Learning model selection at a device

Abstract

Various aspects of the present disclosure relate to learning model selection at a device. A first device, e.g., a user equipment (UE) and / or a network equipment (NE), obtains a set of data samples to provide as input to a set of learning models stored at the first device. The first device generates respective confidence scores associated with outputs from the set of learning models by providing the set of data samples as input to the set of learning models. The first device selects a learning model of the set of learning models for execution at the first device or at a second device. For example, the first device selects a learning model having a confidence score that satisfies a selection criterion, the confidence score generated using an output from the learning model.

Description

[0001] Related applications

[0002] This application claims priority to U.S. Provisional Application No. 63 / 633,524, filed April 12, 2024, entitled “Learning Model Selection at a Device,” the disclosure of which is incorporated herein by reference in its entirety. Technical Field

[0003] This disclosure relates to wireless communication, and more specifically, to learning model techniques for classification. Background Technology

[0004] A wireless communication system may include one or more network communication devices (which may also be referred to as network equipment (NE)) that support wireless communication with one or more user communication devices (which may also be referred to as user equipment (UE) or other suitable terms). The wireless communication system can support wireless communication with one or more user communication devices by utilizing the resources of the wireless communication system (e.g., time resources (e.g., symbols, time slots, subframes, frames, etc.) or frequency resources (e.g., subcarriers, carriers, etc.)). Furthermore, the wireless communication system may support wireless communication across various wireless access technologies, including third-generation (3G), fourth-generation (4G), fifth-generation (5G), and other suitable wireless access technologies above 5G (e.g., sixth-generation (6G)). Summary of the Invention

[0005] The article “a” preceding an element is not limited and should be understood to refer to “at least one” or “one or more” of those elements. The terms “a,” “at least one,” “one or more,” and “at least one of one or more” are interchangeable. As used herein, including as used in the claims, the word “or” used in a list of items (e.g., a list of items preceding phrases such as “at least one” or “one or more” or “one or two”) indicates a list of inclusion, such that a list of at least one of, for example, A, B, or C means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, as used herein, the phrase “based on” should not be construed as referring to a closed set of conditions. For example, without departing from the scope of this disclosure, an example step described as “based on condition A” may be based on both condition A and condition B. In other words, as used herein, the phrase “based on” should also be construed as “at least partially based on.” Moreover, as used herein, including as used in the claims, a “set” may contain one or more elements.

[0006] A first device for wireless communication is described. The first device may be configured, capable, or operable to perform one or more operations described herein. For example, the first device may be configured, capable, or operable to: obtain a set of data samples as input to a set of learning models stored at the first device; generate a corresponding confidence score associated with a set of outputs from the set of learning models based on providing the set of data samples as input to the set of learning models; and select a learning model for execution at the first device or a second device based on the confidence score associated with the output of a learning model in the set of learning models satisfying a selection criterion.

[0007] A processor (e.g., a standalone processor chipset or a component of a first device) is described. The processor is configured, capable, or operable to perform one or more operations described herein. For example, the processor is configured, capable, or operable to: obtain a set of data samples as input to a set of learning models stored at the first device; generate a corresponding confidence score associated with a set of outputs from the set of learning models based on providing the set of data samples as input to the set of learning models; and select a learning model for execution at the first device or a second device based on the confidence score associated with the output of a learning model in the set of learning models satisfying a selection criterion.

[0008] A method for wireless communication, performed by or potentially performed by a first device, is described. The method may include: obtaining a set of data samples to provide as input to a set of learning models stored at the first device; generating a corresponding confidence score associated with a set of outputs from the set of learning models based on providing the set of data samples as input to the set of learning models; and selecting a learning model for execution at the first device or a second device based on the confidence score associated with the output of a learning model in the set of learning models satisfying a selection criterion.

[0009] In some embodiments of the first device, processor, and method described herein, the first device, processor, and method may be further configured, capable, or operable to transmit a message indicating the selected learning model to a second device. In some embodiments of the first device, processor, and method described herein, in order to obtain a set of data samples, the first device, processor, and method may be further configured, capable, or operable to receive a message indicating the set of data samples from the second device. In some embodiments of the first device, processor, and method described herein, in order to obtain the set of data samples, the first device, processor, and method may be further configured, capable, or operable to receive a set of reference signals from the second device and obtain information associated with the set of reference signals, said information including one or more of a reference signal received power (RSRP) measurement, a reference signal received quality (RSRQ) measurement, a channel quality indicator (CQI), a precoding matrix indicator (PMI), a modulation and coding scheme (MCS) index, or a rank indicator (RI). In some embodiments of the first device, processor, and method described herein, the first device, processor, and method may be further configured, capable, or operable to receive a message triggering the selection of a learning model from the second device. In some embodiments of the first device, processor, and method described herein, in order to generate a corresponding confidence score, the first device, processor, and method may be further configured, capable of, or operable to generate a corresponding uncertainty measure associated with the output set from the learning model set based on a set of data samples, wherein the selection criterion includes the minimum uncertainty measure among the corresponding uncertainty measures. In some embodiments of the first device, processor, and method described herein, the corresponding uncertainty measure includes at least one of the marginal entropy of the output set based on the set of data samples or the conditional entropy of the output set.

[0010] In some embodiments of the first device, processor, and method described herein, in order to generate a corresponding confidence score, the first device, processor, and method may be further configured, capable of, or operable to generate a corresponding first uncertainty measure associated with the output set from the learning model set and a corresponding second uncertainty measure associated with the output set from the learning model set, based on a set of data samples, wherein the selection criterion includes the maximum difference between the corresponding first uncertainty measure and the corresponding second uncertainty measure. In some embodiments of the first device, processor, and method described herein, in order to generate a corresponding confidence score, the first device, processor, and method may be further configured, capable of, or operable to generate a corresponding first weighted uncertainty measure associated with the output set from the learning model set and a corresponding second weighted uncertainty measure associated with the output set from the learning model set, based on a set of data samples, wherein the corresponding first weighted uncertainty measure and the corresponding second weighted uncertainty measure are weighted based on values ​​between zero and one. In some embodiments of the first device, processor, and method described herein, in order to generate a corresponding confidence score, the first device, processor, and method may be further configured, capable of, or operable to generate an information gain associated with the output set from the learning model set based on a set of data samples.

[0011] In some embodiments of the first device, processor, and method described herein, in order to execute the learning model, the first device, processor, and method may be further configured, capable, or operable to: generate a prediction of at least one of a message set, symbol set, or bit set as an output from the learning model based on providing a set of data samples as input to the selected learning model, wherein the set of data samples includes one or more channel output signals; and generate a prediction of at least one beam for transmitting or receiving signaling as an output from the learning model based on providing a set of data samples as input to the selected learning model, wherein the set of data samples includes at least one beam measurement set or auxiliary information associated with one or more beams. One less; a prediction of at least one beam for transmitting or receiving signaling in a future time slot set, generated as the output of a selected learning model based on providing a set of data samples as input, wherein the set of data samples contains at least one of a set of beam measurements or auxiliary information associated with one or more beams; or a prediction of channel state information (CSI) generated as the output of a selected learning model based on providing a set of data samples as input, wherein the set of data samples contains at least one of a set of reference signal measurements or a set of channel vectors, and wherein the CSI contains at least one of a CQI, PMI, MCS index, or RI associated with a set of time-frequency resources. In some embodiments of the first device, processor, and method described herein, the set of channel vectors is associated with a first subset of time-frequency resources in the set of time-frequency resources, and wherein the output from the learning model is associated with a second subset of time-frequency resources or the set of time-frequency resources. In some embodiments of the first device, processor, and method described herein, the first device is an NE. In some embodiments of the first device, processor, and method described herein, the first device is a UE. Attached Figure Description

[0012] Figure 1 and 2 Examples of wireless communication systems according to various aspects of this disclosure are shown.

[0013] Figure 3 Examples of signaling diagrams according to various aspects of this disclosure are shown.

[0014] Figure 4 Examples of UEs according to various aspects of this disclosure are shown.

[0015] Figure 5 Examples of processors according to various aspects of this disclosure are shown.

[0016] Figure 6 Examples of NEs according to various aspects of this disclosure are shown.

[0017] Figure 7A flowchart illustrating a method performed by a device (e.g., UE and / or NE) according to various aspects of this disclosure is shown. Detailed Implementation

[0018] Wireless communication systems may include one or more devices, such as UEs and NEs, as well as other devices for transmitting and receiving signaling. The devices may implement one or more learning models, also referred to as machine learning (ML) models and / or artificial intelligence (AI) models, to perform signaling-related classification tasks. The learning models can identify patterns in input data that indicate different categories or labels. Example signaling-related classification tasks include, but are not limited to: predicting the type of message from a received signal; predicting one or more beam indices with the maximum signal strength and / or quality from one or more beam measurements; and predicting signaling-related information (e.g., CQI, PMI, RI, or MCS indices), as well as other classification tasks. In some examples, the NE or another device trains the learning model by determining one or more parameters, which may include weights, biases, and / or other parameters defining the respective layers of the learning model. During training of the learning model, parameters are selected such that the learning model maps (e.g., correlates) input values ​​to output values ​​(e.g., categories or labels in a discrete number of categories and labels). The learning model is executed (e.g., implemented) by one or more devices using the determined or selected parameters.

[0019] In some variations, the device may input data with statistical properties different from the dataset on which the learning model is trained. In some cases, the NE may train multiple different learning models for the corresponding domain on which the learning model is based for inference or prediction, where the domain represents data for network context, network configuration, one or more network parameters, physical propagation medium, and / or device behavior. For example, at the time of data collection or during the simulation of generating the data in the dataset, the domain may include datasets for network configuration (e.g., defined beamcodebooks, sets of defined multiple-input multiple-output (MIMO) configuration parameters, types of scheduled devices, different types of link adaptation and power control), datasets for device behavior (e.g., device mobility, device orientation, quality of service (QoS) supporting communication at the device), datasets for different traffic patterns, datasets for physical conditions of the propagation medium (e.g., channels with rich multipath versus sparse channels, indoor versus outdoor), and other examples. The NE may store different learning models for transmission to one or more other devices (e.g., UEs with reduced memory storage capacity when compared with the NE). Alternatively, one or more other devices may store different learning models.

[0020] Devices (e.g., other devices and / or NEs) may use appropriate learning models for different domains. However, conventional techniques for allowing devices to select stored learning models to use can result in additional signaling overhead and / or processing. For example, a device may exchange signaling containing labeled data samples of a data domain experienced by the device as input to a stored learning model. The device may compare the output from the learning model with the expected output (e.g., the label of the data sample) and may select the learning model with the closest output. The device exchanging signaling containing labeled data samples results in increased signaling overhead. Alternatively, the device may obtain data samples of the data domain experienced by the device (e.g., by measuring and / or receiving data samples from the device), and the device may determine the statistical properties of the data samples. The device may select a learning model trained using training data with similar statistical properties to the data samples. However, processing the data samples to determine the statistical properties and / or obtaining the statistical properties of the training data can result in increased processing, latency, memory footprint, and signaling overhead.

[0021] As described herein, to reduce the signaling overhead and processing associated with selecting learning models for different domains, a device can autonomously select a learning model. This includes selecting a learning model without receiving signaling from or transmitting signaling to another device. For example, a device can use data samples obtained as input to a learning model to generate confidence scores for different learning models. The device can generate confidence scores that vary with the corresponding conditional entropy and marginal entropy of different learning models for a given data sample. Entropy refers to a measure of uncertainty or disorder in a set of data samples. The conditional entropy of a learning model is an indication of the confidence level of an individual prediction made by the learning model, where a lower conditional entropy reflects lower uncertainty in the prediction and correspondingly a higher confidence score for the learning model. The marginal entropy of a learning model is an indication of the confidence level that the marginal distribution of one or more outputs of the learning model (e.g., predicted labels of input data samples) is uniform, where the marginal distribution refers to the distribution of a single variable or the learning model's prediction of a single variable without considering the values ​​of other variables or features.

[0022] In some variations, the confidence score of the learning model is the difference between the conditional entropy and the marginal entropy of the learning model, and / or the difference between the weighted conditional entropy and the weighted marginal entropy of the learning model. The device compares the confidence scores with selection criteria and selects the learning model that satisfies the criteria. Example selection criteria include, but are not limited to: the confidence score of the learning model being the maximum (e.g., highest) confidence score in the set of learning models stored at the device; the difference between the conditional entropy and the marginal entropy of the learning model being the maximum difference in the set of learning models; or the conditional entropy of the learning model being the lowest (e.g., minimum) value in the set of learning models, and other selection criteria. In some examples, the device transmits an indication of the selected learning model to another device (e.g., to a device that implements or executes the learning model). In some other examples, the device executes the selected learning model to obtain one or more predictions for a sample of input data.

