Communication method and apparatus, ai model training method and apparatus, and device and storage medium

Abstract

The present application belongs to the technical field of wireless communications. Disclosed are a communication method and apparatus, an AI model training method and apparatus, and a device and a storage medium. The communication method in an embodiment of the present application comprises: a first device acquiring first input information; on the basis of an AI model and the first input information, the first device obtaining uplink adjustment information, the uplink adjustment information being used for adjusting an uplink transmission state of a terminal; the first device sending to the corresponding terminal the uplink adjustment information obtained by means of inference by the AI model; and on the basis of the uplink adjustment information, the terminal adjusting the uplink transmission state.

Description

Communication method, AI model training method, device, equipment and storage medium

[0001] Priority information

[0002] The present application claims priority to the Chinese patent application No. 202411015909.4, filed on July 26, 2024, and entitled "Communication method, AI model training method, device, equipment and storage medium", the whole content of which is incorporated herein by reference. TECHNICAL FIELD

[0003] The present application belongs to the field of communication technology, and particularly relates to a communication method, an AI model training method, a device, an equipment and a storage medium. BACKGROUND

[0004] In the existing communication system, when a terminal performs uplink transmission, the terminal needs to obtain timing advance (TA) and frequency offset and the like information to adjust the uplink transmission state. Taking TA as an example, the TA is usually sent by a base station to a terminal, and is used to indicate the advance amount of time adjustment when the terminal sends an uplink (UP) symbol. The terminal sends the uplink symbol in advance according to the TA, so as to ensure that the uplink transmissions of different terminals arrive at the base station at substantially the same time.

[0005] In the related art, the base station usually measures the uplink signal and obtains TA and frequency offset and the like information for adjusting the uplink transmission state. However, in some scenarios, the accuracy of the information for adjusting the uplink transmission state measured by the base station is not high, thereby resulting in low accuracy and reliability of the uplink transmission. For example, for TA, in some scenarios, neither the base station nor the terminal has high accuracy of the measured TA, which cannot meet the requirements of the uplink transmission. SUMMARY

[0006] The embodiments of the present application provide a communication method, an AI model training method, a device, an equipment and a storage medium, which can solve the problem of low accuracy of TA and frequency offset and the like information for adjusting the uplink transmission state.

[0007] In a first aspect, a communication method is provided, which is performed by a first device, and the method comprises:

[0008] obtaining first input information;

[0009] obtaining uplink adjustment information according to an artificial intelligence (AI) model and the first input information, the uplink adjustment information being used to adjust an uplink transmission state of a terminal, and the uplink adjustment information comprising at least one of the following:

[0010] uplink adjustment information of a specific terminal;

[0011] uplink adjustment information of a group of terminals;

[0012] uplink adjustment information in a specific area;

[0013] uplink adjustment information of at least one beam direction;

[0014] uplink adjustment information of a specific reference point;

[0015] uplink adjustment information corresponding to a specific path loss range or a specific reference signal receiving power (RSRP) range;

[0016] uplink adjustment information corresponding to a specific sequence or sequence format;

[0017] common uplink adjustment information;

[0018] uplink adjustment information between a specific terminal and at least one network side device;

[0019] uplink adjustment information of other terminals in a terminal group;

[0020] a group of candidate uplink adjustment information and / or a probability value associated with the group of candidate uplink adjustment information.

[0021] In a second aspect, a method for training an AI model is provided, and the method is executed by a second device, and the method comprises:

[0022] obtaining a training sample, wherein the training sample comprises second input information and at least one label;

[0023] training a model using the training sample to obtain the AI model used in the first aspect, wherein the AI model outputs uplink adjustment information, and the uplink adjustment information is used to adjust an uplink transmission state of a terminal, and wherein the second input information at least comprises first input information used for inference of the AI model, and the uplink adjustment information comprises at least one of the following:

[0024] uplink adjustment information of a specific terminal;

[0025] uplink adjustment information of a group of terminals;

[0026] uplink adjustment information in a specific area;

[0027] uplink adjustment information of at least one beam direction;

[0028] uplink adjustment information of a specific reference point;

[0029] uplink adjustment information corresponding to a specific path loss range or a specific reference signal receiving power (RSRP) range;

[0030] uplink adjustment information corresponding to a specific sequence or sequence format;

[0031] common uplink adjustment information;

[0032] uplink adjustment information between a specific terminal and at least one network side device;

[0033] uplink adjustment information of other terminals in a terminal group;

[0034] a group of candidate uplink adjustment information and / or a probability value associated with the group of candidate uplink adjustment information.

[0035] In a third aspect, a communication apparatus is provided, and the apparatus comprises:

[0036] a processing module, configured to obtain first input information;

[0037] The processing module is further configured to obtain uplink adjustment information according to an artificial intelligence (AI) model and the first input information, wherein the uplink adjustment information is used to adjust an uplink transmission state of a terminal, and the uplink adjustment information comprises at least one of the following:

[0038] uplink adjustment information of a specific terminal;

[0039] uplink adjustment information of a group of terminals;

[0040] uplink adjustment information in a specific area;

[0041] uplink adjustment information of at least one beam direction;

[0042] uplink adjustment information of a specific reference point;

[0043] uplink adjustment information corresponding to a specific path loss range or a specific reference signal received power (RSRP) range;

[0044] uplink adjustment information corresponding to a specific sequence or sequence format;

[0045] common uplink adjustment information;

[0046] uplink adjustment information between a specific terminal and at least one network side device;

[0047] uplink adjustment information of other terminals in a terminal group;

[0048] a group of candidate uplink adjustment information and / or a probability value associated with the group of candidate uplink adjustment information.

[0049] In a fourth aspect, a training apparatus of an AI model is provided, and the apparatus comprises:

[0050] a processing module, configured to obtain a training sample, wherein the training sample comprises second input information and at least one label;

[0051] The processing module is further configured to perform model training using the training samples to obtain the AI model of the third aspect, the AI model outputs uplink adjustment information, and the uplink adjustment information is used to adjust the uplink transmission state of the terminal, wherein the second input information at least includes the first input information used for inference of the AI model, and the uplink adjustment information includes at least one of the following:

[0052] Uplink adjustment information of a specific terminal;

[0053] Uplink adjustment information of a group of terminals;

[0054] Uplink adjustment information in a specific area;

[0055] Uplink adjustment information of at least one beam direction;

[0056] Uplink adjustment information of a specific reference point;

[0057] Uplink adjustment information corresponding to a specific path loss range or a specific reference signal received power (RSRP) range;

[0058] Uplink adjustment information corresponding to a specific sequence or sequence format;

[0059] Common uplink adjustment information;

[0060] Uplink adjustment information between a specific terminal and at least one network side device;

[0061] Uplink adjustment information of other terminals in a terminal group;

[0062] A group of candidate uplink adjustment information and / or a probability value associated with the group of candidate uplink adjustment information.

[0063] In a fifth aspect, a communication device is provided, and the device is configured to perform the steps of the method of the first aspect.

[0064] In a sixth aspect, a training device of an AI model is provided, and the device is configured to implement the steps of the method of the second aspect.

[0065] In a seventh aspect, a communication device is provided, and the device includes a processor and a memory, the memory stores programs or instructions executable on the processor, and the programs or instructions are executed by the processor to implement the steps of the method of the first aspect.

[0066] In an eighth aspect, a communication device is provided, including a processor and a communication interface, wherein the processor is configured to: obtain first input information; and obtain uplink adjustment information according to an AI model and the first input information, the uplink adjustment information being used to adjust an uplink transmission state of a terminal, and the communication interface is configured to communicate with other devices.

[0067] The uplink adjustment information includes at least one of:

[0068] uplink adjustment information of a specific terminal;

[0069] uplink adjustment information of a group of terminals;

[0070] uplink adjustment information in a specific area;

[0071] uplink adjustment information of at least one beam direction;

[0072] uplink adjustment information of a specific reference point;

[0073] uplink adjustment information corresponding to a specific path loss range or a specific reference signal receiving power (RSRP) range;

[0074] uplink adjustment information corresponding to a specific sequence or sequence format;

[0075] common uplink adjustment information;

[0076] uplink adjustment information between a specific terminal and at least one network side device;

[0077] uplink adjustment information of other terminals in a terminal group;

[0078] a group of candidate uplink adjustment information and / or a probability value associated with the group of candidate uplink adjustment information.

[0079] In a ninth aspect, a communication device is provided, including a processor and a memory, the memory storing programs or instructions executable on the processor, and the programs or instructions are executed by the processor to implement the steps of the method according to the second aspect.

[0080] In a tenth aspect, a communication device is provided, including a processor and a communication interface, wherein the processor is configured to: obtain training samples, the training samples including second input information and at least one label; and perform model training using the training samples to obtain the AI model according to the first aspect, the AI model outputting uplink adjustment information, the uplink adjustment information being used to adjust an uplink transmission state of a terminal; and the communication interface is configured to communicate with other devices.

[0081] The second input information includes at least the first input information used for inference of the AI model, and the uplink adjustment information includes at least one of:

[0082] uplink adjustment information of a specific terminal;

[0083] uplink adjustment information of a group of terminals;

[0084] uplink adjustment information in a specific area;

[0085] uplink adjustment information of at least one beam direction;

[0086] uplink adjustment information of a specific reference point;

[0087] uplink adjustment information corresponding to a specific path loss range or a specific reference signal received power (RSRP) range;

[0088] uplink adjustment information corresponding to a specific sequence or sequence format;

[0089] common uplink adjustment information;

[0090] uplink adjustment information between a specific terminal and at least one network side device;

[0091] uplink adjustment information of other terminals in a terminal group;

[0092] a group of candidate uplink adjustment information and / or a probability value associated with the group of candidate uplink adjustment information.

