Methods and devices for power saving with artificial intelligence and machine learning model
AI/ML models are used to predict and adapt communication resource usage in wireless systems, addressing power consumption challenges and enhancing efficiency and performance in user equipment and base stations.
Patent Information
- Application Number
- PCT/CN2024/117857
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2024-09-10
- Publication Date
- 2025-07-31
AI Technical Summary
Existing wireless communication systems face challenges in reducing power consumption when utilizing artificial intelligence and machine learning models, particularly in user equipment and base stations, which is crucial for meeting the demands of high-speed, low-latency, and ultra-reliable new generation wireless services.
Implementing AI/ML models for power saving by measuring current communication resources, predicting future resources, and transmitting measurement reports to base stations, utilizing AI/ML models to adaptively adjust time, frequency, and spatial domain resources based on traffic predictions, and splitting AI/ML models into two-sided configurations to balance power consumption between user equipment and network devices.
Enhances power efficiency, improves wireless communication performance, and optimizes resource utilization by reducing unnecessary power consumption in user equipment and base stations, enabling efficient power saving strategies.
Smart Images

Figure CN2024117857_31072025_PF_FP_ABST
Abstract
Description
METHODS AND DEVICES FOR POWER SAVING WITH ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING MODELTECHNICAL FIELD
[0001] The present disclosure is directed generally to wireless communications. Particularly, the present disclosure relates to methods and devices for power saving with artificial intelligence or machine learning (AI / ML) models / functions in a mobile communication system.BACKGROUND
[0002] Wireless communication technologies are moving the world toward an increasingly connected and networked society. High-speed and low-latency wireless communications rely on efficient network resource management and allocation between user equipment and wireless access network nodes (including but not limited to base stations) . A new generation network is expected to provide high speed, low latency and ultra-reliable communication capabilities and fulfill the requirements from different industries and users.
[0003] In some wireless communication systems, various data processing models / functions may be adopted to provide data prediction / inference efficiently, for example, artificial intelligence or machine learning (AI / ML) models / functions. There are some issues / problems associates with adopting these data processing models / functions by the wireless communication systems. For one non-limiting example, one issue / problem may be how to reduce a power consumption of a user equipment / base station with AI / ML models / functions.
[0004] The present disclosure describes various embodiments for power saving with AI / ML models / functions in a mobile communication system, addressing at least one of the issues / problems discussed in the present disclosure, thus increasing efficiency of power reduction, enabling future wireless communication system to provide improved performance on power reduction to meet various demands of new generation wireless services in wireless communication systems.SUMMARY
[0005] This document relates to methods, systems, and devices for wireless communication, and more specifically, for power saving with artificial intelligence or machine learning (AI / ML) models / functions in a mobile communication system. The various embodiments in the present disclosure may be beneficial to enhance efficiency and performance of AI / ML models / functions, increase the power reduction efficiency, and / or boost performance of the wireless data service via wireless communication., increase the overall transmission efficiency and speed, and / or boost performance of the wireless communication.
[0006] In one embodiment, the present disclosure describes a method for power saving in a wireless communication system, performed by a wireless communication device. The method includes measuring, by a user equipment (UE) , current communication resource in a current active state before a power-saving state, to obtain a current measurement; obtaining, by the UE, a measurement for future communication resource in a future active state after the power-saving sate, wherein the measurement for future communication resource is determined based on the current measurement; and transmitting, by the UE to a base station in the current active state, a measurement report indicating the measurement for future communication resource.
[0007] In one embodiment, the present disclosure describes another method for power saving in a wireless communication system, performed by a wireless communication node. The method includes receiving, by a base station from a user equipment (UE) , a measurement report corresponding to a measurement for future communication resource in a future active state after a power-saving sate, wherein the measurement for future communication resource is determined based on a current measurement, and the current measurement is obtained by measuring current communication resource in a current active state before the power-saving state.
[0008] In some other embodiments, an apparatus for wireless communication may include a memory storing instructions and at least one processing circuitry in communication with the memory. When the at least one processing circuitry executes the instructions, the at least one processing circuitry is configured to carry out any of the methods above and / or in the present disclosure.
[0009] In some other embodiments, a device for wireless communication may include a memory storing instructions and at least one processing circuitry in communication with the memory. When the at least one processing circuitry executes the instructions, the at least one processing circuitry is configured to carry out any of the methods above and / or in the present disclosure.
[0010] In some other embodiments, a computer-readable medium comprising instructions which, when executed by a computer, cause the computer to carry out the methods above and / or in the present disclosure. The computer-readable medium may be a non-transitory computer-readable medium.
[0011] In some other embodiments, a computer program product comprising a computer-readable program medium code stored thereupon, the computer-readable program medium code, when executed by at least one processor, causing the at least one processor to implement any of the methods above and / or in the present disclosure. The computer program product may be a non-transitory computer program product. The computer-readable program medium code may be a non-transitory computer-readable program medium code.
[0012] The above and other aspects and their implementations are described in greater detail in the drawings, the descriptions, and the claims.BRIEF DESCRIPTION OF THE DRAWINGS
[0013] FIG. 1 shows an example of a wireless communication system include at least one wireless network node and one or more user equipment.
[0014] FIG. 2 shows an example of a network node.
[0015] FIG. 3 shows an example of a user equipment.
[0016] FIG. 4A shows a flow diagram of a method for wireless communication.
[0017] FIG. 4B shows a flow diagram of another method for wireless communication.
[0018] FIG. 5A shows a schematic diagram of one exemplary embodiment in the present disclosure.
[0019] FIG. 5B shows a schematic diagram of another exemplary embodiment in the present disclosure.
[0020] FIG. 5C shows a schematic diagram of another exemplary embodiment in the present disclosure.
[0021] FIG. 6 shows a schematic diagram of another exemplary embodiment in the present disclosure.
[0022] FIG. 7 shows a schematic diagram of another exemplary embodiment in the present disclosure.
[0023] FIG. 8 shows a schematic diagram of another exemplary embodiment in the present disclosure.
[0024] FIG. 9 shows a schematic diagram of another exemplary embodiment in the present disclosure.DETAILED DESCRIPTION
[0025] The present disclosure will now be described in detail hereinafter with reference to the accompanied drawings, which form a part of the present disclosure, and which show, by way of illustration, specific examples of embodiments. Please note that the present disclosure may, however, be embodied in a variety of different forms and, therefore, the covered or claimed subject matter is intended to be construed as not being limited to any of the embodiments to be set forth below.
[0026] Throughout the specification and claims, terms may have nuanced meanings suggested or implied in context beyond an explicitly stated meaning. Likewise, the phrase “in one embodiment” or “in some embodiments” as used herein does not necessarily refer to the same embodiment and the phrase “in another embodiment” or “in other embodiments” as used herein does not necessarily refer to a different embodiment. The phrase “in one implementation” or “in some implementations” as used herein does not necessarily refer to the same implementation and the phrase “in another implementation” or “in other implementations” as used herein does not necessarily refer to a different implementation. It is intended, for example, that claimed subject matter includes combinations of exemplary embodiments or implementations in whole or in part.
[0027] In general, terminology may be understood at least in part from usage in context. For example, terms, such as “and” , “or” , or “and / or, ” as used herein may include a variety of meanings that may depend at least in part upon the context in which such terms are used. Typically, “or” if used to associate a list, such as A, B or C, is intended to mean A, B, and C, here used in the inclusive sense, as well as A, B or C, here used in the exclusive sense. In addition, the term “one or more” or “at least one” as used herein, depending at least in part upon context, may be used to describe any feature, structure, or characteristic in a singular sense or may be used to describe combinations of features, structures or characteristics in a plural sense. Similarly, terms, such as “a” , “an” , or “the” , again, may be understood to convey a singular usage or to convey a plural usage, depending at least in part upon context. In addition, the term “based on” or “determined by” may be understood as not necessarily intended to convey an exclusive set of factors and may, instead, allow for existence of additional factors not necessarily expressly described, again, depending at least in part on context.