[0023] Although the learning model is described as being trained and implemented for wireless communication, it can also be trained and implemented for any task using classification. Example tasks using classification include, but are not limited to, wireless communication, image recognition, natural language processing, and biomedical imaging, as well as other tasks. The learning model can be switched or transferred from one device to another for implementation to perform the task.

[0024] This document refers to the receipt, transmission, or delivery of data or information, such as signaling communication resources and / or communications transmitted or received between devices. It should be understood that other terms are used interchangeably with "communicating," such as signaling, transmitting, receiving, outputting, forwarding, retrieving, obtaining, etc. Similarly, other terms are used interchangeably with "transmitting" (e.g., transmitting, signaling, outputting, forwarding, etc.), and other terms are used interchangeably with "receiving" (e.g., transmitting, retrieving, obtaining, etc.).

[0025] Various aspects of this disclosure are described in the context of wireless communication systems.

[0026] Figure 1Examples of wireless communication systems 100 according to various aspects of this disclosure are shown. Wireless communication system 100 may include one or more NEs 102, one or more UEs 104, and a core network (CN) 106. Wireless communication system 100 may support various wireless access technologies. In some embodiments, wireless communication system 100 may be a 4G network, such as an LTE network or an LTE-A network. In some other embodiments, wireless communication system 100 may be an NR network, such as a 5G network, a 5G-A network, or a 5G Ultra Wideband (5G-UWB) network. In other embodiments, wireless communication system 100 may be a combination of 4G and 5G networks, or other suitable wireless access technologies, including IEEE 802.11 (Wi-Fi), IEEE 802.16 (WiMAX), and IEEE 802.20. Wireless communication system 100 may support wireless access technologies beyond 5G, such as 6G. In addition, the wireless communication system 100 can support technologies such as Time Division Multiple Access (TDMA), Frequency Division Multiple Access (FDMA), or Code Division Multiple Access (CDMA).

[0027] One or more NEs 102 may be distributed throughout a geographic area to form a wireless communication system 100. One or more of the NEs 102 described herein may be, include, or be referred to as a network node, base station, access point (AP), network element, network function, network entity, network infrastructure (or infrastructure), radio access network (RAN), NodeB, eNodeB (eNB), next-generation NodeB (gNB), or other suitable terms. NEs 102 and UEs 104 may communicate via a communication link, which may be a wireless or wired connection. For example, NEs 102 and UEs 104 may perform wireless communication (e.g., receiving signaling, transmitting signaling) via a Uu interface.

[0028] NE 102 can provide a geographic coverage area, wherein NE 102 can support services for one or more UE 104s within the geographic coverage area. For example, NE 102 and UE 104 can support signal wireless communication associated with services (e.g., voice, video, packet data, messaging, broadcasting, etc.) based on one or more radio access technologies. In some embodiments, NE 102 can be mobile, for example, a satellite associated with a non-terrestrial network (NTN). In some embodiments, different geographic coverage areas associated with the same or different radio access technologies may overlap, but different geographic coverage areas may be associated with different NE 102s.

[0029] One or more UEs 104 may be distributed throughout the geographic area of ​​the wireless communication system 100. UE 104 may include or be referred to as a remote unit, mobile device, wireless device, remote device, subscriber device, transmitter device, receiver device, or some other suitable term. In some embodiments, UE 104 may be referred to as a unit, station, terminal, or client, and other examples. Additionally or alternatively, UE 104 may be referred to as an Internet of Things (IoT) device, an Internet of Everything (IoE) device, or a Machine Type Communication (MTC) device, and other examples.

[0030] UE 104 may be able to support direct wireless communication with other UE 104s via a communication link. For example, UE 104 may support direct wireless communication with another UE 104 via a device-to-device (D2D) communication link. In some implementations, such as vehicle-to-vehicle (V2V) deployments, vehicle-to-everything (V2X) deployments, or cellular V2X deployments, the communication link may be referred to as a side link. For example, UE 104 may support direct wireless communication with another UE 104 via a PC5 interface.

[0031] NE 102 may support communication with CN 106 or with another NE 102, or both. For example, NE 102 may interface with other NE 102 or CN 106 via one or more backhaul links (e.g., S1, N2, N6, or other network interfaces). In some embodiments, NE 102 may communicate directly with each other. In some other embodiments, NE 102 may communicate indirectly with each other (e.g., via CN 106). In some embodiments, one or more NE 102 may include sub-components, such as access network entities, which may be examples of access node controllers (ANCs). The ANC may communicate with one or more UE 104s via one or more other access network transport entities, which may be referred to as radio heads, smart radio heads, or transmit-receive points (TRPs).

[0032] CN 106 can support user authentication, access authorization, tracking, connectivity, and other access, routing, or mobility functions. CN106 can be an evolved packet core (EPC) or a 5G core (5GC), which may include control plane entities that manage access and mobility (e.g., Mobility Management Entity (MME), Access and Mobility Management Function (AMF)), and user plane entities that route packets or interconnect to external networks (e.g., Serving Gateway (S-GW), Packet Data Network (PDN) Gateway (P-GW), or User Plane Function (UPF)). In some implementations, the control plane entities may manage non-access stratum (NAS) functions of one or more UEs 104 served by one or more NEs 102 associated with CN 106, such as mobility, authentication, and bearer management (data bearers, signaling bearers, etc.).

[0033] CN 106 can communicate with the packet data network via one or more backhaul links (e.g., via S1, N2, N6, or other network interfaces). The packet data network may contain an application server. In some implementations, one or more UEs 104 can communicate with the application server. UE 104 can establish a session (e.g., a Protocol Data Unit (PDU) session, etc.) with CN 106 via NE 102. CN 106 can use the established session (e.g., an established PDU session) to route services (e.g., control information, data, etc.) between UE 104 and the application server. A PDU session can be an example of a logical connection between UE 104 and CN 106 (e.g., one or more network functions of CN 106).

[0034] In the wireless communication system 100, NE 102 and UE 104 can use the resources of the wireless communication system 100 (e.g., time resources (e.g., symbols, time slots, subframes, frames, etc.) or frequency resources (e.g., subcarriers, carriers)) to perform various operations (e.g., wireless communication). In some embodiments, NE 102 and UE 104 may support different resource structures. For example, NE 102 and UE 104 may support different frame structures. In some embodiments, such as in 4G, NE 102 and UE 104 may support a single frame structure. In some other embodiments, such as in 5G and other suitable radio access technologies, NE 102 and UE 104 may support various frame structures (i.e., multiple frame structures). NE 102 and UE 104 may support various frame structures based on one or more basic parameters (numerology).

[0035] One or more basic parameters can be supported in the wireless communication system 100, and the basic parameters may include subcarrier spacing and cyclic prefix. The first basic parameter (e.g., This can be associated with the first subcarrier spacing (e.g., 15 kHz) and the normal cyclic prefix. In some implementations, the first fundamental parameter (e.g., ...) associated with the first subcarrier spacing (e.g., 15 kHz) is... One time slot can be used per subframe. The second basic parameter (e.g., This can be associated with the second subcarrier spacing (e.g., 30 kHz) and the normal cyclic prefix. The third fundamental parameter (e.g., This can be associated with the third subcarrier spacing (e.g., 60 kHz) and the normal or extended cyclic prefix. The fourth fundamental parameter (e.g., This can be associated with the fourth subcarrier spacing (e.g., 120 kHz) and the normal cyclic prefix. The fifth fundamental parameter (e.g., This can be associated with the fifth subcarrier spacing (e.g., 240 kHz) and the normal cyclic prefix.

[0036] Time intervals for resources (e.g., communication resources) can be organized according to frames (also known as radio frames). Each frame may have a duration, such as 10 milliseconds (ms). In some embodiments, each frame may contain multiple subframes. For example, each frame may contain 10 subframes, and each subframe may have a duration, such as 1 millisecond. In some embodiments, each frame may have the same duration. In some embodiments, each subframe of a frame may have the same duration.

[0037] Alternatively, the time intervals of resources (e.g., communication resources) can be organized according to time slots. For example, a subframe may contain a certain number (e.g., quantity) of time slots. The number of time slots in each subframe may also depend on one or more fundamental parameters supported in the wireless communication system 100. For example, the first, second, third, fourth, and fifth fundamental parameters (i.e., ...) associated with the corresponding subcarrier spacings of 15 kHz, 30 kHz, 60 kHz, 120 kHz, and 240 kHz... , , , , The subframe can utilize one time slot per subframe, two time slots per subframe, four time slots per subframe, eight time slots per subframe, and 16 time slots per subframe, respectively. Each time slot can contain a certain number (e.g., quantity) of symbols (e.g., OFDM symbols). In some implementations, the number of time slots in a subframe (e.g., quantity) can depend on the underlying parameters. For a normal cyclic prefix, a time slot can contain 14 symbols. For an extended cyclic prefix (e.g., applicable to a 60 kHz subcarrier spacing), a time slot can contain 12 symbols. The relationship between the number of symbols per time slot for normal and extended cyclic prefixes, the number of time slots per subframe, and the number of time slots per frame can depend on the underlying parameters. It should be understood that the first underlying parameter (e.g., quantity) associated with the first subcarrier spacing (e.g., 15 kHz) is determined by the underlying parameters. The references to ) can be used interchangeably between subframes and time slots.

[0038] In wireless communication system 100, the electromagnetic (EM) spectrum can be divided into various categories, bands, channels, etc., based on frequency or wavelength. For example, wireless communication system 100 may support one or more operating bands, such as frequency ranges specified as FR1 (410 MHz to 7.125 GHz), FR2 (24.25 GHz to 52.6 GHz), FR3 (7.125 GHz to 24.25 GHz), FR4 (52.6 GHz to 114.25 GHz), FR4a or FR4-1 (52.6 GHz to 71 GHz), and FR5 (114.25 GHz to 300 GHz). In some embodiments, NE 102 and UE 104 may perform wireless communication on one or more of the operating bands. In some embodiments, FR1 may be used by NE 102 and UE 104, as well as other equipment or devices for cellular communication services (e.g., control information, data). In some implementations, FR2 may be used by NE 102 and UE 104, as well as other equipment or devices for short-range high data rate capability.

[0039] FR1 can be associated with one or more basic parameters (e.g., at least three basic parameters). For example, FR1 can be associated with the following: the first basic parameter (e.g., ), which includes a 15 kHz subcarrier spacing; a second fundamental parameter (e.g., ), which includes a 30 kHz subcarrier spacing; and a third fundamental parameter (e.g., ), which includes a 60 kHz subcarrier spacing. FR2 can be associated with one or more fundamental parameters (e.g., at least two fundamental parameters). For example, FR2 can be associated with the following: a third fundamental parameter (e.g., ), which includes a 60 kHz subcarrier spacing; and a fourth fundamental parameter (e.g., It contains a 120 kHz subcarrier spacing.

[0040] In some examples, the device (e.g., NE 102 and / or UE 104) may implement one or more learning models for classification tasks. Example classification tasks include, but are not limited to: predicting the type of a message from a received signal; predicting one or more beam indices with the maximum signal strength and / or quality from one or more beam measurements; and predicting signaling-related information (e.g., CQI, PMI, RI, or MCS indices), as well as other classification tasks. The classifier learning model and / or the learning model implemented (e.g., performed) for the classification task can identify patterns in the input data that indicate at least one category or label from a finite set of categories or labels.

[0041] In some cases, devices (e.g., NE 102 and / or UE) are equipped with multiple antennas, for example, in the form of one or more antenna array panels with corresponding antenna elements. In some variations, devices may use one or more beams for communication, which are formed using multiple antenna elements available at the device. The beams are inherently directional (e.g., providing directional gain) and may have narrow beamwidths (e.g., beamwidths less than a threshold). To achieve a reliable communication link between devices, one or more beams in the device may be selected for use. That is, the device may select a beam pair comprising a receive beam for the receiving device to receive signaling and a transmit beam for the transmitting device to transmit signaling, such that the beam pair produces a signal strength at the receiving device that meets a threshold.

[0042] In some examples, the process of selecting beam pairs may be referred to as a beam search procedure or a beam selection procedure. For signaling from UE 104 to NE 102 (e.g., downlink signaling), beam selection involves selecting a transmit beam at NE 102 and a corresponding receive beam at UE 104. For signaling from NE 102 to UE 104 (e.g., uplink signaling), beam selection involves selecting a transmit beam at UE 104 and a corresponding receive beam at NE 102. The beam corresponds to an NZP CSI-RS resource in at least one non-zero power (NZP) CSI reference signal (CSI-RS) resource set, wherein the NZP CSI-RS resource set is configured with values ​​of higher-layer parameters set to repeat.

