Machine learning based power control

Abstract

Devices, methods, apparatuses, and computer-readable storage media for machine learning based power control are provided. In one method, a first apparatus transmits, to a second apparatus, a configuration for transmitting a transmit power control (TPC), the configuration indicating a set of power adjustment values specific to a machine learning model. The first apparatus determines, from the set of power adjustment values, a power adjustment value for the second apparatus by applying the machine learning model. The first apparatus transmits, to the second apparatus, a first TPC command including an indication of the determined power adjustment value.

Description

Cross-references to related applications

[0001] This application claims priority and benefit to EP patent application No. 25168248.0, filed on April 3, 2025, the contents of which are incorporated herein by reference in their entirety. Technical Field

[0002] The various example embodiments disclosed herein relate generally to the telecommunications field, and more specifically to devices, methods, apparatuses, and computer-readable storage media for machine learning-based power control. Background Technology

[0003] Transmit Power Control (TPC) commands are instructions sent from the base station to the User Equipment (UE) to adjust the UE's transmit power. TPC commands can help maintain optimal signal quality and reduce interference by dynamically increasing or decreasing the UE's transmit power based on channel conditions and network requirements. Summary of the Invention

[0004] In a first aspect of this disclosure, a first apparatus is provided. The first apparatus includes: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the first apparatus to at least: send a configuration for transmit power control (TPC) to a second apparatus, the configuration indicating a set of power adjustment values ​​specific to a machine learning model; determine a power adjustment value for the second apparatus from the set of power adjustment values ​​by applying the machine learning model; and send a first TPC command to the second apparatus including an indication of the determined power adjustment value.

[0005] In a second aspect of this disclosure, a second apparatus is provided. The second apparatus includes: at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the second apparatus to at least: receive from a first apparatus a configuration for transmit power control (TPC), the configuration indicating a set of power adjustment values ​​specific to a machine learning model; receive from the first apparatus the first TPC command including an indication of power adjustment values ​​in the set of power adjustment values; and adjust the transmit power of the second apparatus based on the first TPC command.

[0006] In a third aspect of this disclosure, a method is provided at a first device. The method includes: sending a configuration for transmit power control (TPC) to a second device, the configuration indicating a set of power adjustment values ​​specific to a machine learning model; determining a power adjustment value for the second device from the set of power adjustment values ​​by applying the machine learning model; and sending a first TPC command to the second device including an indication of the determined power adjustment value.

[0007] In a fourth aspect of this disclosure, a method is provided at a second device. The method includes: receiving from a first device a configuration for transmit power control (TPC), the configuration indicating a set of power adjustment values ​​specific to a machine learning model; receiving from the first device the first TPC command including an indication of power adjustment values ​​in the set of power adjustment values; and adjusting the transmit power of the second device based on the first TPC command.

[0008] In a fifth aspect of this disclosure, a first apparatus is provided. The first apparatus includes: means for transmitting to a second apparatus a configuration for transmit power control (TPC), the configuration indicating a set of power adjustment values ​​specific to a machine learning model; means for determining a power adjustment value for the second apparatus from the set of power adjustment values ​​by applying the machine learning model; and means for transmitting to the second apparatus a first TPC command including an indication of the determined power adjustment value.

[0009] In a sixth aspect of this disclosure, a second apparatus is provided. The second apparatus includes: means for receiving from a first apparatus a configuration for transmit power control (TPC), the configuration indicating a set of power adjustment values ​​specific to a machine learning model; means for receiving from the first apparatus the first TPC including an indication of the power adjustment values ​​in the set of power adjustment values; and means for adjusting the transmit power of the second apparatus based on the first TPC command.

[0010] In a seventh aspect of this disclosure, a computer-readable medium is provided. The computer-readable medium includes instructions stored thereon for causing a device to perform at least the methods according to the third and fourth aspects.

[0011] It should be understood that the summary portion is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0012] Some example implementation schemes will now be described with reference to the accompanying drawings, in which: Figure 1 An example communication environment is shown in which an example implementation of the present disclosure is possible; Figure 2 The signaling flow of an example process for machine learning-based power control according to some embodiments of this disclosure is shown; Figure 3 An example flowchart of an example process for machine learning-based power control according to some example embodiments of the present disclosure is shown; Figure 4An example flowchart of an example process for machine learning-based power control according to some example embodiments of the present disclosure is shown; Figure 5 An example flowchart of an example process for machine learning-based power control according to some example embodiments of the present disclosure is shown; Figure 6 A schematic diagram illustrating the trajectory of average user throughput for different methods according to some example embodiments of this disclosure is shown; Figure 7 A schematic diagram of a power distribution shaped by a machine learning model according to some example embodiments of the present disclosure is shown; Figure 8 A schematic diagram of power distribution in a simulated scenario according to some example embodiments of the present disclosure is shown; Figure 9 A schematic diagram illustrating the relationship between geometric mean throughput and P0 according to some example embodiments of this disclosure is shown; Figure 10 A schematic diagram of the cumulative distribution function of uplink user throughput according to some example embodiments of this disclosure is shown; Figure 11 A schematic diagram illustrating the relationship between the geometric mean throughput and different value tables for TPC commands according to some example embodiments of this disclosure is shown. Figure 12 A flowchart is shown showing an example method implemented at a first device according to some example embodiments of the present disclosure; Figure 13 A flowchart is shown illustrating an example method implemented at a second device according to some example embodiments of the present disclosure; Figure 14 A simplified block diagram of an apparatus suitable for implementing an example embodiment of this disclosure is shown; and Figure 15 A block diagram of an example computer-readable medium according to some example embodiments of the present disclosure is shown.

[0013] In all the accompanying drawings, the same or similar reference numerals denote the same or similar elements. Detailed Implementation

[0014] The principles of this disclosure will now be described with reference to some example embodiments. It should be understood that these embodiments are described for illustrative purposes only and to assist those skilled in the art in understanding and implementing this disclosure, and do not imply any limitation on the scope of this disclosure. The embodiments described herein can be implemented in various ways other than those described below.

[0015] In the following description and claims, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0016] References to "an embodiment," "an embodiment," "an example embodiment," etc., in this disclosure indicate that the described embodiment may include a particular feature, structure, or characteristic, but not every embodiment is required to include that particular feature, structure, or characteristic. Furthermore, such phrases do not necessarily refer to the same embodiment. Additionally, when a particular feature, structure, or characteristic is described in connection with an embodiment, it is believed that in conjunction with other embodiments, whether explicitly described or not, influencing such feature, structure, or characteristic is within the knowledge of those skilled in the art.

[0017] It should be understood that although terms such as "first," "second," etc., preceding nouns may be used herein to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish one element from another, and they do not restrict the order of nouns. For example, without departing from the scope of the example embodiments, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element. As used herein, the term "and / or" includes any and all combinations of one or more of the listed terms.

[0018] As used herein, “at least one of the following: a list of two or more elements” and “at least one of the following: a list of two or more elements” and similar wording (where the list of two or more elements is connected by “and” or “or”) means at least any one of the elements, or at least any two or more of the elements, or at least all of the elements.

[0019] As used herein, unless explicitly stated otherwise, the execution of the step “in response to A” or “in response to A” does not indicate that the step is performed immediately after “A” occurs, and may include one or more intermediate steps.

[0020] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments. As used herein, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that, when used herein, the terms “comprises,” “comprising,” “has,” “having,” “includes,” and / or “including” specify the presence of the stated features, elements, and / or components, but do not exclude the presence or addition of one or more other features, elements, components, and / or combinations thereof.

[0021] As used in this application, the term "circuit" may refer to one or more of the following: (a) Hardware circuit implementation only (such as implementation in analog and / or digital circuits only) and (b) A combination of hardware circuitry and software, such as (if applicable): (i) A combination of analog and / or digital hardware circuitry with software / firmware, and (ii) Any part of a hardware processor having software (including digital signal processors, software, and memory that work together to enable devices such as mobile phones or servers to perform various functions), and (c) (Multiple) hardware circuits and / or (multiple) processors, such as (multiple) microprocessors or a portion thereof, which require software (e.g., firmware) to operate, but may not exist when the software is not required to operate.

[0022] This definition of "circuit" applies to all uses of the term in this application (including in any claim). As another example, as used herein, the term "circuit" also encompasses only hardware circuitry or a processor (or processors) or a portion thereof, and its accompanying software and / or firmware implementation. The term "circuit" also encompasses, for example and if applicable to a particular claim element, baseband integrated circuits or processor integrated circuits for mobile devices, or similar integrated circuits in servers, cellular network devices, or other computing or network devices.

