and eigenvectors on subbands; and eigenvector values, as well as eigenvalues and eigenvector values corresponding to CSI data of other UEs in the target group are input into the trained channel prediction model, and the optimized output of the trained channel prediction model is obtained.
  • CSI data is further configured to perform SVD decomposition on the channels in the CSI data to obtain eigenvalues and eigenvectors on subbands; and eigenvector
  • the device further includes:
  • the second correction module is configured to use the subband channel quality indication and the UE scheduling subband weight in the CSI data to correct the feature value and feature vector corresponding to the CSI data.
  • the device further includes:
  • a receiving parameter module configured to receive the trained channel prediction model and grouping parameters selected for the target grouping sent by the base station;
  • a setting module configured to set the reporting period of the CSI data according to the reporting configuration information.
  • Embodiments of the present disclosure also provide a computer-readable storage medium that stores a computer program, wherein the computer program is configured to execute the steps in any of the above method embodiments when running.
  • the computer-readable storage medium may include but is not limited to: U disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM) , mobile hard disk, magnetic disk or optical disk and other media that can store computer programs.
  • ROM read-only memory
  • RAM random access memory
  • mobile hard disk magnetic disk or optical disk and other media that can store computer programs.
  • Embodiments of the present disclosure also provide an electronic device, including a memory and a processor.
  • a computer program is stored in the memory, and the processor is configured to run the computer program to perform the steps in any of the above method embodiments.
  • the above-mentioned electronic device may further include a transmission device and an input-output device, wherein the transmission device is connected to the above-mentioned processor, and the input-output device is connected to the above-mentioned processor.
  • modules or steps of the present disclosure can be implemented using general-purpose computing devices, and they can be concentrated on a single computing device, or distributed across a network composed of multiple computing devices. They may be implemented in program code executable by a computing device, such that they may be stored in a storage device for execution by the computing device, and in some cases may be executed in a sequence different from that shown herein. Or the described steps can be implemented by making them into individual integrated circuit modules respectively, or by making multiple modules or steps among them into a single integrated circuit module. As such, the present disclosure is not limited to any specific combination of hardware and software.

Landscapes

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

Abstract

本公开实施例提供了一种反馈处理方法、装置、存储介质及电子装置。该方法包括：接收目标分组中UE根据对应的上报配置信息上报的CSI数据；若在上报配置信息指示的预设上报时隙中未接收到所述目标分组中目标UE上报的所述CSI数据，基于训练好的信道预测模型，使用目标分组中除目标UE之外的其他UE上报的CSI数据与目标UE预设时间段内上报的CSI数据进行信道预测，得到优化后的CSI数据；根据优化后的CSI数据确定下行数据发送使用的下行资源。