[0043] Conventionally, for beam selection, devices (e.g., UE 104 and / or NE 102) may search within a set of possible beams and select the beam with the highest signal strength; this is known as an exhaustive search. In some examples, such as for downlink signaling, NE 102 may use a set of possible transmit beams to transmit one or more reference signals. UE 104 may perform one or more measurements to obtain the received signal strength (e.g., RSRP, RSRQ, and / or Signal-to-Interference-plus-Noise Ratio (SINR)) of the beams in the set of possible transmit beams. UE 104 may select the transmit beam that provides the highest RSRP and / or SINR. UE 104 may transmit a message containing an indication of the beam index of the selected transmit beam to NE 102. Alternatively, UE 104 may select a receive beam by fixing the transmit beam at NE 102 and measuring the received signal strength of a set of possible receive beams (e.g., sweeping the receive beam at UE 104). Therefore, exhaustive search involves selecting a beam (e.g., the best beam, the beam with the highest signal strength), but it increases the latency and signaling overhead associated with transmitting and / or receiving signaling on the set of possible beams used to perform the measurement.

[0044] In some examples, the device (e.g., UE 104 and / or NE 102) may implement one or more learning models for beam prediction. For example, the device may train a learning model to output predictions of transmit and / or receive beams for reliable transmission or reception of signaling. The device may input one or more beam measurements into the learning model to obtain predictions of transmit and / or receive beams. Example beam measurements include, but are not limited to, beamforming gain measurements, beamwidth measurements, RSRP measurements, and SINR measurements, as well as other measurements. Beam prediction may include spatial beam prediction, where the device implements a learning model to determine beams in the available or candidate beam set in the wireless network using a relatively small number (e.g., less than a threshold number) of beam measurements. For example, the number of beam measurements may be less than the number of available or candidate beams. Alternatively, beam prediction may include temporal beam prediction, in which the device implements a learning model to determine the beam to be used for transmission during one or more subsequent time slots using a relatively small number (e.g., less than a threshold number) of historical beam measurements or identifiers of beams historically selected for transmission and / or reception of signaling.

[0045] In some variations, the device implements a learning model to predict information (e.g., CQI, PMI, RI, or MCS index). For example, the device implements a learning model to estimate the CSI across a space-time grid based on pilot and / or reference signals transmitted on a relatively small number (e.g., less than a threshold number) of resource elements in the space-time grid. The device implements a learning model to perform CSI prediction, such that the learning model outputs predictions of the CSI for one or more future time slots (e.g., CQI, PMI, RI, or MCS index) based on the estimated CSI and / or additional information from one or more previous time slots. Alternatively or concurrently, UE 104 and / or NE102 implement a learning model to act as a receiver. For example, the learning model outputs predictions of transmitted messages, symbols, or bits based on one or more signals received as output from a radio channel. In some examples, the device may implement one or more learning models for a positioning procedure. For example, the device implements a learning model to determine the location of a node based on the channel impulse response (CIR) experienced by the node (e.g., UE 104, NE 102, or any other device in the wireless communication system 100). The CIR defines how a signal propagates through the channel over time.

[0046] Devices can train learning models using training datasets obtained through simulation or measurement (e.g., considering one or at most a finite set of physical cell sites, network configurations, and / or channel characteristics). A dataset is called a labeled dataset when it contains both input data samples and corresponding output data samples of the data samples in the dataset. A dataset is called an unlabeled dataset when it contains input data samples but no corresponding output data samples. In some examples, devices may implement supervised learning techniques to train the learning model. For example, a device may train a learning model to map input data to output labels based on example input-output pairs provided during training. In supervised learning, the learning model computes a mapping function from input features to output labels by observing a dataset containing labeled examples, allowing the learning model to generalize the mapping to accurately predict new, unseen data. Although supervised learning techniques have been described, devices may additionally or alternatively implement any other type of training technique to train the learning model, including but not limited to unsupervised and semi-supervised learning techniques, as well as other techniques. In unsupervised learning, the learning model detects patterns, structures, or relationships within the training data. Unlike supervised learning, no explicit output labels are provided during training. Semi-supervised learning utilizes both labeled and unlabeled data during the training process. The learning model uses labeled examples while also employing the structure of unlabeled data to improve the model's performance.

[0047] In some examples, and Indicates that it comes from the input. The learning model takes input samples and corresponding labels or desired output data samples. In some variations, Called input Predictions and / or inferences. In some cases, It is a scalar or a one-dimensional or multi-dimensional vector, and It can be a scalar or a one-dimensional or multi-dimensional vector. and Let these represent the input sample space and the output sample space (e.g., the label space), respectively. If... Take a finite number of values ​​(e.g., when) (set When the cardinality (of the learning model) is finite, then the learning model is called a classifier learning model. If Take continuous values ​​(for example, when and If the learning model is such that the model is called a regression model, then the learning model is called a regression model.

[0048] Learning model defines mappings or functions ,in Here, This represents the set of learning model parameters learned during the training process. Example learning model parameters include, but are not limited to, weights, biases, and activation function parameters, as well as other parameters. Weight parameters represent the connection strength between neurons in different layers of the learning model. Bias are additional parameters added to neurons in the learning model, enabling the model to capture offsets or biases in the input data. Activation function parameters may include slope parameters or parameters defining the shape of the activation function in a parameterized activation function. Determining one or more values ​​of model parameters for a defined use case is referred to as training or learning the model. A general procedure for developing learning models involves updating one or more parameters based on a training dataset containing labeled or unlabeled samples to minimize a cost function, thereby causing supervised learning or training, or unsupervised learning or training, respectively. For classifier learning models, the labels of the learning model (e.g., desired output, prediction, and / or inference) are discrete and belong to a finite cardinality set (e.g., ...). and / or ).

[0049] In some examples, the classifier learning model can be probabilistic, where the learning model produces a probability distribution about one or more predictions. The probability distribution can be conditioned on the input data (e.g., data samples provided as input to the learning model) and determined by... Parameterization to obtain Differentiable conditional distributions. Therefore, training a classification learning model (e.g., training based on one or more source domains) produces a probability distribution. Due to the parameters of the learning model The probability distribution of predictions generated by the learned model is fixed at the end of training, therefore it can be expressed as: (For example, none) And it is conditional on the input data. Because Let be discrete variables, where ,therefore For when Provided as input to the learning model hour for The probability of the predicted label (e.g., using learning model parameters fixed at the end of training the learning model). In other words, To be based on the learning model The predictions and / or inferences made Given input The probability of the label. In some variations, or ,in .

[0050] In some examples, a device may implement or deploy a trained learning model to make predictions or inferences. By providing the desired inferences or predictions seen during the training and testing phases of the learning model prior to deployment, the learning model can be expected to perform with the same level of accuracy and / or precision. However, in real-world wireless networks (e.g., in wireless communication system 100), the learning model may use input data to make predictions or inferences that have different statistical properties than the dataset on which the learning model was trained. A learning model using input data with different statistical properties than the dataset on which the learning model was trained may output incorrect (e.g., inaccurate) inferences or predictions. To improve the accuracy of inferences and / or predictions, a device may train a learning model with generalization capabilities to make inferences and / or predictions for a variety of different data distributions (e.g., data distributions different from the training data distributions encountered during inference). However, developing a learning model that generalizes to possible domains and outputs the desired performance across said domains can be difficult for the device, especially in the context of wireless networks where the statistical properties of the data distributions are constantly changing.

[0051] In some examples, there exists Given a variety of data distributions, the learning model provides inferences and / or predictions for said data distributions that meet threshold performance requirements (e.g., precision and / or accuracy). Based on statistical distributions... Distributed data sample set Also known as a data domain or domain. Therefore, the term domain means data with a defined distribution. For example, It can be called the first Data field or the first A source domain is the domain on which the learning model is trained, and a target domain is the domain on which the learning model is executed to provide inferences and / or predictions. In some examples, different target domains may exist for different network conditions, device parameters, channel characteristics, or any other factors affecting the data distribution of one or more data samples. In some cases, the device may target different target domains. Data distribution (e.g., The device can train corresponding learning models for each target domain. It can also train or develop models for specific tasks, such as beam prediction, CSI prediction and / or CSI-based localization, and other classification tasks. A variety of learning models. The device can store the learning model and / or can transfer instructions of the learning model to another device for storage. At any time, the device and / or the other device can select and execute... One of the models is used for inference and / or prediction.

[0052] However, conventional techniques for enabling devices to select which stored learning models to use can result in additional signaling overhead and / or processing. Devices may exchange signaling containing labeled data samples of the data domains the device has experienced, providing them as input to the stored learning models. For example, a device hosting multiple learning models may obtain labeled data samples. ,in It is the input data sample and It corresponds to the label. The device can... The learning model is provided as input to the corresponding learning model stored at the device. The device can use the output of the learning model to select an output. The device can compare the output from the learning model with the expected output (e.g., the label of the data sample) and can select the learning model with the closest output. The device exchanging signaling containing the labeled data sample results in increased signaling overhead. Alternatively, the device can obtain data samples of the data domain experienced by the device (e.g., by measuring and / or receiving data samples from the device), and the device can determine the distribution and / or statistical properties of the data samples. The device can select a learning model trained using training data with similar distribution and / or statistical properties to the data samples. However, processing the data samples to determine the statistical properties and / or obtaining the statistical properties of the training data may result in increased processing, latency, memory footprint, and signaling overhead.

[0053] Therefore, when the device is from When selecting a learning model from among several learning models that meets the threshold performance requirements for the current data distribution and / or domain, conventional techniques may lead to inefficiencies (increased signaling overhead, increased processing, etc.). For example, the device may obtain data from available (e.g., stored) models. The learning model selected from among the learning models produces the highest inference and / or prediction performance (including but not limited to accuracy and precision) for the input data samples obtained by the device. Because the data distribution in the wireless communication system 100 is dynamic (e.g., changes relatively frequently) due to the time-varying nature of the physical propagation medium and other network parameters and / or settings that change over different time intervals, the task of learning model selection is not a one-off task. That is, the device can perform learning model selection when network conditions and / or contexts, including channel characteristics, change and / or when the performance of the previously selected learning model degrades beyond a satisfactory level. Example channel characteristics may include, but are not limited to, attenuation, propagation delay, multipath fading, noise and interference, bandwidth, channel capacity, and Doppler shift, as well as other characteristics. Example network conditions include, but are not limited to, device mobility, environmental factors of the channel (e.g., weather conditions and obstruction by objects, as well as other environmental factors), and / or changes in antenna configurations at one or more devices.

[0054] According to various embodiments, one or more of NE 102 and UE 104 are operable to implement various aspects of the techniques described with reference to this disclosure. In some examples, the device may implement techniques for autonomous model selection. The device may select a learning model for a target domain from a set of learning models stored at the device without receiving signaling and / or transmitting signaling to another device (e.g., signaling containing labeled data samples, distribution and / or other statistical information and / or reference signals, and other information about the target domain). The learning model stored at the device may be used as a classifier learning model (e.g., Develop and / or train a classifier learning model. The classifier learning model classifies the corresponding input data samples into classes. One of several categories, where M is a positive integer greater than one. Example classifier learning models include, but are not limited to, a learning model implemented as a receiver, a learning model for beam prediction (e.g., spatial beam prediction and / or temporal beam prediction), and a learning model for predicting information about signaling (e.g., CQI, PMI, RI, and / or MCS indexes, and other information), provided that the output of the learning model (e.g., CQI, PMI, RI, and / or MCS indexes) is discrete and has a finite number of possible values. The device may select a learning model at the receiver to infer and / or predict transmitted messages, symbols, and / or bits from the signal at the channel output. Alternatively or additionally, the device may select a learning model for predicting information about signaling (e.g., for a target domain). Alternatively or additionally, the device may select a learning model (e.g., for a target domain) to predict one or more beams for signaling.

[0055] Figure 2Examples of wireless communication system 200 according to various aspects of this disclosure are shown. In some examples, wireless communication system 200 implements various aspects of wireless communication system 100. For example, wireless communication system 200 includes device 202-a and device 202-b, which may be referenced... Figure 1 Examples of one or more UEs 104 and / or one or more NEs 102 are described. In some examples, devices (e.g., devices 202-a and 202-b) may exchange signaling with each other. For example, device 202-a transmits signaling to device 202-b via communication link 204-a, which may be a reference link. Figure 1 An example of the described communication link. In some other examples, device 202-b transmits signaling to device 202-a via communication link 204-b, which may be a reference link. Figure 1 An example of the described communication link. Signaling between device 202-a and device 202-b may include control signaling and / or data transmission.