[0028] The present disclosure describes methods and devices for power saving with artificial intelligence or machine learning (AI / ML) models / functions in a mobile communication system.
[0029] Wireless technologies are moving the world toward an increasingly connected and networked society. High-speed and low-latency wireless communication system rely on efficient network resource management and allocation between user equipment and wireless access network nodes (including but not limited to base stations and / or core networks) . New generation networks are expected to provide various data services, in addition to conventional communication service, with high speed, low latency, and highly reliable capabilities, so as to fulfill requirements under various circumstances.
[0030] In some wireless communication systems, various data processing models / functions may be adopted to provide data prediction / inference efficiently, for example, artificial intelligence or machine learning (AI / ML) models / functions. There are some issues / problems associates with adopting these data processing models / functions by the wireless communication systems. For one non-limiting example, one issue / problem may be how to reduce a power consumption of a user equipment / base station with AI / ML models / functions.
[0031] In some implementations, AI, in general, may refer to (or include) devices, components, software, and modules that have self-learning, such as machine learning (ML) , deep learning, reinforcement learning, migration learning, deep reinforcement learning, and meta-learning. In some implementations, AI may be implemented by using an artificial intelligence network (or referred to as a neural network) . The neural network includes multiple layers, and each layer includes at least one node. Typically, the neural network includes an input layer, an output layer, and at least one hidden layer. The neural network of each layer includes but is not limited to using at least one of a full connection layer, a dense layer, a convolutional layer, a transposed convolutional layer, a direct connection layer, an activation function, a normalization layer, a pooling layer. In some implementations, each layer of the neural network may include one sub-neural network, such as a Residual Network block (or Resnet block) , a dense network block (Densenet Block) , a cyclic network (e.g., Recurrent Neural Network (RNN) ) .
[0032] In some implementations the artificial intelligence network includes a neural network model and / or a neural network parameter corresponding to the neural network model. The neural network model may be referred to as a network model in short, and the neural network parameter may be referred to as a network parameter in short. A network model may define an architecture of a network such as a quantity of layers of a neural network, a size of each layer, an activation function, a link status, a convolution kernel and a convolution step of a size, a convolution type (for example, 1D convolution, 2D convolution, 3D convolution, hollow convolution, transposed convolution, divided convolution, packet convolution, or extended convolution) . A network parameter is a weight and / or an offset of a network of each layer in the network model and a value of the weight and / or the offset of the network of each layer in the network model. One network model may correspond to a plurality of different sets of neural network parameter values to adapt to different scenarios. The value of the network parameter may be obtained through offline training and / or online training. For example, the neural network parameter is obtained by training the neural network model by inputting at least one sample and a label.
[0033] In some wireless communication systems, AI / ML may be a promising enhancement direction for mobile communication system, e.g., 5G, 5G-A, 6G, and etc. With the introduction of AI / ML technology into the mobile communication system, the system operating efficiency is expected to be improved, e.g., reducing the overhead of reference signal via AI / ML inference and prediction, and / or enhancing the accuracy of terminal positioning. Various other possible benefits for adopting AI / ML into mobile communication system or completely merged into mobile communication system are expected as well.
[0034] In the present disclosure, “model” may be a general term to describe a device in mobile system being capable of doing a processing method, a functionality, a feature, or a feature group; and / or “model” may refer to functionality, function, functionality module, function module, processing method, information processing method, implementation, feature, feature group, configuration, configuration set, dataset (e.g., for model training) or data-driven algorithms. In some implementations, for the communication system with AI / ML technology, AI / ML model is included / adopted.
[0035] In some implementations, AI / ML models / functions may be adopted for mobile communication system, e.g., 5G (the fifth generation) , 5G-A (5G-Advanced) , 6G (the sixth generation) , and other telecommunication standards. AI model is the key for AI / ML technology with the development of model training, model interference, model monitoring, model management or other possible model functionalities, methods, algorithms, configurations, procedures, capabilities, etc.
[0036] In some implementations, AI / ML models / functions may be used for power saving of user equipment (UE) or mobile terminal. Power saving of UE is very challenging to keep the performance of throughput and latency at the same time. The battery capacity of a mobile terminal is limited, the operation of the communication function must be strictly controlled to keep the terminal battery life as long as possible. There are lots of regular ways to reduce the power consumption of UE but AI assisted power saving is still a new thing for mobile communication system and UE.
[0037] The present disclosure describes various embodiments for data preprocessing for power saving with artificial intelligence or machine learning (AI / ML) models / functions in a mobile communication system, addressing at least one of the issues / problems discussed in the present disclosure, thus increasing efficiency of power reduction, enabling future wireless communication system to provide improved performance on power reduction to meet various demands of new generation wireless services in wireless communication systems.
[0038] FIG. 1 shows a wireless communication system 100 including a core network (CN) 110, a radio access network (RAN) 130, and one or more user equipments (UEs) (152, 154, and 156) . The RAN 130 may include one or more base stations. The base stations may include at least one evolved NodeB (eNB) for 4G Long Term Evolution (LTE) , or a Next generation NodeB (gNB) for 5G New Radio (NR) , or a NodeB for 6G, or any other type of signal transmitting / receiving device such as a UMTS NodeB. In one implementation, the core network 110 may include a 5G core network (5GC) , and the interface 125 may include a new generation (NG) interface. The core network 110 further includes at least one location management function (LMF) , and / or at least one session management function (SMF) , and / or at least one user plane function (UPF) and / or at least one access and mobility management Function (AMF) , and / or etc.
[0039] Referring to FIG. 1, a first UE 152 may receive one or more downlink communication 142 from the RAN 130 and send one or more uplink communication 141 to the RAN 130. Likewise, a second UE 154 may receive downlink communication 144 from the RAN 130 and send uplink communication 143 to the RAN 130; and a third UE 156 may receive downlink communication 146 from the RAN 130 and send uplink communication 145 to the RAN 130. For example but not limited to, a downlink communication may include a physical downlink (DL) shared channel (PDSCH) or a physical downlink control channel (PDCCH) , and a uplink (UL) communication may include a physical uplink shared channel (PUSCH) or a physical uplink control channel (PUCCH) .
[0040] FIG. 2 shows an example of electronic device 200 to implement a network base station. The example electronic device 200 may include radio transmitting / receiving (Tx / Rx) circuitry 208 to transmit / receive communication with UEs and / or other base stations. The electronic device 200 may also include network interface circuitry 209 to communicate the base station with other base stations and / or a core network, e.g., optical or wireline interconnects, Ethernet, and / or other data transmission mediums / protocols. The electronic device 200 may optionally include an input / output (I / O) interface 206 to communicate with an operator or the like.
[0041] The electronic device 200 may also include system circuitry 204. System circuitry 204 may include processor (s) 221 and / or memory 222. Memory 222 may include an operating system 224, instructions 226, and parameters 228. Instructions 226 may be configured for the one or more of the processors 124 to perform the functions of the network node. The parameters 228 may include parameters to support execution of the instructions 226. For example, parameters may include network protocol settings, bandwidth parameters, radio frequency mapping assignments, and / or other parameters.
[0042] FIG. 3 shows an example of an electronic device to implement a terminal device 300 (for example, user equipment (UE) ) . The UE 300 may be a mobile device, for example, a smart phone or a mobile communication module disposed in a vehicle. The UE 300 may include communication interfaces 302, a system circuitry 304, an input / output interfaces (I / O) 306, a display circuitry 308, and a storage 309. The display circuitry may include a user interface 310. The system circuitry 304 may include any combination of hardware, software, firmware, or other logic / circuitry. The system circuitry 304 may be implemented, for example, with one or more systems on a chip (SoC) , application specific integrated circuits (ASIC) , discrete analog and digital circuits, and other circuitry. The system circuitry 304 may be a part of the implementation of any desired functionality in the UE 300. In that regard, the system circuitry 304 may include logic that facilitates, as examples, decoding and playing music and video, e.g., MP3, MP4, MPEG, AVI, FLAC, AC3, or WAV decoding and playback; running applications; accepting user inputs; saving and retrieving application data; establishing, maintaining, and terminating cellular phone calls or data connections for, as one example, internet connectivity; establishing, maintaining, and terminating wireless network connections, Bluetooth connections, or other connections; and displaying relevant information on the user interface 310. The user interface 310 and the inputs / output (I / O) interfaces 306 may include a graphical user interface, touch sensitive display, haptic feedback or other haptic output, voice or facial recognition inputs, buttons, switches, speakers and other user interface elements. Additional examples of the I / O interfaces 306 may include microphones, video and still image cameras, temperature sensors, vibration sensors, rotation and orientation sensors, headset and microphone input / output jacks, Universal Serial Bus (USB) connectors, memory card slots, radiation sensors (e.g., IR sensors) , and other types of inputs.