[0023] As used herein, the term "communication network" refers to a network that conforms to any suitable communication standard, such as New Radio (NR), Long Term Evolution (LTE), LTE-A Advanced (LTE-A), Wideband Code Division Multiple Access (WCDMA), High-Speed ​​Packet Access (HSPA), Narrowband Internet of Things (NB-IoT), etc. Furthermore, communication between user equipment and network equipment in a communication network can be performed according to any suitable generation of communication protocol, including but not limited to first-generation (1G), second-generation (2G), 2.5G, 2.75G, third-generation (3G), fourth-generation (4G), 4.5G, fifth-generation (5G), 5G Advanced, sixth-generation (6G) communication protocols, wireless local network communication protocols (such as IEEE 802.11), and / or any other currently known or future-developed protocols. Furthermore, communication can utilize any suitable wireless communication technology, including but not limited to: Code Division Multiple Access (CDMA), Frequency Division Multiple Access (FDMA), Time Division Multiple Access (TDMA), Frequency Division Duplex (FDD), Time Division Duplex (TDD), Multiple Input Multiple Output (MIMO), Orthogonal Frequency Division Multiple Access (OFDM), Discrete Fourier Transform Extended OFDM (DFT-s-OFDM), and / or any other currently known or future-developed technology. Embodiments of this disclosure can be applied to a wide variety of communication systems. Given the rapid development of communication, there will naturally be future types of communication technologies and systems that embody the nature of this disclosure. The scope of this disclosure should not be construed as limited to the aforementioned systems.

[0024] As used herein, the term "network device" refers to a node in a communications network through which user equipment accesses the network and receives services. Network devices can refer to base stations (BS) or access points (APs), such as Node B (NodeB or NB), evolved Node B (eNodeB or eNB), NR NB (also known as gNB), Remote Radio Unit (RRU), Radio Header End (RH), Remote Radio Header End (RRH), relay, Integrated Access and Backhaul (IAB) node, low-power nodes (such as femtoseconds, picoseconds), non-terrestrial network (NTN) or non-terrestrial network equipment (such as satellite network equipment), low Earth orbit (LEO) satellites and geostationary Earth orbit (GEO) satellites, spacecraft network equipment, etc., depending on the terminology and technology applied. In some example implementations, the Radio Access Network (RAN) split architecture includes a centralized unit (CU) and a distributed unit (DU) at the IAB donor node. The IAB node includes a mobile terminal (IAB-MT) portion that behaves as a UE toward the parent node, and the DU portion of the IAB node behaves as a base station toward the next-hop IAB node.

[0025] The term "user equipment" refers to any terminal device capable of wireless communication. By way of example and not limitation, user equipment may also be referred to as user equipment (UE), subscriber station (SS), portable subscriber station, mobile station (MS), or access terminal (AT). User equipment can include, but is not limited to, mobile phones, cellular phones, smartphones, Voice over IP (VoIP) phones, wireless local loop phones, tablets, wearable user devices, personal digital assistants (PDAs), portable computers, desktop computers, image capture user devices (such as digital cameras), gaming user devices, music storage and return devices, vehicular wireless user devices, wireless endpoints, mobile stations, LEE (Laptop embedded device), LME (Laptop mounted device), Universal Serial Bus (USB) dongles, smart devices, wireless customer premises equipment (CPE), Internet of Things (IoT) devices, watches or other wearable devices, head-mounted displays (HMDs), vehicles, drones, medical devices and applications (e.g., remote surgery), industrial devices and applications (e.g., robots and / or other wireless devices operating in the context of industrial and / or automated processing chains), consumer electronics devices, devices operating on commercial and / or industrial wireless networks, etc. User equipment may also correspond to the mobile terminal (MT) portion of an IAB node (e.g., a relay node). In the following description, the terms "user equipment," "terminal equipment," "terminal," "user equipment," and "UE" are used interchangeably.

[0026] As used herein, the terms “resource,” “transmission resource,” “resource block,” “physical resource block” (PRB), “uplink resource,” or “downlink resource” can refer to any resource used to perform communication, such as communication between a user equipment and a network device, including resources in the time domain, frequency domain, spatial domain, code domain, or any other combination of time, frequency, spatial, and / or code domain resources used to achieve communication. In the following, unless explicitly stated otherwise, resources in the frequency and time domains will be used as examples of transmission resources used to describe some exemplary embodiments of this disclosure. Note that exemplary embodiments of this disclosure are equally applicable to other resources in other domains.

[0027] The term "Artificial Intelligence (AI) / Machine Learning (ML) Model" refers to a data-driven algorithm that applies AI / ML techniques to generate a set of outputs based on a set of inputs. The term "AI / ML Model Delivery" is a general term related to the delivery of AI / ML models from one entity to another in any way. Note that an entity can refer to a network node / function (e.g., gNB, Location Management Function (LMF), etc.), user equipment, proprietary servers, etc. The term "AI / ML Model Inference" refers to the process of using a trained AI / ML model to produce a set of outputs based on a given set of inputs.

[0028] The term "AI / ML model testing" refers to the training sub-process that evaluates the performance of a final AI / ML model using a dataset different from the dataset used for model training and validation. Unlike AI / ML model validation, AI / ML model testing does not assume subsequent adjustments to the model. The term "AI / ML model training" refers to the process of training an AI / ML model in a data-driven manner (by learning input / output relationships) and obtaining a trained AI / ML model for inference. The term "AI / ML model transfer" refers to delivering an AI / ML model over the air interface in a manner opaque to 3GPP signaling, whether the parameters of the model structure are known at the receiving end or it is a new model with parameters. Delivery may contain a complete model or a partial model. The term "AI / ML model validation" refers to the training sub-process that evaluates the quality of an AI / ML model using a dataset different from the dataset used for model training, which helps in selecting model parameters that generalize to datasets other than those used for model training.

[0029] The term "data collection" refers to the process by which network nodes, management entities, or user devices collect data for use in AI / ML model training, data analysis, and inference. The term "federated learning / federated training" refers to an ML technique that trains AI / ML models across multiple decentralized edge nodes (e.g., user devices, gNBs), with each decentralized edge node performing local model training using local data samples. This technique requires multiple interactions between the models but does not require the exchange of local data samples.

[0030] The term "model download" refers to the transfer of a model from a network device to a user device. The term "model identification" refers to the process / method of identifying an AI / ML model for mutual understanding between network devices and user devices. Note that the process / method of model identification may or may not be applicable. Note that information about the AI / ML model may be shared during model identification. The term "model monitoring" refers to the process of monitoring the inference performance of an AI / ML model. The term "model update" refers to the process of updating model parameters and / or the model structure. The term "model upload" refers to the transfer of a model from a user device to a network device.

[0031] The term "network-side AI / ML model" refers to an AI / ML model whose inference is performed entirely at the network device. The term "UE-side AI / ML model" refers to an AI / ML model whose inference is performed entirely at the user equipment. The term "dual-side AI / ML model" refers to a paired AI / ML model on which joint inference is performed, where joint inference includes AI / ML inference jointly performed across the user equipment and network device; that is, the first part of the inference is performed by the UE, and then the remaining part is performed by the gNB, or vice versa.

[0032] The term "offline training" refers to an AI / ML training process where a model is trained based on a collected dataset, and the trained model is later used or delivered for inference. Note that this definition is for guidance only. There may be cases that do not perfectly fit this definition but can still be classified as offline training by generally accepted conventions. The term "online training" refers to an AI / ML training process where the model used for inference is trained (nearly) in real-time as new training samples arrive. Note that the concepts of (near) real-time and non-real-time are context-dependent and relative to the inference timescale. It should also be noted that this definition is for guidance only. There may be cases that do not perfectly fit this definition but can still be classified as online training by generally accepted conventions. It should also be noted that fine-tuning / retraining can be done via either online or offline training.

[0033] The term "reinforcement learning (RL)" refers to the process of training an AI / ML model in an environment where the model interacts with it, using inputs (called states) and feedback signals (called rewards) generated by outputs (also called actions). The term "semi-supervised learning" refers to the process of training a model using a mixture of labeled and unlabeled data. The term "supervised learning" refers to the process of training a model from inputs and their corresponding labels. The term "unsupervised learning" refers to the process of training a model without labeled data.