[0093] In an eleventh aspect, a readable storage medium is provided, and the readable storage medium stores a program or instructions, which are executed by a processor to implement the steps of the method according to the first aspect or the steps of the method according to the second aspect.

[0094] In a twelfth aspect, a wireless communication system is provided, and the wireless communication system includes a first device and a second device, the first device is configured to implement the steps of the method according to the first aspect, and the second device is configured to implement the steps of the method according to the second aspect.

[0095] In a thirteenth aspect, a chip is provided, and the chip includes a processor and a communication interface, the communication interface is coupled to the processor, and the processor is configured to run a program or instructions to implement the method according to the first aspect or the method according to the second aspect.

[0096] In a fourteenth aspect, a computer program / program product is provided, and the computer program / program product is stored in a storage medium, and the computer program / program product is executed by at least one processor to implement the steps of the communication method according to the first aspect or the steps of the AI model training method according to the first aspect.

[0097] In the embodiments of the present application, the first device obtains first input information, and the first device obtains uplink adjustment information according to the AI model and the first input information, the uplink adjustment information being used to adjust an uplink transmission state of a terminal. The first device sends the uplink adjustment information inferred by the AI model to a corresponding terminal, and the terminal adjusts the uplink transmission state according to the uplink adjustment information. The accuracy of the uplink adjustment information inferred by the AI model is high, thereby ensuring the accuracy and reliability of data transmission. BRIEF DESCRIPTION OF DRAWINGS

[0098] FIG. 1 shows a block diagram of a wireless communication system to which embodiments of the present application can be applied;

[0099] FIG. 2 is a schematic diagram of the relationship between the transmission time of an uplink frame and that of a downlink frame;

[0100] FIG. 3 is a schematic diagram of a multi-layer cell-free network structure;

[0101] FIG. 4 is a schematic diagram of an operation framework of an AI model;

[0102] FIG. 5 is a flowchart of a communication method according to an embodiment of the present application;

[0103] FIG. 6 is a flowchart of a data processing method according to an embodiment of the present application;

[0104] FIG. 7 is a flowchart of a training method of an AI model according to an embodiment of the present application;

[0105] FIG. 8 is a signaling flowchart of a RACH-less cell switching procedure according to an embodiment of the present application;

[0106] FIG. 9 is a signaling flowchart of a RACH-less cell switching procedure according to an embodiment of the present application;

[0107] FIG. 10 is a signaling flowchart of a communication method in a cell-free scenario according to an embodiment of the present application;

[0108] FIG. 11 is a signaling flowchart of a communication method in a cell-free scenario according to an embodiment of the present application;

[0109] FIG. 12 is a signaling flowchart of a communication method according to an embodiment of the present application;

[0110] FIG. 13 is a schematic block diagram of a communication apparatus according to an embodiment of the present application;

[0111] FIG. 14 is a schematic block diagram of a training apparatus of an AI model according to an embodiment of the present application;

[0112] FIG. 15 is a schematic structural diagram of a communication device according to an embodiment of the present application;

[0113] FIG. 16 is a schematic diagram of a hardware structure of a terminal device according to an embodiment of the present application;

[0114] FIG. 17 is a schematic diagram of a hardware structure of a network side device according to an embodiment of the present application. DETAILED DESCRIPTION

[0115] The technical solutions in the embodiments of the present application will be clearly described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments in the present application, all the other embodiments obtained by those of ordinary skill in the art belong to the scope of protection of the present application.

[0116] The terms "first", "second", and the like in the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described herein, and the objects distinguished by "first", "second" are generally a category and do not limit the number of objects, for example, the first object can be one or more. In addition, "or" in the present application means at least one of the connected objects. For example, the protection scope of "A or B" at least covers three schemes, namely, scheme one: including A and not including B; scheme two: including B and not including A; scheme three: including A and including B. In addition, the terms "A and / or B", "at least one of A and B", "at least one of A or B" also at least cover the above three schemes, respectively. The character " / " generally represents that the objects before and after are in an "or" relationship.

[0117] The term "indication" in the present application can be a direct indication (or explicit indication) or an indirect indication (or implicit indication). The direct indication can be understood as that the sender explicitly informs the receiver of the specific information, the operation to be performed or the request result, etc. in the indication sent by the sender. The indirect indication can be understood as that the receiver determines the corresponding information according to the indication sent by the sender, or judges and determines the operation to be performed or the request result, etc. according to the judgment result.

[0118] It is worth noting that the technology described in the embodiments of the present application is not limited to Long Term Evolution (LTE) / LTE-Advanced (LTE-A) systems, but can also be used in other wireless communication systems, such as Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA), Frequency Division Multiple Access (FDMA), Orthogonal Frequency Division Multiple Access (OFDMA), Single-carrier Frequency-Division Multiple Access (SC-FDMA) or other systems. The terms "system" and "network" in the embodiments of the present application are often used interchangeably, and the described technology can be used in the above-mentioned systems and radio technologies, as well as in other systems and radio technologies. The following description describes a New Radio (NR) system for example purposes, and NR terminology is used in most of the following description, but these technologies can also be applied to systems other than NR systems, such as 6th Generation (6G) communication systems. th

[0119] ​FIG. 1 shows a block diagram of a wireless communication system to which embodiments of the present application can be applied. The wireless communication system includes a terminal 11 and a network-side device 12. The terminal 11 can be a terminal-side device such as a mobile phone, a Tablet Personal Computer, a Laptop Computer, a notebook computer, a Personal Digital Assistant (PDA), a palmtop computer, a netbook, an Ultra-mobile Personal Computer (UMPC), a Mobile Internet Device (MID), an Augmented Reality (AR) device, a Virtual Reality (VR) device, a robot, a Wearable Device, a flight vehicle, a Vehicle User Equipment (VUE), a shipboard device, a Pedestrian User Equipment (PUE), a smart home (a home device with a wireless communication function such as a refrigerator, a television, a washing machine, or furniture), a game machine, a Personal Computer (PC), a kiosk, or a self-service machine. The Wearable Device includes a smart watch, a smart bracelet, a smart earphone, smart glasses, smart jewelry (a smart bracelet, a smart necklace, a smart ring, a smart necklace, a smart anklet, a smart necklace, and the like), a smart wristband, smart clothes, and the like. The vehicle-mounted device can also be referred to as a vehicle-mounted terminal, a vehicle-mounted controller, a vehicle-mounted module, a vehicle-mounted component, a vehicle-mounted chip, or a vehicle-mounted unit. It should be noted that the specific type of the terminal 11 is not limited in the embodiments of the present application.

[0120] The network-side device 12 can include an access network device or a core network device, wherein the access network device can also be referred to as a radio access network (RAN) device, a radio access network function or a radio access network unit. The access network device can include a base station, a wireless local area network (WLAN) access point (AP) or a wireless fidelity (WiFi) node, etc. Among them, the base station can be referred to as a node B (NB), an evolved node B (eNB), a next generation node B (gNB), a new radio node B (NR node B), an access point, a relay base station (RBS), a serving base station (SBS), a base transceiver station (BTS), a radio base station, a radio transceiver, a basic service set (BSS), an extended service set (ESS), a home node B (HNB), a home evolved node B, a transmit / receive point (TRP), or some other suitable term in the art, as long as the same technical effect is achieved. The base station is not limited to a specific technical term, and it should be noted that in the embodiments of the present application, only the base station in the NR system is taken as an example for introduction, and the specific type of the base station is not limited.

[0121] The core network device can also be referred to as a core network node, a core network function, or a core network network element, etc., which includes but is not limited to at least one of the following: a mobility management entity (MME), an access and mobility management function (AMF), a session management function (SMF), a user plane function (UPF), a policy control function (PCF), a policy and charging rules function (PCRF), an edge application server discovery function (EASDF), a unified data management (UDM), a unified data repository (UDR), a home subscriber server (HSS), a centralized network configuration (CNC), a network repository function (NRF), a network exposure function (NEF), a local NEF (L-NEF), a binding support function (BSF), an application function (AF), a location management function (LMF), a gateway mobile location center (GMLC), a network data analytics function (NWDAF), etc. It should be noted that only the core network device in the NR system is taken as an example for introduction in the embodiments of the present application, and the specific type of the core network device is not limited. If the name of the core network device mentioned in the embodiments of the present application changes in the subsequent protocol version (for example, 6G), it is also within the protection scope of the present application.

[0122] Optionally, the core network device can be implemented by one or more function modules in one device, or can be implemented by multiple devices jointly, and the embodiments of the present application do not make a specific limitation in this regard. It can be understood that the above function modules can be network elements in a hardware device, can be software function modules running on a dedicated hardware, or can be virtualized function modules instantiated on a platform (for example, a cloud platform).

[0123] Optionally, the wireless communication system further includes a server 13, which can be a device specially used for training or inference or providing AI-related information, or can be a third-party server, or can be a device provided by an Over The Top (OTT) service provider, a third-party service provider or the Internet, etc.

[0124] When the terminal 11 performs uplink transmission, it needs to obtain TA and frequency offset and other information to adjust the uplink transmission state. Taking TA as an example, the terminal 11 uses TA to determine the transmission time of the uplink signal, so as to ensure that the signal of the terminal 11 uplink transmission reaches the base station at the expected time. At present, the base station usually measures the radio frequency transmission time delay caused by the distance to estimate the TA, and notifies the terminal of the TA through a Timing Advance Command (TAC). In the NR system, there are mainly two ways to deliver TA to the UE, which are: (1) the base station issues TAC through the RAR message in the random access process; (2) after the random access is completed, the base station issues TAC through MAC CE. In some scenarios, the terminal can also measure and calculate TA by itself without the need for the base station to issue it.