Description

一种反馈处理方法、装置、存储介质及电子装置
相关公开的交叉引用
本公开要求在2022年8月31日提交国家知识产权局、公开号为CN202211068191.6、发明名称为“一种反馈处理方法、装置、存储介质及电子装置”的中国专利申请的优先权，该申请的全部内容通过引用结合在本公开中。
技术领域
本公开实施例涉及但不限于通信领域，具体而言，涉及一种反馈处理方法、装置、存储介质及电子装置。
背景技术
5G系统采用了大规模天线阵列以及波束赋型技术来提高系统性能。基于大规模天线阵列的波束赋型技术要求在发送端能够准确的获得信道状态信息(Channel State Information，简称为CSI)，并依据该信道信息选择最优的波束对数据进行发射。信道状态信息需要通过处于接收端的UE(User Equipment)来向发送端进行反馈，这些信息包括信道状态信息-参考信号(CSI Reference Signals，简称为CSI-RS)、资源指示CRI(CSI-RS Resource Indicator)、信道矩阵的秩(rank indicator，简称为RI)、预编码码本索引(Precoding Matrix Index，简称为PMI)和信道质量指示(Channel Quality Indicator，简称为CQI)。UE利用信道估计获得信道系数矩阵H及噪声系数No，再根据基站配置的码本类型和反馈参数计算并上报对应的参数。
随着基站对UE反馈精度的要求提升，越来越复杂的码本被设计出来，但是这也同时增加了反馈上报的负载。一种解决方法是使用人工智能的方法(自适应学习)对反馈需要上报的数据进行编码，并在基站进行解码。该方法一定程度上解决了反馈上报负载的问题，但是在高精度要求下减小负载的程度有限。同时在反馈流程异常时(特别是在毫米波下，出现探测CSI-RS波束被阻挡，上行发送数据被遮挡等)造成的短时反馈中断也会造成基于反馈的通信系统稳定性降低。当前系统为了恢复基于反馈的通信而采用的波束恢复方法流程耗时较长，也会造 成系统鲁棒性的降低。
针对相关技术中在反馈流程异常时造成的短时反馈中断会造成基于反馈的通信系统稳定性降低的问题，尚未提出解决方案。
发明内容
本公开实施例提供了一种反馈处理方法、装置、存储介质及电子装置，以至少解决相关技术中在反馈流程异常时造成的短时反馈中断会造成基于反馈的通信系统稳定性降低的问题。
根据本公开的一个实施例，提供了一种反馈处理方法，应用于基站，所述方法包括：接收目标分组中UE根据对应的上报配置信息上报的CSI数据，其中，所述上报配置信息用于指示所述目标分组内的UE在预设上报时隙中上报所述CSI数据；若在所述预设上报时隙中未接收到所述目标分组中目标UE上报的所述CSI数据，基于训练好的信道预测模型，使用所述目标分组中除所述目标UE之外的其他UE上报的CSI数据与所述目标UE预设时间段上报的CSI数据进行信道预测，得到优化后的CSI数据；根据所述优化后的CSI数据确定下行数据发送使用的下行资源。
根据本公开的另一个实施例，还提供了一种反馈处理方法，应用于目标UE，所述方法包括：确定信道状态信息CSI数据，其中，所述CSI数据为计算出的CSI数据或对所述计算出的CSI进行优化后的CSI数据；根据所在目标分组对应的上报配置信息向基站上报所述CSI数据，以使所述基站若在所述上报配置信息指示的预设上报时隙中未接收到所述CSI数据，基于训练好的信道预测模型，使用所述目标分组中除所述目标UE之外的其他UE上报的CSI数据与预设时间段上报的CSI数据进行信道预测，得到优化后的CSI数据；根据所述优化后的CSI数据确定下行数据发送使用的下行资源，其中，所述上报配置信息用于用于指示所述目标分组内的UE在所述预设上报时隙中上报所述CSI数据。
根据本公开的另一个实施例，还提供了一种反馈处理装置，应用于基站，所述装置包括：接收数据模块，被配置成接收目标分组中UE根据对应的上报配置信息上报的CSI数据，其中，所述上报配置信息用于指示所述目标分组内的UE在预设上报时隙中上报所述CSI数据；第一信道预测模块，被配置成若在所述预设上报时隙中未接收到所述目标分组中目标UE上报的所述CSI数据，基于训练好的信道预测模型，使用所述目标分组中除所述目标UE之外的其他UE上报的 CSI数据与所述目标UE预设时间段上报的CSI数据进行信道预测，得到优化后的CSI数据；确定资源模块，被配置成根据所述优化后的CSI数据确定下行数据发送使用的下行资源。
根据本公开的另一个实施例，还提供了一种UE分组处理装置，应用于目标UE，所述装置包括：确定数据模块，被配置成确定信道状态信息CSI数据，其中，所述CSI数据为计算出的CSI数据或对所述计算出的CSI进行优化后的CSI数据；上报模块，被配置成根据所在目标分组对应的上报配置信息向基站上报所述CSI数据，以使所述基站若在所述上报配置信息指示的预设上报时隙中未接收到所述CSI数据，基于训练好的信道预测模型，使用所述目标分组中除所述目标UE之外的其他UE上报的CSI数据与预设时间段上报的CSI数据进行信道预测，得到优化后的CSI数据；根据所述优化后的CSI数据确定下行数据发送使用的下行资源，其中，所述上报配置信息用于用于指示所述目标分组内的UE在预设上报时隙中上报所述CSI数据。
根据本公开的又一个实施例，还提供了一种计算机可读的存储介质，所述存储介质中存储有计算机程序，其中，所述计算机程序被设置为运行时执行上述任一项方法实施例中的步骤。
根据本公开的又一个实施例，还提供了一种电子装置，包括存储器和处理器，所述存储器中存储有计算机程序，所述处理器被设置为运行所述计算机程序以执行上述任一项方法实施例中的步骤。
附图说明
图1是本公开实施例的反馈处理方法的移动终端的硬件结构框图；
图2是根据本公开实施例的反馈处理方法的流程图；
图3是根据本公开另一实施例的反馈处理方法的流程图；
图4是根据本公开实施例的5G系统的主要应用场景的示意图；
图5是根据本公开实施例的WIFI与接收机系统的框图；
图6是根据本公开实施例的分组决策与分组信道预测的过程的流程图；
图7是根据本公开实施例的反馈处理装置的框图；
图8是根据本公开另一实施例的反馈处理装置的框图。
具体实施方式
下文中将参考附图并结合实施例来详细说明本公开的实施例。
需要说明的是，本公开的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。
本公开实施例中所提供的方法实施例可以在移动终端、计算机终端或者类似的运算装置中执行。以运行在移动终端上为例，图1是本公开实施例的反馈处理方法的移动终端的硬件结构框图，如图1所示，移动终端可以包括一个或多个(图1中仅示出一个)处理器102(处理器102可以包括但不限于微处理器MCU或可编程逻辑器件FPGA等的处理装置)和用于存储数据的存储器104，其中，上述移动终端还可以包括用于通信功能的传输设备106以及输入输出设备108。本领域普通技术人员可以理解，图1所示的结构仅为示意，其并不对上述移动终端的结构造成限定。例如，移动终端还可包括比图1中所示更多或者更少的组件，或者具有与图1所示不同的配置。
存储器104可用于存储计算机程序，例如，应用软件的软件程序以及模块，如本公开实施例中的反馈处理方法对应的计算机程序，处理器102通过运行存储在存储器104内的计算机程序，从而执行各种功能应用以及业务链地址池切片处理，即实现上述的方法。存储器104可包括高速随机存储器，还可包括非易失性存储器，如一个或者多个磁性存储装置、闪存、或者其他非易失性固态存储器。在一些实例中，存储器104可进一步包括相对于处理器102远程设置的存储器，这些远程存储器可以通过网络连接至移动终端。上述网络的实例包括但不限于互联网、企业内部网、局域网、移动通信网及其组合。
传输装置106用于经由一个网络接收或者发送数据。上述的网络具体实例可包括移动终端的通信供应商提供的无线网络。在一个实例中，传输装置106包括一个网络适配器(Network Interface Controller，简称为NIC)，其可通过基站与其他网络设备相连从而可与互联网进行通讯。在一个实例中，传输装置106可以为射频(Radio Frequency，简称为RF)模块，其用于通过无线方式与互联网进行通讯。
在本实施例中提供了一种运行于上述移动终端或网络架构的反馈处理方法，图2是根据本公开实施例的反馈处理方法的流程图，如图2所示，应用于基站，该方法包括如下步骤：
步骤S202，接收目标分组中UE根据对应的上报配置信息上报的CSI数据，其中，所述上报配置信息用于指示所述目标分组内的UE在预设上报时隙中上报所述CSI数据；
步骤S204，若在预设上报时隙中未接收到所述目标分组中目标UE上报的所述CSI数据，基于训练好的信道预测模型，使用目标分组中除目标UE之外的其他UE上报的CSI数据与目标UE预设时间段上报的CSI数据进行信道预测，得到优化后的CSI数据；
步骤S206，根据优化后的CSI数据确定下行数据发送使用的下行资源。
通过上述步骤S202至S206，可以解决相关技术中在反馈流程异常时造成的短时反馈中断会造成基于反馈的通信系统稳定性降低的问题。在未接收到目标UE上报的CSI数据时，通过预设时间段内上报的CSI数据与分组内其他UE上报的CSI数据进行信道预测，通过信道预测结果配置下行资源，避免基站再次配置反馈资源，有效降低5G系统反馈上报负载，提升系统整体鲁棒性。