[0034] Figure 1 An example communication environment 100 in which an example implementation of this disclosure can be carried out is shown. For example... Figure 1 As shown, the communication environment 100 may include multiple communication devices, including a first device 110 and a second device 120. The first device 110 may function as a network device (e.g., a BS or gNB), and the second device 120 may operate as a terminal device (e.g., a UE).

[0035] The service area provided by the first device 110 is referred to as cell 102. The second device 120 can communicate with the first device 110 within cell 102. Cell 102 currently serving the second device 120 can be considered the serving cell.

[0036] It should be understood that Figure 1 The number or type of devices or equipment and their connections shown are given for illustrative purposes and do not imply any limitation. Communication environment 100 may include any suitable number or type of devices or equipment configured to implement the exemplary embodiments of this disclosure. Although not shown, it should be understood that one or more additional terminal devices may be located in cell 102, and / or one or more additional cells may be provided by first device 110.

[0037] In some example implementations, if the first device 110 is a network device or is included in a network device, and the second device 120 is a terminal device or is included in a terminal device, then the link from the second device 120 to the first device 110 is referred to as an uplink (UL), and the link from the first device 110 to the second device 120 is referred to as a downlink (DL). In the DL, the first device 110 is a transmitting (TX) device (or transmitter), and the second device 120 is a receiving (RX) device (or receiver). In the UL, the second device 120 is a TX device, and the first device 110 is an RX device.

[0038] In communication systems, there are two types of TPC commands: cumulative TPC commands and absolute TPC commands. Cumulative TPC commands can increase or decrease the cumulative Physical Uplink Shared Channel (PUSCH) power control adjustment state for a terminal device (e.g., a UE) with transmit power values ​​between [-1, 0, 1, 3] dB. Cumulative TPC commands use a cumulative method, where each new command is added to or subtracted from the previous power control adjustment state. For example, if the previous cumulative power control adjustment state was 1 dB and a new cumulative TPC command indicates an additional 2 dB, the current cumulative power control adjustment state becomes 3 dB.

[0039] Absolute TPC commands can increase or decrease the absolute PUSCH power control adjustment state of the terminal device's transmission power by values ​​between [-4, -1, 1, 4] dB. Absolute TPC commands directly specify the absolute power control adjustment state, such as -4 dB or 1 dB. Some example TPC command values ​​are provided in Table 1 below: Table 1

[0040] Table 1 shows the absolute value and summation of the TPC command field in the downlink control information (DCI) format for scheduling PUSCH transmissions, or in DCI format 2_2 or DCI format 2_3 with cyclic redundancy check (CRC) scrambled by the TPC-PUSCH-Radio Network Temporary Identifier (RNTI). value or Value mapping. The TPC command field is a field in the DCI format that carries instructions for adjusting the transmit power of the second device 120. As shown in Table 1, cumulative TPC command values ​​include -1, 0, 1, and 3. Absolute command values ​​include -4, -1, 1, and 4.

[0041] The example implementation of this disclosure proposes applying a machine learning model to determine power adjustment values ​​based on TPC command values, allowing the UE to adjust its transmit power. However, the TPC command values ​​provided in Table 1 may not always be optimal for machine learning-based power control, as there may be a bias in the direction of adjustment toward the terminal device's transmit power. Furthermore, conventional power control algorithms struggle to outperform open-loop power control (OLPC) parameters obtained through exhaustive search, and cell edge power enhancement remains their primary application.

[0042] Furthermore, a key challenge in power control is eliminating the need for slow, system-wide tuning of OLPC parameters (P0, path loss a) via RRC signaling. Machine learning-based power control can optimize power levels individually to achieve higher throughput and better uplink distribution, but previous models relied too heavily on closed-loop power control (CLPC) to adjust states, thus limiting generalization.

[0043] In some example implementations, for OLPC, AI / ML models can be used to achieve near-optimal performance in selecting, for example, sufficient P0 and path loss α for power control. Furthermore, cell edge enhancement algorithms using reinforcement learning can be used for CLPC.

[0044] Next-generation communication systems define TPC commands for increasing or decreasing the transmit power of terminal devices. However, the currently defined values ​​(e.g., those defined in Table 1) may not always be optimal for machine learning. Furthermore, any power control algorithm struggles to outperform an exhaustive search of open-loop power control parameterization. TPC commands can only be used to increase cell edge power as needed.

[0045] Furthermore, one issue to address is eliminating the need for dynamic optimization of open-loop power control parameters such as P0 (the parameter for nominal power offset) and path loss α (the parameter for path loss compensation factor). Higher throughput gains can be achieved by individually adjusting the power level for each end device, as machine learning can more accurately adjust (or regulate) the UL power distribution within a cell or network. The problem with adjusting P0 and path loss α is that it is a slow and cumbersome process via system information updates or RRC signaling. Therefore, adjusting P0 and path loss α is not suitable for machine learning (e.g., reinforcement learning) and rapid power level regulation for end devices. However, it is worth noting that even with machine learning-based CLPC, using P0 that at least indicates a reasonable range of the transmission power of the end device is beneficial.

[0046] Based on some example implementations, a solution for machine learning-based power control is proposed. In some example implementations, a first device (e.g., a network device) sends a configuration for transmit power control (TPC) to a second device (e.g., a terminal device), the configuration indicating a set of power adjustment values ​​specific to a machine learning model. The first device determines a power adjustment value for the second device from the set of power adjustment values ​​by applying the machine learning model. Subsequently, the first device sends a first TPC command to the second device, which includes an indication of the determined power adjustment value.

[0047] Using these embodiments of the present disclosure, a set of power adjustment values ​​suitable for a machine learning model can be configured for the second device. The optimal power adjustment value for the second device can then be determined from this set using the machine learning model, which maintains optimal signal quality at the receiver and ensures reliable communication. In this way, applying the machine learning model can optimally shape the uplink power distribution across the second device to maximize cell or network-level uplink performance.

[0048] Figure 2 The signaling flow of an example process 200 for machine learning-based power control according to some embodiments of this disclosure is illustrated. Process 200 involves referencing... Figure 1 The first device 110 and the second device 120 are described.

[0049] like Figure 2 As shown, the first device 110 sends a configuration (205) for the TPC to the second device. This configuration indicates a set of power adjustment values ​​specific to the machine learning model. Accordingly, the second device 120 receives (210) the configuration for the TPC. Different sets of power adjustment values ​​may exist for different machine learning models. The set of power adjustment values ​​specific to the machine learning model may also be referred to as the action space for the machine learning model. In some examples, the machine learning model may be configured to determine the power adjustment values ​​from its action space.

[0050] In some example implementations, the first device 110 may be or can be included in a network device (e.g., gNB or BS), and the second device may be or can be included in a terminal device (e.g., UE).

[0051] In some example implementations, the configuration of the power adjustment value set can be sent via system information or Radio Resource Control (RRC) signaling. In some examples, the power adjustment value set can be pre-specified for the network provider, and the network provider can select the power adjustment value set suitable for a machine learning model. Furthermore, the power adjustment value set can be configured with an identifier within the system information or RRC configuration. The configuration can be sent via other signaling or messages. In examples, the power adjustment value set can specify a table (e.g., a value table) for TPC commands, where the table can be in the format of the aforementioned Table 1, and the relationship between the index and the power adjustment value set is stored in the table. In some example implementations, the first device 110 can send, for example, the preferred number of rows in the table and the value of each column to the second device 120 via system information or RRC signaling.

[0052] In some example implementations, the set of power adjustment values ​​may include a zero value as well as an equal number of positive and negative values. In some examples, one value in the set of power adjustment values ​​may indicate 0 dB, and the set of power adjustment values ​​may also include an equal number of positive and negative values ​​with the same magnitude. For example, the set of power adjustment values ​​may be [-3, 0, 3], [-2, 0, 2], or [-3, -2, -1, 0, 1, 2, 3]. The unit of measurement for the set of power adjustment values ​​may be dB. The set of power adjustment values ​​is other similar combinations of dB values ​​whose sum is zero. In this way, deviations in certain adjustment directions can be avoided.

[0053] After sending (205) the configuration for the TPC, the first device 110 determines (215) the power adjustment value for the second device 120 from a set of power adjustment values ​​by applying a machine learning model. For example, if the set of power adjustment values ​​is [-3, 0, 3], then 3 can be determined by the machine learning model. The determined power adjustment value can be used to adjust the transmission power of the second device 120.