[0125] The terminal will send the uplink signal (UL symbols) in advance according to the TA indicated by the base station to ensure that the uplink transmissions of different terminals arrive at the base station at substantially the same time, and the uplink symbols include channel and Sounding Reference Signal (SRS), Physical Uplink Control Channel (PUCCH) and Physical Uplink Shared Channel (PUSCH).

[0126] TA is controlled by the MAC layer and actually executed by the physical layer. The value of TA depends on the propagation delay of the signal from the base station to the terminal, and different terminals located at different positions have different TA values. The timing control procedure of the uplink transmission is initiated by the MAC layer and transmitted to the physical layer for time adjustment.

[0127] Figure 2 is a schematic diagram of the relationship between the transmission time of an uplink frame and a downlink frame. As shown in Figure 2, the start time of the uplink frame and the start time of the downlink frame differ by a time value TA, which is calculated by the following equation (1): TA = (N TA +N TA,offset )T c (1)

[0128] where N TA is a measurement value sent to the terminal as part of the TAC, N TA,offset is determined according to different frequency bands and subcarrier spacings, and T c is referred to as a basic time unit in the NR system and has a value of 0.509 nanoseconds (ns).

[0129] The terminal performs TA synchronization using the TAC, which includes the initial uplink synchronization and uplink synchronization update processes. The initial uplink synchronization is achieved through a random access procedure, and the base station determines the TA value by measuring the received preamble and sends a 12-bit TAC in the random access response (RAR).

[0130] After the initial access is completed, the terminal adjusts the uplink transmission according to the TA value transmitted in the MAC CE, and the NR system uses the same mechanism as the LTE system, i.e., closed-loop adjustment of the uplink TA. If a particular terminal needs to be corrected, the base station sends a TAC to the terminal to adjust the uplink transmission time, and the TAC is sent to the terminal through the TAC MAC CE. The TAC MAC CE is represented by a MAC PDU subheader with a logical channel identification (LCID) value of 111101, and the LCID is changed from 5 bits in the LTE system to 6 bits.

[0131] The TAC MAC CE has a fixed 8-bit structure, with the high 2 bits being the timing advance group identity (TAG ID), which includes the TAG of the special cell (SpCell) with a TAG ID of 0. The low 6 bits are the TAC, which represents the index value TA used to control the MAC entity to apply timing adjustment, and the index value TA has a value range of 0, 1, 2, …, 63,

[0132] TAG is a concept introduced for carrier aggregation scenarios. Due to different time delays introduced by multiple carriers, or large differences in the geographical locations of Pcells (primary cells) and Scells (secondary cells) of different carriers, a uniform TA cannot be used for processing. Therefore, the concept of TAG is introduced. The same TAG corresponds to the same TA, and different TAGs correspond to different TA values.

[0133] Currently, in some scenarios, the accuracy of the information such as TA and frequency offset measured by the base station or the terminal for adjusting the uplink transmission state may not be sufficient to meet the requirements of uplink transmission. Taking TA as an example, the accuracy of TA cannot meet the requirements in the following scenarios:

[0134] Scenario one, cell handover scenario using RACH-less (i.e. no RACH) handover or 2-step RACH handover.

[0135] Scenario two, scenario using short preamble sequence in random access process.

[0136] Scenario three, Cell-free (i.e. no cell or no cellular network) scenario.

[0137] (1) RACH-less handover technology

[0138] The main purpose of RACH-less (Random Access Channel) technology is to reduce the service interruption caused by handover. One of the main factors causing such interruption duration is that the random access (RA) process must be performed in the target cell. According to statistical data, the handover process takes about 40-50 ms (milliseconds) to complete on average, while the random access channel (RACH) attempt process during handover usually takes 10 to 12 ms. If the RA step can be omitted, the interruption duration and overall handover (HO) execution time can be significantly reduced, and the impact on user experience can be improved.

[0139] One of the main purposes of the RACH procedure in the handover process is to obtain the TA of the target cell. In the absence of the RACH procedure, when the source cell and the target cell are time-synchronized, the terminal can obtain the TA of the target cell without an explicit TAC. The RACH-less technology introduces a method for calculating the TA of the target cell. The TA of the target cell can be calculated by the following formula (2): target TA source - 2 (T1-T2)

[0140] wherein TA target represents the TA of the target cell, and TA sourceT1 is the downlink transmission time between the source cell and the terminal, T2 is the downlink transmission time between the target cell and the terminal, and T1-T2 is the downlink propagation time difference between the source cell and the target cell. The prerequisite for calculating the target cell TA using formula (2) is to assume that the uplink propagation delay is the same as the downlink propagation delay, based on which the terminal can derive the target cell TA from the source cell TA through formula (2).

[0141] The RACH-less technology can be used in the Handover and Secondary Node (SN) change scenarios, and the RACH-less technology is only applicable to N TA = 0 or the source N TA value scenario.

[0142] The RACH-less technology can not only be applied to the handover of traditional cells, but also be used for the handover of Non Terrestrial Networks (NTNs). The RRC Reconfiguration message triggering the RACH-less handover includes a timing adjustment indication and a configured grant or beam indication for accessing the target cell. The terminal synchronizes to the target cell based on the timing adjustment indication, and uses the configured uplink grant (if included) to send the RRC Reconfiguration Complete message. When there is no valid configured uplink grant, the terminal can fall back to RACH. If the configured uplink grant is not included, the terminal obtains the uplink grant by monitoring the Physical Downlink Control Channel (PDCCH) according to the beam indication.

[0143] (2) Cell-free system

[0144] The Cell-free massive Multiple-Input Multiple-Output (MIMO) system can be considered as a deconstruction of the traditional massive MIMO system. The distribution of the antenna set of the traditional massive MIMO system is in one station (base station), and the terminals are distributed around the base station. In the massive MIMO system, a large number of antennas are deployed in each base station, thus providing a high array gain and spatial resolution. Multiple terminals can be served simultaneously on the same time-frequency resource, providing high throughput, high reliability and high energy efficiency.

[0145] A cell-free massive MIMO system breaks the concept of a cell, and a large number of antennas are distributed in a wide area, and terminals are also distributed in the wide area. These distributed antennas are called transmit-receive points (TRPs) or access points (APs). In theory, each terminal can communicate with each AP, and with the help of a front-end network and a central processing unit (CPU), a large number of geographically distributed TRPs can jointly serve some terminals, and the CPU uses channel statistical information to perform joint detection. The cell-free network is expected to be applied to the next generation of indoor and hotspot coverage scenarios, such as smart factories, train stations, shopping centers, stadiums, subways, hospitals, community centers, or university campuses, etc. In practice, the cell-free network in the hotspot area can be regarded as a super cell containing multiple TRPs, where multiple TRPs use the same cell ID, and cooperative transmission can be achieved between TRPs.

[0146] In a traditional centralized massive MIMO network, all antenna units are deployed on a macro base station. In contrast, in a cell-free network, antenna units are distributed in different locations in a TRP in a distributed manner, thereby obtaining better diversity gain. From the perspective of network deployment, there are two typical network architectures for a cell-free network:

[0147] Architecture 1, single-layer cell-free network

[0148] All TRPs are deployed in the same layer, and each TRP is directly connected to the CPU through a front-end link for data transmission and resource allocation. The TRP is responsible for transmitting signals to the terminal and receiving signals from the terminal, while the CPU is responsible for allocating, combining, precoding, and processing data from different TRPs, and updating the TRP cluster serving different terminals.

[0149] Architecture 2, multi-layer cell-free network

[0150] FIG. 3 is a schematic diagram of a multi-layer cell-free network structure, referring to FIG. 3, the cell-free network includes two layers of networks: a first layer network and a second layer network.

[0151] The first layer network can be used to implement the initial access and mobility management of the terminal, complete the low-delay control signaling exchange between the terminal and the network, and provide high coverage performance. For example, the first layer network can be a hyper cell (e.g., a macro TRP) based on a single frequency network (SFN) technology or a dynamic point selection (DPS) technology, or a wide coverage cell based on low frequency band communication (e.g., using existing 2G / 3G infrastructure or spectrum resources), or a satellite / HAPS (high altitude platform station) cell in satellite communication.

[0152] The second layer network can implement MIMO transmission by dynamically selecting one or more transmission nodes (e.g., small TRPs) for each terminal, so as to obtain higher spatial multiplexing gain and provide higher data transmission rate. For example, the second layer network can use a non-SFN mode, or use a higher frequency band than the first layer network, or use a low earth orbit satellite in satellite communication.

[0153] The synchronization signal / reference signal of the first layer network node can be associated with the synchronization signal / reference signal of the second layer network node in the same area. For example, the first layer network can use a wide beam reference signal to maintain stable connection of the terminal, and the second layer network can use multiple narrow beam reference signals related to the wide beam to implement high-speed transmission of the terminal.

[0154] (3) 2-step RACH (two-step random access channel) switching

[0155] 2-step RACH is a simplified random access process introduced in wireless communication, especially in the NR system. Compared with the traditional four-step random access process (4-Step RACH), 2-step RACH reduces the number of information interactions to reduce latency and control signaling overhead.

[0156] The first step is to send MsgA by the terminal to the base station, and MsgA is composed of two parts: MsgA-PRACH (preamble signal) and MsgA-PUSCH (data signal). The second step is to send MsgB by the base station to the terminal, and MsgB is a response to MsgA. MsgB contains RAR and contention resolution messages (if needed), and the RAR can contain the ID, TA, and other information of the terminal.

[0157] For scenario one, in some scenarios (such as asynchronous networks), the accuracy of the TA calculated by the TA determination method introduced by the RACH-less technology may not be enough, which is affected by many factors, including but not limited to: network synchronization error between the source cell and the target cell, downlink (DL) synchronization error of the source cell, and DL synchronization error of the target cell.

