Device capability and performance monitoring for the model

Abstract

Techniques for using artificial intelligence / machine learning (AI / ML) models to improve the accuracy of channel state information (CSI) are described. Due to limited model generalization and dynamic wireless environments, several methods are proposed to monitor the performance of the models. The wireless communication method includes transmitting, by a communication device, a capability message indicating that the communication device is capable of using the model and performing one or more wireless communication operations.

Description

[Technical Field]

[0001] This disclosure generally pertains to digital wireless communications. [Background technology]

[0002] Wireless communication technologies are driving the world towards an increasingly connected and networked society. Compared to existing wireless networks, next-generation systems and wireless communication techniques will need to support a much wider range of use case characteristics and provide a more complex and advanced range of access requirements and flexibility.

[0003] Long-Term Evolution (LTE) is a wireless communication standard for mobile devices and data terminals developed by the 3rd Generation Partnership Project (3GPP®). LTE Advanced (LTE-A) is a wireless communication standard that improves upon the LTE standard. The fifth generation of wireless systems, known as 5G, is tasked with evolving the LTE and LTE-A wireless standards to support higher data rates, more connections, ultra-low latency, high reliability, and other emerging business needs. [Overview of the project] [Means for solving the problem]

[0004] Techniques for reporting UE capabilities, including those of AI / ML models, are disclosed. Several solutions for monitoring model performance in several embodiments are discussed.

[0005] A wireless communication method includes a communication device transmitting a capability message, which indicates that the communication device is capable of performing one or more wireless communication operations using a model provided by the communication device.

[0006] In some embodiments, the capability message includes a time requirement for the model. In some embodiments, the time requirement is associated with a model identifier.

[0007] In some embodiments, the model can be received from a network by a communication device.

[0008] In some embodiments, the capability message includes one of the following: computation time, computation time offset, or model activation time. In some embodiments, computation time is the time required for the communication device to perform a certain operation, computation time offset is the additional time required for the communication device to perform a certain operation, and model activation time is the time required for the communication device to activate a certain operation.

[0009] In some embodiments, the computation time can be a constant independent of the selected model. In some embodiments, the value of the computation time offset depends on the model. In some embodiments, the capacity message indicates that the measurement report is generated after a time elapsed T based on a reference signal received by the communication device, where T depends on the computation time offset and the computation time.

[0010] In some embodiments, the method further includes a communication device transmitting an identifier message indicating that the model is activated.

[0011] In some embodiments, identifier messages can be transmitted through PUCCH.

[0012] In some embodiments, the method further includes receiving a response from a communication device indicating that the identifier message has been successfully received.

[0013] In some embodiments, the method further includes deactivating the model by the communication device when a response is not received by the communication device.

[0014] In some embodiments, the capability message includes the capacity of the communication device, and the capacity includes the parallel capabilities related to model and non-model.

[0015] Another wireless communication method includes reporting, by the communication device, a performance indication of the model.

[0016] In some embodiments, the performance indication includes the relationship between the auxiliary information and the actual measurement information performed by the communication device.

[0017] In some embodiments, the relationship can be a similarity or a distance, the similarity includes cosine similarity, generalized cosine similarity, or squared generalized cosine similarity, and the distance includes Euclidean distance or normalized mean squared error.

[0018] In some embodiments, the communication device is at least configured with dedicated opportunities for the performance indication of the model. In some embodiments, the performance indication includes the predicted measurement value and the actual measurement value of the dedicated opportunity.

[0019] In some embodiments, the actual measurement value includes the RSRP of a plurality of reference signals, and the predicted measurement value includes the predicted RSRP of the reference signals based on the model.

[0020] In some embodiments, the actual measurement value includes a codebook-based precoding matrix indicator (PMI).

[0021] In some embodiments, the performance indication further includes the relationship between the output of the model and the input of the model.

[0022] In some embodiments, the performance indication includes a relationship between M - N of the remaining measurement results out of N measurement results and a part of the model output, where the N measurement results are based on a reference signal, the M measurement results are used for the input of the model based on the N measurement results, and M < N.

[0023] In some embodiments, the relationship can be a similarity or a distance, the similarity can be at least one of the following: cosine similarity, generalized cosine similarity, or squared generalized cosine similarity, and the distance can be at least one of the following: Euclidean distance or normalized mean squared error.