[0056] In some variations, device 202-b is an example of an NE, while device 202-a is an example of a UE. In some other variations, device 202-b is an example of a UE, while device 202-a is an example of an NE. The example UE may include, but is not limited to, cellular devices, laptop computers, desktop computers, wearable devices, Internet of Things (IoT) devices, and / or other equipment capable of transmitting and / or receiving communications.

[0057] Devices in the wireless communication system 200 (e.g., device 202-a and / or device 202-b) may implement one or more learning models (e.g., one or more of learning models 206-a, 206-b, etc., up to learning model 206-T). In some examples, learning models 206-a to 206-T may be examples of ML models and / or AI models. For example, learning models 206-a to 206-T may be examples of neural networks with multiple layers (e.g., deep neural networks (DNNs)). A neural network is a computational model comprising multiple layers of artificial neurons organized into an input layer 208, one or more feature extraction layers 210, one or more classification layers 212, and an output layer 214, said artificial neurons may be referred to as nodes or units. The input layer 208 passes input data to subsequent layers and may not contain learnable parameters. The output layer 214 represents the predictions generated by the neural network and may not contain learnable parameters. Feature extraction layer 210 and / or classification layer 212 may contain one or more hidden layers. Nodes in the hidden layers receive input signals, perform mathematical operations on the input data, and produce output signals, which are then passed to other nodes in subsequent layers. Connections between nodes are represented by weighted edges, which determine the strength of the connections between nodes. During training, the weights are adjusted using input-output pairs from the training dataset with the goal of minimizing a defined loss or error function. Neural networks are capable of learning complex patterns and representations from data, enabling them to perform a wide range of tasks, including classification, regression, clustering, pattern recognition, and sequence generation. Example neural networks include, but are not limited to, autoencoders, convolutional neural networks (CNNs), recurrent neural networks (RNNs), generative adversarial networks (GANs), long short-term memory (LSTM) networks, or any other type of neural network.

[0058] In some examples, learning models 206-a to 206-T may include site-specific learning models. For example, one or more target domains may correspond to site-specific contexts, resulting in site-specific learning models and / or device-specific learning models for device-specific contexts. For site-specific models, different learning models are developed for different cell sites or geographic locations, leading to different data distributions due to their different physical characteristics. For example, cell sites in rural locations may have different channel characteristics than cell sites in dense urban environments, resulting in two different data distributions. In some examples, learning models may include device-specific learning models. Data distributions may vary due to device-specific parameters and / or contexts. For example, the wireless channels experienced by a device in an outdoor environment may have different statistical characteristics than those experienced by a device in an indoor environment. Device mobility and physical orientation can create different data distributions. Furthermore, different devices with different capabilities (e.g., device measurement sensitivity) may benefit from using multiple learning models.

[0059] The equipment can be developed or trained. Each learning model (e.g., learning models 206-a to 206-T) has threshold performance (e.g., threshold accuracy and precision) within at least one target domain of the training dataset. Devices 202-a and / or 202-b may implement the learning model for classification tasks, such as spatial beam prediction, temporal beam prediction, acting as a receiver, and / or predicting information for signaling (e.g., CQI, RI, PMI, and / or MCS indexes). The classification task may include any task of classifying or labeling input data into defined categories or classes based on features or attributes of the input data. The learning model for spatial beam prediction can use the current set of beam measurements to predict the single beam or multiple beams with the highest performance in the beam set. The learning model for temporal beam prediction can use current and past (e.g., historical) data to predict the single beam or multiple beams with the highest performance for a future set of time slots. Current and past data may include current and past beam measurements and / or current and past beam indices with the highest performance. The learning model at the receiver infers or predicts the transmitted messages, symbols, and / or bits based on the signal at the output of the wireless channel. If the output of the learning model is a discrete value (e.g., there is a finite number of values), then the learning model can predict information (e.g., CQI, RI, PMI, and / or MCS index).

[0060] In some variations, learning models 206-a to 206-T include corresponding feature components and corresponding classifier components. The feature components and classifier components may represent subnetworks of a neural network and may contain one or more layers. For example, a feature component may include an input layer 208 and one or more feature extraction layers 210, which may be examples of hidden layers. In some variations, one or more final layers of the learning model (e.g., output layer 214 and / or layers preceding output layer 214) may include at least one classification layer 212. In some cases, the classifier component may include a classification layer 212 and an output layer 214. In some other cases, the classification layer 212 may be part of the output layer 214, such that the classifier component may include an output layer 214 containing the classification layer 212. That is, the classification layer 212 may be a dedicated component within the output layer 214 that performs the final classification based on learned features from the feature components. Classification layer 212 maps one or more learned features from one or more previous layers (e.g., features extracted by feature extraction layer 210 of the feature component) to one or more output categories or labels of output layer 214. Output layer 214 may contain neurons representing different categories or classes, and the activation of these neurons indicates the likelihood or probability that the input data belongs to a certain category. In some variations, the output from the feature component is used as input data and is provided to the classifier component of the learning model. The output of the classifier component provides a classification (e.g., the likelihood or probability that the input data belongs to a certain category) of the input data to input layer 208 of the feature component.

[0061] Although learning models 206-a to 206-T are shown as containing a single classification layer 212, they can contain any number of classification layers 212. Similarly, although learning models 206-a to 206-T are shown as containing a classification layer 212 separate from the output layer 214, both the classification layer 212 and the output layer 214 can be single layers. Furthermore, learning models 206-a to 206-T can contain any number of feature extraction layers 210. The number of layers in learning models 206-a to 206-T can be different for each learning model, and / or the same for each learning model.

[0062] In some examples, device 202-a and / or device 202-b can train one or more of learning models 206-a to 206-T to obtain or generate parameters for the hidden layers of learning models 206-a to 206-T. Example parameters for learning models 206-a to 206-T include, but are not limited to, weights, biases, and activation function parameters, as well as other parameters. Weight parameters represent the connection strength between neurons in different layers of learning models 206-a to 206-T. Bias is an additional parameter added to neurons in learning models 206-a to 206-T, enabling them to capture offsets or biases in the input data. Activation function parameters may include slope parameters or parameters that define the shape of the activation function in a parameterized activation function.

[0063] In some variations, device 202-b can train learning models 206-a to 206-T, while device 202-a and / or device 202-b implements (e.g., executes) at least one of learning models 206-a to 206-T. For example, device 202-b can generate parameters defining learning models 206-a to 206-T and can transfer these parameters to device 202-a, wherein device 202-a implements at least one learning model 206-a to 206-T that includes said parameters. Alternatively or additionally, device 202-a can train learning models 206-a to 206-T, while device 202-a and / or device 202-b implements (e.g., executes) at least one of learning models 206-a to 206-T.

[0064] Devices 202-a and / or 202-b may develop (e.g., train) multiple learning models (e.g., learning models 206-a to 206-T) for a defined task, wherein the respective learning models have different parameters for different layers. Devices 202-a and / or 202-b may develop learning models for different target domains for the defined task. Different target domains may represent specific contexts, configurations, network parameters, physical propagation media, or device behavior. That is, the target domain represents data on contexts, configurations, network parameters, physical propagation media, and / or device behavior. Devices 202-a and / or 202-b may store the parameters of learning models 206-a to 206-T for transmission to one or more other devices (e.g., a UE or other devices). Alternatively or additionally, one or more other devices may receive and store the parameters of learning models 206-a to 206-T. Other devices may use the corresponding learning models for different target domains. For example, a device can provide data samples from a target domain as input to at least one learning model. The learning model can then provide the output used by the device for the task.

[0065] In some cases, and This represents the input and output data samples for learning models 206-a to 206-T. Since learning models 206-a to 206-T are classifier learning models, the outputs, predictions, and / or inferences... These are discrete values. When learning models 206-a to 206-T are developed for the same task (e.g., for estimating transmitted messages, symbols, and / or bits, for spatial beam prediction, for temporal beam prediction, or for CSI prediction), learning models 206-a to 206-T may have the same output space (e.g., (where M is a positive integer greater than one). In some variations, the first... Classification learning model (where) ) to input To generate a probability distribution for the prediction based on the conditions, denoted as: For the corresponding number Classification learning model (where) ),predict Let be discrete variables, where ,and According to the model , for The probability of predicting the label. That is, To learn based on the classification model The predictions and / or inferences made Given a sample of input data The probability of the label, where To learn the model parameters. In some examples, or ,in .

[0066] In some examples, at 216, device 202-b (e.g., and / or device 202-a) may select one of learning models 206-a to 206-T based on the confidence score of the learning model satisfying the selection criteria. The selected learning model may produce the maximum accuracy and / or precision (e.g., inference and / or prediction performance) for one or more incoming data samples. Device 202-b may perform learning model selection without labeled data samples, reference signals, and / or control signals (e.g., or any other side information and / or auxiliary information used to assist in learning model selection) from another node and / or device in the wireless communication system 200. Device 202-b selects a learning model using a set of unlabeled input data samples to be classified, or equivalently, data samples whose labels are to be predicted and / or inferred by one of learning models 206-a to 206-T. Therefore, learning model selection may be referred to as autonomous learning model selection. Alternatively, autonomous learning model selection can be used in conjunction with labeled data, reference signals, or any other side information and / or auxiliary information (e.g., characteristics of the input samples) that may be relevant to the learning model selection. Therefore, autonomous learning model selection can be used as a standalone method or as a complement to other learning model selection methods.

[0067] In some variations, the autonomous learning model selection can be triggered, initiated, and / or executed by a device of the managed learning models 206-a to 206-T (e.g., a user equipment containing a UE or other user equipment, or an NE) or another device and / or entity in the wireless communication system 200.

[0068] Device 202-b can obtain a set of data samples to be classified or inferred using a learning model selected from learning models 206-a to 206-T. These data samples may be referred to as data samples from the target domain. The data distribution of the incoming input data samples at device 202-b may change periodically or continuously across a certain time period, causing device 202-b to frequently perform learning model selection (e.g., performing it more than a threshold number of times within a certain time period). Data samples from the target domain may be unlabeled samples. Therefore, in order to select... One of the learning models (e.g., learning models 206-a to 206-T), device 202-b uses a set of input data samples from the target domain. (For example, no input sample is required) ( The corresponding tag for ) ( )).

[0069] In some examples, for input data samples ( Equipment 202-b selection Learning model among learning models Learning Model ( Conditional distribution Equipment 202-b can be used for... and calculate Device 202-b can be based on Equation 1. Input data sample Calculate conditional entropy:

[0070] (1)

[0071] in Let be the probability distribution of the input data samples for learning model k. In some examples, the probability distribution of the input data samples is available to the learning model or can be provided to device 202-b (e.g., the node and / or entity performing model selection). For example, the probability distribution may be based on the set of training data samples used to train the learning model. If the input data samples are equally likely, then the conditional entropy can be expressed according to Equation 2. Input data sample The given first The average entropy of the learning model's prediction (y):

[0072] (2)

[0073] In some examples, the logarithm (log) can be the natural logarithm, or the logarithm to base 2. Conditional entropy An indicator of the confidence level of an individual prediction made by the learning model (e.g., lower entropy means lower uncertainty in the prediction, which indicates higher confidence in the learning model).

[0074] Device 202-b can calculate the empirical marginal distribution of the predicted labels according to Equation 3:

[0075] (3)

[0076] In some examples, the input data samples for learning model k The probability distribution can be obtained from the learning model or provided to device 202-b (e.g., the node and / or entity performing model selection). The probability distribution can be based on the set of training data samples used to train the learning model. If the input data samples are equally likely, then the distribution of the data samples can be calculated according to Equation 4:

[0077] (4)

[0078] In some examples, the calculated distribution Is as The true marginal distribution An approximation. In some examples, instead of calculating the marginal distribution of predicted labels, device 202-b may additionally obtain the marginal probability distribution of the learned model, for example, by receiving an indication of the marginal probability distribution. For instance, the probability distribution of predicted labels may be based on the set of training data samples used to train the model and the training labels.

[0079] Device 202-b can calculate the entropy of the predicted tag according to Equation 5:

[0080] (5)

[0081] In some variations, A higher value indicates a balance in cross-label predictions, or suggests that the marginal distribution of predicted labels is close to a uniform distribution (e.g., this is for a reasonable number of data samples). ( (The desired properties).