[0158] For scenario two, using a short preamble sequence can achieve the purpose of energy saving and resource saving, but the TA estimation accuracy and size of the short preamble sequence are worse than those of the long preamble sequence, and the performance requirements may not be met, so only small cells can be supported, and the short preamble sequence cannot be applied to other larger cells.

[0159] For scenario three, the geographical positions of the distributed TRPs are quite different, and the coverage ranges of different TRPs are different, so the propagation delay of the terminal based on multiple TRPs for transmission can be quite different, and the terminal needs to perform large TA adjustment / switching when switching the TRP. In addition, for such distributed TRP deployment, how to quickly and accurately obtain the TA used for transmission between the terminal and multiple TRPs is also a problem to be solved.

[0160] To solve the problems in the related art, the embodiments of the present application provide a method for determining uplink adjustment information based on an artificial intelligence (AI) model, the uplink adjustment information being used to adjust the uplink transmission state of a terminal. The AI model is used to determine the uplink adjustment information, thereby improving the accuracy of the uplink adjustment information and ensuring the accuracy and reliability of the uplink transmission. The training, inference and supervision processes of the AI model are described in detail.

[0161] The AI model has various implementation manners, including but not limited to: neural network, decision tree, support vector machine, Bayesian classifier, etc. AI can also be represented as machine learning (ML).

[0162] FIG. 4 is a schematic diagram of the operation framework of the AI model. As shown in FIG. 4, the life cycle management of the AI model includes multiple functional modules: data collection, model training, model management, model inference and model storage, etc.

[0163] The data collection module is used to collect training data, supervision data and inference data required for model training, model supervision and model inference.

[0164] The model training module performs AI model training, verification and testing, and can generate model performance indicators that can be used as part of the model testing process. If necessary, the model training module is also responsible for data preparation (e.g., data preprocessing and cleaning, formatting and conversion) on the training data provided by the data collection module.

[0165] The model training module also performs model updating, delivering the trained, validated and tested AI model to the model storage function, or delivering an updated version of the model to the model storage module, which is used to store multiple versions of the AI model.

[0166] The model management module supervises the AI model, issues AI function related information, and feeds back model monitoring performance, etc. The module is also responsible for making decisions on the data received from the data collection module and the inference module to ensure correct inference operations.

[0167] The model management module issues AI function related information to the model inference module through management instructions, which includes selecting / activating / deactivating / switching models based on AI models or AI-based functions, falling back to non-AI operations (i.e. not dependent on inference process), etc.

[0168] The model management module requests the required AI model from the model storage module through a model transmission request. The model management module inputs the required information to the model training module through a performance feedback request or a retraining request, for example, the purpose of model retraining or updating.

[0169] The model inference module is used to provide the output of the AI model using the inference data provided by the data collection module as input. If necessary, the module is also responsible for data preparation (e.g. data preprocessing and cleaning, formatting and conversion) of the inference data provided by the data collection module.

[0170] The AI model in the embodiments of the present application can also be referred to as an AI unit, an AI structure, etc., or the AI model can be a processing unit capable of implementing specific algorithms, formulas, processing flows, capabilities, etc. related to AI, or the AI model can be a processing method, algorithm, function, module or unit for a specific data set, or the AI model can be a processing method, algorithm, function, module or unit running on AI-related hardware such as Graphics Processing Unit (GPU), Neural network Processing Unit (NPU), Tensor Processing Unit (TPU), Application Specific Integrated Circuit (ASIC), etc., which is not limited in the present application.

[0171] The identifier (ID) of the AI model in the embodiments of the present application can be a model identifier, an AI structure identifier, an AI algorithm identifier, or an identifier of a specific data set associated with the AI model, or an identifier of a specific scene, environment, channel feature, or device related to AI, or an identifier of a function, feature, capability, or module related to AI, which is not limited in the present application.

[0172] In the embodiments of the present application, the ID (or index) of the AI model can be represented in various ways, such as a functionality ID or a model ID of the model, a model physical ID, a model logical ID, a model global ID, or a model local ID. It can be understood that an AI model can have multiple functions, and accordingly, the functionality ID of the model can have multiple IDs. The model global ID can be a unique and global ID defined within all networks or all model providers provided by the model, which can uniquely identify a model, and the model local ID is used to identify a model in a specific network or a specific model provider provided by the model.

[0173] The RACH or preamble involved in the embodiments of the present application can also be referred to as any module containing at least one of a synchronization signal, a random access signal / channel, an uplink control signal / channel, or other control channels for access.

[0174] The cells involved in the embodiments of the present application can be replaced by TRPs, network nodes (or network side devices), carriers / bands, subbands, bandwidth parts (BWP), etc.

[0175] The communication method and the training method of the AI model provided by the embodiments of the present application will be described in detail below in combination with the accompanying drawings and some embodiments and application scenarios.

[0176] Embodiment One

[0177] FIG. 5 is a flowchart of the communication method provided by the first embodiment of the present application, which illustrates the inference process of the AI model. The inference process of the AI model involves the following aspects: input and output of model inference, trigger condition of model inference, terminal behavior (i.e., UE behavior) after triggering model inference, and definition of the accuracy of model inference. As shown in FIG. 5, the method provided by the present embodiment includes the following steps.

[0178] S101, the first device obtains first input information.

[0179] S102, the first device obtains uplink adjustment information according to the AI model and the first input information, the uplink adjustment information being used to adjust an uplink transmission state of the terminal.

[0180] The first device obtaining the uplink adjustment information according to the AI model and the first input information can also be described as: the first device inputs the first input information into the AI model to obtain the uplink adjustment information, or the AI model obtains the uplink adjustment information based on / acccording to the first input information. The first input information is input into the AI model, and the AI model obtains the uplink adjustment information by inference according to the first input information.

[0181] The uplink adjustment information is used to adjust the uplink transmission state of the terminal, and the uplink adjustment information output by the AI model is used to adjust the uplink transmission state of one terminal or a type of terminal. The type of terminal can be all terminals in a certain cell or all terminals satisfying a certain condition.

[0182] In this embodiment, the execution subject of model inference is the first device, which can be a terminal, a network side device or a server in the system shown in FIG. 1. For example, the network side device is a base station, a TRP or a core network device (such as a core network device specially used for model training). The base station can be a base station of a current camping cell of the terminal, a base station of a current access cell of the terminal, a base station of a target cell after handover of the terminal, a base station of a target cell after reselection of the terminal, a base station of a Pcell currently accessed by the terminal or a base station of a Scell currently accessed by the terminal, etc.

[0183] The AI model can be trained by the first device itself, or can be trained by a second device and sent to the first device. The second device is a training device of the AI model. When the first device is a terminal, the second device can be another terminal, a network side device or a server; when the first device is a network side device, the second device can be a terminal or a server; when the first device is a server, the second device can be a terminal or a network side device.

[0184] The input of the AI model is the first input information, and the output of the AI model is the uplink adjustment information. For example, the first input information includes at least one of the following information:

[0185] (1) signal strength information between the terminal and the network side device of the first cell

[0186] In this embodiment, the first cell includes a source cell, a target cell, a current camping cell, a current access cell, a candidate cell, a Pcell or a Scell of the terminal, and the signal strength information includes reception strength information of uplink signals or reception strength information of downlink signals between the terminal and the network side device of the first cell.

[0187] The reception strength information of the downlink signal is used for reasoning at the terminal side and is obtained by terminal measurement. The reception strength information of the downlink signal can be Reference Signal Receiving Power (RSRP), Reference Signal Receiving Quality (RSRQ), Received Signal Strength Indicator (RSRI) of a Synchronization Signal / PBCH Block (SSB), and the like.

[0188] The reception strength information of the uplink signal is used for reasoning at the network side and is obtained by network side device measurement. The reception strength information of the uplink signal can be RSRP, RSRQ, RSSI, and the like obtained based on Sounding Reference Signal (SRS) measurement.

[0189] (2) Signal quality information between the terminal and the network side device of the first cell

[0190] The signal quality information can include signal quality information of the uplink signal or the downlink signal between the terminal and the network side device of the first cell, and the signal quality information includes, but is not limited to, Signal to Interference plus Noise Ratio (SINR), Signal-to-Noise Ratio (SNR), latency, and the like.

[0191] (3) Path loss information

[0192] The path loss can include the path loss between the terminal and the network side device of the first cell, and can also include the path loss between the source cell and the target cell of the terminal or the path loss between the Pcell and the Scell of the terminal.

[0193] (4) Distance information

[0194] The distance information can include the distance between the terminal and the network side device of the first cell, and can also include the distance between the source cell and the target cell of the terminal or the distance between the Pcell and the Scell of the terminal.

[0195] (5) Absolute TA value between the terminal and the network side device of the first cell

[0196] The absolute TA value can be a current or one or more absolute TA values in a period of time or recently used, for example, an absolute TA value (e.g., RAR TA) last used between the terminal and the network-side device of the first cell.

[0197] (6) Relative TA value between the terminal and the network-side device of the first cell

[0198] The relative TA value can be a current or one or more relative TA values in a period of time or recently used, for example, a relative TA value (e.g., MAC CE TA) last used between the terminal and the network-side device of the first cell.

[0199] (7) Identifier of TRP

[0200] For example, the identifier of at least one of the source TRP, the target TRP, the currently selected or camped TRP, the currently accessed TRP, the primary TRP, and the secondary TRP of the terminal can be included.

[0201] (8) Identifier of TRP group

[0202] For example, the group identifier of at least one of the source TRP, the target TRP, the currently selected or camped TRP, the currently accessed TRP, the primary TRP, and the secondary TRP of the terminal can be included.

[0203] (9) Cell identifier of the first cell

[0204] For example, the identifier of at least one of the source cell, the target cell, the currently camped cell, the currently accessed cell, the Pcell, and the Scell of the terminal can be included.