[0054] The transmission power (e.g., PUSCH transmission power) of the second device 120 can be defined as follows:

[0055] in It is open-loop power control parameterization, and It is the closed loop part. This represents the maximum allowed transmission power of carrier f and bandwidth portion c in subframe i. This indicates the power offset value (e.g., P0) for PUSCH transmission. This represents the number of resource blocks allocated for the PUSCH transport. This represents the path loss compensation factor (e.g., path loss a) of the second device 120. This represents the path loss estimate for the second device 120. This indicates the power adjustment value. This indicates a PUSCH power control adjustment state (also known as a CLPC adjustment state, e.g., a cumulative adjustment state or an absolute adjustment state) based on the specific needs of transmission in subframe i and layer l, which adjusts the transmission power of the second device 120.

[0056] In some examples, machine learning models can be used to adjust the PUSCH power control adjustment state based on a determined power adjustment value. In one example, regarding cumulative TPC commands, if the previous PUSCH power control adjustment state was 1dB and the determined power adjustment value was 2dB, the current PUSCH power control adjustment state could become 3dB. In another example, regarding absolute TPC commands, regardless of the previous PUSCH power control adjustment state, if the determined power adjustment value was 2dB, the current PUSCH power control adjustment state could become 2dB. By applying machine learning models, uplink power distribution can be optimally shaped even with unoptimized or deliberately aggressive open-loop power control parameterization.

[0057] In some example implementations, the first device 110 may determine a power adjustment value by applying a machine learning model based on at least one of the following inputs: the number of resource blocks to be allocated, a measurement or estimate of the UL signal-to-interference-plus-noise ratio (SINR), the transmit power of the second device 120, and a measurement or estimate of the reference signal received power (RSRP) of the UL reference signal. In one example, the transmit power of the second device 120 may be calculated using the cumulative adjustment state. The transmit power may be the sum of the basic uplink transmit power of the second device and the cumulative adjustment state. The basic uplink transmit power of the second device 120 may be calculated using a finite number of parameters, such as the value of P0, path loss α, path loss estimation, etc. It should be noted that the above inputs are merely examples of equivalent values ​​that the first device 110 or the second device 120 may obtain for utilizing the machine learning model. Therefore, new signaling and implementations of the first device 110 are required to obtain the inputs for the machine learning model.

[0058] The machine learning model can be trained on different entities. In some example implementations, the machine learning model can be trained offline by a network provider, and the trained machine learning model can be provided to the first device 110. Alternatively, the machine learning model can be trained online by the first device 110. In some examples, the inputs to the machine learning model used in model training are the same as the inputs used in model inference. By using the total transmitted power (including the accumulated adjusted state) instead of just the accumulated adjusted state as input during model training, the machine learning model can learn power control more effectively with fewer training samples and minimal tuning.

[0059] In model training, it has been observed that inputs such as path loss, which may be related to optimal power but are unaffected by power adjustment, are detrimental to training generalized machine learning models. For example, in the experiments conducted in this disclosure, the performance of a machine learning model trained with path loss may not improve if the value of P0 is changed. However, the performance of a machine learning model trained without path loss can be improved by utilizing other values ​​of P0.

[0060] In some example implementations, the machine learning model may include a reinforcement learning (RL) model. In the example, the RL model may be built based on distributed soft actor-judge discrete (DSACD), which can be used to tune the optimal power level of the second device 120. DSACD can be further improved with Munchausen RL. It should be noted that other machine learning algorithms may also be used. By using DSACD, it is possible to train an ultra-lightweight actor model for real-time computation and distributed learning during training using automatic entropy control. Distributed learning with a lightweight actor model is made possible through an actor-judge design, where distributed learning occurs only on the judge's side. Automatic entropy control eliminates the need for manual exploration and fine-tuning. Therefore, it is well-suited for tuning UL power control without requiring extensive hyperparameter tuning or any specially designed exploration strategy.

[0061] Similar to other off-policy deep RL algorithms, DSACD learns by sampling mini-batches of states, actions, rewards, and next-state experiences from its experience replay buffer. For replay memory sampling, prioritization can be used, which prioritizes new samples and utilizes two judging networks to create samples with high average estimation error. The example state space is the aforementioned input to the machine learning model, such as the number of resource blocks to be allocated, the measurement or estimate of UL SINR, etc. The example action space is the set of power adjustment values ​​in Table 1 or configured by the first device 110.

[0062] The original Munchausen RL instructs how to apply Munchausen RL to algorithms derived from Deep Q Networks (DQN). This modification involves adding a scaled logarithmic policy to the immediate reward. However, based on experiments conducted in this disclosure, this modification does not work for algorithms based on the Soft Actor Critics (SAC).

[0063] To leverage the Munchhausen RL and improve its convergence with the Soft Actor Critics (SAC), a scaled log-shearing policy can be added to the policy objective function. Therefore, in the case of (distributed) SAC, all Huber losses... Discrete modification error of quantiles as follows:

[0064] in Representing state Q value, action , Indicates the state , China takes action The instant rewards received This represents a discount factor for future rewards. Indicates the next state The policy function, the next action , Indicates the next state Q value, next action , Indicates the next state and the next action The update strategy function. In some examples, and It can be set to -1. It can be the learning temperature coefficient of SAC, and R This is the scaling factor for the logarithmic strategy. In the example, It can be set to 0.9. It can be observed that adding a scaled, pruned log-based policy to the policy loss function is beneficial. Therefore, the policy loss function for all actions is similarly modified using an array of scaled, pruned log-based policy values ​​as follows:

[0065] in Representing state The policy function below, and It represents the expected reward for taking a specific action in a given state.

[0066] Learning from a multi-step guided objective with an additional step is achieved as the following evaluation learning:

[0067] superscript and These represent time steps t+1 and t+2, respectively. This helps in learning the long-term effects of transmit power variations used for power control. For example, it allows us to learn how neighboring cells react and how interference increases or decreases.

[0068] It should be noted that the proposed Munchausen RL modification is also performed on the policy objective function, i.e., it does not only add a scaled logarithmic policy to the reward. The Munchausen RL modification can also be applied to distributed or non-distributed soft actor-judgment discrete (SACD) scenarios.

[0069] In some example implementations, rewards can be used during the training of a machine learning model (e.g., an RL model). The RL model can be trained based on a training objective configured to increase or maximize the reward. In some examples, the geometric mean of throughput increments can be used as the reward. Alternatively, SINR or other throughput-based metrics, packet latency, energy efficiency, etc., can be used as the reward.

[0070] The goal of the machine learning model is to maximize long-term future rewards. When the power of neighboring cells also increases, the machine learning model can learn that excessively high UL power can lead to high interference. On the other hand, too low UL power can lead to poor spectral power density. In the simulations conducted by this disclosure, it is assumed that a centralized entity collects training samples from multiple base stations into its experience replay memory (as an example of the first device 110). All base stations use the same machine learning model (e.g., an actor model) to make decisions during training. The trained model is then provided to all base stations after training. Therefore, real-time execution only requires inference of the machine learning model, and training is not time-critical. During the model inference phase, all base stations can use the machine learning model independently without further coordination. It is anticipated that for real-world network use, the machine learning model can be pre-trained in simulations (e.g., at a network provider) and fine-tuned within the real network (e.g., at the first device 110). For any suitable central processing unit (CPU), the inference time of the machine learning model can be within microseconds.

[0071] In some examples, the update interval of the cumulative TPC command can be adjusted based on available computing resources. A machine learning model is applied to determine whether the power adjustment value can be executed at each transmission time interval (TTI) or in more sparser time periods. If sparse adjustment or learned periodicity is used, 0 dB cumulative power can be indicated within the DCI during periods when the ML model is not inferred.

[0072] To avoid exceeding the power limit of the second device 120 during the model training phase, action masking can be used to mask some power adjustment values ​​that might cause the transmission power of the second device 120 to exceed the specified maximum transmission power (e.g., approximately 23 dBm) or fall below the minimum transmission power, where the adjustment of the transmission power is meaningless. Therefore, the output of the machine learning model (e.g., power adjustment values) that causes the transmission power of the second device 120 to reach the maximum or minimum transmission power can be invalid, and the output can be selected from the remaining options (e.g., the remaining values ​​in the set of power adjustment values).

[0073] Furthermore, the measured UL SINR can be used to limit excessive increases or decreases in the transmission power of the second device 120. For example, if the SINR is less than, say, -7 dB, the second device 120 may not achieve much data throughput. If the SINR is greater than 35 dB, it can be considered whether such high power is wasting energy, even if the second device 120 may not cause much interference.