[0082] Device 202-b can assign reliability scores to the corresponding learning models (e.g., learning models 206-a to 206-T). For example, device 202-b can set the first [equation] according to Equations 6 and 7. The confidence score of the model (represented as) ):

[0083] (6)

[0084] (7)

[0085] In some examples, and (in , This assigns weights (e.g., importance) to the corresponding items in the confidence score. Device 202-b may select and / or determine the values ​​of the weights based on the use case of the learning model and / or other available side information at the device performing the model selection procedure. Alternatively or additionally, device 202-b may receive an indication of the weights (e.g., from another device or node in the wireless communication system 200). ,So ,in This represents information gain, which is the sum of the inputs to the learning model. When it is known, the first Model output Information gain is a measure of the degree to which entropy or uncertainty is reduced by device 202-b. In other words, information gain is the amount of information gained by device 202-b through knowing... , And obtained about The measure of the amount of information, in which express and (No. The learning model (learning the interaction information between its input and output). In some cases, device 202-b can fix the interaction information based on user-provided input. , The user-provided input is based on the use cases of the learning model, known physical principles related to the wireless communication system 200, and other factors.

[0086] In some cases, such as for spatial beam prediction, the learning model classifies a corresponding input data sample (e.g., a set of beam measurements) into one of M candidate beams, such that the inferred and / or predicted beam is the beam with the highest performance (e.g., signal quality and / or signal power, and other metrics). The beam with the highest performance for a device (e.g., a UE) within a defined geographical location can be part of a subset of the available beam set. Therefore, the learning model output is a subset of values ​​from the available beam indices, which produces... The lower the value. In some other examples, such as for a learning-based receiver, the learning model classifies the input data samples received from the wireless channel output into one of the possible sets of transmitted messages, symbols, and / or bits. In some variations, the information bits are uniformly distributed, and the output of the learning model can contain corresponding different possible symbols (e.g., all different possible symbols), even with a relatively small number of data samples. ( The same applies. Therefore, The value can be relatively high (e.g., greater than a threshold). Device 202-b can assign values ​​to the weights such that when the number of data samples available from the target domain is relatively small (e.g., when...), the value is relatively high. When it is less than the threshold, When n is a relatively large value (e.g., greater than a threshold), device 202-b can assign values ​​to the weights such that... .

[0087] Device 202-b can select the learning model with the highest confidence score (e.g., by comparing the confidence scores of learning model 206-a to learning model 206-T). For example, device 202-b can store... Number of learning models , Device 202-b can be configured based on the target domain and weights. , The value is obtained from unlabeled data samples In some examples, device 202-b may receive signaling 218 indicating data samples from another device (e.g., device 202-a). For example, device 202-a may perform one or more measurements (e.g., beam measurements, reference signal measurements, and / or channel measurements, and other measurements) and may transmit the measurements to device 202-b for use as data samples. Alternatively or additionally, device 202-b may perform measurements to obtain data samples. Example beam measurements may include, but are not limited to, received signal strength and / or quality from a beam set. Example reference signal measurements and / or channel measurements may include, but are not limited to, channel gain data, phase shift data, frequency response data, delay spread data, and Doppler shift data, and other information.

[0088] Equipment 202-b is for data Obtain an index of the learning model with the highest performance (e.g., accuracy and / or precision). For example, for Equipment 202-b in use and To target Sure Equipment 202-b calculates conditional entropy. and ,in Equipment 202-b calculates the first... Model confidence score , Device 202-b selects a learning model that meets the selection criteria. For example, the selection criteria might include the learning model with the highest confidence score (e.g., Alternatively, the selection criteria may include: the difference between the conditional entropy and marginal entropy of the learning model is the maximum difference in the set of learning models. Alternatively, the selection criteria may include: the conditional entropy of the learning model is the lowest value (e.g., the minimum value) in the set of learning models.

[0089] In some examples, the distribution of the input data samples can change relatively quickly (e.g., within a duration less than a threshold, on a short timescale). For example, the distribution of the data samples can change after a certain number of data samples (e.g., per few samples or per pair of samples). Device 202-b can select weights based on the number of data samples in the target distribution (e.g., and The value of ). For example, device 202-b can select the value of the weight so that when the number of data samples available from the target domain is relatively small (e.g., when When it is less than the threshold, .if Then, the selectable values ​​for device 202-b are: and Alternatively or in lieu, if If a learning model is available, then device 202-b can calculate the confidence score of the corresponding learning model for the incoming data sample to be classified. Device 202-b can select the output of the learning model with the highest confidence score. Therefore, device 202-b can use multiple learning models simultaneously, allowing the device to use the learning model with the highest confidence score for each incoming data sample to be classified (e.g., at each time step).

[0090] In some examples, device 202-b may implement autonomous learning model selection for information prediction of signals (e.g., CQI prediction). For example, for CQI prediction, the input to the learning model contains channel vectors over a subset of time and / or frequency resource sets. The output of the learning model contains the CQI for the entire time and / or frequency resource set. In some cases, the input to the learning model contains channel estimates for a subset of the set of subbands of the bandwidth portion (BWP) of the channel. The output of the learning model contains subband CQI values ​​for both the subset of the subband set and the remaining subbands within the subband set. In some other cases, the input to the learning model contains channel estimates for a first set of time slots (e.g., time resources) preceding the CSI reference resources used for CSI reporting settings. The output of the learning model contains CQI values ​​for a second set of time slots following the CSI reference resources. Device 202-b may be configured with CSI resource settings where one or more CSI-RS with CQI prediction type one are transmitted over a subset of resource blocks of the BWP. A subset of resource blocks for the BWP is selected by device 202-b, configured by the network, or set by rules. The selection of the subset of resource blocks may depend on the selected learning model. The UE may implement the learning model. In other cases, the input to the learning model contains CQI values ​​for a subset of the time and / or frequency resource sets. The output of the learning model contains CQI values ​​for the remaining time and / or frequency resources. In other cases, the input to the learning model contains subband CQI values ​​for a subset of the subband set of the BWP for the channel. The output of the learning model contains subband CQI values ​​for the remaining subbands in the subband set. In other cases, the input to the learning model contains CQI values ​​for a first set of time slots preceding the CSI reference resources used for CSI reporting settings. The output of the learning model contains CQI values ​​for a second set of time slots following the CSI reference resources. Device 202-b may be configured with CSI reporting settings, wherein the CSI reporting settings define the reporting of a subset of subband CQI values ​​containing the subband set of the BWP. A subset of the subband set can be a configured network and can depend on the chosen learning model. NE can implement the learning model.

[0091] In some examples, device 202-b may transmit an indication 220 of the selected learning model to device 202-a. For example, device 202-b may transmit control signaling indicating the indication 220 of the selected learning model via communication link 204-a. In some variations, the indication 220 of the selected learning model may include an index of the selected learning model, and / or may include one or more parameters defining the selected learning model.

[0092] Figure 3An example of signaling diagram 300 according to various aspects of this disclosure is shown. In some examples, signaling diagram 300 may implement aspects of wireless communication system 100 and wireless communication system 200. Signaling diagram 300 may show an example of device 302-a and / or device 302-b performing autonomous model selection. In some cases, device 302-a may be an example of device 202-a and device 302-b may be an example of device 202-b, as referenced. Figure 2 As described. For example, devices 302-a and 302-b may be examples of NE and / or UE. Alternative examples of the following may be implemented, some of which may be performed in a different order than described or not performed at all. In some cases, procedures may include additional features not mentioned below, or additional procedures may be added.

[0093] In some examples, at 304, device 302-a may transmit signaling that triggers the selection of a learning model to device 302-b. For example, device 302-a may transmit control signaling to device 302-b, the control signaling containing a field (e.g., a one-bit indicator) that triggers the selection of a learning model.

[0094] In some cases, at 306, device 302-a may transmit signaling indicating one or more data samples to device 302-b. For example, device 302-a may perform one or more measurements to obtain data samples. Device 302-a may transmit control signaling and / or data transmission containing data samples. Example measurements may include, but are not limited to, the received signal strength of one or more beams, quality measurements from the beam set, channel gain data, phase shift data, frequency response data, delay spread data, and Doppler shift data, as well as other information. In some examples, signaling indicating data samples may also trigger the selection of a learning model (e.g., signaling containing data samples implicitly instructs device 302-b to trigger the selection of a learning model, and / or the signaling contains a field explicitly instructing device 302-b to trigger the selection of a learning model).

[0095] At 308, device 302-b may acquire one or more data samples. The data samples may include one or more data samples (e.g., unlabeled data samples) used as input to a learning model. For example, the learning model may be used to classify the data samples into one or more labels and / or categories. At 304, device 302-b may acquire the data samples via signaling. Alternatively or additionally, device 302-b may perform a measurement to acquire the data samples.

[0096] In some cases, device 302-b may receive one or more reference signals from another device (e.g., from device 302-a). Device 302-b may obtain information by performing measurements on the reference signals. The information may include one or more of RSRP measurements, RSRQ measurements, CQI, PMI, MCS index, or RI.

[0097] At 310, device 302-b can generate confidence scores for the learning model set. The device can generate confidence scores by providing a set of data samples as input to the learning model set. For example, the confidence scores can vary with the conditional entropy and marginal entropy of the learning model. In some cases, device 302-b generates a corresponding uncertainty measure of the output from the learning model based on the set of data samples (e.g., conditional entropy and / or marginal entropy). For example, the corresponding uncertainty measure includes the marginal entropy of the output, or the conditional entropy of the output given an input (e.g., the set of data samples). In some examples, device 302-b can apply one or more weights (e.g., having values ​​between zero and one) to the conditional entropy and marginal entropy. In some examples, the confidence score includes the information gain from the output of the learning model given an input (e.g., the set of data samples). Information gain is a measure of the amount of information device 302-b gains about the output by knowing the input.

[0098] At 312, device 302-b may select a learning model based on the confidence score of the learning model satisfying a selection criterion. Device 302-b selects a learning model to execute at device 302-b and / or at device 302-a. The selection criterion may include the minimum conditional entropy (e.g., an uncertainty measure) among the corresponding conditional entropies of different learning models. Alternatively, the selection criterion may include the maximum difference between conditional entropy and marginal entropy (e.g., or weighted conditional entropy and weighted marginal entropy). Alternatively, the selection criterion may include the highest confidence score among the confidence scores computed (e.g., generated, obtained) for different learning models.

[0099] In some examples, at 314, device 302-b may transmit an indication of the selected learning model to device 302-a. The indication of the selected learning model may include an indication of the index of the learning model, and / or may include one or more parameters defining the learning model. Device 302-a may include the indication of the learning model in control signaling and / or in any other transmission sent to device 302-a (e.g., signaling dedicated to the learning model indication, and / or one or more new fields in signaling not dedicated to the learning model indication).

[0100] In some variations, at 316, device 302-b can execute (process, implement, deploy, etc.) the learning model. For example, device 302-b can produce a prediction as the output of the learning model. Alternatively, at 318, device 302-a can execute the learning model. For example, device 302-a can produce a prediction as the output of the learning model.

[0101] For a learning model implemented as a receiver, devices 302-b and / or 302-a may input a signal output from a channel (e.g., a wireless channel) into a selected learning model, and the selected learning model may output a prediction of at least one of a message, symbol set, or bit set. For a learning model implemented for beam prediction, devices 302-b and / or 302-a may input a set of beam measurements and / or auxiliary information for one or more beams into the learning model, and the learning model may output a prediction of at least one beam (e.g., the beam with the highest signal quality and / or power, and other performance metrics) for transmitting or receiving signaling. Alternatively or additionally, devices 302-b and / or 302-a may input a set of beam measurements or auxiliary information for one or more current and historical beams into the learning model, and the learning model may output a prediction of at least one beam for transmitting or receiving signaling in a future time slot set. Beam measurements may include, but are not limited to, RSRP measurements, RSRQ measurements, and / or SINR measurements. Auxiliary information may include, but is not limited to, information about the channel (e.g., CSI), quality of service (QoS), and information about the communication environment, as well as other information. For a learning model implemented to predict information about the signal and / or channel (e.g., CSI, including CQI, MPI, RI, and / or MCS indices), devices 302-b and / or 302-a may input a set of reference signal measurements or a set of channel vectors into the learning model, and the learning model may output a prediction of the information about the signal and / or channel. The set of channel vectors defines an initial subset of time-frequency resources in the time-frequency resource set, and the output from the learning model defines another subset of time-frequency resources and / or the entire time-frequency resource set.