[0205] (10) Group identifier of the cell group in which the first cell is located

[0206] (11) Group identifier of the TAG in which the first cell is located

[0207] (12) Tracking area (TA) identifier in which the first cell is located

[0208] (13) Radio access network notification area (RNA) identifier in which the first cell is located

[0209] (14) Frequency domain information in which the first cell operates

[0210] The frequency domain information can include frequency band, frequency band, frequency point, carrier frequency, frequency layer, or BWP information in which the first cell operates.

[0211] (15) Reference signal index, beam index or beam direction transmitted between the terminal and the network-side device of the first cell

[0212] Exemplarily, the reference signal index, beam index or beam direction can include one or a group of reference signal indexes, beam indexes or beam directions.

[0213] (16) Transmission delay between the terminal and the network-side device of the first cell

[0214] The transmission delay can also be referred to as propagation delay, and can be a signal transmission delay between uplink, downlink and sidelink between the terminal and the network-side device of the first cell, and can also be a transmission delay between network-side devices.

[0215] (17) Round-Trip Time (RTT) between the terminal and the network-side device of the first cell

[0216] (18) Transmission power information of the network-side device of the first cell

[0217] The network-side device of the first cell can be a base station or a TRP, and the transmission power information can be the actual transmission power of the network-side device.

[0218] (19) Position information of the network-side device of the first cell

[0219] The position information can be physical position information of the network-side device, and the network-side device can be a base station or a TRP.

[0220] (20) Position information or distribution information of the terminal

[0221] The position information of the terminal can be a specific geographic position coordinate (such as a GPS coordinate), or approximate position range information of the terminal (such as range information of which street), or position information of the terminal relative to a camping cell or an access cell or a certain TRP or a certain group of TRPs (such as in the east direction of the camping cell).

[0222] The distribution information of the terminal can include quantity information of the terminal in different areas (such as a camping cell or an access cell or a certain TRP or a certain group of TRPs).

[0223] (21) Moving direction and / or moving speed of the terminal, satellite or network-side device of the first cell

[0224] The moving direction can be an absolute direction, such as east 40 degrees south; or a relative direction, such as a direction relative to a certain base station or TRP.

[0225] (22) Energy consumption status and / or power status of the terminal

[0226] The energy consumption status of the terminal refers to the current energy consumption status of the terminal, for example, the terminal is in high energy consumption or low energy consumption. The power status of the terminal refers to the current power of the battery of the terminal, for example, how much power is left in the terminal.

[0227] (23) Operator information and network type information supported by the terminal

[0228] (24) Antenna orientation information of the terminal, satellite or network side device of the first cell

[0229] The antenna orientation information can be the orientation information of the panel of the antenna, specifically, it can be the orientation angle of the antenna, or the orientation direction of the antenna (for example, east or south).

[0230] (25) Energy consumption status and / or power status of the first cell

[0231] The energy consumption status of the first cell can be the source type of the electrical energy of the network side device of the first cell, and the source of the electrical energy can be divided into clean energy or non-clean energy. In some implementations, AI inference can not be allowed for non-clean energy.

[0232] (26) Perception information of the terminal

[0233] The perception information of the terminal includes communication environment information, scene information, channel state information, the number of terminals, etc. The channel state information can include Line of Sight (LOS) state or Non-Line of Sight (NLOS) state or blocking, etc.

[0234] For example, the terminal can perceive whether there is an obstacle under a specific beam and / or the number of obstacles, or perceive the number of terminals or devices covered by a specific beam.

[0235] (27) Scene information of the network in which the terminal is located

[0236] The scene information of the network can be the propagation model of the network, the propagation model suitable for 0.5-100GHz in the NR system, including but not limited to: Uma (urban macro station), UMi (rural macro station), RMa (urban micro station) and InH (indoor hotspot).

[0237] The scene information of the network can also be the network architecture type, for example, the network architecture includes homogeneous network or heterogeneous network (with or without overlapping coverage).

[0238] (28) Environment information (such as weather information) in which the terminal is located

[0239] (29) Ephemeris information

[0240] Ephemeris refers to the precise position or trajectory of a celestial body over time in GPS measurements. It is a function of time. In the field of wireless communication, ephemeris is used to describe the position and velocity of space flight bodies such as satellites, spacecraft, etc.

[0241] (30) Reference location or moving trajectory information of a cell in a Non-Terrestrial Network (NTN) scenario

[0242] NTN uses satellites or other space platforms to achieve seamless coverage worldwide, especially in remote areas, oceans, and airspace where ground networks are difficult to reach.

[0243] The two common orbit types in satellite communication are Low-Earth Orbit (LEO) and Geosynchronous Equatorial Orbit (GEO). LEO is suitable for communication applications that require low latency, high bandwidth, and fast coverage. GEO is mainly used for broadcast and communication satellite services such as satellite television, satellite telephone, and satellite Internet.

[0244] In the LEO scenario, as the satellite moves, the cell on Earth is moving, so the moving trajectory information of the cell can be obtained. In the GEO scenario, as the satellite moves, the cell on Earth is fixed, and it can be considered that there is a reference location, so the reference location of the cell can be obtained.

[0245] (31) Multipath information of a channel

[0246] This multipath information can be the first path or the strongest path information of the terminal and different base stations / TRPs.

[0247] (32) Time information

[0248] The time can be a specific time, such as a time accurate to the second, e.g., 13:25:38. It can also be a time range, such as 13:00 to 14:00, morning or afternoon, day or night.

[0249] This time information can be timing information obtained through other Radio Access Technologies (RATs). RATs can be Bluetooth, Wi-Fi, and 3G, 4G, or 5G, etc.

[0250] (33) one or more of the following information currently used or used within a certain time window: timing advance offset, timing pre-compensation, common timing advance offset, dedicated timing advance offset, frequency compensation, frequency offset, common frequency offset, and dedicated frequency offset.

[0251] When the first input information includes multiple pieces of information, at least part of the information in the first input information can be sent to the first device by the other device, where the at least part of the information in the first input information sent by the other device includes the following three cases: part of the information in the first input information is sent by the other device, and the rest of the information is measured by the first device; all of the information in the first input information is measured by the first device; and all of the information in the first input information is sent by the other device.

[0252] When the first device is a terminal, the terminal can receive at least part of the first input information sent by the network side device, i.e., all or part of the first input information is sent by the network side device, and when part of the first input information is sent by the network side device, the rest of the information in the first input information can be measured by the terminal itself.

[0253] In some implementations, the first input information sent by the network side device to the terminal can be sent through at least one of the following signals or channels: MAC CE, RRC message, Non-Access Stratum (NAS) message, user plane data, Downlink Control Information (DCI) information, System Information Block (SIB), PDCCH, Physical Downlink Share Channel (PDSCH), MSG2 information, MSG4 information, or MSGB information. Among them, MSG2 and MSG4 are the second and fourth messages in the four-step random access process, and MSGB is the second message in the two-step random access process.

[0254] When the first device is a network side device, the network side device receives at least part of the first input information sent by the terminal, i.e., all or part of the first input information is sent by the terminal, and when part of the first input information is sent by the terminal, the rest of the information in the first input information can be measured by the network side device itself.

[0255] In some implementable manners, the first input information reported by the terminal to the network-side device can be transmitted through at least one of the following signals or channels: a MAC CE, an RRC message, a Non-Access Stratum (NAS) message, user plane data, MSG1 information, MSG3 information, MSGA information, a Physical Uplink Control Channel (PUCCH), a Physical Uplink Shared Channel (PUDCH), or a Physical Random Access Channel (PRACH), an SRS, or other uplink reference signals (such as a Wake-Up Signal (WUS)). Among them, MSG1 and MSG3 are the first and third messages in a four-step random access procedure, and MSGA is the first message in a two-step random access procedure.

[0256] When the first device is a server, the server receives at least part of the first input information sent by the terminal and / or the network-side device, that is, all or part of the first input information is sent by the terminal, or all or part of the first input information is sent by the network-side device, or part of the first input information is sent by the terminal and part of the first input information is sent by the network-side device. The network-side device and the terminal can send all or part of the first input information through an Over-The-Top (OTT) message.

[0257] The uplink adjustment information output by the AI model includes at least one of the following: TA-related information and frequency-related information.

[0258] In some implementable manners, the TA-related information includes any one of the following: an absolute TA value, a relative TA value, a timing advance offset, a timing pre-compensation, a common timing advance offset, and a specific timing advance offset.

[0259] In some implementable manners, the frequency-related information includes at least one of the following: a frequency error, a frequency compensation, a frequency offset, a common frequency offset, and a specific frequency offset.

[0260] In some implementable manners, the uplink adjustment information output by the AI model includes at least one of the following:

[0261] (1) Uplink adjustment information of a specific terminal

[0262] The uplink adjustment information for a specific terminal refers to the uplink adjustment information used by the specific terminal. The uplink adjustment information for a specific terminal can be the uplink adjustment information between the terminal and the network node of the target cell / Scell. For example, the AI model can output the absolute TA value or the relative TA value between the terminal and the network node of the target cell / Scell.

[0263] (2) The uplink adjustment information for a group of terminals

[0264] For example, the uplink adjustment information for all terminals under the coverage of a certain beam or TRP. For example, the AI model can output the TA value used by all terminals under the coverage of the beam or TRP.

[0265] (3) The uplink adjustment information in a specific area

[0266] The uplink adjustment information in a specific area refers to the uplink adjustment information used by the terminals located in the specific area or the uplink adjustment information of the terminals in the specific area. The specific area can be an area within a certain location range.