[0024] In yet another exemplary aspect, the method described above is embodied in the form of processor - executable code and stored in a non - transitory computer - readable storage medium. The code included in the computer - readable storage medium, when executed by a processor, causes the processor to perform the method described in this patent document.

[0025] In yet another exemplary embodiment, a device configured or operable to perform the method described above is disclosed.

[0026] The above and other aspects and their implementations are described in more detail in the drawings, the description, and the claims. The present invention provides, for example, the following: (Item 1) A wireless communication method, wherein the method is This includes transmitting capability messages via a communication device. A method for indicating that the capability message indicates that the communication device is capable of using a model to perform one or more wireless communication operations. (Item 2) The capability message is the method described in item 1, including the time requirements of the model. (Item 3) The aforementioned time requirement is the method described in item 2, associated with the model identifier. (Item 4) The model described above can be received from a network by the communication device, according to the method described in item 3. (Item 5) The capability message includes one of the following: computation time, computation time offset, or model activation time. The method according to item 2, wherein the computation time is the time required for the communication device to perform a certain operation, the computation time offset is the additional time required for the communication device to perform a certain operation, and the model activation time is the time required for the communication device to activate a certain operation. (Item 6) The method according to item 5, wherein the computation time can be a constant independent of the selected model. (Item 7) The value of the computation time offset depends on the model, as described in item 5. (Item 8) The capacity message indicates that the measurement report is generated after a time elapsed T based on a reference signal received by the communication device, where T depends on the calculation time offset and the calculation time, according to the method of item 5. (Item 9) The method according to item 1, further comprising transmitting an identifier message indicating that the model is activated by the communication device. (Item 10) The identifier message can be transmitted via PUCCH, as described in item 9. (Item 11) The method according to item 9, further comprising receiving a response from the communication device indicating that the identifier message has been successfully received. (Item 12) The method according to item 11, further comprising deactivating the model when the communication device does not receive the response. (Item 13) The method according to item 1, wherein the capability message includes the capacity of the communication device, and the capacity includes parallel capabilities related to model-related and non-model-related aspects. (Item 14) A wireless communication method, the method including reporting, by a communication device, a performance indication of a model. (Item 15) The method according to item 14, wherein the performance indication includes a relationship between auxiliary information and actual measurement information performed by the communication device. (Item 16) The method according to item 15, wherein the relationship can be a similarity or a distance, the similarity includes a cosine similarity, a generalized cosine similarity, or a squared generalized cosine similarity, and the distance includes a Euclidean distance or a normalized mean squared error. (Item 17) The method according to item 14, wherein the communication device is configured with at least a dedicated opportunity for the performance indication of the model. (Item 18) The method according to item 14, wherein the performance indication includes a predicted measurement value and an actual measurement value of the dedicated opportunity. (Item 19) The method according to item 18, wherein the actual measurement value includes the RSRP of a plurality of reference signals, and the predicted measurement value includes the predicted RSRP of the reference signals based on the model. (Item 20) The method according to item 18, wherein the actual measurement value includes a codebook-based precoding matrix indication (PMI). (Item 21) The method according to item 14, wherein the performance indication further includes a relationship between an output of the model and an input of the model. (Item 22) The method according to item 14, wherein the performance indication includes a relationship between the remaining M - N measurement results out of N measurement results and a part of the model output, the N measurement results are based on the reference signals, the M measurement results are based on the N measurement results and are used for the input of the model, and M < N. (Item 23) The method according to item 21 or 22, wherein the relationship can be a similarity or a distance, the similarity can be at least one of a cosine similarity, a generalized cosine similarity, or a squared generalized cosine similarity, and the distance can be at least one of a Euclidean distance or a normalized mean squared error. (Item 24) An apparatus for wireless communication, the apparatus including a processor configured to implement the method described in one or more of items 1 - 24. (Item 25) A non-transient, computer-readable program storage medium storing code, wherein the code, when executed by a processor, causes the processor to perform one or more of the methods described in items 1-24. [Brief explanation of the drawing]

[0027] [Figure 1] Figure 1 shows an example of an artificial intelligence / machine learning (AI / ML) model.