在一实施例中，若在所述预设上报时隙中接收到所述目标分组中目标UE上报的CSI数据，基于训练好的信道预测模型，使用所述目标分组中除所述目标UE之外的其他UE上报的CSI数据与所述目标UE的CSI数据进行信道预测，得到所述优化后的CSI数据；和/或若在所述预设上报时隙中接收到所述目标分组中目标UE上报的CSI数据，根据所述CSI数据判断所述目标UE是否保留在所述目标分组中；在判断结果为所述目标UE不保留在所述目标分组中的情况下，将所述目标UE从所述目标分组内剔除，以更新所述目标分组；在判断结果为所述目标UE保留在所述目标分组中的情况下，将所述目标UE保留在所述目标分组中。
在一可选的实施例中，上述根据所述CSI数据判断所述目标UE是否保留在所述目标分组中具体可以包括：根据所述目标分组对应的训练好的信道预测模型对所述目标UE上报的所述CSI数据进行信道预测，得到所述优化后的CSI数据；进一步的，将所述CSI数据中的信道进行SVD分解，得到所述目标UE在子带上的特征值与特征向量，可选的，还可以使用所述目标UE上报的CSI数据中的子带信道质量指示以及UE调度子带权重，对所述目标UE的CSI数据对应的特征值与特征向量进行修正；将所述目标UE的CSI数据对应的特征值与特征向量，以及所述目标分组内其他UE的CSI数据对应的特征值和特征向量值输入所述训练好的信道预测模型中，得到所述训练好的信道预测模型输出的所述优化后的 CSI数据，所述优化后的CSI数据为优化后的所述目标UE在子带上的特征值与特征向量；根据所述优化后的CSI数据判断所述目标UE是否保留在所述目标分组中。
在一可选的实施例中，上述根据所述CSI数据判断所述目标UE是否保留在所述目标分组中具体可以包括：根据所述CSI数据确定所述目标UE的分组，进一步的，分别根据所述目标UE的CSI数据与除所述目标UE之外的每个UE的CSI数据确定所述目标UE与除所述目标UE之外的每个UE组成的UE对的加权相关值，确定所述相关值大于预设阈值的UE对组成的分组为所述目标UE的分组；或者将所述CSI数据输入训练好的分组决策模型中，得到所述训练好的分组决策模型输出的所述目标UE的分组；判断所述目标UE的分组与所述目标分组是否相同；在所述目标UE的分组与所述目标分组相同的情况下，确定所述判断结果为所述目标UE保留在所述目标分组中；在所述目标UE的分组与所述目标分组不相同的情况下，确定所述判断结果为所述目标UE不保留在所述目标分组中。
在一可选的实施例中，所述方法还包括：在所述判断结果为目标UE不保留在所述目标分组中的情况下，将所述目标UE设置为未分组用户；在所述判断结果为目标UE保留在所述目标分组中的情况下，根据所述目标UE的CSI数据与所述目标分组内除所述目标UE之外其他UE的CSI数据对所述训练好的分组决策模型进行训练，以更新所述训练好的分组决策模型。
在另一实施例中，所述方法还包括：若在所述预设上报时隙中接收到所述目标UE上报的CSI数据，根据所述目标UE的CSI数据与存储的所述目标分组内除所述目标UE之外其他UE的CSI数据对所述训练好的信道预测模型进行训练，以更新所述训练好的信道预测模型。
在另一实施例中，所述方法还包括：在所述目标UE不属于任一分组的情况下，根据所述目标UE上报的CSI数据确定所述目标UE的分组结果，其中，所述分组结果包括：所述目标UE加入已有的某一分组，或者所述目标UE与其他未分组用户组成新的分组。
在另一实施例中，在上述步骤S202之前，所述方法还包括：获取覆盖范围内的多个UE上报的CSI数据，其中，所述目标UE为所述多个UE中的任一UE；根据所述多个UE的CSI数据对所述多个UE进行初始分组，得到所述多个分组，其中，所述目标分组为所述多个分组中的任一分组；分别为所述多个分组选择对 应的训练好的信道预测模型，并将所述训练好的信道预测模型、分组参数及所述上报配置信息反馈给所述多个UE。
根据本公开的另一个实施例，还提供了一种反馈处理方法，图3是根据本公开另一实施例的反馈处理方法的流程图，如图3所示，应用于目标UE，所述方法包括：
步骤S302，确定信道状态信息CSI数据，其中，所述CSI数据为计算出的CSI数据或对所述计算出的CSI进行优化后的CSI数据；
步骤S304，根据所在目标分组对应的上报配置信息向基站上报CSI数据，以使基站若在上报配置信息指示的预设上报时隙中未接收到CSI数据，基于训练好的信道预测模型，使用目标分组中除所述目标UE之外的其他UE上报的CSI数据与预设时间段上报的CSI数据进行信道预测，得到优化后的CSI数据；根据所述优化后的CSI数据确定下行数据发送使用的下行资源，其中，所述上报配置信息被配置成指示所述目标分组内的UE在预设上报时隙中上报CSI数据。
通过上述步骤S302至S304，可以解决相关技术中在反馈流程异常时造成的短时反馈中断会造成基于反馈的通信系统稳定性降低的问题。在未接收到目标UE上报的CSI数据时，通过预设时间段内上报的CSI数据与分组内其他UE上报的CSI数据进行信道预测，通过信道预测结果配置下行资源，避免基站再次配置反馈资源，有效降低5G系统反馈上报负载，提升系统整体鲁棒性。
在一实施例中，在上述步骤S302之前，所述方法还包括：接收所述基站下发的CSI-RS探测信号；判断所述CSI-RS探测信息的接收质量是否满足预设条件；在判断结果为是的情况下，计算CSI数据，得到所述计算出的CSI数据，根据所述CSI数据与所述基站下发的所述目标分组内其他UE的CSI数据对训练好的信道预测模型进行训练，以更新所述训练好的信道预测模型；在判断结果为否的情况下，计算CSI数据，得到所述计算出的CSI数据，根据训练好的信道预测模型对所述计算出的CSI数据进行信道预测，得到所述优化后的CSI数据。
进一步的，根据训练好的信道预测模型对所述计算出的CSI数据进行预测，得到所述优化后的CSI数据具体可以包括：对所述CSI数据中的信道进行SVD分解，得到在子带上的特征值和特征向量；将所述CSI数据对应的特征值和特征向量值，以及所述目标分组内其他UE的CSI数据对应的特征值和特征向量值输入所述训练好的信道预测模型中，得到所述训练好的信道预测模型输出的所述优 化后的CSI数据。可选的，还可以使用所述CSI数据中的子带信道质量指示以及UE调度子带权重对所述CSI数据对应的特征值与特征向量进行修正。
在一实施例中，所述方法还包括：接收所述基站下发的为所述目标分组选择的所述训练好的信道预测模型与分组参数；和/或根据所述上报配置信息设置所述CSI数据的上报周期。
本实施例基于自适应学习与UE分组技术的反馈方法，在传统的AI编译码器反馈流程的基础上通过基站根据UE反馈上报结果进行基于自适应学习的UE信道相关性分组；基站再进一步的针对每个UE分组下发group-DCI控制每个UE分组进行周期轮流上报；之后基站对于每个分组中正常上报的反馈UE数据放入训练数据集合中，对于上报异常或本次无上报的UE使用组中其余UE上报数据基于自适应学习进行上报重构；相应的UE也可以根据基站决定的分组在UE侧进行前述的模型训练以及上报重构，从而进一步降低上报负载，提高系统反馈鲁棒性。
图4是根据本公开实施例的5G系统的主要应用场景的示意图。如图4所示，一个gNB覆盖范围内可以存在多个UE，gNB与每个UE进行基于3GPP协议的物理层数据传输、反馈、控制等流程。每个UE的反馈等相关行为由gNB进行控制与调度。
目前主流的5G系统反馈方法为基于码本的反馈上报、基于AI编译码器的压缩CSI上报等。这些方法的基本流程是基站针对每个UE发送相应的NZP CSI-RS探测信号，UE根据该信号的信道估计结果使用不同方法(码本，AI编译码器等)计算CSI上报信息并通过上行信道将这些信息传输给gNB，gNB得到这些CSI上报信息后对每个用户进行自适应调度(包含波束、预编码、MCS调度等)。
考虑到信道的时频特性，CSI上报往往是一个周期或者出发式的非周期过程，对UE调度的精度要求越高，gNB就需要越频繁的要求UE进行CSI上报，这样往往会给上行带来较大的负担。同时在毫米波频段，探测及上报往往也是基于波束赋形进行的传输，短时的遮挡可能造成反馈流程的中断，在现用协议框架下，这种中断往往需要gNB与UE间进行重建波束连接流程后，再重新进入CSI上报流程，这又增加了系统开销，同时在频繁中断后可能造成系统不稳定。
图5是根据本公开实施例的WIFI与接收机系统的框图。如图5所示，该系统主要模块功能为：
CSI数据集合/UE分组集合模块501，主要负责存储更新各个UE解压缩后的CSI信息；根据分组结果、UECSI接收评估结果推送对应UECSI集合数据到模块503-506。
CSI解压缩编码模块502，主要负责将接收到的压缩CSI信息进行解压缩。
UE分组决策模块503，根据模块501推送的UE CSI集合数据计算当前UE集合的分组决策结果，输出UE分组、组内CSI压缩编解码模型等。
UE分组决策模型训练模块504，根据模块501推送的数据对UE分组决策模型进行训练，将训练后的升级模型推送给模块503。
组内信道预测模型训练模块505，根据模块501推送的数据与分组结果对本组的信道预测模型进行训练，将训练后的升级模型推送给模块506。
组内信道预测计算模块506，根据模块501推送的本组CSI集合数据计算当前UE的预测CSI，输出CSI预测结果。
信道预测结果保存模块507，保存预测结果，用于后需下行发送预编码计算，下行CSI压缩传输等。
分组与模型选择保存模块508，保存模块503的输出，并按照各个模块需求将对应的分组和/或模型选择结果传输给需要的模块。
CSI压缩编码模块509，根据模块508选择的压缩模型对模块507提供的信道数据进行压缩并传输至模块515。
下行数据发送模块510，按照协议发送相应UE的数据或者参考信号等。
CSI压缩编码模块511，主要负责将接收到的CSI信息进行压缩。