[0074] After determining (215) the power adjustment value, the first device 110 sends (220) a first TPC command to the second device 120, which includes an indication of the determined power adjustment value. Accordingly, the second device 120 receives (225) the first TPC command. The first TPC command can be configured to instruct the second device 120 to adjust its transmission power.

[0075] In some example implementations, the first TPC command may include a cumulative TPC command and / or an absolute TPC command. The power adjustment value set may include a set of cumulative values ​​for the cumulative TPC command and / or a set of absolute values ​​for the absolute TPC command. The cumulative TPC command may utilize the set of cumulative values ​​to increase or decrease the cumulative adjustment state of the transmission power of the second device 120. The absolute TPC command may utilize the set of absolute values ​​to increase or decrease the absolute adjustment state of the transmission power of the second device 120. In this way, the transmission power of the second device 120 can be adjusted to any value using the first TPC command.

[0076] In some example implementations, the first TPC command can be sent via DCI. The first TPC command can also be sent via other signaling. The number of bits indicating the first TPC command within the DCI can be based on the minimum number of bits required to represent the power adjustment value set. For example, if the minimum number of bits required to represent the power adjustment value set is 3, then the number of bits indicating the first TPC command within the DCI is at least 3. In this way, if the power adjustment value set is configured by the first device 110, the DCI format used for PUSCH scheduling can support TPC commands of different bit counts.

[0077] In some example implementations, the first TPC command may include an index for indexing the determined power adjustment value from a set of power adjustment values. In one example, the set of power adjustment values ​​is represented by a value table, which stores the relationship between the index and the set of power adjustment values. For example, the index corresponding to the value -3 is 1, the index corresponding to the value 0 is 2, and so on. If the determined power adjustment value is -3, then the index used to index that power adjustment value is 1. The first TPC command may include index 1.

[0078] Upon receiving (225) the first TPC command, the second device 120 adjusts (230) its transmission power based on the first TPC command. In some examples, the first TPC command may include an index, and the second device 120 may determine an indicated power adjustment value corresponding to that index. The second device 120 may then adjust its transmission power using the indicated power adjustment value.

[0079] In some example implementations, the second device 120 may determine the cumulative adjustment state of its transmit power based on an indicated power adjustment value, and adjust its transmit power based on the cumulative adjustment state. In an example of a cumulative TPC command, the second device 120 may determine the cumulative adjustment state based on the indicated power adjustment value. In this example, if the previous cumulative adjustment state was 1 dB and the indicated power adjustment value was 2 dB, the current cumulative adjustment state could be 3 dB. The second device 120 may then adjust its transmit power based on the current cumulative adjustment state.

[0080] In addition to the machine learning model applied by the first device 110, the second device 120 may also apply a machine learning model. In some example embodiments, the second device 120 may send capability information supporting the application of the machine learning model to the first device 110. Accordingly, the first device 110 may receive the capability information. In some examples, the second device 120 may be able to apply the machine learning model and may provide the capability information to request the machine learning model from the first device 110 to determine the power adjustment value.

[0081] After receiving the capability information, the first device 110 can send a machine learning model to the second device 120 for application by the second device 120. Correspondingly, the second device 120 can receive the machine learning model. In this way, the second device 120 can autonomously adjust its transmission power by applying the machine learning model, and can avoid TPC command signaling between the first device 110 and the second device 120, thereby reducing signaling overhead.

[0082] In some example implementations, the second device 120 may determine the power adjustment value by applying a machine learning model based on at least one input. The input may be the same as that used by the first device 110. In examples, the input to the machine learning model may include the number of resource blocks to be allocated, a measurement or estimate of the UL SINR, the transmit power of the second device, and a measurement or estimate of the RSRP of the UL reference signal.

[0083] In some example implementations, the first device 110 may send an instruction to the second device 120 to stop the application of the machine learning model by the second device. Accordingly, the second device 120 may receive the instruction and, based on the instruction, stop the application of the machine learning model. In some examples, after the machine learning model has been applied by the second device, if the transmission power of the second device is incorrect (e.g., greater than the maximum transmission power or less than the minimum transmission power), the first device 110 may instruct the second device 120 to stop applying the machine learning model. For example, the second device 120 may adjust its transmission power based on system-wide adjustment of the OLPC parameter (P0, path loss a). In some example implementations, the machine learning model can be trained using segments, each segment starting from a cumulative adjustment state of 0 dB. During each event, the cumulative adjustment state can be accumulated using a certain number of TPC commands or over a certain number of TTIs. After each training segment, the cumulative adjustment state can be reset. Therefore, during the model inference phase, a new TPC command may be needed to reset the cumulative adjustment state, or the first device 110 can be configured for periodic automatic reset of the cumulative adjustment state.

[0084] In some example implementations, the first device 110 may send a second TPC command to the second device 120 to reset the cumulative adjustment state of the second device's transmit power. Accordingly, the second device 120 may receive the second TPC command and reset the cumulative adjustment state based on the second TPC command. Because the machine learning model is being trained using snippets, the second device 120 can reset the cumulative adjustment state during model inference when commanded. In this way, the performance of the machine learning model can be improved.

[0085] Alternatively, the first device 110 may send a second TPC command to the second device 120 to periodically reset the cumulative adjustment state of the transmission power of the second device. Accordingly, the second device 120 may receive the second TPC command and periodically reset the cumulative adjustment state based on the second TPC command. In the example, the first device 110 may be configured for periodic automatic reset of the cumulative adjustment state, and the second device 120 may reset the cumulative adjustment state every specific number of TTIs. In some examples, the table for the TPC command may be designed to include reset commands and a set of power adjustment values. Table 2 is an example of such a table, as shown below: Table 2

[0086] As shown in Table 2, there are positive and negative power adjustment values ​​of equal magnitude. Furthermore, a reset option is available, which allows the accumulated adjustment state to be reset, for example, after each training segment or during the model inference phase.

[0087] In some example implementations, a second TPC command is sent via DCI or Media Access Control (MAC) control element (CE). In the example in Table 2, a second TPC command is sent via DCI by indicating "3" in the TPC command field of DCI.

[0088] The use of machine learning models can be based on network load. In some example implementations, if the network load suddenly decreases, the machine learning model may not be used. The first device 110 can reset the transmit power of the second device 120 to provide high instantaneous UL throughput by using a high P0 value. When the network load increases, the first device 110 can apply the machine learning model to adjust the transmit power of the second device 120, for example, during busy periods, where interference and spectral power density optimization become more critical.

[0089] In some example implementations, the set of power adjustment values ​​used during the training process of the machine learning model differs from the set of power adjustment values ​​used during the inference process. For example, the set of power adjustment values ​​used during training might be [-3, 0, 3], while the set of power adjustment values ​​used during inference might be [-1, 0, 1]. This difference may not affect the performance of the machine learning model because machine learning models have generalization capabilities and can perform well on unseen data.

[0090] Reference Figure 3 This section introduces an example process for power control. Figure 3An example flowchart of an example process 300 for machine learning-based power control according to some example embodiments of this disclosure is shown. In this example, gNB 301 is an example of a first device 110, and UE 302 is an example of a second device 120. Figure 3 In the example, the entity for which machine learning is applied is gNB 301.

[0091] like Figure 3 As shown, at box 305, the network provider can train a machine learning model for power control, which is configured to adjust UL power (e.g., the transmit power of UE 302). Furthermore, the network provider can select a set of power adjustment values ​​(e.g., TPC command values) suitable for the machine learning model. The network provider can provide the machine learning model to gNB 301.

[0092] At box 310, gNB 301 can maintain the machine learning model for later application. At box 315, gNB 301 can send a TPC command configuration to UE 302, for example, via RRC signaling. In some examples, in the TPC command configuration, gNB 301 indicates a set of power adjustment values. The TPC command configuration can be an index of a pre-specified table of TPC command values ​​(indicating power adjustment values). Alternatively, the TPC command configuration can be an index that generates the precise dB adjustment values ​​for the desired TPC command value table.

[0093] At box 320, gNB 301 may apply a machine learning model to determine a power adjustment value for UE 302 from a set of power adjustment values ​​and send a TPC command (also referred to as a first TPC command) to UE 302 via DCI. The TPC command may include an indication of the determined power adjustment value. In the example, the TPC command may include an index for indexing the determined power adjustment value.