[0043] Referring to FIG. 3, the communication interfaces 302 may include a Radio Frequency (RF) transmit (Tx) and receive (Rx) circuitry 316 which handles transmission and reception of signals through one or more antennas 314. The communication interface 302 may include one or more transceivers. The transceivers may be wireless transceivers that include modulation / demodulation circuitry, digital to analog converters (DACs) , shaping tables, analog to digital converters (ADCs) , filters, waveform shapers, filters, pre-amplifiers, power amplifiers and / or other logic for transmitting and receiving through one or more antennas, or (for some devices) through a physical (e.g., wireline) medium. The transmitted and received signals may adhere to any of a diverse array of formats, protocols, modulations (e.g., QPSK, 16-QAM, 64-QAM, or 256-QAM) , frequency channels, bit rates, and encodings. As one specific example, the communication interfaces 302 may include transceivers that support transmission and reception under the 2G, 3G, BT, WiFi, Universal Mobile Telecommunications System (UMTS) , High Speed Packet Access (HSPA) +, 4G / Long Term Evolution (LTE) , 5G standards, 6G, and / or any further generation standards. The techniques described below, however, are applicable to other wireless communications technologies whether arising from the 3rd Generation Partnership Project (3GPP) , GSM Association, 3GPP2, IEEE, or other partnerships or standards bodies.
[0044] Referring to FIG. 3, the system circuitry 304 may include one or more processors 321 and memories 322. The memory 322 stores, for example, an operating system 324, instructions 326, and parameters 328. The processor 321 is configured to execute the instructions 326 to carry out desired functionality for the UE 300. The parameters 328 may provide and specify configuration and operating options for the instructions 326. The memory 322 may also store any BT, WiFi, 3G, 4G, 5G, 6G, or other data that the UE 300 may send, or has received, through the communication interfaces 302. In various implementations, a system power for the UE 300 may be supplied by a power storage device, such as a battery or a transformer.
[0045] The present disclosure describes various embodiment for power saving with AI / ML models / functions in a mobile communication system, which may be implemented, partly or totally, by one or more network base station and / or one or more user equipment described above in FIGs. 2-3. The various embodiments in the present disclosure may enable efficient wireless transmission in the telecommunication system, which may increase the resource utilization efficiency and / or boost wireless communication performance.
[0046] Referring to FIG. 4A, the present disclosure describes various embodiments of a method 400 for power saving in wireless communication. The method 400 may be performed by a wireless communication device (e.g., a user equipment) . The method 400 may include a portion or all of the following: step 410, measuring, by a user equipment (UE) , current communication resource in a current active state before a power-saving state, to obtain a current measurement; step 420, obtaining, by the UE, a measurement for future communication resource in a future active state after the power-saving sate, wherein the measurement for future communication resource is determined based on the current measurement; and / or step 430, transmitting, by the UE to a base station in the current active state, a measurement report indicating the measurement for future communication resource.
[0047] Referring to FIG. 4B, the present disclosure describes various embodiments of a method 450 for power saving in wireless communication. The method 450 may be performed by a wireless communication node (e.g., a base station or a radio access network (RAN) ) . The method 450 may include step 460, receiving, by a base station from a user equipment (UE) , a measurement report corresponding to a measurement for future communication resource in a future active state after a power-saving sate, wherein: the measurement for future communication resource is determined based on a current measurement, and the current measurement is obtained by measuring current communication resource in a current active state before the power-saving state.
[0048] In various embodiment in the present disclosure, power saving may be performed with the help of AI / ML models / functions in a mobile communication system. The current measurement may refer to measurement result at a current active state, e.g., before entering a power saving state. The measurement for future communication resource, which is predicted by AI / ML models / functions based on the current measurement and / or history measurement data, may refer to a predicted measurement for communication resource at a future active state, after a power saving state.
[0049] In some implementations, optionally or additionally to any one or any combinations of one or more implementations or embodiments in the present disclosure, the current communication resource and the further communication resource comprise at least one of the following: a first time domain resource and a second time domain resource, a first frequency domain resource and a second frequency domain resource, or a first spatial domain resource and a second spatial domain resource, respectively.
[0050] In some implementations, optionally or additionally to any one or any combinations of one or more implementations or embodiments in the present disclosure, the measurement for future communication resource is determined based on the current measurement and / or one or more history measurement; the current measurement comprises at least one of the following: a current active time period, a current bandwidth part (BWP) , and / or a set of current antenna ports; and / or the measurement for future communication resource comprises at least one of the following: a future active time period, a future BWP, and / or a set of future antenna ports.
[0051] In some implementations, optionally or additionally to any one or any combinations of one or more implementations or embodiments in the present disclosure, the future BWP in the measurement for future communication resource comprises time domain information and frequency domain information; and / or the set of future antenna ports in the measurement for future communication resource comprises time domain information and spatial domain information.
[0052] In some implementations, optionally or additionally to any one or any combinations of one or more implementations or embodiments in the present disclosure, the base station transmits configuration information to the UE, and / or the configuration information comprises at least one of the following: the future active time period, the future BWP, or the set of future antenna ports.
[0053] In some implementations, optionally or additionally to any one or any combinations of one or more implementations or embodiments in the present disclosure, the measurement for future communication resource is reported by the UE in an uplink transmission corresponding to at least one of the following: a current active time period, a current bandwidth part (BWP) , or a set of current antenna ports.
[0054] In some implementations, optionally or additionally to any one or any combinations of one or more implementations or embodiments in the present disclosure, the current measurement and the measurement for future communication resource comprise a current time stamp and a future time stamp, respectively; and / or the measurement for future communication resource is determined based on the future time stamp, or the future time stamp is determined based on the future communication resource.
[0055] In some implementations, optionally or additionally to any one or any combinations of one or more implementations or embodiments in the present disclosure, the future active time period comprises at least one of the following: a starting time and a duration of future active time period or inactivation time, stopping time on reception or synchronization, or a skipping indication indicating that a reception of active time period or reception of downlink (DL) channels or signaling is skipped; the future BWP comprises a switching time and a target BWP; and / or the set of future antenna ports comprises a switching time and a target set of antenna ports.
[0056] In some implementations, optionally or additionally to any one or any combinations of one or more implementations or embodiments in the present disclosure, the future active time period is determined based on the traffic prediction; the future BWP is determined based on the traffic prediction; and / or the set of future antenna ports is determined based on the traffic prediction; wherein a traffic prediction comprising at least one of the following: a traffic load, a traffic flow, and / or a traffic distribution.
[0057] In some implementations, optionally or additionally to any one or any combinations of one or more implementations or embodiments in the present disclosure, the method may further include receiving, by the UE, a configuration from the base station, wherein the configuration is configured by the base station based on the measurement report for the future communication resource in the future active state.
[0058] In some implementations, optionally or additionally to any one or any combinations of one or more implementations or embodiments in the present disclosure, the measurement report comprises at least one of the following: a time stamp of the future active state, a predicted buffer status report (BSR) , a predicted channel state information (CSI) , a current power consumption, a predicted power consumption, an efficiency of energy, a pattern of power usage, information of computing power, a status of battery, an effect on service quality of service (QoS) due to power saving, a possibility of failure of a prediction function, or likelihood of validity of prediction.