[0028] [Figure 2] Figure 2 shows an example with computation time and computation time offset.

[0029] [Figure 3] Figure 3 shows an example of the relationship between auxiliary channel information and actual channel measurements.

[0030] [Figure 4] Figure 4 shows an example of a dedicated channel measurement opportunity for monitoring model performance.

[0031] [Figure 5] Figure 5 shows an example of employing a relationship between a portion of the AI ​​model output and the AI ​​model input for monitoring model performance.

[0032] [Figure 6] Figure 6 shows an illustrative block diagram of a hardware platform that may be part of a network device or communication device.

[0033] [Figure 7] Figure 7 shows an example of wireless communication including a base station (BS) and user equipment (UE) based on several implementations of the disclosed technology. [Modes for carrying out the invention]

[0034] (Introduction) Section headings in this publication are used solely to improve readability and do not limit the scope of the embodiments and techniques disclosed within each section to that section only. Furthermore, some embodiments are described with reference to the 3rd Generation Partnership Project (3GPP®) New Radio (NR) Standard ("5G") for ease of understanding, and the techniques described may be implemented in different radio systems implementing protocols other than the 5G protocol. In addition, the AI / ML models are illustrative scenarios, and the technical solutions described herein may be generalizable or applicable to any model determining the relationship between inputs and outputs.

[0035] (Introduction Notes)

[0036] Artificial intelligence / machine learning (AI / ML) is being researched and applied in various fields.

[0037] Several studies in artificial intelligence and machine learning are being conducted, particularly to improve the efficiency of wireless communication systems at the physical layer. For example, AI / ML models can be used to improve the accuracy of channel status information (CSI). In addition, AI / ML models can predict beam information. To support channel measurements by using AI / ML models, UEs in the network should have the corresponding capabilities.

[0038] This application discloses several procedures proposed for reporting the UE capabilities of AI / ML models. Due to limited model generalization and dynamic environments, models can degrade in performance over time. Therefore, several solutions for monitoring model performance are discussed in this application.

[0039] (I. Introduction)

[0040] AI / ML is being studied and used in various fields to extract features that cannot be derived by other mathematical methods. Generally, an AI / ML model is a data-driven algorithm that applies AI / ML techniques to generate a set of outputs based on a set of inputs and includes at least three parts, as shown in Figure 1. - AI model input: Data that is fed into the AI / ML model. - AI model output: Output of AI / ML model - AI Model: An algorithm for deriving the relationship between AI model input and AI model output.

[0041] To facilitate the discussion, the following technical terms will be introduced along with some general explanations.

[0042] AI model training is the process of training an AI / ML model by learning input / output relationships in a data-driven manner, and obtaining a trained AI / ML model for inference.

[0043] AI model inference is the process of employing a trained AI / ML model to produce a set of outputs based on a set of inputs.

[0044] In measurement reporting, such as CSI reporting or beam reporting, the UE may use AI models to derive the measurement reports. To support channel measurements by using AI / ML models, the UE should have the corresponding capability.

[0045] This application proposes several procedures for reporting the UE capabilities of AI / ML models. Furthermore, due to limited model generalization and dynamic environments, models can degrade in performance over time. In other words, the data used for AI model training and that used for AI model inference can be quite different, potentially preventing the AI ​​model from obtaining the expected AI model output. This application discusses several solutions for monitoring model performance.

[0046] (II. Exemplary Embodiments)

[0047] (A. Capability signal transmission and auxiliary signal transmission)

[0048] In some embodiments, the UE may report its model (e.g., AI / ML model)-related information to the network through UE capability signaling or auxiliary information signaling.

[0049] In some embodiments, model-related information may be reported after the network sends a request message to the UE, and the request message may include a model identifier.

[0050] In some embodiments, model-related information can be reported without first requesting a message from the network.

[0051] In some embodiments, model-related information includes time requirements for the model.

[0052] In some embodiments, model-related information may include a model identifier for relating to the corresponding time requirement:

[0053] The time requirement may include one of the following:

[0054] (1) Calculation time)

[0055] In one example, computation time may include the time required for the UE to perform / operate / run the model, and / or the time required for the UE to prepare the measurement reports associated with the model.

[0056] In some embodiments, the computation time is the same for all models.