组内信道预测计算模块512，根据模块513推送的本组CSI集合数据计算当前UE的预测CSI，输出CSI预测结果。
组内CSI数据集合模块513，保存被UECSI估计结果与gNB下发的组内CSI信息，同时控制数据输出给对应模块。
组内信道预测模型训练模块514，根据模块513推送的本组CSI集合数据训练UE CSI预测模型，输出模型结果。
下行数据接收/处理模块515，接收解码gNB下发的数据并输出结果，接收gNB下发的CSI参考信号并输出CSI估计结果。
上行数据发送模块516，发送下行数据解码结果/上行数据。
上行数据接收/处理模块517，接收处理上行数据。
链路质量评估模块518，根据UE上报的解码结果等评估UE的链路质量。
本实施例基于自适应学习与UE分组技术的反馈方法，图6是根据本公开实施例的分组决策与分组信道预测的过程的流程图，如图6所示，该过程包括以下步骤：
步骤601，短周期上报。gNB调度覆盖范围内的UE进行CSI上报，该上报结果作为UE分组选择的初始数据。
步骤602，gNB分组决策与分组配置。具体的，根据初始数据对所有UE进行基于自适应学习的初始分组，并选择每个分组的分组反馈学习模型，每个分组内每个UE的上报周期与上报偏移。
步骤603，长周期上报。gNB通知每个UE其分组反馈学习模型，上报周期与上报偏移的等分组参数，UE根据上述参数使用对应的学习模型在特定的时间进行周期上报。
gNB收到某个UE的基于自适应学习的反馈上报后，将上报信息作为其所在组内信道预测模型的训练数据，用于判断分组是否继续成立。如果成立则该数据还将用于升级组内信道预测模型并作为全UE分组模型的训练数据升级分组模型；如果不成立，则该数据作为全UE分组模型的输入数据，gNB使用该数据对当前没有分组的UE重新进行分组。
步骤604，UE侧分组信道预测。UE在收到gNB下发的分组参数以及组内其他UE的CSI信息后，可以使用该CSI信息通过UE侧组内信道预测模型对本UE CSI估计进行加强或训练UE侧预测模型。
步骤605，gNB侧分组信道预测。具体的，如果gNB在某个UE的CSI上报时隙没有收到该UE的上报信息，则使用该UE所在分组内其他UE以及该UE最近一次CSI上报信息作为组内信道预测模型输入，输出该UE的预测信道结果，并使用预测结果继续后续波束、码本选择流程。如果使用预测信道结果还是没有保持该UE的正常数据通信，则进入协议规定的波束失败流程重选波束，同时将该UE从分组中删除，待波束重建后使用新的CSI上报进行重新分组。
UE在应该进行反馈CSI的时隙没有接收到对应的在CSI-RS信号，此时使用最新的组内CSI信息和本UE CSI信息输入UE侧组内信道预测模型，输出对应的CSI信息作为本次上报的CSI。
具体的，以一个如图4所示的一个gNB范围内存在6个UE为例对本实施例的UE分组和组内预测方法进行说明。
在某个时刻，gNB配置UE0-5进行CSI上报，UE0-5按照gNB配置的时间、压缩模型等在特定时刻使用压缩模型(可使用协议规定的码本，基于AI的自适应压缩编译码等)对CSI进行压缩后上报给gNB，gNB接收到对应的CSI信息后进行基于压缩模型的解压缩处理，并将解压缩后的CSI信息存入模块501中。
gNB收集完UE0-5的CSI信息后，将所有UE的CSI信息输入UE分组决策模块，该模块根据已经训练完成的UE分组决策模型对输入的CSI信息进行处理，输出UE分组决策结果，将UE信道关联性高于一定门限的UE分到一组内，同时为每个UE分组选择一个信道预测模型。gNB完成UE分组后将分组结果传递给模块501，将分组结果和预测模型选择结果模块505，同时通过下行DCI将对应的分组结果、预测模型、CSI上报配置等信息通知给UE模块512。该DCI使用UE指定的CSI DCI format进行传输，CSI DCI format包含的字段如下表1所示。
相应的将该UE分组内除了本UE外其他的UE的CSI信息通过模块509进行压缩后传输给UE。gNB通过组播方式将组内最新更新UE的压缩CSI信息传输给组内所有UE。
UE接收分组结果、预测模型、CSI上报配置等信息后，解析对应信息，设 置模块512预测模型，同时设置上报周期等。UE接收本组其他UE的CSI信息后先解压缩对应CSI，再将对应的数据存入模块513中。
经过上述步骤，gNB和UE完成了初始CSI信息存储，初始UE分组以及gNB和UE间的信令互传，本实施例下，UE0为独立UE，UE1-3为用户分组0，UE4-5为UE分组2。
UE0不会得到分组相关信息，其CSI上报流程使用传统流程。
用户分组0中包含3个UE(UE1-3)，gNB配置分组0内的UE使用相同的周期在不同的时隙进行CSI上报，例如上报周期为10，三个UE分别在时隙号％10＝0,3,6这三个不同时隙偏移进行上报CSI，同时gNB通知分组0内每个UE其预测模型选择结果、分组中其他UE ID以及对应UE的CSI信息，例如gNB会通知UE1其预测模型，告知组内还有UE2与UE3，并发送UE2与UE3的CSI给UE1。
用户分组1中包含2个UE(UE4-5)，gNB配置分组1内的UE使用相同的周期在不同的时隙进行CSI上报，例如上报周期为20，2个UE分别在时隙号％10＝0,10这2个不同时隙偏移进行上报CSI，同时gNB通知分组1内每个UE其预测模型选择结果、分组中其他UE ID以及对应UE的CSI信息，例如gNB会通知UE4其预测模型，告知组内还有UE5，并发送UE5的CSI给UE4。
在接下来进入正常周期CSI上报流程。
未加入分组的UE0接收到gNB下发的CSI-RS探测信号且本次CSI-RS接收质量良好(可已探测信号SNR为参考，设置门限，高于某个门限认为质量较好)则计算出对应的CSI信息存入模块513，模块513将本次计算出来的CSI与之前保存的UE0的CSI信息输入模块514进行信道预测模型训练，模块514输出升级后模型到模块512，后续CSI压缩上报也直接使用本次计算出来的CSI信息。
用户分组0中的UE1接收到gNB下发的CSI-RS探测信号且本次CSI-RS接收质量良好则计算出对应的CSI信息存入模块513，模块513将本次计算出来的UE1CSI信息、之前保存的UE1及用户分组0内其他用户(UE2/3，gNB下发给UE1)的CSI信息输入模块514进行信道预测模型训练，模块514输出升级后模型到模块512，后续CSI压缩上报也直接使用本次计算出来的CSI信息。如果本次CSI-RS接收质量较差，则模块513将本次计算出来的UE1CSI、之前保存的UE1及用户分组0内其他用户(UE2/3，gNB下发给UE1)的CSI信息输入模块512进行信道预测与修正，模块512输出计算后的CSI信息到模块511进行压缩 上报。用户分组0中的UE1接收到gNB下发的分组内其他UE(例如UE2)的CSI压缩信息后，首先进行CSI解压缩，将解压缩后的CSI信息存入模块513，模块513将之前保存的UE1CSI、UE3CSI输入模块514进行信道预测模型训练，模块514输出升级后模型到模块512。
用户分组0中的其他用户接收后gNB下发的CSI-RS探测信号或分组内其他UE的CSI压缩信息后处理方法同UE1。
用户分组1中的UE4接收到gNB下发的CSI-RS探测信号且本次CSI-RS接收质量良好则计算出对应的CSI信息存入模块513，模块513将本次计算出来的UE4CSI信息、之前保存的UE4及用户分组1内其他用户(UE5，gNB下发给UE4)的CSI信息输入模块514进行信道预测模型训练，模块514输出升级后模型到模块512，后续CSI压缩上报也直接使用本次计算出来的CSI信息.用户分组1中的UE4接收到gNB下发的分组内其他UE(例如UE5)的CSI压缩信息后，首先进行CSI解压缩，将解压缩后的CSI信息存入模块513，模块513将之前保存的UE4CSI输入模块514进行信道预测模型训练，模块514输出升级后模型到模块512。
gNB接收UE0的上报CSI后，先进行CSI解压缩，由于UE0不在任何用户分组内，可以直接将接收的CSI信息存储到模块501中。后续UE0的下行数据发送使用的预编码矩阵，层数,MCS等也直接使用其最新上报结果计算得到。由于UE0不属于任何用户分组，此时模块501会将UE0本次CSI信息与当前分组策略输入模块503中进程分组决策，决定UE0是否加入当前某个用户分组或者和其他未分组用户组成新的用户分组。
gNB接收UE1的上报CSI后，先进行CSI解压缩，由于UE1在用户分组0内，gNB先将接收的CSI信息存储到模块501中。101模块判断选择使用更新后的数据进行训练还是测试。如果选择进行测试则将更新后的UE1CSI信息与存储的UE2/3CSI信息同时输入模块506，根据模型预测结果判断当前UE0是否能够继续保留在用户分组0中，如果不能保留，则通知模块501刷新分组策略，将用户分组0中UE调整为UE2/3,将UE1放入未分组用户；如果保留，则通知模块501该结果，模块501会将UE0的最新CSI数据作为模块504的训练数据，升级分组决策模型。如果选择进行训练则将更新后的UE1CSI信息与存储的UE2/3CSI信息同时输入模块505，输出训练后模型。UE1的下行数据发送使用的预编码矩阵、层数、MCS等也直接使用其最新上报结果计算得到。
gNB接收到UE2/3的上报CSI后，后续步骤同UE1接收。