[0059] In some implementations, optionally or additionally to any one or any combinations of one or more implementations or embodiments in the present disclosure, the pattern of power usage comprises at least one of the following: a time ratio of active periods or sleep periods in history, an average power consumption in each level of active, sleep, a transition mode, a power consumption during wake-up signal monitoring, a probability of missing detection on wake-up signal, and / or an amount of energy saved over a period due to using power saving solution; and / or the effect on service QoS due to power saving comprises at least one of the following: how using power saving affects the UE's overall QoS, a latency or issue in power saving mode when receiving or transmitting data or signals, and / or a deterioration degree of performance or latency.
[0060] In some implementations, optionally or additionally to any one or any combinations of one or more implementations or embodiments in the present disclosure, each of the current power consumption and the predicted power consumption comprises at least one of the following: an overall power consumption, a power consumption of an AI model, and / or a power consumption of a non-AI portion.
[0061] In some implementations, optionally or additionally to any one or any combinations of one or more implementations or embodiments in the present disclosure, the UE transmits the measurement report in response to a condition associated with a threshold being satisfied; and / or the base station determines the received measurement report as valid in response to the condition associated with the threshold being satisfied, wherein the threshold is either configured by the base station or pre-defined. For a non-limiting example, when a reference signal received power (RSRP) is higher than the threshold, the quality of prediction is fine, then the measurement report is supported.
[0062] In some implementations, optionally or additionally to any one or any combinations of one or more implementations or embodiments in the present disclosure, the measurement for future communication resource is determined from an AI model and the AI model is sliced into two portions, so as to be a two-side AI model; and / or one portion of the AI model is disposed in the UE, and the rest portion of the AI model is disposed in the base station.
[0063] In some implementations, optionally or additionally to any one or any combinations of one or more implementations or embodiments in the present disclosure, a slice pattern of the AI model is determined based on slice-determination information comprising at least one of the following: a channel capacity between the UE and the base station, a channel latency between the UE and the base station, carrier information between the UE and the base station, a number of active UEs, a strategy of slicing, a ratio of power division between the UE and the base station.
[0064] In some implementations, optionally or additionally to any one or any combinations of one or more implementations or embodiments in the present disclosure, a core network obtains the AI model after the AI model is trained; the core network receives the slice-determination information from the base station; the core network determines the slice pattern of the AI model based on the slice-determination information; and / or the core network transmits a command of slice to the base station and the UE.
[0065] In some implementations, optionally or additionally to any one or any combinations of one or more implementations or embodiments in the present disclosure, the AI model is trained in one of the following: the UE, the base station, or a third-party device; and / or the command of slice comprises at least one of the following: an indicator on whether to slice, a slice location, a way to slice, a result of slice, information for assisting slice, or one or more sliced model.
[0066] The present disclosure describes various exemplary embodiments for power saving with AI / ML models / functions in a mobile communication system, and the exemplary embodiments merely serve as examples and do not pose limitations. Any steps and / or operations in one same embodiment / implementation or more than one different embodiments / implementation in the present disclosure may be combined or arranged in any amount or order, as desired. Two or more of the steps and / or operations may be performed in parallel. Embodiments and implementations in the disclosure may be used separately or combined in any order. Further, each of the methods (or embodiments) may be implemented by processing circuitry (e.g., one or more processors or one or more integrated circuits) .
[0067] Embodiment Set I
[0068] The present disclosure describes various embodiments for methods about measurement prediction for power saving with AI / ML models / functions.
[0069] In some implementations, power consumption in the active time period, frequency segment or antenna port is extremely higher than the power consumption in the idle time period, frequency segment or closed antenna port. The adaptive usage on the time, frequency and antenna port is the key to save the power consumption based on the traffic load. When the traffic load is lower, the time, frequency, and / or spatial domain (e.g., one or more antenna ports) resources may be assigned or scheduled less. On the contrary, when the traffic load is higher, the time, frequency, spatial domain resources can be assigned or scheduled more.
[0070] In some implementations, the typical primitive time domain adaptive method to reduce the power consumption is DRX (Discontinuous Reception) / DTX (Discontinuous Transmission) for a UE. The DRX / DTX allows a UE entering into a sleep mode or low power mode. The DRX can reduce the unnecessary reception of PDCCH or other downlink channels, the DTX can reduce the unnecessary transmission of uplink channels. In some implementations, a network configures the DRX cycle through RRC signaling, each cycle consists of two time periods: an active period and a non-active period (or referred as active state / period and power saving state / period) . The state in active period is On-Duration or Wakeup, the state in non-active period is Off-Duration or sleep. During the active period, the terminal listens to the PDCCH, while during the non-active period, the terminal does not listen to the PDCCH, nor does it receive the PDSCH, nor does it respond to PUSCH scheduling. As well as DRX, DTX proceeds the similar wakeup and / or sleep behavior on uplink transmissions.
[0071] In some implementations, multiple sleep (inactivity) levels can be defined, for example, micro sleep, light sleep, deep sleep, with deeper sleep levels involving larger groups of components. Crucially, the depth of sleep modes correlates with the extent of component deactivation. However, a trade-off exists, as deeper sleep modes entail longer delays in transitioning away from that sleep mode. Similar definition can be for wakeup level, the deeper wakeup level needs more components active and more power. The configurations for DRX / DTX try to align or match the time distribution of service traffic, but actually as the service is very flexible, the semi-static configurations are not so adaptive for the service traffic. In some implementations, the outcome of power saving of DRX / DTX is not that good.
[0072] In some implementations, a typical frequency domain adaptive method to reduce the power consumption is based on the bandwidth adjustment of frequency segment. The narrow bandwidth of frequency segment consumes lower power but provides less traffic load, while wide bandwidth frequency segment consumes higher power but provides more traffic load. When the need of traffic load is lower, narrow bandwidth is assigned or scheduled, and vice versa. The granularity of frequency segment is typically the BWP (bandwidth part) or carrier. To adapt the service traffic, the candidates of BWP and / or carrier have different bandwidth, and BWP and / or carrier switching among different bandwidths is allowed or configured to adapt the change of traffic. The BWP and / or carrier switching has the same flaw which is not very flexible for change of traffic due to the semi-static configuration.
[0073] In some implementations, a typical spatial domain adaptive method to reduce the power consumption is based on the adjustment of number of antenna ports used. Fewer (or less) antenna ports consume lower power but provide less traffic load, while more antenna ports consume higher power but provide more traffic load. When the need of traffic load is lower, fewer (or less) antenna port is assigned or scheduled, and vice versa.
[0074] In some implementations, an alternative solution is to switch the resident BWP as different BWPs configured with different limitation on the maximum antenna ports. For a non-limiting example, a first BWP (BWP1) supports 2 maximum antenna ports, a second BWP (BWP2) supports 4 maximum antenna ports; and when the traffic is very large, UE can switch to BWP2 from BWP1 to facilitate the increment of traffic.
[0075] FIG. 5A shows a schematic diagram illustrating a primitive adaption for traffic on time, frequency, and spatial domains.
[0076] In some implementations, the common flaw for the ways of DRX / DTX, BWP / carrier switching, limitation on the maximum antenna ports to adapt the traffic is not so flexible for change of traffic and the overhead of configuration signaling cannot be ignored. The power saving gain may be further enhancement by more flexible and more adaptive solutions.
[0077] In some implementations, the AI / ML model is native for traffic prediction. The traffic flow, distribution in the future can be precisely predicted by AI / ML interference process in an AI / ML model. The AI / ML model is trained with the input of data of traffic.
[0078] In some implementations, when AI / ML model for traffic prediction is used in UE, the fixed configuration of DRX cycle, configured BWP / carrier switching, configured maximum antenna ports are not needed. UE can autonomously activate or inactivate the reception or transmission in time domain, can autonomously decide which BWP / carrier is selected and activated, whatever the bandwidth of BWP / carrier is narrow or wide, and / or can autonomously decide the maximum antenna ports needed, all of the decisions are based on the future traffic prediction. When the traffic prediction is accurate enough compared with the reality, the adaptation may be close to the ideal state and the effect of power saving may reach the peak.