[0057] In some embodiments, the computation time differs with respect to different reporting quantities in the measurement report. In other words, the computation time depends on the content included in the measurement report. For example, the computation time for a beam report may differ from that for a CSI report.

[0058] In some embodiments, computation time is not required to be reported by the UE. The computation time reuses a value defined for non-model-based measurement reporting.

[0059] (2) Calculation time offset)

[0060] In one example, the computation time offset may include the additional time the UE needs to do / operate / execute for the model.

[0061] In some embodiments, each model has its own computation time offset.

[0062] In some embodiments, the computation time offset is not required to be reported by the UE. In this scenario, the value of the computation time offset is set to zero by default. Alternatively, the UE may report a zero value for the computation time offset.

[0063] In some embodiments, as shown in Figure 2, the UE reports only measurement values ​​based on a reference signal, the reference signal is received T time units earlier than the time the measurement value report is transmitted, and T time units is the calculation time plus the calculation time offset.

[0064] (3) Model activation / application time)

[0065] In some embodiments, when a UE receives a media access control (MAC) signal transmission containing an instruction command for using a model, the UE may require some time to activate the model. For example, when the UE transmits a PUCCH with HARQ-ACK information corresponding to a PDSCH carrying an instruction command at time n, the UE assumes that the corresponding model, which is ready for inference / to be activated, should be applied at time n + model activation time.

[0066] In some embodiments, when a UE receives a Downlink Control Instruction (DCI) containing an instruction command for using a model, the UE may require some time to activate the model. For example, if the UE receives a PDCCH carrying an instruction command at time n, the UE assumes that the corresponding model, which is ready for inference / to be activated, should be applied at time n + model activation time.

[0067] In some embodiments, the UE can send a signal to the network to indicate that a model is ready / activated for inference. In some embodiments, the UE can send a signal to the network to indicate that a certain model is ready / activated for inference, and if the UE is indicated to use this model to perform an operation, the UE does not require any extra time (e.g., activation time) to activate this model.

[0068] In some embodiments, signal transmission can be carried out by PUCCH, and each model may be associated with a dedicated PUCCH resource.

[0069] In some embodiments, signal transmission can be carried out by a PRACH, and each model may be associated with a dedicated PRACH.

[0070] In some embodiments, the UE can receive a response from the network confirming that the signal transmission has been successfully received by the network. In some embodiments, the response is transmitted by MAC signal transmission. In some embodiments, the UE confirms that the signal transmission has been successfully received by the network if it detects a PDCCH scrambled by a dedicated RNTI.

[0071] In some embodiments, if the UE does not receive a response, the UE may automatically deactivate the model.

[0072] In some embodiments, the UE may automatically deactivate the model if the waiting time from when the UE starts transmitting a signal exceeds a time threshold.

[0073] In some embodiments, the UE may report parallel / mixed UE capabilities for both model-related and non-model-related functions.

[0074] In some embodiments, the UE may report that it can consist of a number of measurement reports that use the model and a number of measurement reports that do not use the model.

[0075] For example, a UE might report that if it consists of M CSI reports that use a model in a bandwidth partial / component carrier (BWP / CC), it is not expected that the UE will consist of more than N other CSI reports that do not use a model. Here, the value of N can be either not requested to be reported or can be reported as zero, both of which indicate that if the UE consists of CSI reports that use a model, it cannot consist of other CSI reports that do not use a model.

[0076] In another example, a UE might report that if it consists of M beam reports that use a model in BWP / CC (Bandwidth Part / Component Carrier), it is not expected to consist of more than N other beam reports that do not use a model. Here, the value of N can be either not requested to be reported or reported as zero, both of which indicate that if the UE consists of beam reports that use a model, it cannot consist of other beam reports that do not use a model.

[0077] In some embodiments, the UE may report that it can process measurement reports using the model and measurement reports not using the model simultaneously.

[0078] For example, a UE can simultaneously process M beam reports using a model and N beam reports not using a model within a CC (or across all CCs). The value of N can be either not requested to be reported or reported as zero.

[0079] For example, a UE can simultaneously process M beam reports using a model and N CSI reports not using a model within a CC (or across all CCs). The value of N can be either not requested to be reported or reported as zero.