gNB接收UE4的上报CSI后，先进行CSI解压缩，由于UE4在用户分组1内，gNB先将接收的CSI信息存储到模块501中。101模块判断选择使用更新后的数据进行训练还是测试。如果选择进行测试则将更新后的UE4CSI信息与存储的UE5CSI信息同时输入模块506，根据模型预测结果判断当前UE4是否能够继续保留在用户分组1中，如果不能保留，则通知模块501刷新分组策略，由于用户分组1仅有UE4/5两个用户，此时101模块会解散用户分组1，将UE4/5均放入未分组用户集合；如果保留，则通知模块501该结果，模块501会将UE4的最新CSI数据作为模块504的训练数据，升级分组决策模型。如果选择进行训练则将更新后的UE4CSI信息与存储的UE5CSI信息同时输入模块505，输出训练后模型。UE4的下行数据发送使用的预编码矩阵、层数、MCS等也直接使用其最新上报结果计算得到。
如果在周期上报过程中出现UE接收CSI-RS探测信号异常：
未加入分组的UE0如果在周期CSI-RS子帧接收到的CSI-RS信号质量较差，则UE0进入异常处理流程，模块513将之前保存的UE0的CSI信息输入模块512进行信道预测与修正，模块512输出计算后的CSI信息到模块511进行压缩上报。
用户分组0中的UE1如果在周期CSI-RS子帧接收到的CSI-RS信号质量较差，则模块513将之前保存的UE1及用户分组0内其他用户(UE2/3，gNB下发给UE1)的CSI信息输入模块512进行信道预测与修正，模块512输出计算后的CSI信息到模块511进行压缩上报。
用户分组0中的其他用户接收后gNB下发的CSI-RS探测信号或分组内其他UE的CSI压缩信息后处理方法同UE1。
用户分组1中的UE4如果在周期CSI-RS子帧接收到的CSI-RS信号质量较差，则模块513将之前保存的UE4及用户分组1内其他用户(UE5，gNB下发给UE4)的CSI信息输入模块512进行信道预测与修正，模块512输出计算后的CSI信息到模块511进行压缩上报。
用户分组1中的其他用户接收后gNB下发的CSI-RS探测信号或分组内其他UE的CSI压缩信息后处理方法同UE4。
gNB如果在规定的CSI信息上报时隙中没有接收到UE0的上报CSI信息，则模块501将UE0之前保存的CSI信息输入模块506进行UE0CSI信息的预测，输出预测CSI结果。UE0的下行数据发送使用的预编码矩阵、层数、MCS等也 直接使用其最新预测结果计算得到。
gNB如果在规定的CSI信息上报时隙中没有接收到UE1的上报CSI信息，由于UE1属于用户分组0，则模块501将UE1之前保存的CSI信息与用户分组0中其他用户(UE2/3)最新上报的CSI信息同时输入模块506进行UE1CSI信息的预测，输出预测CSI结果。UE1的下行数据发送使用的预编码矩阵，层数,MCS等也直接使用其最新预测结果计算得到。
gNB如果在规定的CSI信息上报时隙中没有接收到UE2/3的上报CSI，后续步骤同UE0。
gNB如果在规定的CSI信息上报时隙中没有接收到UE4的上报CSI信息，由于UE4属于用户分组1，则模块501将UE4之前保存的CSI信息与用户分组1中其他用户(UE5)最新上报的CSI信息同时输入模块506进行UE4CSI信息的预测，输出预测CSI结果。UE4的下行数据发送使用的预编码矩阵、层数、MCS等也直接使用其最新预测结果计算得到。
gNB如果在规定的CSI信息上报时隙中没有接收到UE5的上报CSI，后续步骤同UE4。
本实施例中的gNB侧用户分组决策可以使用基于UE信道特征向量的相关性决策，包括：
1.首先对所有UE解压缩后的信道H进行SVD分解，具体的在本实施例中gNB使用16port NZP CSI-RS对UR0-5进行探测，则UE i在子带s上解压缩后的H为：
进行SVD分解后，得到则UE i在子带s上的特征值和特征向量：
其中，Rx,i表示UEi的接收天线数，Tx,gNB表示gNB发送的探测信号端口数。
2.对每个UE的上报信号进行上述计算后，再针对进行两两加权相关计算，比如对于UEi和j，两者加权相关值Ri,j为:
3.对于相关值大于门限(门限可通过仿真得到)的UE对，则认为可分为一 组，如果分组间出现重叠，比如UE0与UE1相关值大于门限，且UE1与UE2相关值也大于门限，则UE0,1,2分为一组。
还可以使用基于UE信道特征向量的修正相关性决策，包括：
1.首先对所有UE解压缩后的信道H进行SVD分解，具体的在本实施例中gNB使用16port NZP CSI-RS对UR0-5进行探测，则UE i在子带s上解压缩后的H为：
进行SVD分解后，得到则UE i在子带s上的特征值和特征向量：
其中，Rx,i表示UE i的接收天线数，Tx,gNB表示gNB发送的探测信号端口数。
2.对每个UE的上报信号进行上述计算后，再针对进行两两修正加权相关计算，使用UE上报的CSI信息中的子带CQIqs以及UE调度子带权重ws修正比如对于UE i和j，两者加权相关值Ri,j为:
其中，qi,s表示子带CQI等效权重，可以使用下列公式得到，设当前子带CQI对应的调制系数为x，其中x取值如下表2所示。
表2
设当前子带CQI对应的码率为e，则：
其中，wi,s表示本次计算前T秒内UEi在子带s内平均调度率，在(0,1]区间 内。
对于相关值大于门限(门限可通过仿真得到)的UE对，则认为可分为一组，如果分组间出现重叠，比如UE0与UE1相关值大于门限，且UE1与UE2相关值也大于门限，则UE0,1,2分为一组。
还可以使用基于自适应学习的用户分组决策，包括：
首先对所有UE解压缩后的信道H进行预处理(一般为SVD分解提取特征值)。具体的在本实施例中gNB使用16port NZP CSI-RS对UR0-5进行探测，则UE i在子带s上解压缩后的H为：进行SVD分解后，得到则UE i在子带s上的特征值和特征向量：
其中，Rx,i表示UE i的接收天线数，Tx,gNB表示gNB发送的探测信号端口数。
对每个UE的信道信息进行预处理后将预处理的特征值S，特征向量V，(也可以包含其他的权重参数，比如子带质量参数，调度参数等)，将这些参数输入训练后的分组决策模型，模型计算得到最终的分组结果。
分组决策模型算法可以使用自适应机器学习算法、自适应强化学习算法、自适应深度学习算法、策略优化增强学习算法等自适应学习算法；分组决策模型可以使用部分观察的马尔可夫决策过程(POMDP)、人工神经网络(包含深度信念网络(DBN)、深度卷积网络(DCN)、循环神经网络(RNN)、多层感知神经网络(MLPCN)、卷积神经网络(CNN)等)。
分组决策模型的计算可以通过仿真环境进行离线训练产生，也可以通过系统实时接收到的CSI信息进行在线训练升级。
本实施例中的UE侧用户分组信道预测方法包括：
首先对本UE信道估计后得到的信道H进行预处理(一般为SVD分解提取特征值)，具体的在本实施例中gNB使用16port NZP CSI-RS进行探测，则UE i 在时隙t子带s上信道估计后的H为：进行SVD分解后，得到则UE i在子带s上的特征值和特征向量：
其中，Rx,i表示UE i的接收天线数，Tx,gNB表示gNB发送的探测信号端口数，t表示时隙号。
再接收gNB下发的其他UE的压缩CSI信息，解压缩后得到H，SVD分解后保存UE j的信道特征值和特征向量：
如果在时隙t+n，本UE i接收到的gNB下发的探测CSI-RS质量很差，则将UE i时隙t+n,t,t-n,....上保存的特征值与特征向量，分组中其他UE(UE j)最新保存的特征值与特征向量(比如最新在时隙t+x保存了UE j的CSI)，将上述数据：
输入到分组信道预测计算中，得出UE i本次预测修正后的信道特征值与特征向量。
分组信道预测算法可以使用自适应机器学习算法、自适应强化学习算法、自适应深度学习算法、策略优化增强学习算法等自适应学习算法；分组信道预测可以使用部分观察的马尔可夫决策过程(POMDP)、人工神经网络(包含深度信念网络(DBN)、深度卷积网络(DCN)、循环神经网络(RNN)、多层感知神经网络(MLPCN)、卷积神经网络(CNN)等)。
分组信道预测的计算可以通过仿真环境进行离线训练产生，也可以通过系统实时接收到的CSI信息进行在线训练升级。
本方法所述的gNB侧用户分组信道预测方法描述如下：
首先对分组内所有UE解压缩后的信道H进行预处理(一般为SVD分解提取特征值)，具体的在本实施例中gNB使用16port NZP CSI-RS进行探测，则UE i在时隙t子带s上解压缩CSI后的H为：
进行SVD分解后，得到则UE i在子带s上的特征值和特征向量：
其中，Rx,i表示UE i的接收天线数，Tx,gNB表示gNB发送的探测信号端口数，t表示时隙号。
如果在时隙t+n，未接收到UE i上报的CSI信息，则将UE i时隙t,t-n,....上保存的特征值与特征向量，分组中其他UE(UE j)最新保存的特征值与特征向量(比如最新在时隙t+x保存了UE j的CSI)，将上述数据：
输入到分组信道预测计算中，得出UE i本次预测修正后的信道特征值与特征向量。
分组信道预测算法可以使用自适应机器学习算法、自适应强化学习算法、自适应深度学习算法、策略优化增强学习算法等自适应学习算法；分组信道预测可以使用部分观察的马尔可夫决策过程(POMDP)、人工神经网络(包含深度信念网络(DBN)、深度卷积网络(DCN)、循环神经网络(RNN)、多层感知神经网络(MLPCN)、卷积神经网络(CNN)等)。
gNB侧分组信道预测的计算可以通过仿真环境进行离线训练产生，也可以通过系统实时接收到的CSI信息进行在线训练升级。
gNB侧分组信道预测还支持使用接收到的UE的CSI信息测试，设UE i在时隙t+n子带s上解压缩CSI后的H为：
进行SVD分解后，得到UEi对应子带的特征值与特征向量。
此时使用则将UE i时隙t,t-n,....上保存的特征值与特征向量，分组中其他UE(UE j)最新保存的特征值与特征向量(比如最新在时隙t+x保存了UE j的CSI)，将上述数据：