[0267] (4) The uplink adjustment information for at least one beam direction

[0268] The at least one beam direction includes a specific beam direction or multiple beam directions. The uplink adjustment information for a specific beam direction refers to the uplink adjustment information used by the terminals located in the coverage of the specific beam direction. The uplink adjustment information for multiple beam directions refers to the uplink adjustment information used by the terminals located in the coverage of the multiple beam directions.

[0269] (5) The uplink adjustment information for a specific reference point

[0270] The specific reference point can be a specific location or an area range. For example, the reference point can be a point (or an area) inside or outside the cell. When the specific reference point is located inside a cell, the first device can notify the uplink adjustment information of the specific reference point to all or part of the terminals in the cell.

[0271] Specifically, when the specific reference point is a point inside the cell, the first device can notify the uplink adjustment information of the specific reference point to all terminals in the cell or the terminals within a certain distance from the specific reference point. When the specific reference point is an area inside the cell, the uplink adjustment information of the specific reference point can be notified to all terminals in the cell or the terminals in the area.

[0272] (6) The uplink adjustment information corresponding to a specific path loss range or a specific RSRP range

[0273] The uplink adjustment information corresponding to a specific pathloss / RSRP range refers to the uplink adjustment information used by the terminal whose pathloss / RSRP value is within the specific pathloss / RSRP range.

[0274] The pathloss / RSRP within the specific pathloss / RSRP range refers to the pathloss / RSRP between the terminal and the base station, which can be the base station of the source cell, the target cell, the currently camped cell, the currently accessed cell, the candidate cell, the Pcell or the Scell.

[0275] (7) Uplink adjustment information corresponding to a specific sequence or sequence format

[0276] The uplink adjustment information corresponding to a specific preamble or preamble format refers to the uplink adjustment information used by the terminal configured or using the specific preamble or preamble format.

[0277] (8) Common uplink adjustment information

[0278] The common uplink adjustment information can be inferred by the network side device and broadcast to all terminals, and the common uplink adjustment information can be adjusted subsequently. Taking the TA related information as an example, the common uplink adjustment information can be Common TA, which is inferred by the network side device and broadcast to all terminals, and the terminal first uses the Common TA to send uplink signals / channels, and adjusts the TA based on the subsequently sent uplink signals / channels.

[0279] (9) Uplink adjustment information between a specific terminal and at least one network side device

[0280] For example, the AI model outputs the uplink adjustment information between the specific terminal and multiple base stations / TRPs.

[0281] (10) Uplink adjustment information of other terminals in a terminal group

[0282] The terminal group includes multiple terminals, and the first device can infer the TA related information of other terminals in the terminal group based on the information of a certain terminal in the terminal group. When the first device is a certain terminal in the terminal group, the terminal can infer the TA related information of other terminals in the terminal group based on its own information.

[0283] Taking the TA related information as an example, a certain terminal can infer the absolute TA value or relative TA value between the TA of other terminals in the terminal group and multiple base stations / TRPs through the AI model.

[0284] (11) a set of candidate uplink adjustment information and / or probability values associated with the set of candidate uplink adjustment information.

[0285] The set of candidate uplink adjustment information includes a plurality of candidate uplink adjustment information, and each candidate uplink adjustment information is associated with a probability value representing a probability that the corresponding uplink adjustment information is finally used by the terminal.

[0286] Taking the TA-related information as an example, the set of candidate uplink adjustment information can be a set of candidate absolute TA values. The base station or the TRP can infer a set of candidate absolute TA values through the AI model and notify the terminals within the coverage of the base station or the TRP of the set of candidate absolute TA values.

[0287] In this embodiment, after the first device infers the uplink adjustment information, the first device sends the uplink adjustment information to the terminal using the uplink adjustment information. When the first device is the terminal and the uplink adjustment information output by the AI model is used by the terminal, the terminal does not need to send the uplink adjustment information to the terminal using the uplink adjustment information. Except for this case, the first device needs to send the uplink adjustment information to the terminal using the uplink adjustment information.

[0288] After the terminal receives the uplink adjustment information sent by the first device, the terminal adjusts its own uplink transmission state according to the uplink adjustment information, wherein the adjustment of the uplink transmission state of the terminal includes at least one of the following adjustment operations: adjusting the time of uplink transmission (which can also be described as compensating for the timing offset of uplink transmission); adjusting the frequency of uplink transmission (which can also be described as compensating for the frequency offset of uplink transmission); adjusting the phase of uplink transmission.

[0289] The uplink adjustment information received by the terminal includes TA-related information and / or frequency-related information. The terminal can adjust the time of uplink transmission according to the TA-related information, adjust the frequency of uplink transmission according to the frequency-related information, and adjust the phase of uplink transmission according to the TA-related information and / or the frequency-related information.

[0290] In some implementable manners, the first device further obtains first configuration information of the AI model, and performs inference of the AI model according to the first configuration information.

[0291] In some implementable manners, the first configuration information includes at least one of the following:

[0292] (1) an identifier of the AI model used for AI inference

[0293] (2) an application range of inference of the AI model

[0294] The application scope of the AI model inference can include at least one of the following ranges: a frequency domain range, a cell identifier or a cell list, a range of a location where the terminal is located, a distance range between the terminal and the base station, and a maximum range of TA between the terminal and the base station.

[0295] (3) Validity duration of AI model inference

[0296] The validity duration can be a start time and an end time of the inference, or a specific time length (for example, 1 hour, 2 hours, etc.).

[0297] (4) Period of AI model inference

[0298] Within the validity duration of the AI model inference, the first device performs periodic inference according to the period of the AI model inference. For example, the validity duration is 1 hour, and the first device can perform inference every 1 minute within the 1 hour, and the inference period is 1 minute.

[0299] (5) Trigger condition of AI model inference

[0300] The trigger condition is used to trigger the AI model inference. It should be noted that the trigger condition is not the only condition for triggering or activating the AI model inference, and the AI model inference can also be triggered or activated in combination with other information.

[0301] (6) Configuration parameters of the AI model, including the type of the first input information of the AI model and / or the type of the output information of the AI model The type of the first input information of the AI model is used to represent which information is included in the first input information, and the specific content is described above. The type of the output information of the AI model is used to represent which information is included in the output information, and the specific content is described above.

[0302] (7) Whether to support joint inference of multiple devices.

[0303] The joint inference means that at least part of the first input information of the AI model is provided by other devices when the first device uses the AI model for inference. For example, when the terminal uses the AI model for inference, at least part of the first input information of the AI model is sent by the network side. When the network side device uses the AI model for inference, at least part of the first input information of the AI model is reported by the terminal. When the server uses the AI model for inference, at least part of the first input information of the AI model is sent by the terminal and / or the network side device.

[0304] When the AI model is trained by the second device, the second device can deliver the AI model and the first configuration information to the first device together, of course, the second device can also deliver the AI model and the first configuration information to the first device independently.

[0305] In some implementations, the trigger condition of the AI model inference includes at least one of the following:

[0306] (1) AI related timer timeout

[0307] The AI related timer can be a newly defined specific timer for triggering AI inference. When the first device detects that the specific timer is timed out, the AI model inference is triggered. It can be understood that the AI related timer can also reuse an existing timer.

[0308] (2) enable RACH-less technology

[0309] The first device triggers the AI model inference when it determines that the RACH-less technology is enabled for a certain cell or certain terminals.

[0310] (3) the terminal is configured to use short preamble for PRACH transmission

[0311] In the network, some terminals are configured to use short preamble, and in FR2, only short preamble with subcarrier spacing of 60KHz and 120KHz is supported. When the subcarrier spacing of the cell accessed by the terminal is 60KHz and 120KHz, the terminal is configured to use preamble for PRACH transmission. When the first device detects that the terminal uses short preamble for PRACH transmission, the AI model inference is triggered.

[0312] (4) the network side device configures PRACH resource dedicated for traditional RACH as fallback mechanism, and the network side configures AI for RACH mechanism

[0313] In the embodiments of the present application, the terminal can use the AI model to perform AI inference during cell switching. During the cell switching process, the terminal accesses the target cell through the random access procedure. The network side device can configure the AI for RACH mechanism, wherein the AI for RACH mechanism refers to that in the random access process, the base station or the terminal uses the AI model to infer the uplink adjustment information of the terminal, and the uplink adjustment information at least includes TA related information.

[0314] The network-side device can further configure PRACH resources dedicated to the traditional RACH as a fallback mechanism after configuring the AI-based RACH mechanism. The traditional RACH refers to a random access mode through a four-step process. When the terminal cannot access the target cell through the AI-based RACH mechanism, the terminal can access the target cell through the traditional RACH based on the PRACH resources dedicated to the traditional RACH configured by the network-side device.

[0315] (5) The accuracy of the uplink adjustment information is less than or equal to a target threshold or less than or equal to the target threshold for more than a certain time length

[0316] In the embodiments of the present application, the network-side device can define a target threshold for the accuracy of the uplink adjustment information. When the first device detects that the accuracy of the uplink adjustment information is less than or equal to the target threshold or less than or equal to the target threshold for more than a certain time length, it indicates that the accuracy of the uplink adjustment information is too low to meet the demand. At this time, AI model reasoning can be triggered to determine the uplink adjustment information using an AI model. The uplink adjustment information can be determined using a non-AI method or an AI method.

[0317] (6) The accuracy of the uplink adjustment information calculated using a non-AI method is less than a specific threshold

[0318] (7) The number of failures or the number of consecutive failures of signal transmission using the uplink adjustment information calculated using a non-AI method exceeds a target threshold

[0319] The terminal adjusts the uplink transmission state using the uplink adjustment information calculated using a non-AI method and performs uplink signal transmission. If the number of failures within a certain time length exceeds a target threshold, AI model reasoning can be triggered, or if the number of consecutive failures exceeds the target threshold, AI model reasoning can be triggered.