[0094] At block 325, UE 302 can adjust its transmit power based on TPC commands. At block 330, UE 302 can use the adjusted transmit power to transmit data to gNB 301 via PUSCH.

[0095] At box 335, gNB 301 can determine additional power adjustment values ​​for UE 302 and send a TPC command to UE 302 via DCI. The TPC command may include an indication of the additional power adjustment values. Boxes 340 and 345 are similar to boxes 325 and 330, and their details will not be repeated.

[0096] Reference Figure 4 This section introduces an example process for power control. Figure 4An example flowchart of an example process 400 for machine learning-based power control according to some example embodiments of this disclosure is shown. In this example, gNB 401 is an example of a first device 110, and UE 402 is an example of a second device 120. Figure 4 In the example, the entity for which the machine learning model is applied is UE 402.

[0097] Boxes 405 to 415 are similar to boxes 305 and 315, and their details will not be repeated. At box 420, UE 402 may send capability information about applications supporting machine learning models to gNB 401. gNB 401 needs to know whether UE 402 is capable of running machine learning models, such as machine learning models for power control or general machine learning models based on capability information.

[0098] If UE 402 is capable of running a machine learning model, then at box 425, gNB 401 can provide machine learning to UE 402. At box 430, UE 402 can apply the machine learning model to determine a power adjustment value for UE 402 from a set of power adjustment values, and adjust its transmit power based on the determined power adjustment value.

[0099] At box 435, gNB 401 can send an instruction to UE 402 to stop applying the machine learning model (e.g., fall back to conventional power control). In some examples, after UE 402 applies the machine learning model, if UE 402's transmit power is incorrect, gNB 401 can instruct UE 402 to stop applying the machine learning model and fall back to conventional power control. In the legacy power control mechanism, UE 402 can adjust its transmit power based on system-wide tuning of OLPC parameters (P0, path loss α).

[0100] Reference Figure 5 This section introduces an example process for power control. Figure 5 An example flowchart of an example process 500 for machine learning-based power control according to some example embodiments of this disclosure is shown. In this example, gNB 501 is an example of a first device 110, and UE 502 is an example of a second device 120. Figure 5 In the example, the entity for which machine learning is applied is gNB 501.

[0101] Boxes 505 to 540 are similar to boxes 310 and 345, and their details will not be repeated. At box 545, gNB 501 may send a command to UE 502 to reset the cumulative adjustment state (also known as a second TPC command), for example, after each training period or during the model inference phase. At box 550, UE 502 may reset the cumulative adjustment state based on the command.

[0102] Alternatively, UE 502 may have a configuration for periodically resetting the cumulative adjustment state. At block 555, UE 502 may periodically reset the cumulative adjustment state, for example, every certain number of TTIs.

[0103] This disclosure includes simulations to verify the benefits of the proposed solution. A table of values ​​for the TPC commands used in the simulations (as an example of a set of power adjustment values ​​specific to the machine learning model) is shown in Table 3 below: Table 3

[0104] In Table 3, reserving an additional bit for the TPC command can create options applicable to machine learning, which helps to automatically explore, accelerate the finding of the best strategy, and ultimately achieve better results, especially when utilizing pre-trained machine learning models.

[0105] Figure 6 A schematic diagram illustrating the trajectory of average user throughput for different methods according to some example embodiments of this disclosure is shown. Figure 6 As shown, the proposed solution 605, which uses a machine learning model with a TPC command table suitable for machine learning, can outperform the exhaustive search open-loop power control method 610 almost immediately from the very beginning of the simulation. The machine learning model can also learn to use all currently specified values ​​in the 5G. However, because the currently specified values ​​create a bias towards power increases, automatic entropy causes the untrained model to increase the power to its maximum value, and the machine learning model spends time figuring out how to adjust the power in the opposite direction and optimizing the policy using the action space biased towards power increases. To remove this bias, the machine learning model can be configured to use only the first 3 indices [-1, 0, 1] dB from the 5G TPC commands. In this way, automatic entropy can learn an appropriate power regulation policy. A better choice of regulation values ​​can help learn a better power regulation policy.

[0106] Figure 7 A schematic diagram of a power distribution shaped by a machine learning model according to some example embodiments of this disclosure is shown. Figure 7 As shown, the machine learning model learns to optimize the power level of the second device 120 individually and uniquely shapes the power distribution in a way that the second device 120 can benefit from. The machine learning model can learn from scratch using the action space to which machine learning is applicable (e.g., a set of power adjustment values ​​specific to the machine learning model). It takes samples from fewer than 100 TTIs to optimize uplink power control.

[0107] Figure 8A schematic diagram of power distribution in a simulated scenario according to some example embodiments of this disclosure is shown. For example... Figure 8 As shown, in this simulated scenario, the power distribution 805 with P0 of -90 dBm and path loss α of 0.8 provides results without machine learning-based power control. However, even without optimized closed-loop parameters, the proposed solution applying a machine learning model can dynamically shape the power distribution using TPC commands and achieve better performance.

[0108] Table 4 summarizes the gains of the proposed solutions for the values ​​of P0 and α in the exhaustive search: Table 4

[0109] The results in Table 4 show that, with the proposed solution, the adjustment of P0 and path loss α values ​​becomes obsolete, and the power level can be automatically shaped using the TPC command.

[0110] Simulations also verified that a pre-trained machine learning model can be loaded from a file at the start of the simulation, and that it can immediately accumulate the PUSCH power control adjustment state of the second device 120 to outperform exhaustive search open-loop power control (without any further training).

[0111] This disclosure allows for some other simulations with P0 = -76 dBm and path loss α = 1. Figure 9 A schematic diagram illustrating the relationship between geometric mean throughput and P0 according to some example embodiments of this disclosure is shown. Figure 9 As shown, the exhaustive search method attempts all combinations of values ​​and searches for the value that provides the best performance. The exhaustive search method is performed with a 2dB granularity to make it even more exhaustive. This results in finding the optimal P0 for this particular dense urban macro scene to be -86dBm. Furthermore, this result is provided using a machine learning model (e.g., a model pre-trained with a new Munchausen RL-powered SACD). It is shown that the machine learning model can outperform the exhaustive search method regardless of whether P0 and path loss α are configured with optimal or suboptimal values. The geometric mean throughput can be used as a reward for training the machine learning model.

[0112] Figure 10 A schematic diagram of the cumulative distribution function (CDF) of uplink user throughput according to some example embodiments of this disclosure is shown. Figure 10 As shown, a performance gain can be observed in 90% of the second device 120 with different P0 configurations and the same pre-trained machine learning model. In the example, the machine learning model can be trained using the proposed Munchausen RL modification to SACD.

[0113] It was also found that the same machine learning model can be used for different value tables for TPC commands, simply by changing the size of the values ​​used. Figure 11 A schematic diagram illustrating the relationship between the geometric mean throughput and different value tables for TPC commands according to some example embodiments of this disclosure is shown. Figure 11 As shown, the non-standard [-3, 0, 3] dB adjustment happens to provide a slightly better gain than the standard-compliant [-1, 0, 1] dB adjustment. The proposed solution can improve geometric mean throughput if both optimal P0 (e.g., -86 dBm) and non-optimal P0 (e.g., -76 dBm) are configured.

[0114] Figure 12 A flowchart of an example method 1200 implemented at a first device according to some example embodiments of the present disclosure is shown. For discussion purposes, [the following will be discussed]. Figure 1 Method 1200 is described by the angle of the first device 110 in the middle.

[0115] In block 1210, the first device 110 sends a configuration for transmit power control (TPC) to the second device 120, which indicates a set of power adjustment values ​​specific to the machine learning model.

[0116] At box 1220, the first device 110 determines the power adjustment value for the second device from the set of power adjustment values ​​by applying a machine learning model.

[0117] In block 1230, the first device 110 sends a first TPC command to the second device 120, which includes an indication of the determined power adjustment value.

[0118] In some example implementations, the configuration of the power adjustment value set can be sent via system information or radio resource control signaling.

[0119] In some example implementations, the first TPC command may include at least one of a cumulative TPC command or an absolute TPC command, and the power adjustment value set may include at least one of a cumulative value set for the cumulative TPC command or an absolute value set for the absolute TPC command.

[0120] In some example implementations, the set of power adjustment values ​​may include zero values ​​as well as an equal number of positive and negative values.