[0080] (B. Model monitoring procedure)

[0081] On the UE side, the performance of deployed models may not always be good. Due to limited model generalization and the dynamic environment, model performance can degrade. Therefore, both the network and the UE should constantly monitor model performance.

[0082] Here, we propose three solutions for monitoring model performance.

[0083] Method 1: The network provides auxiliary channel information to assist in model performance monitoring.

[0084] In some embodiments, the UE may report the relationship between auxiliary channel information and actual channel measurements.

[0085] In the example shown in Figure 3, the auxiliary channel information and the actual channel measurement are, respectively, H assist And shown by H. The output of the first AI model is,

number

number

[0086] One metric for evaluating model performance is similarity, such as cosine similarity, generalized cosine similarity, or squared generalized cosine similarity. assist The similarity between and H is S input And,

number

number

[0087] Another metric for evaluating model performance is distance, such as Euclidean distance or normalized mean squared error. assistThe distance between and H is D input and [Number] and [Number] the distance between and is D output Assume that it is so. In one example, when the value of D input is equal to (or approximately equal to) that of D output the AI model still operates well for a certain guaranteed performance. <​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​​Method 2: A dedicated channel measurement opportunity is used to monitor model performance.

[0093] In some embodiments, the network may provide at least one dedicated channel measurement opportunity for monitoring model performance.

[0094] In one example where an AI model is employed for channel prediction, the AI ​​model input is based on actual channel measurements from several previous opportunities (e.g., observation opportunities). The AI ​​model output is the predicted channel measurement, which can be used in several future opportunities (e.g., prediction opportunities). Therefore, the UE does not need to obtain the actual channel measurement for the prediction opportunity. However, if the opportunity is presented as a dedicated channel measurement opportunity (e.g., monitoring opportunity) for monitoring model performance, as shown in Figure 4, the UE may need to obtain both the actual and predicted channel measurements. The UE / network can then compare the actual and predicted channel measurements to check whether the channel prediction is sufficiently accurate.

[0095] In one embodiment, the UE needs to report both the actual channel measurement and the predicted channel measurement of the monitoring opportunity in the measurement report.

[0096] In one example, an AI model is used for beam prediction.

[0097] In some embodiments, the actual channel measurement includes the RSRP of multiple reference signals, and the predicted channel measurement includes the RSRP of multiple reference signals.

[0098] In some embodiments, the network may show the number of RSRPs included in the actual channel measurements and the predicted channel measurements.

[0099] In some embodiments, the number of RSRPs included in the actual channel measurement is the same as that included in the predicted channel measurement.

[0100] In some embodiments, each RSRP is associated with a reference signal index.

[0101] In another example, an AI model could be employed for CSI prediction.

[0102] In some embodiments, the actual channel measurement includes a conventional codebook-based precoding matrix indication (PMI).

[0103] Method 3: The AI ​​module's dedicated module is for monitoring model performance.

[0104] In some embodiments, a portion of the AI ​​model output can be used for model performance monitoring.

[0105] In some embodiments, a portion of the AI ​​model output includes an indicator that shows at least whether the AI ​​model is valid. The indicator may be reported in the measurement report.

[0106] In some embodiments, the relationship between a portion of the AI ​​model output and the AI ​​model input can be used for monitoring model performance.

[0107] In the example shown in Figure 5, the AI ​​model input is channel information H, and the AI ​​model output of the first submodule is compressed channel information

number

number

[0108] In some embodiments, the measurement report may include an indication of a relationship.

[0109] In some embodiments, the relationship can be a similarity / distance between channel information and reconstructed channel information.

[0110] In some embodiments, the measurement report may include both an indication of a relationship and compressed channel information.

[0111] In some embodiments, the UE obtains N measurement results based on the received reference signals. The AI model input may include only M of the N measurement results (where M < N). In this scenario, the remaining N - M measurement results can be used for model performance monitoring.

[0112] In some embodiments, the N measurement results are N RSRP values, and each RSRP value corresponds to a beam between the network and the UE. The AI model input includes only M of the N RSRP values (where M < N). Thus, the remaining N - M RSRP values can be used for model performance monitoring. The remaining N - M RSRP values correspond to a set of beams between the network and the UE. Further, a part of the AI model output includes predicted N - M RSRP values, and the predicted N - M RSRP values correspond to a set of beams between the network and the UE.