[0320] (8) Default use in a specific scenario

[0321] For example, in a Cell-free scenario, AI model reasoning is used by default.

[0322] (9) Cell or TRP reselection

[0323] (10) Cell or TRP handover

[0324] (11) Satisfying the conditions for cell handover

[0325] For example, the conditions for cell handover can include at least one of the following:

[0326] The measurement result (L1 or L3 RSRP\RSRQ\RSSI, etc.) of the source cell is greater than or less than a threshold;

[0327] a measurement result of a target cell (a neighbor cell) is better than a source cell;

[0328] a measurement result of a target cell (a neighbor cell) is greater than a threshold;

[0329] a measurement result of a source cell is less than a threshold 1 and a measurement result of a target cell (a neighbor cell) is greater than a threshold 2;

[0330] a load of a source cell is greater than a threshold.

[0331] (12) the terminal triggers access to a Scell

[0332] (13) the terminal moves to a specific location

[0333] The specific location can be a cell edge, or a designated area in a cell.

[0334] (14) uplink or downlink data arrives

[0335] (15) a specific type of service arrives

[0336] The specific type of service may be, for example, a service with high latency requirements, or a service with high Quality of Service (QoS) requirements.

[0337] (16) receiving at least one item of first input information

[0338] (17) the terminal speed or acceleration exceeds a set threshold

[0339] (18) the terminal or network location changes or changes exceed a set threshold

[0340] (19) the terminal selected beam or spatial domain transmission filter changes

[0341] (20) the change of a first parameter of the terminal relative to a certain time point or within a certain time window exceeds a certain value, the first parameter including at least one of the following parameters: timing advance offset, timing pre-compensation, common timing advance offset, dedicated timing advance offset, frequency compensation, frequency offset, common frequency offset, dedicated frequency offset.

[0342] In some implementable manners, when the trigger condition of the AI model inference is met, the first device determines to perform inference using the AI model.

[0343] In some embodiments, the first device obtains the activation indication information or the deactivation indication information, and determines to use the AI model to perform AI inference according to the activation indication information and a trigger condition of AI model inference, or determines to use the AI model to perform AI inference according to the activation indication information, the trigger condition of AI model inference and AI capability information of the first device, or determines not to use the AI model to perform AI inference according to the deactivation indication information.

[0344] The activation indication information is used to indicate activation of the AI model, and the deactivation indication information is used to indicate deactivation of the AI model. After the AI model is activated, the first device can use the AI model to perform AI inference. In some embodiments, after the AI inference ends, the first device defaults to a non-AI mode of uplink adjustment information determination process. It can be understood that before the AI model is activated, the first device uses the non-AI mode of uplink adjustment information determination process to determine the uplink adjustment information.

[0345] In some embodiments, after the AI model is activated, the AI model is deactivated when the deactivation indication information is received, and the first device defaults to the non-AI mode of uplink adjustment information determination process after the AI model is deactivated.

[0346] In the embodiments of the present application, the non-AI mode of uplink adjustment information determination process refers to a mode of determining uplink adjustment information without using the AI model, and can also be understood as a mode of determining uplink adjustment information using a traditional method. For example, for TA related information, the existing determination process can be used to determine the TA related information.

[0347] The activation indication information or the deactivation indication information can be sent by a network side device. The activation indication information and the deactivation indication information can be enable information of an inference switch of the AI model. The enable information can be enable and disable. When the enable information of the inference switch of the AI model is enable, it indicates that the inference of the AI model is activated. When the enable information of the inference switch of the AI model is disable, it indicates that the inference of the AI model is deactivated.

[0348] The AI capability information of the first device includes: whether the first device has the capability of using the AI model to perform AI inference, and / or whether the first device supports reporting one or more types of assistance information for AI model inference.

[0349] In some implementable manners, the first device determines to use the AI model for AI inference according to the activation indication information, the trigger condition of the AI model inference, and the AI capability information of the first device. Specifically, when the trigger condition of the AI model inference is met, the first device determines whether to use the AI model for AI inference according to the activation indication information and the AI capability information of the first device. If the first device receives the activation indication information and the AI capability information of the first device indicates that the first device has the capability to use the AI model for AI inference, it is determined that the AI model is used for AI inference. If the trigger condition of the AI model inference is not met, the step of determining whether to use the AI model for AI inference according to the activation indication information and the AI capability information of the first device will not be performed.

[0350] In some implementable manners, the trigger condition of the AI model inference is met if at least one of the trigger conditions is met.

[0351] In some implementable manners, when the first device is a terminal, after the terminal obtains the uplink adjustment information according to the AI model and the first input information, the terminal performs at least one of the following operations according to the uplink adjustment information:

[0352] (1) reporting the first information to the training device

[0353] The first information is used for updating the AI model, and the first information includes at least one of the following information: input information of this AI inference, output information of this AI inference, and current state information of the terminal. The current state information of the terminal can include the position of the terminal, the cell accessed by the terminal, and the like.

[0354] After the training device receives the first information, the AI model is updated according to the first information, wherein the updating of the AI model includes but is not limited to: fine tuning of the existing AI model, replacement of the AI model, adjustment of the parameters of the AI model, and training of the AI model.

[0355] (2) falling back to a non-AI manner of determining uplink adjustment information

[0356] In some implementable manners, after the current AI inference ends, the terminal can fall back to a non-AI manner of determining uplink adjustment information by default.

[0357] (3) triggering updating of the AI model

[0358] The terminal can also update the AI model by itself, for example, fine tuning of the AI model by the terminal, switching of the AI model, and the like. For example, in some scenarios, the training device trains multiple AI models, and then the terminal can select one of the multiple AI models.

[0359] (4) Triggering the change of the input information of the AI model

[0360] The first input information of the AI model can include multiple different types of information. Referring to the foregoing description, the terminal can select to increase, decrease, or change the type of input information. For example, the first input information includes 6 items of information before the change, and the first input information includes 4 items after the change, a total of 2 items of information are reduced; or the first input information includes 2 items of information before the change, and the first input information includes 5 items of information after the change, a total of 3 items of information are increased; or the first input information includes 4 items of information before and after the change, but the type of the 4 items of information after the change is different from the type of the 4 items of information before the change, and part or all of the 4 items of information before the change can be replaced.

[0361] (5) Triggering the supervision of the AI model

[0362] The terminal can trigger the supervision of the AI model by default after the AI model inference ends, and the terminal can perform the supervision of the AI model according to the supervision configuration information configured by the network side device.

[0363] Optionally, the terminal performs at least one of the above operations (i.e., at least one of reporting the first information to the training device, falling back to the determination process of the uplink adjustment information in the non-AI manner, triggering the update of the AI model, triggering the change of the input information of the AI model, and triggering the supervision of the AI model) when at least one of the following conditions is met:

[0364] The TA related information that meets the accuracy requirement is not obtained after using the AI model to infer for more than M time lengths;

[0365] The AI inference is not successfully completed using the AI model;

[0366] The AI inference is successfully completed using the AI model;

[0367] N times of AI inference are performed using the AI model.

[0368] In this embodiment, the first device obtains the first input information, and the first device obtains the uplink adjustment information according to the AI model and the first input information. The uplink adjustment information is used to adjust the uplink transmission state of the terminal. The first device sends the uplink adjustment information obtained by the AI model inference to the corresponding terminal. The terminal adjusts the uplink transmission state according to the uplink adjustment information. The accuracy of the uplink adjustment information obtained by the AI model inference is high, thereby ensuring the accuracy and reliability of data transmission.

[0369] Embodiment Two

[0370] Embodiment one mainly describes the content related to AI model inference, and embodiment two of the present application provides a data processing method for supervising the AI model described in embodiment one. It can be understood that when the inference environment of the AI model is greatly different from the training environment, the inference performance based on the uplink adjustment information of the AI model will become very poor, for example, the accuracy of the inferred uplink transmission information may not be enough. Therefore, the actual inference performance based on the AI model needs to be supervised, and a series of adjustment measures are triggered according to the supervision result to ensure the reliability of the inference result of the AI model.

[0371] FIG. 6 is a flowchart of the data processing method provided by embodiment two of the present application. As shown in FIG. 6, the method provided by the present embodiment includes the following steps:

[0372] S201, the first device obtains first input information.

[0373] S202, the first device obtains uplink adjustment information according to the AI model and the first input information, and the uplink adjustment information is used to adjust the uplink transmission state of the terminal.

[0374] The related content of steps S201-S202 is described in embodiment one, which will not be repeated here.

[0375] S203, the first device obtains supervision configuration information of the AI model.

[0376] S204, the first device supervises the AI model according to the supervision configuration information.

[0377] The supervision configuration information of the AI model can be sent to the first device by the training device of the AI model, or sent to the first device by other devices. The first device determines when to supervise the model and / or how to supervise the model according to the supervision configuration information.

[0378] In some implementable manners, the supervision configuration information includes at least one of the following 8 items:

[0379] (1) identification of the AI model that needs to be supervised

[0380] (2) model supervision period

[0381] The model supervision period can be the interval length of the model supervision, for example, the model supervision is performed once every 1 day.

[0382] (3) number of model supervision

[0383] The number of model supervision refers to the number of times of model supervision in a supervision period, for example, 10 times of model supervision in a supervision period. Alternatively, the number of model supervision refers to the total number of times of model supervision after receiving the supervision configuration information.

[0384] (4) Length of model supervision

[0385] The length of model supervision refers to the length of time of model supervision in each model supervision in a supervision period. Alternatively, the length of model supervision refers to the effective length of time of the supervision configuration information, for example, there are N model supervision periods in the length of model supervision, or model supervision is only performed in the length of model supervision.