[0079] In some implementations, when AI / ML model for traffic prediction is used in a network device (e.g., a network or a base station) , the UE doesn’t know when to activate or inactivate the time period for future reception. The network may indicate the time information related to activation or inactivation through downlink (DL) signaling in the current active time period. The time information may include the start time and duration of time period when activation in the future is determined. The time information may alternatively include the time to be inactivated or reception stopping, or include the skipping indication which indicates the reception of already configured active time period or DL channels / signaling may be skipped. In some implementations, similar indication may also be applied to switching of frequency domain or spatial domain resources. The network may indicate the target BWP, carrier, and / or antenna port information to be switched to and together with the time related to switching through DL signaling in the current active time period.
[0080] In some implementations, the best prediction of traffic may help the UE to further reduce the power consumption. Some DL measurements, which may include a portion or all of the following: RSRP, channel state information (CSI) , channel quality information (CQI) and / or etc., from SSB / DL RS / special DL signals are also needed in advance to prepare the DL synchronization and / or reception in future time, in the target BWP / carrier to be switched, on the unused antenna ports to be used. Some non-limiting examples are shown in FIG. 5B to illustrate some measurements in advance.
[0081] In some implementations, in the time domain, the measurement occurs in advance the next (future) active period, wherein a UE may trigger the receiver to active state in sleep period and do measurements. The additional power consumption for these measurements is not what is wanted, and is indeed necessary to prepare the reception or synchronization in next active period.
[0082] In some implementations, in the frequency domain, the measurement occurs in the BWP / carrier to be switched. When a UE supports the simultaneously measurements parallel in multiple BWP / carrier, the power consumption of measurement added in target BWP / carrier is needed. When a UE doesn’t support the simultaneously measurements parallel in multiple BWP / carrier, the UE may need suspend reception / transmission in current BWP / carrier and switch to target BWP / carrier to measure. This switching itself need more power consumption and negatively affects the latency of traffic.
[0083] In some implementations, in the spatial domain, the measurement occurs in the antenna ports to be switched or added. When a UE supports the simultaneously measurements parallel in multiple antenna ports, the power consumption of measurement added in new antenna ports is needed. When a UE doesn’t support the simultaneously measurements parallel in multiple antenna ports, the UE may need suspend reception / transmission in current antenna ports and switch to target antenna ports to measure. This switching also needs more power consumption and negatively affects the latency of traffic.
[0084] In some implementations, in preparing procedure for synchronization and / or reception in the next (future) active time period, target BWP / carrier, target antenna ports, the in-advance measurements, for example, the synchronization time or the channel state information are needed, and the time gap between the measurements and next (future) active time start point may be short enough to maintain the synchronization time or the channel state information stable because the synchronization time may drift and the channel state information may change a lot due to the time-varying characteristics of the channel. That means the measurements for preparing the next (future) active time period, target BWP / carrier, target antenna ports, may most highly occur in the sleep time period just before the next active time period. In some implementations, the measurements in last (current) active time period cannot be reused for the next active time period.
[0085] In some implementations, after the in-advance measurement, the measure result may also be reported to a network for decision of DL schedule. For example, the measurement report of CSI / CQI may help the network to determine the scheduling information such as modulation and coding scheme (MCS) , number of codeword, RANK for multiple-input multiple-output (MIMO) , modulation, transport block (TB) size, transmission resources of PDSCH, etc. As the report is carried in uplink transmission, and the current active time period or current BWP / carrier or current antenna port is described from downlink transmission aspect, the uplink transmission time may be not the same with the current downlink active time period, and the BWP, carrier, and / or antenna port may be not the same with current BWP, carrier, and / or antenna port. Then the time, frequency, and / or spatial resources used uplink transmission are just corresponding to the current active time period, BWP, carrier, antenna ports, but maybe not the same. The measurement report depends on the UL transmission which consume lots of UE power.
[0086] In some implementations, to further reduce the power on in-advance measurement and switching, the measurement prediction, such as AI based measurement prediction or other possible prediction way, is recommended to be used, for example, based on the measurements in current and / or history active time period, or on current BWP / carrier, or on current antenna port. In some implementations, even though the time gap between the current active time period and future active time period is very large so that the channel character cannot be kept stable, the measurements in current and / or history active time period may be reused to precisely predict the future time / frequency / spatial domain synchronization time or the channel state information or other possible information to be measured, the detailed measurement items and related report are described in other portions of this disclosure. In some implementations, the aforementioned measurements in the sleep time period, in the target BWP / carrier, and / or in the target antenna ports, are naturally omitted and the power consumption on the measurement is reduced accordingly.
[0087] In some implementations, as shown in FIG. 5C, using the measurements in current active period to predict measurements in next active period is a single dimension prediction on the time domain. Using the measurements in current BWP / carrier to predict measurements in target BWP / carrier is a two-dimension prediction occurring on time domain and frequency domain as the time gap due to BWP / carrier switching. Similarly, using the measurements in current antenna ports to predict measurements in target antenna ports is also a two-dimension prediction occurring on time domain and spatial domain as the time gap due to antenna switching.
[0088] In some implementations, as shown in FIG. 6, the measurement prediction based on a function of AI model inference, not precluding other possible non-AI model based prediction, e.g., Wiener Filter, is for next (future) active period or target BWP / carrier, or target antenna ports. The active period is a general description of time period that a UE does reception / transmission, unnecessary be configured from the DRX signaling. The input of inference function of the AI model 650 includes the actual measurement at current active period 620 and / or one or more measurements at history active time period 610; and the output of the inference function of the AI model 650 may include the measurement prediction for next active period, or target BWP / carrier, or target antenna ports additional with time 630. The time (stamp) of actual measurement and the predicted measurement may be additionally as the function input aiming for prediction, and the time (stamp) of predicted measurement may alternatively be as function output when this time is not as input of function.
[0089] In some implementations, the next (future) active time period start point may be determined based on the traffic prediction or the DRX cycle configuration. Similarly, the time of activation on target BWP / carrier or target antenna ports is also based on the traffic prediction or the configuration. In some implementations, the future time may also be the input of inference function of AI model. When the traffic prediction, such as traffic flow or distribution is made in a UE, the time (stamp) as the function input or output is an implementation issue inside the UE. When the traffic prediction is made in a network, the time (stamp) may be indicated or configured to a UE for measurement prediction.
[0090] In some implementations, when the traffic prediction cannot be known by a UE or traffic prediction is not in the UE, the UE may not know the time to be predicted, and the predefine future time to be predicted can be used, for example, the predefined time offset based on the current measurement time, current active period starting point or etc. Alternatively, the future time to be predicted can also be configured from the network to the UE. The number of future times to be predicted can be only one or multiple. Multiple future times may be a set of times to be predicted.
[0091] In some implementations, the target BWP / carrier, antenna ports may also be configured to a UE from a network as the candidates for UE prediction. Clearly, multiple target BWPs, carriers, and / or antenna ports in a set are preferred as the UE doesn’t know whether the narrower or wider BWP / carrier is needed, more or fewer / less antenna ports are needed in future.
[0092] In some implementations, an AI model for inference may be trained as shown in FIG. 7, wherein some input of training may be same as the input of inference. The input of the training for AI model of measurement prediction 750 may include the actual measurement at current active period 720 and / or one or more measurements at history active time period 710, optionally with the time. The other input, as model training ground truth label, are the measurements on next active period, or target BWP / carrier, or target antenna ports, optionally with the time of measurement 730.
[0093] Some of the described implementations or embodiments in the present disclosure may use DL measurements and prediction of DL as examples, and the prediction is made in a UE and the motivation is to reduce the power consumption in the UE, which are not limiting to the applicability of the present disclosure. More generally, some UL measurements about RSRP, CSI, CQI and etc. from SRS / RACH / RACH-like signal may also needed in advance to prepare the UL synchronization / reception in future time, in the target BWP / carrier to be switched, on the unused antenna ports to be used. Thus, the present disclosure is also applicable to similar prediction to UL measurements and the prediction function is in a network (e.g., a base station) .