输入到分组信道预测计算中，得出UE i本次预测修正后的信道特征值与特征向量，记为：
将其与
计算上述两者G与G’的MMSE，如果误差高于设定门限值则本次测试失败，将UEi从分组中删除；如果低于设定门限，则本次测试成功，保留分组。
本实施例中的UE侧和gNB侧的分组信道预测可以二选一的实现(仅保留一侧)或者两侧均保留，可根据gNB和UE能力灵活配置。
根据本公开的另一个实施例，还提供了一种反馈处理装置，图7是根据本公开实施例的反馈处理装置的框图，如图7所示，应用于基站，所述装置包括：
接收数据模块72，被配置成接收目标分组中UE根据对应的上报配置信息上报的CSI数据，其中，所述上报配置信息用于指示所述目标分组内的UE在预设上报时隙中上报所述CSI数据；
第一信道预测模块74，被配置成若在所述预设上报时隙中未接收到所述目标分组中目标UE上报的所述CSI数据，基于训练好的信道预测模型，使用所述目标分组中除所述目标UE之外的其他UE上报的CSI数据与所述目标UE预设时间段上报的CSI数据进行信道预测，得到优化后的CSI数据；
确定资源模块76，被配置成根据所述优化后的CSI数据确定下行数据发送使用的下行资源。
在一实施例中，所述装置还包括：
第二信道预测模块，被配置成若在所述预设上报时隙中接收到所述目标分组中目标UE上报的CSI数据，基于训练好的信道预测模型，使用所述目标分组中除所述目标UE之外的其他UE上报的CSI数据与所述目标UE的CSI数据进行信道预测，得到所述优化后的CSI数据；和/或
判断模块，被配置成若在所述预设上报时隙中接收到所述目标分组中目标UE上报的CSI数据，根据所述CSI数据判断所述目标UE是否保留在所述目标分组中；在判断结果为所述目标UE不保留在所述目标分组中的情况下，将所述目标UE从所述目标分组内剔除，以更新所述目标分组；在判断结果为所述目标UE保留在所述目标分组中的情况下，将所述目标UE保留在所述目标分组中。
在一实施例中，所述判断模块包括：
信道预测子模块，被配置成根据所述目标分组对应的训练好的信道预测模型对所述目标UE上报的所述CSI数据进行信道预测，得到所述优化后的CSI数据；
第一判断子模块，被配置成根据所述优化后的CSI数据判断所述目标UE是否保留在所述目标分组中。
在一实施例中，所述信道预测子模块，还被配置成将所述CSI数据中的信道进行SVD分解，得到所述目标UE在子带上的特征值与特征向量；将所述目标UE的CSI数据对应的特征值与特征向量，以及所述目标分组内其他UE的CSI数据对应的特征值和特征向量值输入所述训练好的信道预测模型中，得到所述训练好的信道预测模型输出的所述优化后的CSI数据，所述优化后的CSI数据为优化后的所述目标UE在子带上的特征值与特征向量。
在一实施例中，所述装置还包括：
修正子模块，被配置成使用所述目标UE上报的CSI数据中的子带信道质量指示以及UE调度子带权重对所述目标UE的CSI数据对应的特征值与特征向量进行修正。
在一实施例中，所述判断模块包括：
第一确定子模块，被配置成根据所述CSI数据确定所述目标UE的分组；
第二判断子模块，被配置成判断所述目标UE的分组与所述目标分组是否相同；
第二确定子模块，被配置成在所述目标UE的分组与所述目标分组相同的情况下，确定所述判断结果为所述目标UE保留在所述目标分组中；
第三确定子模块，被配置成在所述目标UE的分组与所述目标分组不相同的情况下，确定所述判断结果为所述目标UE不保留在所述目标分组中。
在一实施例中，所述第一确定子模块，还被配置成分别根据所述目标UE的CSI数据与除所述目标UE之外的每个UE的CSI数据确定所述目标UE与除所述目标UE之外的每个UE组成的UE对的加权相关值，确定所述相关值大于预设阈值的UE对组成的分组为所述目标UE的分组；或者将所述CSI数据输入训练好的分组决策模型中，得到所述训练好的分组决策模型输出的所述目标UE的分组。
在一实施例中，所述装置还包括：
设置子模块，被配置成在所述判断结果为目标UE不保留在所述目标分组中的情况下，将所述目标UE设置为未分组用户；
训练子模块，被配置成在所述判断结果为目标UE保留在所述目标分组中的情况下，根据所述目标UE的CSI数据与所述目标分组内除所述目标UE之外其他UE的CSI数据对所述训练好的分组决策模型进行训练，以更新所述训练好的分组决策模型。
在一实施例中，所述装置还包括：
训练模块，被配置成若在所述预设上报时隙中接收到所述目标UE上报的CSI数据，根据所述目标UE的CSI数据与存储的所述目标分组内除所述目标UE之外其他UE的CSI数据对所述训练好的信道预测模型进行训练，以更新所述训练好的信道预测模型。
在一实施例中，所述装置还包括：
确定分组模块，被配置成在所述目标UE不属于任一分组的情况下，根据所述目标UE上报的CSI数据确定所述目标UE的分组结果，其中，所述分组结果包括：所述目标UE加入已有的某一分组，或者所述目标UE与其他未分组用户组成新的分组。
在一实施例中，所述装置还包括：
获取模块，被配置成获取覆盖范围内的多个UE上报的CSI数据，其中，所述目标UE为所述多个UE中的任一UE；
分组模块，被配置成根据所述多个UE的CSI数据对所述多个UE进行初始分组，得到所述多个分组，其中，所述目标分组为所述多个分组中的任一分组；
反馈模块，被配置成分别为所述多个分组选择对应的训练好的信道预测模型，并将所述训练好的信道预测模型、分组参数及所述上报配置信息反馈给所述多个UE。
根据本公开的另一个实施例，还提供了一种UE分组处理装置，图8是根据本公开另一实施例的反馈处理装置的框图，如图8所示，应用于目标UE，所述装置包括：
确定数据模块82，被配置成确定信道状态信息CSI数据，其中，所述CSI数据为计算出的CSI数据或对所述计算出的CSI进行优化后的CSI数据；
上报模块84，被配置成根据所在目标分组对应的上报配置信息向基站上报所述CSI数据，以使所述基站若在所述上报配置信息指示的预设上报时隙中未接收到所述CSI数据，基于训练好的信道预测模型，使用所述目标分组中除所述目标UE之外的其他UE上报的CSI数据与预设时间段上报的CSI数据进行信道预测，得到优化后的CSI数据；根据所述优化后的CSI数据确定下行数据发送使用的下行资源，其中，所述上报配置信息用于用于指示所述目标分组内的UE在预设上报时隙中上报所述CSI数据。
在一实施例中，所述装置还包括：
接收信号模块，被配置成接收所述基站下发的CSI-RS探测信号；
第二判断模块，被配置成判断所述CSI-RS探测信息的接收质量是否满足预设条件；
第三训练模块，被配置成在判断结果为是的情况下，计算CSI数据，得到所述计算出的CSI数据，根据所述CSI数据与所述基站下发的所述目标分组内其他UE的CSI数据对训练好的信道预测模型进行训练，以更新所述训练好的信道预测模型；
第四训练模块，被配置成在判断结果为否的情况下，计算CSI数据，得到所述计算出的CSI数据，根据训练好的信道预测模型对所述计算出的CSI数据进行信道预测，得到所述优化后的CSI数据。
在一实施例中，所述第四训练模块，还被配置成对所述CSI数据中的信道进行SVD分解，得到在子带上的特征值和特征向量；将所述CSI数据对应的特征值和特征向量值，以及所述目标分组内其他UE的CSI数据对应的特征值和特征向量值输入所述训练好的信道预测模型中，得到所述训练好的信道预测模型输出的所述优化后的CSI数据。
在一实施例中，所述装置还包括：
第二修正模块，被配置成使用所述CSI数据中的子带信道质量指示以及UE调度子带权重对所述CSI数据对应的特征值与特征向量进行修正。
在一实施例中，所述装置还包括：
接收参数模块，用被配置成接收所述基站下发的为所述目标分组选择的所述训练好的信道预测模型与分组参数；和/或
设置模块，被配置成根据所述上报配置信息设置所述CSI数据的上报周期。
本公开的实施例还提供了一种计算机可读存储介质，该计算机可读存储介质中存储有计算机程序，其中，该计算机程序被设置为运行时执行上述任一项方法实施例中的步骤。
在一个示例性实施例中，上述计算机可读存储介质可以包括但不限于：U盘、只读存储器(Read-Only Memory，简称为ROM)、随机存取存储器(Random Access Memory，简称为RAM)、移动硬盘、磁碟或者光盘等各种可以存储计算机程序的介质。
本公开的实施例还提供了一种电子装置，包括存储器和处理器，该存储器中存储有计算机程序，该处理器被设置为运行计算机程序以执行上述任一项方法实施例中的步骤。
在一个示例性实施例中，上述电子装置还可以包括传输设备以及输入输出设备，其中，该传输设备和上述处理器连接，该输入输出设备和上述处理器连接。
本实施例中的具体示例可以参考上述实施例及示例性实施方式中所描述的示例，本实施例在此不再赘述。
显然，本领域的技术人员应该明白，上述的本公开的各模块或各步骤可以用通用的计算装置来实现，它们可以集中在单个的计算装置上，或者分布在多个计算装置所组成的网络上，它们可以用计算装置可执行的程序代码来实现，从而，可以将它们存储在存储装置中由计算装置来执行，并且在某些情况下，可以以不同于此处的顺序执行所示出或描述的步骤，或者将它们分别制作成各个集成电路模块，或者将它们中的多个模块或步骤制作成单个集成电路模块来实现。这样，本公开不限制于任何特定的硬件和软件结合。
以上所述仅为本公开的优选实施例而已，并不用于限制本公开，对于本领域的技术人员来说，本公开可以有各种更改和变化。凡在本公开的原则之内，所作的任何修改、等同替换、改进等，均应包含在本公开的保护范围之内。