[0121] In some example implementations, the first TPC command may be sent via downlink control information (DCI), and the number of bits indicating the first TPC command within the DCI may be based on the minimum number of bits required to represent the set of power adjustment values.

[0122] In some example implementations, the first TPC command may include an index for indexing the determined power adjustment value from a set of power adjustment values.

[0123] In some example implementations, the first device 110 may determine the power adjustment value by applying a machine learning model based on at least one of the following: the number of resource blocks to be allocated, a measurement or estimate of the uplink signal-to-interference-plus-noise ratio, the transmit power of the second device, or a measurement or estimate of the reference signal received power of the uplink reference signal.

[0124] In some example implementations, the first device 110 may receive capability information from the second device 120 that supports the application of a machine learning model by the second device 120, and send the machine learning model to the second device 120 for application by the second device 120.

[0125] In some example implementations, the first device 110 may send an instruction to the second device 120 to stop the application of the machine learning model by the second device 120.

[0126] In some example implementations, the first device 110 may send a second TPC command to the second device 120 to reset the cumulative adjustment state of the transmission power of the second device 120, or send a second TPC command to the second device 120 to periodically reset the cumulative adjustment state of the transmission power of the second device 120.

[0127] In some example implementations, the second TPC command may be sent via DCI or Media Access Control (MAC) control element (CE).

[0128] In some example implementations, the set of power adjustment values ​​used during the training process of the machine learning model may be different from the set of power adjustment values ​​used during the inference process of the machine learning model.

[0129] In some example implementations, the first device 110 may be a network device or may be included in a network device, and the second device 120 may be a terminal device or may be included in a terminal device.

[0130] In some example implementations, a first device capable of performing any method 1200 (e.g., Figure 1 The first device 110 may include components for performing corresponding operations of method 1200 and any embodiment thereof. The device may be implemented in any suitable form. For example, the device may be implemented in a circuit or software module. The first device may be implemented as or included in Figure 1 In the first device 110.

[0131] Figure 13A flowchart of an example method 1300 implemented at a second device according to some example embodiments of the present disclosure is shown. For discussion purposes, [the following will be discussed]. Figure 1 Method 1300 is described by the angle of the second device 120 in the middle.

[0132] In block 1310, the second device 120 receives from the first device 110 a configuration for transmit power control (TPC) that indicates a set of power adjustment values ​​specific to a machine learning model.

[0133] In block 1320, the second device 120 receives from the first device 110 a first TPC command including an indication of the power adjustment values ​​in the set of power adjustment values.

[0134] In frame 1330, the second device 120 adjusts its transmission power based on the first TPC command.

[0135] In some example implementations, the second device 120 may determine the cumulative adjustment state of the transmission power of the second device 120 based on the indicated power adjustment value, and adjust the transmission power of the second device 120 based on the cumulative adjustment state.

[0136] In some example implementations, the configuration of the power adjustment value set may be received via system information or radio resource control signaling.

[0137] In some example implementations, the first TPC command may include at least one of a cumulative TPC command or an absolute TPC command, and the power adjustment value set may include at least one of a cumulative value set for the cumulative TPC command or an absolute value set for the absolute TPC command.

[0138] In some example implementations, the set of power adjustment values ​​may include zero values ​​as well as an equal number of positive and negative values.

[0139] In some example implementations, the first TPC command may be sent via downlink control information (DCI), and the number of bits indicating the first TPC command within the DCI may be based on the minimum number of bits required to represent the set of power adjustment values.

[0140] In some example implementations, the first TPC command may include an index for indexing the indicated power adjustment value from a set of power adjustment values.

[0141] In some example implementations, the second device 120 may send capability information to the first device 110 that supports applications using machine learning models developed by the second device 120, and receive machine learning models from the first device 110 for use by the second device 120.

[0142] In some example implementations, the second device 120 may determine the power adjustment value by applying a machine learning model based on at least one of the following: the number of resource blocks to be allocated, a measurement or estimate of the uplink signal-to-interference-plus-noise ratio, the transmit power of the second device, or a measurement or estimate of the received power of the uplink reference signal.

[0143] In some example implementations, the second device 120 may receive an instruction from the first device 110 to stop applying the machine learning model by the second device, and stop applying the machine learning model based on the instruction.

[0144] In some example implementations, the second device 120 may receive a second TPC command from the first device 110 to reset the cumulative adjustment state of the transmit power for the second device 120, and reset the cumulative adjustment state based on the second TPC command.

[0145] In some example implementations, the second device 120 may receive a second TPC command from the first device 110 to periodically reset the cumulative adjustment state of the transmit power for the second device, and periodically reset the cumulative adjustment state based on the second TPC command.

[0146] In some example implementations, the second TPC command may be sent via DCI or Media Access Control (MAC) control element (CE).

[0147] In some example implementations, the set of power adjustment values ​​used during the training process of the machine learning model may be different from the set of power adjustment values ​​used during the inference process of the machine learning model.

[0148] In some example implementations, a second device capable of performing any method 1300 (e.g., Figure 1 The second device 120 may include components for performing corresponding operations of method 1300 and any embodiment thereof. This device may be implemented in any suitable form. For example, it may be implemented in a circuit or software module. The second device may be implemented as or included in... Figure 1 The second device 120 in the middle.

[0149] Figure 14 This is a simplified block diagram of device 1400 suitable for implementing example embodiments of this disclosure. Device 1400 can be provided to implement communication devices, such as... Figure 1 The first device 110 and the second device 120 are shown. As shown, the device 1400 includes one or more processors 1410, one or more memories 1420 coupled to the processors 1410, and one or more communication modules 1440 coupled to the processors 1410.

[0150] Communication module 1440 is used for bidirectional communication. Communication module 1440 has one or more communication interfaces to facilitate communication with one or more other modules or devices. The communication interface can represent any interface required for communication with other network elements. In some example embodiments, communication module 1440 may include at least one antenna.

[0151] As a non-limiting example, processor 1410 can be any type suitable for a local technology network and can include one or more of the following: general-purpose computer, special-purpose computer, microprocessor, digital signal processor (DSP), and processor based on a multi-core processor architecture. Device 1400 can have multiple processors, such as application-specific integrated circuit chips that are time-dependent on a clock synchronized with a main processor.

[0152] Memory 1420 may include one or more non-volatile memories and one or more volatile memories. Examples of non-volatile memories include, but are not limited to, read-only memory (ROM) 1424, electrically programmable read-only memory (EPROM), flash memory, hard disk, optical disc (CD), digital video disc (DVD), optical disc, laser disc, and other magnetic and / or optical storage. Examples of volatile memories include, but are not limited to, random access memory (RAM) 1422 and other volatile memories that will not persist for the duration of a power outage.

[0153] Computer program 1430 includes computer-executable instructions that are executed by an associated processor 1410. The instructions of program 1430 may include instructions for performing operations / actions of some example embodiments of this disclosure. Program 1430 may be stored in memory (e.g., ROM 1424). Processor 1410 can perform any suitable actions and processes by loading program 1430 into RAM 1422.

[0154] The exemplary implementation of this disclosure can be achieved through program 1430, enabling device 1400 to perform as described in the reference. Figures 1 to 13 Any process discussed in this disclosure. Example embodiments of this disclosure may also be implemented by hardware or by a combination of software and hardware.

[0155] In some example embodiments, program 1430 may be tangibly contained in a computer-readable medium, which may be included in device 1400 (such as memory 1420) or other storage device accessible by device 1400. Device 1400 may load program 1430 from the computer-readable medium into RAM 1422 for execution. In some example embodiments, the computer-readable medium may include any type of non-transitory storage medium, such as ROM, EPROM, flash memory, hard disk, CD, DVD, etc. As used herein, the term "non-transitory" refers to a limitation on the medium itself (i.e., tangible, not tactile), rather than a limitation on the persistence of data storage (e.g., RAM and ROM).

[0156] Figure 15 An example of a computer-readable medium 1500 is shown, which may be in the form of a CD, DVD, or other optical storage disc. Program 1430 is stored on the computer-readable medium 1500.