[0094] Various embodiments in the present disclosure may reduce the power consumption due to measurement / report / switching on frequency or antenna port for preparing the next / future active time period reception / synchronization. The prediction of measurement on future active time period (time domain) and / or on target BWP / carrier (frequency domain) and / or on target antenna port (spatial domain) may be based on the measurements on current and / or history active time period, or on current BWP / carrier, or on current antenna port. The outcome of prediction is the measurement of the future time / frequency / spatial domain synchronization time or the channel state information or other possible information to be measured.
[0095] In some implementations, the prediction of measurement is reported in uplink transmission corresponding to the current active time period or current BWP / carrier or current antenna port. In some implementations, the prediction can be based on AI model, specifically on the inference function of AI model. In some implementations, the prediction of the measurements of target BWP / carrier is a two-dimension prediction occurring both on time domain and frequency domain. The prediction of measurements of target antenna port is also a two-dimension prediction occurring on both time domain and spatial domain. In some implementations, the time (stamp) of actual measurement or the predicted measurement may additionally be as the input of function for prediction. Or the time (stamp) of predicted measurement may alternatively be as output of function for prediction. In some implementations, the starting time and duration of future active time period are determined based on the traffic prediction or the DRX cycle configuration. The target BWP / carrier or target antenna ports together with time related to switching are also based on traffic prediction or corresponding configuration.
[0096] In some implementations, the prediction of traffic, such as traffic load, flow or distribution, is made in either the UE or the base station. When the traffic prediction is made in the network, some signaling are needed to transmit from the network to the UE. The future time information related to starting and duration of future active time period, target BWP / carrier / antenna ports information together with the time related to switching should be indicated to the UE through DL signaling in the current active time period for measurement prediction occurring in the UE. Alternatively, the time information may include the time of inactivation or stopping on reception / synchronization, or include the skipping indication which indicates the reception of active time period or reception of DL channels / signaling may be skipped.
[0097] In some implementations, the future time (set) , target BWP / carrier (set) , target antenna ports (set) may be configured to UE from network in order to be predicted. Or these candidates can be predefined. The similar consideration as described above may be applied to uplink measurement accordingly.
[0098] Embodiment Set II
[0099] The present disclosure describes various embodiments for methods about report for power saving with AI / ML models / functions.
[0100] In some implementations, the report may include a portion or all of the following: a predicted BSR (buffer status report) , a predicted CSI (channel state information) , an overall power consumption, a power consumption of AI model, a power consumption of non-AI model, an efficiency of energy, a pattern of power usage, information of computing power, a status of battery, an effects on service quality of service (QoS) due to power saving, a possibility of failure of AI model, a likelihood of validity of prediction in the report. The AI model may be used for traffic prediction, and / or measurement prediction.
[0101] In some implementations, the BSR here may not be the legacy BSR, not the buffer status at current report time, but the prediction of buffer status in future. The predicted BSR may be together with a certain time in future which reflects the buffer status.
[0102] In some implementations, the CSI may not be the legacy CSI, and it is the prediction of CSI from prediction of measurement. The predicted CSI may include the predicted RSRP, CQI, RI (rank indicator) , PMI and so on, even including results of CSI after compression.
[0103] In some implementations, the status of power consumption also includes the current power consumption and predicted power consumption. The unit of power consumption may be Watt or Joule. The types of power consumptions may also be differentiated by overall power consumption, power consumption of AI model, power consumption of non-AI model. These can reflect the power used by AI models themselves. Similar with above report, when the report is the prediction, it is recommended to report with the associated time stamp.
[0104] In some implementations, the efficiency of energy is used to evaluate the throughput per Watt, or the consumption of power per bit. The report of efficiency of energy may be the history statistics or a prediction of future.
[0105] In some implementations, the concept of the pattern of power usage is wide. It may mean the time ratio of active periods or time ratio of sleep periods in history, or the average power consumption in each level of active / sleep / transition mode, or power consumption during wakeup signal monitoring, or probability of missing detection on wakeup signal, or overall power saving (total amount of energy saved over a period due to the use of power saving solution) . This report may help the network to make a decision on the strategy of power saving.
[0106] In some implementations, the information of computing power may include a portion or all of the following. Computing power in the terminal is very import to determine whether it can support AI / ML models processing or what kind of scale of the AI / ML model can be supported. The UE may report the remaining computing power periodically or aperiodically (e.g, following a request command) , the remaining computing power may also be named as computing power headroom. The computing power may refer to the FLOS (Floating Point Operations Per Second) of graphics processing unit (GPU) , Operations Per Second of neural processing unit (NPU) , MIPS (Million Instructions Per Second) of central processing unit (CPU) or other possible processing units.
[0107] In some implementations, the status of battery reports the remaining capacity of battery, drain rate of battery and so on.
[0108] In some implementations, the effects on service QoS due to power saving is reported on how the use of power saving affects the UE's overall Quality of Service (QoS) , such as the latency or issues in power saving mode when receiving / transmitting data or signals. The effects can focus on the negative aspect, for example, the deterioration degree of performance, latency.
[0109] In some implementations, the possibility of failure of prediction may include a portion or all of the following. A metric reports whether the prediction, for example using AI model to predict for power saving, is successful or failed, for example, 0 means successful, 1 means failed, and the value between 0 and 1 means the possibility of failure of prediction.
[0110] In some implementations, the likelihood of validity of prediction in the report may include a portion or all of the following. When the report consists of the measurement prediction, especially predicted by the AI mode, the report can be associated with a likelihood of validity of prediction. The likelihood of validity may also be called confidence level, credibility, or reliability. This term describes the percentage / degree of confidence that the prediction may be trusted. The network can determine whether to use the prediction in report according to the likelihood of validity together with the prediction report.
[0111] In some implementations, the report may be constructed according to a portion or all of the following.
[0112] For normally the report supporting the current measurement, e.g., the measurement result of current CSI, now the prediction of measurement may be added into the report as a whole, or reports separately.
[0113] In some implementations, the separate report structure of signaling may simply follow the current structure, just with a flag of indicating it is a prediction, or with a time stamp which implies it is a prediction in the future.
[0114] In some implementations, when the prediction is added into the legacy report as a whole, the structure of signaling can also be reused as much as possible. For example, there is a current CSI measurement and a predicted CSI measurement, the structure of multiple CSI report can be reused. The only thing to be done is added with a flag of indicating it is a prediction, or with a time stamp which implies it is a prediction.
[0115] In some implementations, the threshold for report may include a portion or all of the following. A report may consist of the prediction of measurement in future, in target BWP, carrier, and / or antenna port, and the prediction is likely to be invalid or untrusted when the current measurement level is lower / higher than a value. The value is a threshold which implies whether the prediction should be turned off. The threshold is configured by a network to a UE or predefined in specification. In some implementations, when the current measurement result is lower than the threshold, the default behavior for UE device is to turn off the prediction report; and / or when the current measurement result is higher than the threshold, the UE device may additionally transmit the prediction report. In some implementations, when the network checks that the current measurement is lower than the threshold, network assumes the prediction report is invalid, even when the prediction report is received (or exists) . For example, the network can configure the RSRP threshold to aid the UE to determine whether to turn off the prediction report. When the RSRP measurement is lower than corresponding threshold, the UE may obey the default behavior which is turning off the prediction report. On the contrary, when the current measurement result is higher than the threshold, the default behavior for UE device may be to turn off the prediction report, for example, the threshold is configured on Path Loss. Larger path loss means the UE is farther away from base station. Furthermore, it is possible to configure a scope of measurement result to help the UE to determine whether to turn off the prediction report, for example, the scope of the measurement on variation rate of SIR (first derivative of SIR, ) , when the measurement result is within the scope configured, the UE may turn on the prediction report, otherwise, the UE may turn off the prediction report.
[0116] Various embodiments in the present disclosure may be implemented to reduce the power consumption due to supply the items of report, structure of report, and threshold of report.