[0113] In some embodiments, the UE may report a validity indicator based on the remaining N - M RSRP values and the predicted N - M RSRP values.

[0114] In some embodiments, the validity indicator can be a similarity between the remaining N - M RSRP values and the predicted N - M RSRP values.

[0115] In some embodiments, the effectiveness indicator can be the difference (i.e., distance) between the remaining N - M RSRP values and the predicted N - M RSRP values.

[0116] In some embodiments, the UE may report both the remaining N - M RSRP values and the predicted N - M RSRP values in the measurement report.

[0117] In some embodiments, one of the predicted N - M RSRP values is reported for one of the remaining N - M RSRP values corresponding to the same beam between the network and the UE.

[0118] In some embodiments, the N measurement results are N channel measurement results corresponding to the N - port reference signal. The AI model input includes only M (M < N) of the N channel measurement results. Thus, the remaining N - M channel measurement results can be used for model performance monitoring. The remaining N - M channel measurement results correspond to a set of ports of the N - port reference signal. Further, a part of the AI model output includes the predicted N - M channel measurement results corresponding to the set of ports of the N - port reference signal.

[0119] In some embodiments, the UE may report an effectiveness indicator based on the remaining N - M channel measurement results and the predicted N - M channel measurement results.

[0120] In some embodiments, the effectiveness indicator can be the similarity between the remaining N - M channel measurement results and the predicted N - M channel measurement results.

[0121] In some embodiments, the effectiveness indicator can be the difference (i.e., distance) between the remaining N - M channel measurement results and the predicted N - M channel measurement results.

[0122] Figure 6 shows an exemplary block diagram of a hardware platform 600, which may be part of a network device (e.g., a base station) or a communication device (e.g., a user device (UE)). The hardware platform 600 includes at least one processor 610 and a memory 605 that stores instructions. The instructions configure the hardware platform 600 to perform operations as described in Figures 1-5 and 1-7 and in the various embodiments described in this patent document when executed by the processor 610. A transmitter 615 transmits or sends information or data to another device. For example, a network device transmitter may send a message to a user device. A receiver 620 receives information or data transmitted or sent by another device. For example, a user device may receive a message from a network device.

[0123] Implementations such as those discussed above would be applied to wireless communication. Figure 7 shows an example of a wireless communication system (e.g., a 5G or NR cellular network) including a base station 720 and one or more user devices (UEs) 711, 712, and 713. In some embodiments, the UE accesses a BS (e.g., the network) using a communication link to the network (sometimes referred to as the uplink direction, as depicted by dashed arrows 731, 732, and 733), which then enables subsequent communication from the BS to the UE (e.g., shown in the direction from the network to the UE, sometimes referred to as the downlink direction, as indicated by arrows 741, 742, and 743). In some embodiments, the BS transmits information to the UE (sometimes referred to as the downlink direction, as depicted by arrows 741, 742, and 743), which then enables subsequent communication from the UE to the BS (e.g., shown by dashed arrows 731, 732, and 733, sometimes referred to as the uplink direction, as indicated in the direction from the UE to the BS). UEs can include, for example, smartphones, tablets, mobile computers, machine-to-machine (M2M) devices, and Internet of Things (IoT) devices.

[0124] A wireless communication method includes a communication device transmitting a capability message indicating that the communication device is capable of performing one or more wireless communication operations using a model provided by the communication device.

[0125] In some embodiments, the capability message includes a time requirement for the model. In some embodiments, the time requirement is associated with a model identifier.

[0126] In some embodiments, the model can be received from a network by a communication device.

[0127] In some embodiments, the capability message includes one of the following: computation time, computation time offset, or model activation time. In some embodiments, computation time is the time required for the communication device to perform a certain operation, computation time offset is the additional time required for the communication device to perform a certain operation, and model activation time is the time required for the communication device to activate a certain operation.

[0128] In some embodiments, the computation time can be a constant independent of the selected model. In some embodiments, the value of the computation time offset depends on the model. In some embodiments, the capacity message indicates that the measurement report is generated after a time elapsed T based on a reference signal received by the communication device, where T depends on the computation time offset and the computation time.