[0102] Figure 4 An example of a UE 400 according to various aspects of this disclosure is shown. UE 400 may include a processor 402, a memory 404, a controller 406, and a transceiver 408. The processor 402, memory 404, controller 406, or transceiver 408, or various combinations thereof, or various components thereof, may be examples of means for performing various aspects of this disclosure described herein. These components may be coupled via one or more interfaces (e.g., operatively, communicatively, functionally, electronically, or electrically).

[0103] Processor 402, memory 404, controller 406, or transceiver 408, or various combinations or components thereof, may be implemented in hardware (e.g., a circuit system). The hardware may include a processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), or other programmable logic device, or any combination thereof, configured to or otherwise supporting means for performing the functions described in this disclosure.

[0104] Processor 402 may include intelligent hardware devices (e.g., a general-purpose processor, DSP, CPU, ASIC, field-programmable gate array (FPGA), or any combination thereof). In some embodiments, processor 402 may be configured to operate memory 404. In some other embodiments, memory 404 may be integrated into processor 402. Processor 402 may be configured to execute computer-readable instructions stored in memory 404 to cause UE 400 to perform various functions of this disclosure.

[0105] Memory 404 may include volatile or non-volatile memory. Memory 404 may store computer-readable, computer-executable code containing instructions that, when executed by processor 402, cause UE 400 to perform the various functions described herein. The code may be stored in non-transitory computer-readable media (such as memory 404 or another type of memory). Computer-readable media includes both non-transitory computer storage media and communication media containing any media that facilitates the transfer of computer programs from one place to another. Non-transitory storage media may be any available media accessible by a general-purpose or special-purpose computer.

[0106] In some implementations, processor 402 and memory 404 coupled to processor 402 may be configured or operable to cause UE 400 to perform one or more functions described herein (e.g., instructions stored in memory 404 are executed by processor 402). For example, processor 402 may support wireless communication at UE 400 according to the examples disclosed herein. UE 400 may be configured or operable to support means for performing: obtaining a set of data samples as input to a set of learning models stored at a first device; generating a corresponding confidence score associated with a set of outputs from the set of learning models based on providing the set of data samples as input to the set of learning models; and selecting a learning model for execution at a first device or a second device based on a selection criterion satisfied by a confidence score associated with the output of a learning model from the set of learning models.

[0107] Additionally, UE 400 may be configured or operable to support any one or a combination of the following: transmitting a message indicating the selected learning model to a second device. Alternatively, obtaining a data sample set includes receiving a message from the second device indicating the data sample set. Alternatively, obtaining a data sample set includes receiving a reference signal set from the second device and obtaining information associated with the reference signal set, wherein the information includes one or more of RSRP measurements, RSRQ measurements, CQI, PMI, MCS index, or RI. Alternatively, UE 400 may be configured or operable to support receiving a message from the second device that triggers the selection of a learning model. Alternatively, generating a corresponding confidence score includes generating a corresponding uncertainty measure associated with the output set from the learning model set based on the data sample set, wherein the selection criterion includes the minimum uncertainty measure among the corresponding uncertainty measures. Alternatively, the corresponding uncertainty measure includes at least one of the marginal entropy of the output set or the conditional entropy of the output set based on the data sample set.

[0108] Alternatively, generating the corresponding confidence score includes generating a corresponding first uncertainty measure associated with the output set from the learning model set based on the data sample set and a corresponding second uncertainty measure associated with the output set from the learning model set, wherein the selection criterion includes the maximum difference between the corresponding first uncertainty measure and the corresponding second uncertainty measure. Alternatively, generating the corresponding confidence score includes generating a corresponding first weighted uncertainty measure associated with the output set from the learning model set based on the data sample set and a corresponding second weighted uncertainty measure associated with the output set from the learning model set, wherein the corresponding first weighted uncertainty measure and the corresponding second weighted uncertainty measure are weighted based on values ​​between zero and one. Alternatively, generating the corresponding confidence score includes generating an information gain associated with the output set from the learning model set based on the data sample set.

[0109] Alternatively or concurrently, the execution of the learning model includes: generating a prediction of at least one of a message set, symbol set, or bit set as output from the learning model based on providing a set of data samples as input to the selected learning model, wherein the set of data samples includes one or more channel output signals; generating a prediction of at least one beam for transmitting or receiving signaling based on providing a set of data samples as input to the selected learning model, wherein the set of data samples includes at least one of a beam measurement set or auxiliary information associated with one or more beams; generating a prediction of at least one beam for transmitting or receiving signaling in a future time slot set as output from the learning model based on providing a set of data samples as input to the selected learning model, wherein the set of data samples includes at least one of a beam measurement set or auxiliary information associated with one or more beams; or generating a prediction of CSI based on providing a set of data samples as input to the selected learning model, wherein the set of data samples includes at least one of a reference signal measurement set or a channel vector set, and wherein the CSI includes at least one of a CQI, PMI, MCS index, or RI associated with a time-frequency resource set. Alternatively or alternatively, the channel vector set is associated with a first subset of time-frequency resources in the time-frequency resource set, and the output from the learning model is associated with a second subset of time-frequency resources or the time-frequency resource set. Alternatively or alternatively, the first device is an NE. Alternatively or alternatively, the first device is a UE.

[0110] Alternatively, UE 400 may support at least one memory (e.g., memory 404) and at least one processor (e.g., processor 402) coupled to at least one memory, and is configured or operable to cause UE to: obtain a set of data samples to provide as input to a set of learning models stored at a first device; generate a corresponding confidence score associated with a set of outputs from the set of learning models based on providing the set of data samples as input to the set of learning models; and select the learning model for execution at a first device or a second device based on the confidence score associated with the output of a learning model in the set of learning models satisfying a selection criterion.

[0111] Additionally, UE 400 may be configured or operable to support any one or a combination of the following: at least one processor is configured or operable to transmit a message indicating the selected learning model to a second device. Alternatively, to obtain a data sample set, at least one processor is configured or operable to receive a message indicating the data sample set from the second device. Alternatively, to obtain the data sample set, at least one processor is configured or operable to receive a reference signal set from the second device and obtain information associated with the reference signal set, said information including one or more of RSRP measurements, RSRQ measurements, CQI, PMI, MCS index, or RI. Alternatively, at least one processor is configured or operable to receive a message from the second device that triggers the selection of a learning model. Alternatively, to generate a corresponding confidence score, at least one processor is configured or operable to generate a corresponding uncertainty measure associated with the output set from the learning model set based on the data sample set, wherein the selection criterion includes the minimum uncertainty measure among the corresponding uncertainty measures. Alternatively, the corresponding uncertainty measure includes at least one of the marginal entropy of the output set or the conditional entropy of the output set based on the data sample set.

[0112] Alternatively or alternatively, to generate the corresponding confidence score, at least one processor is configured or operable to generate a corresponding first uncertainty measure associated with the output set from the learning model set and a corresponding second uncertainty measure associated with the output set from the learning model set, based on a set of data samples, wherein the selection criterion includes the maximum difference between the corresponding first uncertainty measure and the corresponding second uncertainty measure. Alternatively or alternatively, to generate the corresponding confidence score, at least one processor is configured or operable to generate a corresponding first weighted uncertainty measure associated with the output set from the learning model set and a corresponding second weighted uncertainty measure associated with the output set from the learning model set, based on a set of data samples, wherein the corresponding first weighted uncertainty measure and the corresponding second weighted uncertainty measure are weighted based on values ​​between zero and one. Alternatively or alternatively, to generate the corresponding confidence score, at least one processor is configured or operable to generate an information gain associated with the output set from the learning model set, based on a set of data samples.

[0113] Alternatively or concurrently, in order to execute the learning model, at least one processor is configured or operable to: generate a prediction of at least one of a message set, symbol set, or bit set as the output of the learning model based on providing a set of data samples as input to the selected learning model, wherein the set of data samples includes one or more channel output signals; generate a prediction of at least one beam for transmitting or receiving signaling based on providing a set of data samples as input to the selected learning model, wherein the set of data samples includes at least one of a beam measurement set or auxiliary information associated with one or more beams; based on providing a set of data samples as input to the selected learning model, the processor generates a prediction of at least one beam for transmitting or receiving signaling as the output of the learning model; The collaboration provides input to a selected learning model to generate a prediction of at least one beam for transmitting or receiving signaling in a future time slot set as output from the learning model, wherein the data sample set contains at least one of a beam measurement set or auxiliary information associated with one or more beams; or, based on providing the data sample set as input to the selected learning model, a prediction of CSI is generated as output from the learning model, wherein the data sample set contains at least one of a reference signal measurement set or a channel vector set, and wherein the CSI contains at least one of a CQI, PMI, MCS index, or RI associated with a time-frequency resource set. Alternatively, the channel vector set is associated with a first time-frequency resource subset in the time-frequency resource set, and the output from the learning model is associated with a second time-frequency resource subset or the time-frequency resource set. Alternatively, the first device is an NE. Alternatively, the first device is a UE.

[0114] Controller 406 manages the input and output signals of UE 400. Controller 406 can also manage peripheral devices not integrated into UE 400. In some embodiments, controller 406 may utilize an operating system such as iOS®, Android®, Windows®, or other operating systems. In some embodiments, controller 406 may be implemented as part of processor 402.

[0115] In some embodiments, UE 400 may include at least one transceiver 408. In other embodiments, UE 400 may have more than one transceiver 408. Transceiver 408 may represent a wireless transceiver. Transceiver 408 may include one or more receiver chains 410, one or more transmitter chains 412, or a combination thereof.

[0116] Receiver chain 410 may be configured or operable to receive signals (e.g., control information, data, packets) via wireless media. For example, receiver chain 410 may include one or more antennas for receiving signals in the air or via wireless media. Receiver chain 410 may include at least one amplifier (e.g., a low-noise amplifier (LNA)) configured or operable to amplify the received signal. Receiver chain 410 may include at least one demodulator configured or operable to demodulate the received signal and obtain the transmitted data by reversing the modulation technique applied during signal transmission. Receiver chain 410 may include at least one decoder for decoding the demodulated signal to receive the transmitted data.

[0117] Transmitter chain 412 may be configured or operable to generate and transmit signals (e.g., control information, data, packets). Transmitter chain 412 may include at least one modulator for modulating data onto a carrier signal, preparing the signal for transmission over a wireless medium. At least one modulator may be configured or operable to support one or more techniques, such as amplitude modulation (AM), frequency modulation (FM), or digital modulation schemes like phase shift keying (PSK) or quadrature amplitude modulation (QAM). Transmitter chain 412 may also include at least one power amplifier configured or operable to amplify the modulated signal to an appropriate power level suitable for transmission over a wireless medium. Transmitter chain 412 may also include one or more antennas for transmitting the amplified signal over the air or wireless medium.

[0118] Figure 5 An example of a processor 500 according to various aspects of this disclosure is shown. Processor 500 may be an example of a processor configured to perform various operations according to the examples described herein. Processor 500 may include a controller 502 configured to perform various operations according to the examples described herein. Processor 500 may optionally include at least one memory 504, which may be, for example, an L1 / L2 / L3 cache. Additionally or alternatively, processor 500 may optionally include one or more arithmetic logic units (ALUs) 506. One or more of these components may be electronically communicated or otherwise coupled (e.g., operatively, communicatively, functionally, electronically, or electrically) via one or more interfaces (e.g., buses).

[0119] Processor 500 may be a processor chipset and includes a protocol stack (e.g., a software stack) executed by the processor chipset to perform various operations (e.g., receive, acquire, retrieve, transmit, output, forward, store, determine, identify, access, write, read) according to the examples described herein. The processor chipset may include one or more cores, one or more caches (e.g., memory local to the processor chipset (e.g., processor 500) or contained within the processor chipset) or other memory (e.g., random access memory (RAM), read-only memory (ROM), dynamic RAM (DRAM), synchronous dynamic RAM (SDRAM), static RAM (SRAM), ferroelectric RAM (FeRAM), magnetic RAM (MRAM), resistive RAM (RRAM), flash memory, phase-change memory (PCM), etc.).

[0120] Controller 502 can be configured to manage and coordinate various operations of processor 500 (e.g., signaling, receiving, acquiring, retrieving, transmitting, outputting, forwarding, storing, determining, identifying, accessing, writing, and reading) to enable processor 500 to support various operations according to the examples described herein. For example, controller 502 can act as a control unit of processor 500, generating control signals that manage the operation of various components of processor 500. These control signals include enabling or disabling functional units, selecting data paths, initiating memory accesses, and coordinating the timing of operations.