[0117] In some implementations, the report (e.g., being transmitted from a UE to a network) may include a portion or all of the following: the predicted BSR (buffer status report) , predicted CSI (channel state information) , current / predicted power consumption (overall power consumption, power consumption of AI model, power consumption of non-AI) , the efficiency of energy, the pattern of power usage, the information of computing power, the status of battery, the effects on service QoS due to power saving, the possibility of failure of AI model, the likelihood of validity of prediction, and so on.
[0118] In some implementations, the predicted report may include the associated time stamp of the future.
[0119] In some implementations, the possibility of failure of prediction and the likelihood of validity of prediction may be associated with the corresponding prediction of measurement, the prediction of measurement and associated possibility / likelihood are highly recommended to be reported together.
[0120] In some implementations, the details of pattern of power usage, the effects on service QoS due to power saving can also be in the claims to improve the novelty when possible and practical.
[0121] In some implementations, the prediction report is encouraged to reuse the legacy signaling structure, whatever separate report or inseparable reports. The flag of indicating the report is from prediction, or a time stamp which implies the report is from prediction should be added with the prediction of measurement in report.
[0122] In some implementations, a threshold is configured by network or predefined to indicate the default behavior for UE is to turn off the prediction report or mark the prediction report invalid, when the current measurement is lower / higher than the threshold. Vice versa, the default behavior for UE is to turn on the prediction report or mark the prediction report valid.
[0123] Embodiment Set III
[0124] The present disclosure describes various embodiments for methods about end to end power saving coordination for power saving with AI / ML models / functions.
[0125] In some implementations, the power consumption of the whole wireless communication system at least includes power consumption of a network, power consumption of a UE, and / or power consumption of other third party devices. The power saving or energy saving may be considered as the collaboration between the network and UE for joint energy saving. In some implementations, it is not desirable that in order to reduce the power of UE, the complexity and power consumption of the network is unilaterally increased. It is also not desirable that in order to reduce the power of the network, the complexity and power consumption of the UE is unilaterally increased. The power consumption of the network and UEs may be considered together as a whole, so as to reach a compromise or balance.
[0126] In some implementations, the power consumption of AI / ML model can’t be ignored in the whole power consumption of system. When the AI model in the UE is very complicated, the UE may not afford / bear the power consumption of AI model. For one solution, cutting apart the single side AI model into double-side or two-side AI models and moving one side model of the double-side models into the network may reduce the power consumption in the UE.
[0127] In some implementations, the AI / ML model / function for measurement prediction in various embodiments in the present disclosure may be separated into two parts, one of the two parts can be moved to network.
[0128] Suppose the AI / ML model / function is a MLP (Multilayer Perceptron) artificial neural network, wherein it includes an input layer, an output layer, and a plurality of hidden layers, e.g., two hidden layers, as shown in FIG. 8. Each layer has a plurality of neurons (represented by circles) . The number of layers represents the depth of the AI Neural Network, and the number of neurons in each layer represents the width of layers.
[0129] FIG. 8 shows MLP artificial neural network and different divisions, for non-limiting examples, there may be at least four kinds of division / slice solutions for this MLP artificial neural network (also named as model) . The first division / slice method (with division 1) is cutting apart the model between two hidden layers. The second and third division / slice methods (with division 2 and division 3) are cutting apart the model after input layer and before the output layer, respectively. The four division / slice method (with division 4) is cutting apart the model in width dimension, but not in layer dimension.
[0130] In some implementations, when the single side model is cut into two-side models, one side is disposed in a UE, and the other side is disposed in a network. The connection between the two-side models is needed, and the original model internal link is exposed outside and transmissions of related parameters, forward propagation and back propagation should be carried by the air interface between network and UE directly or transparently. Different divisions have different air interface payload or overhead requirements. The first division or the fourth division may have more payload on parameter, forward propagation and back propagation to be transmitted. The second and third divisions may have less payload.
[0131] In some implementations, at least two aspects may be considered to determine how to slice the model, e.g., determining the location of slicing. One aspect is the resources provided by the tube between the RAN network and a UE, including more specifically, the channel capacity, latency of the L1 / L2 / L3 of RAN network and the UE. Some other information may be also needed such as the carrier information, the number of active UEs in the area. These information may help to evaluate / judge the potential resources to carry the potential data exchange between the two-side models. The other aspect is the tradeoff on division of the power consumption from the two-side models. For example, an optimal power division is 80%for base station, 20%for UE.
[0132] In some implementations, the core network or other third party device may determine the slice location of the AI / ML model / function, which cuts apart the one side model into two-side models, based on the input or gathered information of channel capacity and latency, optionally with carrier information, the number of active UEs and so on. The tradeoff of the power consumption division between the network and the UE is considered for the overall power saving strategy.
[0133] For a non-limiting example, FIG. 9 illustrates a procedure of the slice determination. The procedure may include a portion or all of the steps, and may be performed by a portion or all of the following: a core network 991, a base station 992, a UE 993, and / or a third device 995. The In some implementations, one side model is trained in UE, or base station or core network or the 3rd party devices. The one side mode is transferred to the core network, when the trained model is not in core network. For example, step 910: when the trained AI / ML model (or a portion of the trained model) is in the UE, the UE transfers the trained AI / ML model to the core network; and / or step 912: when the trained AI / ML model (or a portion of the trained model) is in the base station, the base station transfers the trained AI / ML model to the core network; and / or step 914: when the trained AI / ML model (or a portion of the trained model) is in the third device, the third device transfers the trained AI / ML model to the core network.
[0134] In step 920: the core network gathers the information of channel capacity and latency, optionally with carrier information, the number of active UEs or the strategy of slice e.g., the ratio of power division between the network and UE and so on.
[0135] In step 930, the core network determines the slice of one side model to two-side model. The determined details may include a portion or all of the following: an indicator on whether to slice, slice location, the way to slice, the result of slice or other aiding information for slice, specifically the result of slice can be the sliced two-side models. The strategy of slice may be based on the ratio of power division between the network and the UE.
[0136] In steps 940 and 942, the core network transmits the slice command to the base station and / or the UE. The moving or transferring of sliced two-side models to different devices, such as base station and UE can be part of the slice command. Alternatively, the details on the slice such as the indicator on whether to slice, slice location, the way to slice or other aiding information for slice can also be part of the slice command to aid base station and UE to construct the two-side models.
[0137] In some implementations, the core network is the device to gather the model, gather the aid information and make determination on model slice. In some implementations, the third party device except core network can replace the above functions as the center device of model slice.
[0138] In some implementations, the third party device may be a portion of the base station or may be built in the base station. When the initial single side model is trained in base station, there is no need to transfer model to other device outside of base station. The aid information of channel capacity, latency and so on are also gathered inside the base station. The determination is made in base station and slice command is forward from the base station to the UE.
[0139] Various embodiments in the present disclosure may slice the one side model to two-side models to reduce the power consumption in UE. In some implementations, from the aspect of main device to determine the slice, a core network gets the trained one side model from devices, such as base station, UE, or 3rd party devices. The core network gets the information of channel / tube capacity and latency, optionally with carrier information, the number of active UEs or the strategy of slice e.g., the ratio of power division between the network and UE. The core network determines the slice of one side model to two-side model; for example, determine the slice location. The dimension of slice can be in width or in layer. The core network sends the command of slice, with aiding information if any, to other devices which would reside the two-side models. The command of slice may include the indicator on whether to slice, slice location, the way to slice, the result of slice or other aiding information for slice, or directly move / transfer sliced models to other devices. In some implementations, the one side model includes a AI / ML model / function.
[0140] The present disclosure describes methods, apparatus, and computer-readable medium for power saving with AI / ML models / functions in a mobile communication system. The present disclosure addressed the issues with power saving in a wireless communication system. The methods, devices, and computer-readable medium described in the present disclosure may facilitate the performance of power saving with AI / ML models / functions in wireless communication, thus improving efficiency and overall performance. The methods, devices, and computer-readable medium described in the present disclosure may improves the overall efficiency of the wireless communication systems.