Claims (20)

  1. 一种反馈处理方法，应用于基站，所述方法包括：
    接收目标分组中UE根据对应的上报配置信息上报的信道状态信息CSI数据，其中，所述上报配置信息用于指示所述目标分组内的UE在预设上报时隙中上报所述CSI数据；
    若在所述预设上报时隙中未接收到所述目标分组中目标UE上报的所述CSI数据，基于训练好的信道预测模型，使用所述目标分组中除所述目标UE之外的其他UE上报的CSI数据与所述目标UE预设时间段内上报的CSI数据进行信道预测，得到优化后的CSI数据；
    根据所述优化后的CSI数据确定下行数据发送使用的下行资源。
  2. 根据权利要求1所述的方法，其中所述方法还包括：
    若在所述预设上报时隙中接收到所述目标分组中目标UE上报的CSI数据，基于训练好的信道预测模型，使用所述目标分组中除所述目标UE之外的其他UE上报的CSI数据与所述目标UE的CSI数据进行信道预测，得到所述优化后的CSI数据；和/或
    若在所述预设上报时隙中接收到所述目标分组中目标UE上报的CSI数据，根据所述CSI数据判断所述目标UE是否保留在所述目标分组中；在判断结果为所述目标UE不保留在所述目标分组中的情况下，将所述目标UE从所述目标分组内剔除，以更新所述目标分组。
  3. 根据权利要求2所述的方法，其中，根据所述CSI数据判断所述目标UE是否保留在所述目标分组中包括：
    根据所述目标分组对应的训练好的信道预测模型对所述目标UE上报的所述CSI数据进行信道预测，得到所述优化后的CSI数据；
    根据所述优化后的CSI数据判断所述目标UE是否保留在所述目标分组中。
  4. 根据权利要求3所述的方法，其中，根据所述目标分组对应的训练好的 信道预测模型对所述目标UE上报的所述CSI数据进行信道预测，得到所述优化后的CSI数据包括：
    将所述CSI数据中的信道进行SVD分解，得到所述目标UE在子带上的特征值与特征向量；
    将所述目标UE的CSI数据对应的特征值与特征向量，以及所述目标分组内其他UE的CSI数据对应的特征值和特征向量值输入所述训练好的信道预测模型中，得到所述训练好的信道预测模型输出的所述优化后的CSI数据，所述优化后的CSI数据为优化后的所述目标UE在子带上的特征值与特征向量。
  5. 根据权利要求4所述的方法，其中，在将所述CSI数据中的信道进行SVD分解，得到所述目标UE在子带上的特征值与特征向量之后，所述方法还包括：
    使用所述目标UE上报的CSI数据中的子带信道质量指示以及UE调度子带权重对所述目标UE的CSI数据对应的特征值与特征向量进行修正。
  6. 根据权利要求2所述的方法，其中，根据所述CSI数据判断所述目标UE是否保留在所述目标分组中包括：
    根据所述CSI数据确定所述目标UE的分组；
    判断所述目标UE的分组与所述目标分组是否相同；
    在所述目标UE的分组与所述目标分组相同的情况下，确定所述判断结果为所述目标UE保留在所述目标分组中；
    在所述目标UE的分组与所述目标分组不相同的情况下，确定所述判断结果为所述目标UE不保留在所述目标分组中。
  7. 根据权利要求6所述的方法，其中，根据所述CSI数据确定所述目标UE的分组包括：
    分别根据所述目标UE的CSI数据与除所述目标UE之外的每个UE的CSI数据确定所述目标UE与除所述目标UE之外的每个UE组成的UE对的加权相关值，确定所述相关值大于预设阈值的UE对组成的分组为所述目标UE的分组；或者
    将所述CSI数据输入训练好的分组决策模型中，得到所述训练好的分组决策模型输出的所述目标UE的分组。
  8. 根据权利要求7所述的方法，其中，在将所述CSI数据输入训练好的分组决策模型中，得到所述训练好的分组决策模型输出的所述目标UE的分组之后，所述方法还包括：
    在所述判断结果为目标UE不保留在所述目标分组中的情况下，将所述目标UE设置为未分组用户；
    在所述判断结果为目标UE保留在所述目标分组中的情况下，根据所述目标UE的CSI数据与所述目标分组内除所述目标UE之外其他UE的CSI数据对所述训练好的分组决策模型进行训练，以更新所述训练好的分组决策模型。
  9. 根据权利要求1所述的方法，其中，所述方法还包括：
    若在所述预设上报时隙中接收到所述目标UE上报的CSI数据，根据所述目标UE的CSI数据与存储的所述目标分组内除所述目标UE之外其他UE的CSI数据对所述训练好的信道预测模型进行训练，以更新所述训练好的信道预测模型。
  10. 根据权利要求1所述的方法，其中，所述方法还包括：
    在所述目标UE不属于任一分组的情况下，根据所述目标UE上报的CSI数据确定所述目标UE的分组结果，其中，所述分组结果包括：所述目标UE加入已有的某一分组，或者所述目标UE与其他未分组用户组成新的分组。
  11. 根据权利要求1至10中任一项所述的方法，其中，在接收目标分组中UE根据对应的上报配置信息上报的CSI数据之前，所述方法还包括：
    获取覆盖范围内的多个UE上报的CSI数据，其中，所述目标UE为所述多个UE中的任一UE；
    根据所述多个UE的CSI数据对所述多个UE进行初始分组，得到所述多个分组，其中，所述目标分组为所述多个分组中的任一分组；
    分别为所述多个分组选择对应的训练好的信道预测模型，并将所述训练好的 信道预测模型、分组参数及所述上报配置信息反馈给所述多个UE。
  12. 一种反馈处理方法，应用于目标UE，所述方法包括：
    确定信道状态信息CSI数据，其中，所述CSI数据为计算出的CSI数据或对所述计算出的CSI进行优化后的CSI数据；
    根据所在目标分组对应的上报配置信息向基站上报所述CSI数据，以使所述基站若在所述上报配置信息指示的预设上报时隙中未接收到所述CSI数据，基于训练好的信道预测模型，使用所述目标分组中除所述目标UE之外的其他UE上报的CSI数据与所述目标UE预设时间段内上报的CSI数据进行信道预测，得到优化后的CSI数据；根据所述优化后的CSI数据确定下行数据发送使用的下行资源，其中，所述上报配置信息用于指示所述目标分组内UE在预设上报时隙中上报所述CSI数据。
  13. 根据权利要求12所述的方法，其中，在确定信道状态信息CSI数据之前，所述方法还包括：
    接收所述基站下发的CSI-RS探测信号；
    判断所述CSI-RS探测信息的接收质量是否满足预设条件；
    在判断结果为是的情况下，计算CSI数据，得到所述计算出的CSI数据，根据所述CSI数据与所述基站下发的所述目标分组内其他UE的CSI数据对训练好的信道预测模型进行训练，以更新所述训练好的信道预测模型；
    在判断结果为否的情况下，计算CSI数据，得到所述计算出的CSI数据，根据训练好的信道预测模型对所述计算出的CSI数据进行信道预测，得到所述优化后的CSI数据。
  14. 根据权利要求13所述的方法，其中，根据训练好的信道预测模型对所述计算出的CSI数据进行预测，得到所述优化后的CSI数据包括：
    对所述CSI数据中的信道进行SVD分解，得到在子带上的特征值和特征向量；
    将所述CSI数据对应的特征值和特征向量值，以及所述目标分组内其他UE 的CSI数据对应的特征值和特征向量值输入所述训练好的信道预测模型中，得到所述训练好的信道预测模型输出的所述优化后的CSI数据。
  15. 根据权利要求14所述的方法，其中，在对所述CSI数据中的信道进行SVD分解，得到在子带上的特征值和特征向量之后，所述方法还包括：
    使用所述CSI数据中的子带信道质量指示以及UE调度子带权重对所述CSI数据对应的特征值与特征向量进行修正。
  16. 根据权利要求12至15中任一项所述的方法，其中，所述方法还包括：
    接收所述基站下发的为所述目标分组选择的所述训练好的信道预测模型与分组参数；和/或
    根据所述上报配置信息设置所述CSI数据的上报周期。
  17. 一种反馈处理装置，应用于基站，所述装置包括：
    接收数据模块，被配置成接收目标分组中UE根据对应的上报配置信息上报的信道状态信息CSI数据，其中，所述上报配置信息用于指示所述目标分组内的UE在预设上报时隙中上报所述CSI数据；
    第一信道预测模块，被配置成若在所述预设上报时隙中未接收到所述目标分组中目标UE上报的所述CSI数据，基于训练好的信道预测模型，使用所述目标分组中除所述目标UE之外的其他UE上报的CSI数据与所述目标UE预设时间段上报的CSI数据进行信道预测，得到优化后的CSI数据；
    确定资源模块，被配置成根据所述优化后的CSI数据确定下行数据发送使用的下行资源。
  18. 一种UE分组处理装置，应用于目标UE，所述装置包括：
    确定数据模块，被配置成确定信道状态信息CSI数据，其中，所述CSI数据为计算出的CSI数据或对所述计算出的CSI进行优化后的CSI数据；
    上报模块，被配置成根据所在目标分组对应的上报配置信息向基站上报所述CSI数据，以使所述基站若在所述上报配置信息指示的预设上报时隙中未接收到 所述CSI数据，基于训练好的信道预测模型，使用所述目标分组中除所述目标UE之外的其他UE上报的CSI数据与目标UE预设时间段上报的CSI数据进行信道预测，得到优化后的CSI数据；根据所述优化后的CSI数据确定下行数据发送使用的下行资源，其中，所述上报配置信息用于指示所述目标分组内UE在预设上报时隙中上报所述CSI数据。
  19. 一种计算机可读的存储介质，所述存储介质中存储有计算机程序，其中，所述计算机程序被设置为运行时执行所述权利要求1至11、12至16任一项中所述的方法。
  20. 一种电子装置，包括存储器和处理器，所述存储器中存储有计算机程序，所述处理器被设置为运行所述计算机程序以执行所述权利要求1至11、12至16任一项中所述的方法。
PCT/CN2023/113645 2022-08-31 2023-08-18 一种反馈处理方法、装置、存储介质及电子装置 WO2024046140A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202211068191.6A CN117675145A (zh) 2022-08-31 2022-08-31 一种反馈处理方法、装置、存储介质及电子装置
CN202211068191.6 2022-08-31