[0157] Generally, the various embodiments of this disclosure can be implemented in hardware or special-purpose circuitry, software, logic, or any combination thereof. Some aspects can be implemented in hardware, while others can be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device. Although various aspects of the embodiments of this disclosure are illustrated and described as block diagrams, flowcharts, or using some other graphical representation, it should be understood that, as a non-limiting example, the blocks, apparatuses, systems, techniques, or methods described herein can be implemented in hardware, software, firmware, special-purpose circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0158] Some example embodiments of this disclosure also provide at least one computer program product tangibly stored on a computer-readable medium, such as a non-transitory computer-readable medium. The computer program product includes computer-executable instructions that execute in a device on a target physical or virtual processor, such as those included in a program module, to perform any of the methods described above. Typically, a program module includes routines, programs, libraries, objects, classes, components, data structures, etc., that perform a specific task or implement a specific abstract data type. In various embodiments, the functionality of a program module can be combined or split among program modules as needed. The machine-executable instructions for a program module can execute within a local or distributed device. In a distributed device, the program module can reside in both local and remote storage media.

[0159] Program code used to perform the methods of this disclosure may be written in any combination of one or more programming languages. The program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a stand-alone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0160] In the context of this disclosure, computer program code or related data may be carried by any suitable carrier to enable a device, apparatus, or processor to perform the various processes and operations described above. Examples of carriers include signals, computer-readable media, etc.

[0161] Computer-readable media can be computer-readable signal media or computer-readable storage media. Computer-readable media can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any suitable combination thereof. More specific examples of computer-readable storage media will include electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable optical disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0162] Furthermore, although the operations are described in a specific order, this should not be construed as requiring that such operations be performed in the specific order shown or sequentially, or that all the operations shown be performed to achieve the desired result. In some cases, multitasking and parallel processing may be advantageous. Similarly, although several specific implementation details are included in the discussion above, these details should not be construed as limiting the scope of this disclosure, but rather as descriptions of features that may be specific to a particular implementation. Unless explicitly stated otherwise, certain features described in the context of a single implementation may also be implemented in combination in a single implementation. Conversely, unless explicitly stated otherwise, the various features described in the context of a single implementation may also be implemented individually or in any suitable sub-combination in multiple implementations.

[0163] Although this disclosure has been described in language specific to structural features and / or methodological actions, it should be understood that the disclosure as defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are disclosed as exemplary forms for implementing the claims.

[0164] Furthermore, the various implementations of this disclosure can be described with reference to the following terms, and their features can be combined in any reasonable manner.

[0165] Clause 1. A first means for communication, comprising: at least one processor; and at least one memory storing instructions, the instructions, when executed by the at least one processor, causing the first means to at least: send to a second means a configuration for transmit power control (TPC), the configuration indicating a set of power adjustment values ​​specific to a machine learning model; determine a power adjustment value for the second means from the set of power adjustment values ​​by applying the machine learning model; and send to the second means a first TPC command including an indication of the determined power adjustment value.

[0166] Clause 2. The first apparatus according to Clause 1, wherein the configuration indicating the set of power adjustment values ​​is transmitted via system information or radio resource control signaling.

[0167] Clause 3. The first apparatus according to Clause 1, wherein the first TPC command includes at least one of a cumulative TPC command or an absolute TPC command, and wherein the power adjustment value set includes at least one of a cumulative value set for the cumulative TPC command or an absolute value set for the absolute TPC command.

[0168] Clause 4. The first device according to Clause 1, wherein the set of power adjustment values ​​includes zero values ​​and an equal number of positive and negative values.

[0169] Clause 5. The first apparatus according to Clause 1, wherein the first TPC command is transmitted via downlink control information (DCI), and wherein the number of bits indicating the first TPC command within the DCI is based on the minimum number of bits required to represent the set of power adjustment values.

[0170] Clause 6. The first apparatus according to any one of Clauses 1 to 5, wherein the first TPC command includes an index for indexing the determined power adjustment value from the set of power adjustment values.

[0171] Clause 7. The first apparatus according to any one of Clauses 1 to 5, wherein the first apparatus is further configured to: determine the power adjustment value by applying the machine learning model based on at least one of the following: the number of resource blocks to be allocated, a measurement or estimate of the uplink signal to interference plus noise ratio, the transmit power of the second apparatus, or a measurement or estimate of the reference signal receive power of the uplink reference signal.

[0172] Clause 8. The first device according to any one of Clauses 1 to 5, wherein the first device is further configured to: receive from the second device capability information supporting the application of a machine learning model by the second device; and send the machine learning model to the second device for application by the second device.

[0173] Clause 9. The first device according to Clause 8, wherein the first device is further configured to: send an instruction to the second device to stop the application of the machine learning model by the second device.

[0174] Clause 10. The first device according to any one of Clauses 1 to 5, wherein the first device is further configured to: send a second TPC command to the second device to reset the cumulative adjustment state of the transmission power for the second device, or send a second TPC command to the second device to periodically reset the cumulative adjustment state of the transmission power for the second device.

[0175] Clause 11. The first apparatus according to any one of Clauses 1 to 5, wherein the set of power adjustment values ​​used during the training process of the machine learning model is different from the set of power adjustment values ​​used during the inference process of the machine learning model.

[0176] Clause 12. A second means for communication, comprising: at least one processor; and at least one memory storing instructions, which, when executed by the at least one processor, cause the second means to at least: receive from a first means a configuration for transmit power control (TPC), the configuration indicating a set of power adjustment values ​​specific to a machine learning model; receive from the first means a first TPC command including an indication of a power adjustment value in the set of power adjustment values; and adjust the transmit power of the second means based on the first TPC command.

[0177] Clause 13. The second apparatus according to Clause 12, wherein the second apparatus is further configured to: determine a cumulative adjustment state of the transmission power for the second apparatus based on the indicated power adjustment value; and adjust the transmission power of the second apparatus based on the cumulative adjustment state.

[0178] Clause 14. A method for communication, comprising: at a first device, sending to a second device a configuration for transmit power control (TPC), the configuration indicating a set of power adjustment values ​​specific to a machine learning model; determining a power adjustment value for the second device from the set of power adjustment values ​​by applying the machine learning model; and sending to the second device a first TPC command including an indication of the determined power adjustment value.

[0179] Clause 15. A method for communication, comprising: at a second device, receiving from a first device a configuration for transmit power control (TPC), the configuration indicating a set of power adjustment values ​​specific to a machine learning model; receiving from the first device a first TPC command including an indication of power adjustment values ​​in the set of power adjustment values; and adjusting the transmit power of the second device based on the first TPC command.

Claims

1. A first device for communication, comprising: At least one processor; as well as At least one memory storing instructions, which, when executed by the at least one processor, cause the first device to at least: Send a configuration for transmit power control (TPC) to the second device, the configuration indicating a set of power adjustment values ​​specific to the machine learning model; By applying the machine learning model, a power adjustment value for the second device is determined from the set of power adjustment values; and A first TPC command, including an indication of the determined power adjustment value, is sent to the second device.

2. The first apparatus of claim 1, wherein the configuration indicating the set of power adjustment values ​​is transmitted via system information or radio resource control signaling.

3. The first apparatus of claim 1, wherein the first TPC command comprises at least one of a cumulative TPC command or an absolute TPC command, and The set of power adjustment values ​​includes at least one of the set of cumulative values ​​for the cumulative TPC command or the set of absolute values ​​for the absolute TPC command.

4. The first device according to claim 1, wherein the set of power adjustment values ​​includes zero values ​​and an equal number of positive and negative values.

5. The first apparatus of claim 1, wherein the first TPC command is transmitted via downlink control information (DCI), and wherein the number of bits indicating the first TPC command within the DCI is based on the minimum number of bits required to represent the set of power adjustment values.

6. The first apparatus according to any one of claims 1 to 5, wherein the first TPC command includes an index for indexing the determined power adjustment value from the set of power adjustment values.

7. The first device according to any one of claims 1 to 5, wherein the first device is further configured to: The power adjustment value is determined by applying the machine learning model based on at least one of the following: The number of resource blocks to be allocated. Measurement or estimation of the uplink signal-to-interference-plus-noise ratio. The transmission power of the second device, or Measurement or estimation of the received power of the uplink reference signal.

8. The first device according to any one of claims 1 to 5, wherein the first device is further configured to: Receive from the second device information the capability information supporting the application of the machine learning model by the second device; and The machine learning model is sent to the second device for application by the second device.

9. The first device according to claim 8, wherein the first device is further configured to: Send an instruction to the second device to stop the application of the machine learning model by the second device.

10. The first device according to any one of claims 1 to 5, wherein the first device is further configured to: Send a second TPC command to the second device to reset the cumulative adjustment state of the transmission power of the second device, or Send a second TPC command to the second device to periodically reset the cumulative adjustment state of the transmission power for the second device.

Images