[0141] In some other embodiments, a computer-readable medium comprising instructions which, when executed by a computer, cause the computer to carry out the above methods. The computer-readable medium may be referred as non-transitory computer-readable media (CRM) that stores data for extended periods such as a flash drive or compact disk (CD) , or for short periods in the presence of power such as a memory device or random access memory (RAM) . In some embodiments, computer-readable instructions may be included in a software, which is embodied in one or more tangible, non-transitory, computer-readable media. Such non-transitory computer-readable media can be media associated with user-accessible mass storage as well as certain short-duration storage that are of non-transitory nature, such as internal mass storage or ROM. The software implementing various embodiments of the present disclosure can be stored in such devices and executed by a processor (or processing circuitry) . A computer-readable medium can include one or more memory devices or chips, according to particular needs. The software can cause the processor (including CPU, GPU, FPGA, and the like) to execute particular processes or particular parts of particular processes described herein, including defining data structures stored in RAM and modifying such data structures according to the processes defined by the software. In various embodiments in the present disclosure, the term “processor” may mean one processor that performs the defined functions, steps, or operations or a plurality of processors that collectively perform defined functions, steps, or operations, such that the execution of the individual defined functions may be divided amongst such plurality of processors.
[0142] Reference throughout this specification to features, advantages, or similar language does not imply that all of the features and advantages that may be realized with the present solution should be or are included in any single implementation thereof. Rather, language referring to the features and advantages is understood to mean that a specific feature, advantage, or characteristic described in connection with an embodiment is included in at least one embodiment of the present solution. Thus, discussions of the features and advantages, and similar language, throughout the specification may, but do not necessarily, refer to the same embodiment.
[0143] Furthermore, the described features, advantages and characteristics of the present solution may be combined in any suitable manner in one or more embodiments, for non-limiting examples, a portion from one or more embodiment may be combined with another portion of other embodiments. One of ordinary skill in the relevant art will recognize, in light of the description herein, that the present solution can be practiced without one or more of the specific features or advantages of a particular embodiment. In other instances, additional features and advantages may be recognized in certain embodiments that may not be present in all embodiments of the present solution.
Claims
1.A method for power saving in a wireless communication system, comprising:measuring, by a user equipment (UE) , current communication resource in a current active state before a power-saving state, to obtain a current measurement;obtaining, by the UE, a measurement for future communication resource in a future active state after the power-saving sate, wherein the measurement for future communication resource is determined based on the current measurement; andtransmitting, by the UE to a base station in the current active state, a measurement report indicating the measurement for future communication resource.2.A method for power saving in a wireless communication system, comprising:receiving, by a base station from a user equipment (UE) , a measurement report corresponding to a measurement for future communication resource in a future active state after a power-saving sate, wherein:the measurement for future communication resource is determined based on a current measurement, and the current measurement is obtained by measuring current communication resource in a current active state before the power-saving state.3.The method according to any of claims 1 and 2, wherein:the current communication resource and the further communication resource comprise at least one of the following: a first time domain resource and a second time domain resource, a first frequency domain resource and a second frequency domain resource, or a first spatial domain resource and a second spatial domain resource, respectively.4.The method according to any of claims 1 and 2, wherein:the measurement for future communication resource is determined based on the current measurement and / or one or more history measurement;the current measurement comprises at least one of the following: a current active time period, a current bandwidth part (BWP) , or a set of current antenna ports; andthe measurement for future communication resource comprises at least one of the following: a future active time period, a future BWP, or a set of future antenna ports.5.The method according to claim 4, wherein:the future BWP in the measurement for future communication resource comprises time domain information and frequency domain information; andthe set of future antenna ports in the measurement for future communication resource comprises time domain information and spatial domain information.6.The method according to claim 4, wherein:the base station transmits configuration information to the UE, andthe configuration information comprises at least one of the following: the future active time period, the future BWP, or the set of future antenna ports.7.The method according to claim 4, wherein:the measurement for future communication resource is reported by the UE in an uplink transmission corresponding to at least one of the following: a current active time period, a current bandwidth part (BWP) , or a set of current antenna ports.8.The method according to claim 4, wherein:the current measurement and the measurement for future communication resource comprise a current time stamp and a future time stamp, respectively; andthe measurement for future communication resource is determined based on the future time stamp, or the future time stamp is determined based on the future communication resource.9.The method according to claim 4, wherein:the future active time period comprises at least one of the following: a starting time and a duration of future active time period or inactivation time, stopping time on reception or synchronization, or a skipping indication indicating that a reception of active time period or reception of downlink (DL) channels or signaling is skipped;the future BWP comprises a switching time and a target BWP; andthe set of future antenna ports comprises a switching time and a target set of antenna ports.10.The method according to claim 4, wherein:the future active time period is determined based on the traffic prediction;the future BWP is determined based on the traffic prediction; andthe set of future antenna ports is determined based on the traffic prediction;wherein a traffic prediction comprising at least one of the following: a traffic load, a traffic flow, or a traffic distribution.11.The method according to any of claims 1 and 2, further comprising:receiving, by the UE, a configuration from the base station, wherein the configuration is configured by the base station based on the measurement report for the future communication resource in the future active state.12.The method according to any of claims 1 and 2, wherein:the measurement report comprises at least one of the following: a time stamp of the future active state, a predicted buffer status report (BSR) , a predicted channel state information (CSI) , a current power consumption, a predicted power consumption, an efficiency of energy, a pattern of power usage, information of computing power, a status of battery, an effect on service quality of service (QoS) due to power saving, a possibility of failure of a prediction function, or likelihood of validity of prediction.13.The method according to claim 12, wherein:the pattern of power usage comprises at least one of the following: a time ratio of active periods or sleep periods in history, an average power consumption in each level of active, sleep, a transition mode, a power consumption during wake-up signal monitoring, a probability of missing detection on wake-up signal, or an amount of energy saved over a period due to using power saving solution; orthe effect on service QoS due to power saving comprises at least one of the following: how using power saving affects the UE's overall QoS, a latency or issue in power saving mode when receiving or transmitting data or signals, or a deterioration degree of performance or latency.14.The method according to claim 12, wherein:each of the current power consumption and the predicted power consumption comprises at least one of the following: an overall power consumption, a power consumption of an AI model, or a power consumption of a non-AI portion.15.The method according to any of claims 1 and 2, wherein:the UE transmits the measurement report in response to a condition associated with a threshold being satisfied; orthe base station determines the received measurement report as valid in response to the condition associated with the threshold being satisfied,wherein the threshold is either configured by the base station or pre-defined.16.The method according to any of claims 1 and 2, wherein:the measurement for future communication resource is determined from an AI model and the AI model is sliced into two portions, so as to be a two-side AI model; andone portion of the AI model is disposed in the UE, and the rest portion of the AI model is disposed in the base station.17.The method according to claim 16, wherein:a slice pattern of the AI model is determined based on slice-determination information comprising at least one of the following: a channel capacity between the UE and the base station, a channel latency between the UE and the base station, carrier information between the UE and the base station, a number of active UEs, a strategy of slicing, a ratio of power division between the UE and the base station.18.The method according to claim 17, wherein:a core network obtains the AI model after the AI model is trained;the core network receives the slice-determination information from the base station;the core network determines the slice pattern of the AI model based on the slice-determination information; andthe core network transmits a command of slice to the base station and the UE.19.The method according to claim 18, wherein:the AI model is trained in one of the following: the UE, the base station, or a third-party device; orthe command of slice comprises at least one of the following: an indicator on whether to slice, a slice location, a way to slice, a result of slice, information for assisting slice, or one or more sliced model.20.A wireless communications apparatus comprising at least one processor and a memory, wherein the at least one processor is configured to read instructions from the memory and implement the method recited in any of claims 1 to 19.21.A computer-readable medium comprising instructions which, when executed by a computer, causing the computer to carry out the method recited in any of claims 1 to 19.
Citation Information
Patent Citations
Early sleep state for circuitry associated with synchronization wakeup periods
US20230024081A1
AI / ML Data Collection and Usage Possibly for MDTs
US20230044727A1
Method for configuring WUS DCI and terminal using same method
US20230247620A1
Method and apparatus for beam management in communication system
US20230421238A1
Artificial intelligence based discontinuous reception configuration
WO2024016265A1