[0121] Controller 502 may be configured to fetch (e.g., obtain, retrieve, receive) instructions from memory 504 and determine one or more subsequent instructions to be executed to enable processor 500 to support various operations according to the examples described herein. Controller 502 may be configured to track the memory addresses of instructions associated with memory 504. Controller 502 may be configured to decode instructions to determine the operations to be performed and the operands involved. For example, controller 502 may be configured to interpret instructions and determine control signals to be output to other components of processor 500 to enable processor 500 to support various operations according to the examples described herein. Alternatively or additionally, controller 502 may be configured to manage data flow within processor 500. Controller 502 may be configured to control data transfers between registers, ALU 506, and other functional units of processor 500.

[0122] Memory 504 may include one or more caches (e.g., memory local to processor 500 or included in processor 500, or other memory such as RAM, ROM, DRAM, SDRAM, SRAM, MRAM, flash memory, etc.). In some embodiments, memory 504 may reside within or on the processor chipset (e.g., locally to processor 500). In some other embodiments, memory 504 may reside outside the processor chipset (e.g., remotely from processor 500).

[0123] Memory 504 may store computer-readable, computer-executable code containing instructions that, when executed by processor 500, cause processor 500 to perform the various functions described herein. The code may be stored in a non-transitory computer-readable medium (e.g., system memory or another type of memory). Controller 502 and / or processor 500 may be configured to execute the computer-readable instructions stored in memory 504 to cause processor 500 to perform various functions. For example, processor 500 and / or controller 502 may be coupled to or coupled to memory 504, processor 500, and controller 502, and may be configured to perform the various functions described herein. In some examples, processor 500 may include multiple processors, and memory 504 may include multiple memories. One or more of the multiple processors may be coupled to one or more of the multiple memories, which may be individually or collectively configured to perform the various functions described herein.

[0124] One or more ALUs 506 may be configured to support various operations according to the examples described herein. In some embodiments, one or more ALUs 506 may reside within or on a processor chipset (e.g., processor 500). In some other embodiments, one or more ALUs 506 may reside outside the processor chipset (e.g., processor 500). One or more ALUs 506 may perform one or more calculations on data, such as addition, subtraction, multiplication, and division. For example, one or more ALUs 506 may receive input operands and operation codes that determine the operation to be performed. One or more ALUs 506 may be configured with various logic and arithmetic circuitry, including adders, subtractors, shifters, and logic gates, to process and manipulate data according to the operation. Alternatively, one or more ALU 506s may support logical operations such as AND, OR, XOR, NOR, and NAND, thereby enabling one or more ALU 506s to manage conditional operations, comparisons, and bitwise operations.

[0125] Processor 500 may support wireless communication according to the examples disclosed herein. Processor 500 may be configured or operable to support at least one controller (e.g., controller 502) coupled to at least one memory (e.g., memory 504), and configured or operable to cause the processor to: obtain a set of data samples to provide as input to a set of learning models stored at the processor; generate a corresponding confidence score associated with a set of outputs from the set of learning models based on providing the set of data samples as input to the set of learning models; and select said learning model for execution at the processor or at a device based on the confidence score associated with the output of a learning model in the set of learning models satisfying a selection criterion.

[0126] Additionally, the processor 500 may be configured or operable to support any one or a combination of the following: at least one controller is configured or operable to cause the processor to transmit a message indicating the selected learning model to the device. Alternatively, to obtain a set of data samples, at least one controller is configured or operable to cause the processor to receive a message indicating the set of data samples from the device. Alternatively, to obtain a set of data samples, at least one controller is configured or operable to cause the processor to receive a set of reference signals from the device and obtain information associated with the set of reference signals, said information including one or more of RSRP measurements, RSRQ measurements, CQI, PMI, MCS index, or RI. Alternatively, at least one controller is configured or operable to cause the processor to receive a message from the device that triggers the selection of a learning model. Alternatively, to generate a corresponding confidence score, at least one controller is configured or operable to cause the processor to generate a corresponding uncertainty measure associated with the output set from the set of learning models based on the set of data samples, wherein the selection criterion includes the minimum uncertainty measure among the corresponding uncertainty measures. Alternatively, the corresponding uncertainty measure includes at least one of the marginal entropy of the output set based on the set of data samples or the conditional entropy of the output set.

[0127] Alternatively or alternatively, to generate the corresponding confidence score, at least one controller is configured or operable to cause the processor to generate a corresponding first uncertainty measure associated with the output set from the learning model set and a corresponding second uncertainty measure associated with the output set from the learning model set, based on a set of data samples, wherein the selection criterion includes the maximum difference between the corresponding first uncertainty measure and the corresponding second uncertainty measure. Alternatively or alternatively, to generate the corresponding confidence score, at least one controller is configured or operable to cause the processor to generate a corresponding first weighted uncertainty measure associated with the output set from the learning model set and a corresponding second weighted uncertainty measure associated with the output set from the learning model set, based on a set of data samples, wherein the corresponding first weighted uncertainty measure and the corresponding second weighted uncertainty measure are weighted based on values ​​between zero and one. Alternatively or alternatively, to generate the corresponding confidence score, at least one controller is configured or operable to cause the processor to generate an information gain associated with the output set from the learning model set, based on a set of data samples.

[0128] Alternatively or concurrently, in order to execute the learning model, at least one controller is configured or operable to cause the processor to: generate a prediction of at least one of a message set, symbol set, or bit set as the output of the learning model based on providing a set of data samples as input to the selected learning model, wherein the set of data samples includes one or more channel output signals; generate a prediction of at least one beam for transmitting or receiving signaling based on providing a set of data samples as input to the selected learning model, wherein the set of data samples includes at least one of a set of beam measurements or auxiliary information associated with one or more beams; and generate a prediction of at least one beam for transmitting or receiving signaling based on providing a set of data samples as input to the selected learning model, wherein the set of data samples includes at least one of a set of beam measurements or auxiliary information associated with one or more beams; and generate a prediction of at least one beam for transmitting or receiving signaling based on providing a set of data samples as input to the selected learning model. This set of data samples is provided as input to the selected learning model to generate a prediction of at least one beam for transmitting or receiving signaling in a future time slot set, which is the output of the learning model. The data sample set contains at least one of a beam measurement set or auxiliary information associated with one or more beams. Alternatively, a prediction of CSI is generated based on providing the data sample set as input to the selected learning model, which is the output of the learning model. The data sample set contains at least one of a reference signal measurement set or a channel vector set, and the CSI contains at least one of a CQI, PMI, MCS index, or RI associated with a time-frequency resource set. Alternatively or additionally, the channel vector set is associated with a first time-frequency resource subset in the time-frequency resource set, and the output from the learning model is associated with a second time-frequency resource subset or the time-frequency resource set. Alternatively or additionally, the processor is associated with the NE. Alternatively or additionally, the processor is associated with the UE.

[0129] Figure 6An example of an NE 600 according to various aspects of this disclosure is shown. The NE 600 may include a processor 602, a memory 604, a controller 606, and a transceiver 608. The processor 602, memory 604, controller 606, or transceiver 608, or various combinations thereof, or various components thereof, may be examples of means for performing various aspects of this disclosure described herein. These components may be coupled via one or more interfaces (e.g., operatively, communicatively, functionally, electronically, or electrically).

[0130] Processor 602, memory 604, controller 606, or transceiver 608, or various combinations or components thereof, may be implemented in hardware (e.g., a circuit system). The hardware may include processors, DSPs, ASICs, or other programmable logic devices, or any combination thereof, configured to or otherwise supporting means for performing the functions described in this disclosure.

[0131] Processor 602 may include intelligent hardware devices (e.g., a general-purpose processor, DSP, CPU, ASIC, FPGA, or any combination thereof). In some embodiments, processor 602 may be configured to operate memory 604. In some other embodiments, memory 604 may be integrated into processor 602. Processor 602 may be configured to execute computer-readable instructions stored in memory 604 to cause NE 600 to perform various functions of this disclosure.

[0132] Memory 604 may comprise volatile or non-volatile memory. Memory 604 may store computer-readable, computer-executable code containing instructions that, when executed by processor 602, cause NE 600 to perform the various functions described herein. The code may be stored in non-transitory computer-readable media (such as memory 604 or another type of memory). Computer-readable media includes both non-transitory computer storage media and communication media containing any media that facilitates the transfer of computer programs from one place to another. Non-transitory storage media may be any available media accessible by a general-purpose or special-purpose computer.

[0133] In some embodiments, processor 602 and memory 604 coupled to processor 602 may be configured or operable to cause NE 600 to perform one or more functions described herein (e.g., instructions stored in memory 604 are executed by processor 602). For example, processor 602 may support wireless communication at NE 600 according to the examples disclosed herein. NE 600 may be configured or operable to support means for performing: obtaining a set of data samples as input to a set of learning models stored at a first device; generating a corresponding confidence score associated with a set of outputs from the set of learning models based on providing the set of data samples as input to the set of learning models; and selecting a learning model for execution at a first device or a second device based on a selection criterion satisfied by a confidence score associated with the output of a learning model in the set of learning models.

[0134] Additionally, the NE 600 may be configured or operable to support any one or a combination of the following: the method further includes transmitting a message indicating the selected learning model to a second device. Alternatively, obtaining the data sample set includes receiving a message indicating the data sample set from the second device. Alternatively, obtaining the data sample set includes receiving a set of reference signals from the second device and obtaining information associated with the set of reference signals, said information including one or more of RSRP measurements, RSRQ measurements, CQI, PMI, MCS index, or RI. Alternatively, the NE 600 may be configured or operable to support receiving a message from the second device that triggers the selection of a learning model. Alternatively, generating a corresponding confidence score includes generating a corresponding uncertainty measure associated with the output set from the set of learning models based on the data sample set, wherein the selection criterion includes the minimum uncertainty measure among the corresponding uncertainty measures. Alternatively, the corresponding uncertainty measure includes at least one of the marginal entropy of the output set or the conditional entropy of the output set based on the data sample set.

[0135] Alternatively, generating the corresponding confidence score includes generating a corresponding first uncertainty measure associated with the output set from the learning model set based on the data sample set and a corresponding second uncertainty measure associated with the output set from the learning model set, wherein the selection criterion includes the maximum difference between the corresponding first uncertainty measure and the corresponding second uncertainty measure. Alternatively, generating the corresponding confidence score includes generating a corresponding first weighted uncertainty measure associated with the output set from the learning model set based on the data sample set and a corresponding second weighted uncertainty measure associated with the output set from the learning model set, wherein the corresponding first weighted uncertainty measure and the corresponding second weighted uncertainty measure are weighted based on values ​​between zero and one. Alternatively, generating the corresponding confidence score includes generating an information gain associated with the output set from the learning model set based on the data sample set.

[0136] Alternatively or concurrently, the execution of the learning model includes: generating a prediction of at least one of a message set, symbol set, or bit set as output from the learning model based on providing a set of data samples as input to the selected learning model, wherein the set of data samples includes one or more channel output signals; generating a prediction of at least one beam for transmitting or receiving signaling based on providing a set of data samples as input to the selected learning model, wherein the set of data samples includes at least one of a beam measurement set or auxiliary information associated with one or more beams; generating a prediction of at least one beam for transmitting or receiving signaling in a future time slot set as output from the learning model based on providing a set of data samples as input to the selected learning model, wherein the set of data samples includes at least one of a beam measurement set or auxiliary information associated with one or more beams; or generating a prediction of CSI based on providing a set of data samples as input to the selected learning model, wherein the set of data samples includes at least one of a reference signal measurement set or a channel vector set, and wherein the CSI includes at least one of a CQI, PMI, MCS index, or RI associated with a time-frequency resource set. Alternatively or alternatively, the channel vector set is associated with a first subset of time-frequency resources in the time-frequency resource set, and the output from the learning model is associated with a second subset of time-frequency resources or the time-frequency resource set. Alternatively or alternatively, the first device is an NE. Alternatively or alternatively, the first device is a UE.

[0137] Alternatively, the NE 600 may support at least one memory (e.g., memory 604) and at least one processor (e.g., processor 602) coupled to at least one memory, and be configured or operable to cause the NE to: obtain a set of data samples to provide as input to a set of learning models stored at a first device; generate a corresponding confidence score associated with a set of outputs from the set of learning models based on providing the set of data samples as input to the set of learning models; and select the learning model for execution at the first device or a second device based on the confidence score associated with the output of a learning model in the set of learning models satisfying a selection criterion.