Publications (1)

Publication Number Publication Date
WO2024046140A1 true WO2024046140A1 (zh) 2024-03-07

Family

ID=90066958

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/113645 WO2024046140A1 (zh) 2022-08-31 2023-08-18 一种反馈处理方法、装置、存储介质及电子装置

Country Status (2)

Country Link
CN (1) CN117675145A (zh)
WO (1) WO2024046140A1 (zh)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2021072691A1 (en) * 2019-10-17 2021-04-22 Qualcomm Incorporated Configuration of csi reference resource and csi target resource for predictive estimation of channel state information
CN113541892A (zh) * 2020-04-15 2021-10-22 维沃移动通信有限公司 一种csi处理方法、处理装置及ue
CN114760654A (zh) * 2021-01-08 2022-07-15 北京紫光展锐通信技术有限公司 Csi的上报方法及装置
CN114765509A (zh) * 2021-01-15 2022-07-19 大唐移动通信设备有限公司 信息上报、接收方法、终端设备及网络设备

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2021072691A1 (en) * 2019-10-17 2021-04-22 Qualcomm Incorporated Configuration of csi reference resource and csi target resource for predictive estimation of channel state information
CN113541892A (zh) * 2020-04-15 2021-10-22 维沃移动通信有限公司 一种csi处理方法、处理装置及ue
CN114760654A (zh) * 2021-01-08 2022-07-15 北京紫光展锐通信技术有限公司 Csi的上报方法及装置
CN114765509A (zh) * 2021-01-15 2022-07-19 大唐移动通信设备有限公司 信息上报、接收方法、终端设备及网络设备

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
QUALCOMM INCORPORATED: "CSI enhancements: MTRP and FR1 FDD reciprocity", 3GPP DRAFT; R1-2110171, 3RD GENERATION PARTNERSHIP PROJECT (3GPP), MOBILE COMPETENCE CENTRE ; 650, ROUTE DES LUCIOLES ; F-06921 SOPHIA-ANTIPOLIS CEDEX ; FRANCE, vol. RAN WG1, no. e-Meeting; 20211011 - 20211019, 2 October 2021 (2021-10-02), Mobile Competence Centre ; 650, route des Lucioles ; F-06921 Sophia-Antipolis Cedex ; France, XP052059107 *
SAMSUNG: "On mechanisms to improve reliability for RRC_CONNECTED UEs", 3GPP DRAFT; R1-2105337, 3RD GENERATION PARTNERSHIP PROJECT (3GPP), MOBILE COMPETENCE CENTRE ; 650, ROUTE DES LUCIOLES ; F-06921 SOPHIA-ANTIPOLIS CEDEX ; FRANCE, vol. RAN WG1, no. e-Meeting; 20210519 - 20210527, 12 May 2021 (2021-05-12), Mobile Competence Centre ; 650, route des Lucioles ; F-06921 Sophia-Antipolis Cedex ; France , XP052011381 *
ZTE: "Preliminary Simulation Results of Rel-17 MBS", 3GPP DRAFT; R1-2008829, 3RD GENERATION PARTNERSHIP PROJECT (3GPP), MOBILE COMPETENCE CENTRE ; 650, ROUTE DES LUCIOLES ; F-06921 SOPHIA-ANTIPOLIS CEDEX ; FRANCE, vol. RAN WG1, no. e-Meeting; 20201026 - 20201113, 24 October 2020 (2020-10-24), Mobile Competence Centre ; 650, route des Lucioles ; F-06921 Sophia-Antipolis Cedex ; France , XP051946711 *

Also Published As

Publication number Publication date
CN117675145A (zh) 2024-03-08

Similar Documents

Publication Publication Date Title
US11973708B2 (en) Method and apparatus for reporting channel state information
US20230118031A1 (en) Neural network adjustment method and apparatus
US20230051456A1 (en) Information transmission and receiving method and device, storage medium and electronic device
US20230337194A1 (en) Communication method and apparatus
WO2021142836A1 (zh) 通信方法及装置
KR102579905B1 (ko) 대규모 mimo 시스템에서의 전송 방법 전환 방법 및 디바이스
US20230412430A1 (en) Inforamtion reporting method and apparatus, first device, and second device
WO2023246618A1 (zh) 信道矩阵处理方法、装置、终端及网络侧设备
US11431543B2 (en) Facilitating a two-stage downlink control channel in a wireless communication system
WO2023185978A1 (zh) 信道特征信息上报及恢复方法、终端和网络侧设备
WO2024046140A1 (zh) 一种反馈处理方法、装置、存储介质及电子装置
CN106936486B (zh) 一种csi反馈方法及装置
CN117397180A (zh) 用于csi预测控制的装置
WO2023179476A1 (zh) 信道特征信息上报及恢复方法、终端和网络侧设备
WO2024032606A1 (zh) 信息传输方法、装置、设备、系统及存储介质
WO2022151063A1 (zh) 信息发送方法、信息接收方法、装置、设备及介质
WO2023179474A1 (zh) 信道特征信息辅助上报及恢复方法、终端和网络侧设备
WO2023179473A1 (zh) 信道特征信息上报及恢复方法、终端和网络侧设备
WO2023077274A1 (zh) Csi反馈方法、装置、设备及存储介质
WO2024037380A1 (zh) 信道信息处理方法、装置、通信设备及存储介质
WO2024104126A1 (zh) 更新ai网络模型的方法、装置和通信设备
US20240162957A1 (en) Method and device for reporting csi based on ai model in wireless communication system
US20230403587A1 (en) Method and apparatus for monitoring and reporting ai model in wireless communication system
CN115189740A (zh) 信道状态信息反馈系统及方法
CN116961711A (zh) 一种通信方法及装置

Legal Events

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

Ref document number: 23859174

Country of ref document: EP

Kind code of ref document: A1