[0129] In some embodiments, the method further includes a communication device transmitting an identifier message indicating whether the model is to be activated.

[0130] In some embodiments, identifier messages can be transmitted through PUCCH.

[0131] In some embodiments, the method further includes receiving a response from a communication device indicating that the identifier message has been successfully received.

[0132] In some embodiments, the method further includes deactivating the model when a response is not received by the communication device.

[0133] In some embodiments, the capability message includes the capacity of the communication device, and the capacity includes parallel capabilities relating to model-related and non-model-related capabilities.

[0134] Another wireless communication method involves a communication device reporting performance specifications for the model.

[0135] In some embodiments, the performance indication includes a relationship between auxiliary information and actual measurement information performed by a communication device.

[0136] In some embodiments, the relationship can be similarity or distance, where similarity includes cosine similarity, generalized cosine similarity, or squared generalized cosine similarity, and distance includes Euclidean distance or normalized mean squared error.

[0137] In some embodiments, the communication device comprises at least a dedicated device used for model performance instruction. In some embodiments, the performance instruction includes predicted and actual measurements from the dedicated device.

[0138] In some embodiments, the actual measured values ​​include the RSRP of multiple reference signals, while the predicted measured values ​​include the predicted RSRP of a reference signal based on a model.

[0139] In some embodiments, the actual measurements include codebook-based precoding matrix indications (PMIs).

[0140] In some embodiments, the performance instruction further includes a relationship between the model's output and its input.

[0141] In some embodiments, the performance indicator includes a relationship between MN of the remaining measurement results out of N measurement results and a portion of the model output, where N measurement results are used for the model input based on a reference signal, and M measurement results are used based on N measurement results. <Nである。

[0142] In some embodiments, the relationship can be a similarity or a distance, where the similarity can be at least one of the following: cosine similarity, generalized cosine similarity, or squared generalized cosine similarity, and the distance can be at least one of the following: Euclidean distance or normalized mean squared error.

[0143] In yet another exemplary aspect, the method described above is embodied in the form of processor-executable code and stored in a non-transient computer-readable storage medium. When the code contained in the computer-readable storage medium is executed by a processor, it causes the processor to carry out the method described in this patent document.

[0144] In yet another exemplary embodiment, a device configured or operable to carry out the methods described above is disclosed.

[0145] The above and other aspects and their implementations will be described in more detail in the drawings, descriptions and claims.

[0146] In this book, the term "exemplary" is used to mean "an example of" and does not imply an ideal or preferred embodiment unless otherwise noted.

[0147] Some of the embodiments described herein are described in a general context of methods or processes that may be implemented by a computer program product, which in one embodiment is implemented on a computer-readable medium containing computer-executable instructions, such as program code, and is executed by a computer in a networked environment. Computer-readable media may include, but are not limited to, removable and non-removable storage devices, such as read-only memory (ROM), random-access memory (RAM), compact discs (CDs), and digital versatile discs (DVDs). Thus, computer-readable media may include non-transient storage media. Generally, program modules may include routines, programs, objects, components, data structures, etc., that perform a particular task or implement a particular abstract data type. Computer or processor-executable instructions, associated data structures, and program modules represent examples of program code for performing steps of the methods disclosed herein. A particular sequence of such executable instructions or associated data structures represents an example of a corresponding action for implementing the functionality described in such steps or processes.

[0148] Some of the disclosed embodiments can be implemented as devices or modules using hardware circuits, software, or a combination thereof. For example, a hardware circuit implementation may include discrete analog and / or digital components integrated as part of a printed circuit board, for example. Alternatively, or in addition, the disclosed components or modules may be implemented as application-specific integrated circuits (ASICs) and / or field-programmable gate array (FPGA) devices. Some implementations may, in addition, or alternatively, include a digital signal processor (DSP), which is a specialized microprocessor with an architecture optimized for the operational needs of digital signal processing associated with the functionality disclosed herein. Similarly, various components or subcomponents within each module may be implemented in software, hardware, or firmware. Connectivity between modules and / or components within modules may be provided using any one of the connectivity methods and media known in the Art, including, but not limited to, communication over the Internet, wired, or wireless networks using appropriate protocols.