[0138] Additionally, the NE 600 can be configured to support any one or a combination of the following: at least one processor is configured or operable to transmit a message indicating the selected learning model to a second device. Alternatively, to obtain a set of data samples, at least one processor is configured or operable to receive a message indicating the set of data samples from the second device. Alternatively, to obtain the set of data samples, at least one processor is configured or operable to receive a set of reference signals from the second device and obtain information associated with the set of reference signals, said information including one or more of RSRP measurements, RSRQ measurements, CQI, PMI, MCS index, or RI. Alternatively, at least one processor is configured or operable to receive a message from the second device that triggers the selection of a learning model. Alternatively, to generate a corresponding confidence score, at least one processor is configured or operable to generate a corresponding uncertainty measure associated with the output set from the set of learning models based on the set of data samples, wherein the selection criterion includes the minimum uncertainty measure among the corresponding uncertainty measures. Alternatively, the corresponding uncertainty measure includes at least one of the marginal entropy of the output set based on the set of data samples or the conditional entropy of the output set.

[0139] Alternatively or alternatively, to generate the corresponding confidence score, at least one processor is configured or operable to generate a corresponding first uncertainty measure associated with the output set from the learning model set and a corresponding second uncertainty measure associated with the output set from the learning model set, based on a set of data samples, wherein the selection criterion includes the maximum difference between the corresponding first uncertainty measure and the corresponding second uncertainty measure. Alternatively or alternatively, to generate the corresponding confidence score, at least one processor is configured or operable to generate a corresponding first weighted uncertainty measure associated with the output set from the learning model set and a corresponding second weighted uncertainty measure associated with the output set from the learning model set, based on a set of data samples, wherein the corresponding first weighted uncertainty measure and the corresponding second weighted uncertainty measure are weighted based on values ​​between zero and one. Alternatively or alternatively, to generate the corresponding confidence score, at least one processor is configured or operable to generate an information gain associated with the output set from the learning model set, based on a set of data samples.

[0140] Alternatively or concurrently, in order to execute the learning model, at least one processor is configured or operable to: generate a prediction of at least one of a message set, symbol set, or bit set as the output of the learning model based on providing a set of data samples as input to the selected learning model, wherein the set of data samples includes one or more channel output signals; generate a prediction of at least one beam for transmitting or receiving signaling based on providing a set of data samples as input to the selected learning model, wherein the set of data samples includes at least one of a beam measurement set or auxiliary information associated with one or more beams; based on providing a set of data samples as input to the selected learning model, the processor generates a prediction of at least one beam for transmitting or receiving signaling as the output of the learning model; The collaboration provides input to a selected learning model to generate a prediction of at least one beam for transmitting or receiving signaling in a future time slot set as output from the learning model, wherein the data sample set contains at least one of a beam measurement set or auxiliary information associated with one or more beams; or, based on providing the data sample set as input to the selected learning model, a prediction of CSI is generated as output from the learning model, wherein the data sample set contains at least one of a reference signal measurement set or a channel vector set, and wherein the CSI contains at least one of a CQI, PMI, MCS index, or RI associated with a time-frequency resource set. Alternatively, the channel vector set is associated with a first time-frequency resource subset in the time-frequency resource set, and the output from the learning model is associated with a second time-frequency resource subset or the time-frequency resource set. Alternatively, the first device is an NE. Alternatively, the first device is a UE.

[0141] Controller 606 manages the input and output signals of NE 600. Controller 606 can also manage peripheral devices not integrated into NE 600. In some embodiments, controller 606 may utilize an operating system such as iOS®, Android®, Windows®, or other operating systems. In some embodiments, controller 606 may be implemented as part of processor 602.

[0142] In some embodiments, NE 600 may include at least one transceiver 608. In other embodiments, NE 600 may have more than one transceiver 608. Transceiver 608 may represent a wireless transceiver. Transceiver 608 may include one or more receiver chains 610, one or more transmitter chains 612, or a combination thereof.

[0143] Receiver chain 610 may be configured to receive signals (e.g., control information, data, packets) via wireless media. For example, receiver chain 610 may include one or more antennas for receiving signals in the air or via wireless media. Receiver chain 610 may include at least one amplifier (e.g., a low-noise amplifier (LNA)) configured to amplify the received signal. Receiver chain 610 may include at least one demodulator configured to demodulate the received signal and obtain the transmitted data by reversing the modulation technique applied during signal transmission. Receiver chain 610 may include at least one decoder for decoding the demodulated signal to receive the transmitted data.

[0144] Transmitter chain 612 can be configured to generate and transmit signals (e.g., control information, data, packets). Transmitter chain 612 may include at least one modulator for modulating data onto a carrier signal, preparing the signal for transmission over a wireless medium. At least one modulator may be configured to support one or more techniques, such as amplitude modulation (AM), frequency modulation (FM), or digital modulation schemes like phase shift keying (PSK) or quadrature amplitude modulation (QAM). Transmitter chain 612 may also include at least one power amplifier configured to amplify the modulated signal to an appropriate power level suitable for transmission over a wireless medium. Transmitter chain 612 may also include one or more antennas for transmitting the amplified signal over the air or into the wireless medium.

[0145] Figure 7 A flowchart of a method 700 according to various aspects of this disclosure is shown. The operation of the method can be implemented by the devices described herein (e.g., NE and / or UE). In some embodiments, the UE and / or NE can execute a set of instructions to control the functional elements of the UE and / or NE to perform the described functions. It should be noted that the methods described herein describe possible implementations, and the operations and steps may be rearranged or otherwise modified, and other implementations are possible.

[0146] At 702, the method may include obtaining a set of data samples to provide as input to a set of learning models stored at a first device. The operation at 702 may be performed according to the examples described herein. In some embodiments, aspects of the operation at 702 may be referenced. Figure 4 and 6 The described UE and / or NE are used to perform this.

[0147] At 704, the method may include generating a corresponding confidence score associated with the output set from the learning model set based on providing a set of data samples as input to a set of learning models. The operation at 704 can be performed according to the examples described herein. In some embodiments, aspects of the operation at 704 may be referenced. Figure 4 and 6 The described UE and / or NE are used to perform this.

[0148] At 706, the method may include selecting the learning model for execution at a first device or a second device based on a selection criterion that the confidence score associated with the output of a learning model from the set of learning models satisfies the selection criteria. The operation of 706 may be performed according to the examples described herein. In some embodiments, aspects of the operation of 706 may be referenced from... Figure 4 and 6 The described UE and / or NE are used to perform this.

[0149] The description provided herein is intended to enable those skilled in the art to make or use this disclosure. Various modifications to this disclosure will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other variations without departing from the scope of this disclosure. Therefore, this disclosure is not limited to the examples and designs described herein, but is given the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A first device, comprising: At least one memory; as well as At least one processor, coupled to the at least one memory and operable to enable the first device to: A set of data samples is obtained and used as input to multiple learning models stored at the first device; At least in part, confidence scores are generated based on providing the data sample set as input to the plurality of learning models, and are associated with the plurality of outputs from the plurality of learning models. as well as The learning model is selected for execution at the first device or the second device, at least in part, based on the confidence score associated with the output of one of the plurality of learning models satisfying the selection criteria.

2. The first device according to claim 1, wherein the at least one processor is further operable to cause the first device to transmit a message instructing the learning model to the second device.

3. The first device according to claim 1, wherein, in order to obtain the data sample set, the at least one processor is operable to cause the first device to receive a message indicating the data sample set from the second device.

4. The first device according to claim 1, wherein, in order to obtain the data sample set, the at least one processor is operable to cause the first device to: Receive multiple reference signals from the second device; and Obtain information associated with the plurality of reference signals, wherein the information includes one or more of the following: reference signal received power (RSRP) measurement, reference signal received quality (RSRQ) measurement, channel quality indicator (CQI), precoding matrix indicator (PMI), modulation and coding scheme index (MCS) or rank indicator.

5. The first device of claim 1, wherein the at least one processor is further operable to enable the first device to receive a message from the second device that triggers the selection of the learning model.

6. The first device of claim 1, wherein, in order to generate the corresponding confidence score, the at least one processor is operable to cause the first device to generate, at least in part, a corresponding uncertainty measure associated with the plurality of outputs from the plurality of learning models based on the data sample set, wherein the selection criterion includes the minimum uncertainty measure among the corresponding uncertainty measures.

7. The first device of claim 6, wherein the corresponding uncertainty measure includes at least one of the marginal entropy of the plurality of outputs or the conditional entropy of the plurality of outputs, which is at least partially based on the data sample set.

8. The first device of claim 1, wherein, in order to generate the corresponding confidence score, the at least one processor is operable to cause the first device to generate, at least in part, a corresponding first uncertainty measure associated with the plurality of outputs from the plurality of learning models and a corresponding second uncertainty measure associated with the plurality of outputs from the plurality of learning models based on the data sample set, wherein the selection criterion includes the maximum difference between the corresponding first uncertainty measure and the corresponding second uncertainty measure.

9. The first device of claim 1, wherein, in order to generate the corresponding confidence score, the at least one processor is operable to cause the first device to generate, at least in part, a corresponding first weighted uncertainty measure associated with the plurality of outputs from the plurality of learning models and a corresponding second weighted uncertainty measure associated with the plurality of outputs from the plurality of learning models, based at least in part on the data sample set, wherein the corresponding first weighted uncertainty measure and the corresponding second weighted uncertainty measure are weighted at least in part based on values ​​between zero and one.

10. The first device of claim 1, wherein, in order to generate the corresponding confidence scores, the at least one processor is operable to cause the first device to generate information gain associated with the plurality of outputs from the plurality of learning models, at least in part, based on the set of data samples.

11. The first device of claim 1, wherein, in order to execute the learning model, the at least one processor is operable to cause the first device to: The output of the learning model is based at least in part on a prediction of at least one of a message, a symbol set, or a bit set, which is provided as input to the learning model, wherein the data sample set includes one or more channel output signals. The prediction of at least one beam for transmitting or receiving signaling is generated as the output from the learning model, at least in part, based on providing the data sample set as input to the learning model, wherein the data sample set includes at least one of a set of beam measurements or auxiliary information associated with one or more beams; The output of the learning model is at least in part based on the prediction of at least one beam for transmitting or receiving signaling in a future time slot set, generated by providing the data sample set as input to the learning model, wherein the data sample set includes at least one of a set of beam measurements or auxiliary information associated with one or more beams; or The prediction of Channel State Information (CSI) generated by providing the data sample set as input to the learning model is at least partially based on the prediction of the CSI from the learning model, wherein the data sample set includes at least one of a reference signal measurement set or a channel vector set, and wherein the CSI includes at least one of a Channel Quality Indicator (CQI), a Precoding Matrix Indicator (PMI), a Modulation and Coding Scheme (MCS) Index, or a Rank Indicator associated with a time-frequency resource set.

12. The first device of claim 11, wherein the set of channel vectors is associated with a first subset of time-frequency resources in the set of time-frequency resources, and wherein the output from the learning model is associated with a second subset of time-frequency resources in the set of time-frequency resources or the set of time-frequency resources.

13. The first device according to claim 1, wherein the first device is a network equipment (NE).

14. The first device according to claim 1, wherein the first device is a user equipment (UE).

15. A processor comprising: At least one controller coupled to at least one memory and operable to cause the processor to: A set of data samples is obtained and used as input to multiple learning models stored at the processor; At least in part, confidence scores are generated based on providing the data sample set as input to the plurality of learning models, and are associated with the plurality of outputs from the plurality of learning models. as well as The learning model is selected for execution at the processor or device, at least in part, based on the confidence score associated with the output of one of the plurality of learning models satisfying the selection criteria.

16. A method performed by a first device, the method comprising: A set of data samples is obtained and used as input to multiple learning models stored at the first device; At least in part, confidence scores are generated based on providing the data sample set as input to the plurality of learning models, and are associated with the plurality of outputs from the plurality of learning models. as well as The learning model is selected for execution at the first device or the second device, at least in part, based on the confidence score associated with the output of one of the plurality of learning models satisfying the selection criteria.

17. The method of claim 16, further comprising transmitting a message instructing the learning model to the second device.

18. The method of claim 16, wherein obtaining the data sample set comprises receiving a message indicating the data sample set from the second device.

19. The method of claim 16, wherein obtaining the data sample set comprises: Receive multiple reference signals from the second device; as well as Obtain information associated with the plurality of reference signals, wherein the information includes one or more of the following: reference signal received power (RSRP) measurement, reference signal received quality (RSRQ) measurement, channel quality indicator (CQI), precoding matrix indicator (PMI), modulation and coding scheme index (MCS) or rank indicator.

20. The method of claim 16, further comprising receiving from the second device a message that triggers the selection of the learning model.

Images