[0149] This book contains many details, but these should be interpreted as descriptions of features specific to particular embodiments rather than as limitations on the claimed invention or the scope of what can be claimed. Certain features described in this book in the context of a separate embodiment can also be implemented in a single embodiment or in combination. Conversely, various features described in the context of a single embodiment can also be implemented separately in multiple embodiments or in any preferred secondary combination. Furthermore, features are described above as acting in a combination, and may even be initially claimed as such, but one or more features from a claimed combination may, in some cases, be removed from the combination, and the claimed combination may be subject to secondary combinations or variations of secondary combinations. Similarly, while operations are depicted in the drawings in a particular order, this should not be understood as requiring that such operations be performed in a particular order or sequential order as shown, or that all illustrated operations be performed, in order to achieve a desired result.

[0150] Only a few implementations and examples are described, and other implementations, enhancements, and modifications may also be made based on those described and illustrated in this disclosure.

Claims

1. A method of wireless communication, wherein the method is A communication device transmits a capability message, the capability message indicating that the communication device is capable of using a model to perform one or more wireless communication operations, and the capability message includes the time requirements of the model. The communication device transmits an identifier message indicating that the model is activated, and the identifier message is transmitted through a physical uplink control channel (PUCCH). Includes, The aforementioned method, The communication device receives a response indicating that the identifier message has been successfully received, or The communication device deactivates the model when the response is not received by the communication device. Methods that further include the above.

2. The method according to claim 1, wherein the aforementioned time requirement is associated with a model identifier.

3. The aforementioned time requirement includes one of the following: computation time, computation time offset, or model activation time. The aforementioned calculation time is the time required for the communication device to perform a certain operation. The calculation time offset is an additional time in addition to the calculation time required for the communication device to perform a certain operation. The method according to claim 1, wherein the model activation time is the time required for the communication device to activate a certain operation.

4. The method according to claim 3, wherein the computation time is a constant independent of the selected model.

5. The method according to claim 3, wherein the value of the calculation time offset depends on the model.

6. The method according to claim 3, wherein the capability message indicates that the measurement report is generated after a time elapsed T based on a reference signal received by the communication device, where T depends on the calculation time offset and the calculation time.

7. The method according to claim 1, wherein the capability message includes the capacity of the communication device, and the capacity includes parallel capabilities relating to model-related and non-model-related capabilities.

8. A method of wireless communication, wherein the method is A communication device transmits an identifier message indicating that a model for performing one or more wireless communication operations is activated, wherein the identifier message is transmitted through a physical uplink control channel (PUCCH). The communication device reports performance instructions for the model. Includes, The aforementioned method, The communication device receives a response indicating that the identifier message has been successfully received, or The communication device deactivates the model when the response is not received by the communication device. Methods that further include the above.

9. The performance indication includes the relationship between auxiliary information and actual measurement information performed by the communication device. The method according to claim 8, wherein the relationship may be a similarity or a distance, the similarity includes cosine similarity, generalized cosine similarity, or squared generalized cosine similarity, and the distance includes Euclidean distance or normalized mean squared error.

10. The communication device comprises at least a dedicated device used for the performance indication of the model, The method according to claim 8, wherein the performance indication includes predicted and actual measurements of the dedicated machine.

11. The aforementioned actual measurements include codebook-based precoding matrix instructions (PMI), or The method according to claim 10, wherein the actual measured value includes one or more reference signal received powers (RSRPs) of a plurality of reference signals, and the predicted measured value includes one or more predicted RSRPs of the plurality of reference signals based on the model.

12. The performance instruction includes the relationship between the output of the model and the input of the model, The method according to claim 8, wherein the relationship may be a similarity or a distance, the similarity may be at least one of cosine similarity, generalized cosine similarity, or squared generalized cosine similarity, and the distance may be at least one of Euclidean distance or normalized mean squared error.

13. The method according to claim 8, wherein the performance indication includes a relationship between N-M remaining measurement results out of N measurement results and a portion of the model output, the N measurement results being used for the input of the model based on a reference signal and the M measurement results being used based on the N measurement results, and M < N.

14. A device for wireless communication, The device comprises at least one processor, The apparatus wherein the at least one processor is configured to perform the method according to any one of claims 1 to 13.

Images