[0386] (5) Window related information of model supervision

[0387] The window related information of model supervision refers to the length of time or the number of samples (times) of model supervision actually performed in a model supervision period. Alternatively, the window related information of model supervision refers to the length of time of the window, and / or the starting time, and / or the ending time, in which model supervision needs to be performed in a model supervision period.

[0388] (6) Trigger condition of model supervision

[0389] (7) Index of model supervision

[0390] Illustratively, the index of model supervision includes the error between the predicted value of the AI model and the true value, and / or the performance index of the network.

[0391] The predicted value of the AI model refers to the value of the uplink adjustment information inferred by the AI model according to the input information, and the true value of the AI model refers to the value of the uplink adjustment information calculated by other means, for example, the true value of the uplink adjustment information calculated by non-AI means. The error between the predicted value of the AI model and the true value can reflect the accuracy of the predicted value of the AI model. If the error is large, it means that the error of the predicted value of the AI model is large.

[0392] The performance index of the network includes transmission delay, throughput, etc. It can be understood that if the accuracy of the predicted value of the AI model is high, the efficiency, accuracy and reliability of uplink transmission after uplink transmission adjustment based on the predicted value of the AI model will be improved in general, and accordingly, the performance index of the network meets the requirements. However, if the accuracy of the predicted value of the AI model is low, it will affect the efficiency, accuracy and reliability of uplink transmission, so that the performance index of the network cannot meet the requirements, and therefore, the performance index of the network can be used as the index of model supervision.

[0393] (8) Model supervised label

[0394] In the embodiments of the present application, the input information and labels of the AI model are included in the training samples during the training of the AI model. In machine learning and deep learning, the label generally refers to the identification or annotation of the true class or target value of the data sample. The label is used to represent the information that the model should learn and predict.

[0395] The label is a key component in supervised learning tasks and is used to train machine learning models. The model learns patterns and rules by comparing with the true label in order to make predictions or classifications on unseen data. The quality and accuracy of the label are crucial to the performance of the model.

[0396] The model supervised label type belongs to the label type in the AI model training process. There can be multiple different types of labels in the training samples during the AI model training process. During model supervision, the configured model supervised label is a subset of the label type in the training samples or is the same as the label type in the training samples. For example, during the training of the AI model, four labels are included in the training samples. During the model supervision process, four or fewer labels can be supervised.

[0397] In some implementations, the value of the model supervised label is the same as or different from the value of the label in the AI model training process. That is, the value of the same label can be the same or different in the supervision and training stages.

[0398] In some implementations, the triggering condition of the model supervision includes at least one of the following 12 conditions:

[0399] (1) The inference result of the AI model does not meet the accuracy requirement

[0400] The accuracy requirement can be defined for the inference result of the AI model. The model supervision is triggered when the inference result of the AI model does not meet the accuracy requirement, or when the number of times that the inference result of the AI model does not meet the accuracy requirement reaches a set threshold, or when the number of consecutive times that the inference result of the AI model does not meet the accuracy requirement reaches a set threshold.

[0401] (2) At least one of the inference indicators of the AI model does not meet the requirement

[0402] In some implementations, the inference indicators of the AI model include at least one of the following: the complexity of the AI model, the latency of the inference of the AI model, the success rate of the inference of the AI model, and the reliability of the inference result of the AI model.

[0403] The complexity of the AI model can be the number of parameters in the model structure of the AI model, the more parameters in the model structure, the higher or greater the complexity of the AI model. The complexity of the AI model can also be the number of types of input information of the AI model, for example, one AI model input information includes 2 types of information, and another AI model input information includes 5 types of information, then the complexity of the latter AI model is higher than that of the former AI model.

[0404] For example, different AI model complexity indicators can be defined for different types / capabilities of base stations or terminals. For example, for a general terminal, the complexity of the AI model used by the terminal should not exceed a certain value. When the complexity of the AI model used by the general terminal exceeds the certain value, it can be determined that the inference indicator of the AI model does not meet the requirements.

[0405] The latency of the AI model inference can be the time length of the AI model inference, which can be the total time length required from inputting the first input information to the model to outputting the inference result. The time length of the AI model inference can be set not to exceed a certain value. When the first device determines that the time length of the AI model inference used by itself exceeds the certain value, it can be determined that the inference indicator of the AI model does not meet the requirements.

[0406] The success rate of the AI model inference can be the ratio of the number of successful inferences of the AI model to the total number of inferences within a certain time. The success rate of the AI model inference can be set not to be less than or greater than a certain value. When the first device determines that the success rate of the AI model inference used by itself is less than the certain value, it can be determined that the inference indicator of the AI model does not meet the requirements.

[0407] It can be understood that the success rate of the AI model inference corresponds to the failure rate of the AI model inference. The success rate of the AI model inference can be the ratio of the number of failed inferences of the AI model to the total number of inferences within a certain time. The failure rate of the AI model inference = 1 - the success rate of the AI model inference. Therefore, the success rate of the AI model inference can be replaced by the failure rate of the AI model inference. The failure rate of the AI model inference can be set not to be greater than or less than a certain value. For example, when the failure rate of the AI model inference is greater than a certain value, it can be determined that the inference indicator of the AI model does not meet the requirements.

[0408] The reliability of the AI model inference result refers to the probability that the uplink adjustment information inferred using the AI model meets the accuracy requirements. The probability that the uplink adjustment information inferred using the AI model meets the accuracy requirements can be set to be greater than or equal to a certain value. When the first device determines that the probability that the uplink adjustment information inferred using the AI model meets the accuracy requirements is less than the certain value, it can be determined that the inference indicator of the AI model does not meet the requirements.

[0409] (3) The index of model supervision does not meet the requirements

[0410] The index of model supervision includes: the error between the predicted value and the true value of the AI model, and / or the performance index of the network.

[0411] (4) The supervision timer of the AI model is overdue

[0412] A new supervision timer for triggering AI model supervision can be defined, and the supervision of the AI model is triggered when the first device detects that the supervision timer is overdue. It can be understood that the supervision timer can also reuse the existing timer.

[0413] (5) The terminal switches to a new cell, TRP or beam

[0414] (6) After the AI model inference fails

[0415] In this trigger condition, the supervision of the AI model is triggered as long as the AI model inference fails.

[0416] (7) The AI model fails N times in a row

[0417] In this trigger condition, the AI model records the number of consecutive inference failures when it fails for the first time. When the number of consecutive inference failures reaches the preset N times, the supervision of the AI model is triggered. For example, when N is 5, if the number of consecutive inference failures reaches 4, the inference succeeds in the 5th inference, and the supervision of the AI model is not triggered. Assuming that the 6th inference fails, the number of consecutive inference failures is reset.

[0418] (8) The number of AI model inference failures reaches a threshold value

[0419] Unlike the trigger condition in item (6), this trigger condition records the total number of AI model inference failures from a certain time point, rather than the number of consecutive inference failures. When the number of AI model inference failures reaches the threshold value, the supervision of the AI model is triggered. It can be understood that the threshold values in items (6) and (7) can be the same or different.

[0420] (9) Using the AI model for inference

[0421] In this trigger condition, the model supervision is triggered as long as the first device uses the AI model for inference. It can also be understood that the model supervision is triggered once after each AI model inference by the first device.

[0422] (10) The terminal moves to a new cell, tracking area or geographic location

[0423] (11) The moving speed of the terminal changes significantly within a specified time length

[0424] The moving speed of the terminal changes significantly within a specified time length refers to that the change value of the moving speed of the terminal within the specified time length exceeds a set threshold, for example, the moving speed of the terminal suddenly drops within a short time, such as from 250 km / h to 3 km / h within 5 minutes, at this time, the change value of the moving speed of the terminal is 250-3=247 km / h, assuming that the set threshold is 150 km / h, the change value exceeds the set threshold, and it is determined that the moving speed of the terminal changes significantly within a specified time length.

[0425] (12) The external environment in which the inference device of the AI model is located changes.

[0426] The external environment in which the inference device is located changes includes that the channel environment changes, the external weather state changes, the location changes, etc., and the change of the external environment can be determined according to the sensor measurement data.

[0427] In this embodiment, the first device obtains the supervision configuration information of the AI model, and supervises the AI model according to the supervision configuration information. By supervising the AI model, it can be judged whether the performance of the AI model can meet the requirements. If it does not meet the requirements, a series of adjustment measures can be triggered according to the supervision result to ensure the reliability of the AI model.

[0428] Embodiment three

[0429] Before the AI model is used for inference, model training is needed to generate a specific AI model with uplink adjustment information determination capability. In addition, online training of the AI model or model updating is also needed according to the situation to ensure the reliability and accuracy of the AI model inference.

[0430] FIG. 7 is a flowchart of the training method of the AI model provided by embodiment three of the present application. The training method of the present embodiment is used to train the AI model described in embodiments one and two. As shown in FIG. 7, the method provided by the present embodiment includes the following steps.

[0431] S301, the second device obtains a training sample, the training sample including second input information and at least one label.

[0432] S302, the second device uses the training sample to perform model training to obtain an AI model, the AI model outputting uplink adjustment information, the uplink adjustment information being used to adjust the uplink transmission state of the terminal, wherein the second input information at least includes first input information used by the AI model for inference.

[0433] The second device is a training device of the AI model, which can be a terminal, a network-side device, or a server shown in FIG. 1.

[0434] Optionally, before training the AI model, the second device further obtains AI model training related information (or referred to as AI model training configuration information), which includes at least one of the following:

[0435] (1) AI model training input data information

[0436] The input data information is used to indicate the information content or information type included in the second input information required for AI model training, and the second input information at least includes all the first input information used for AI model inference.

[0437] (2) AI model training label (i.e. true value), or model output information

[0438] (3) AI model training loss function

[0439] (4) AI model training AI algorithm or its index

[0440] (5) AI model training reward information, or adjustment / feedback information

[0441] The adjustment / feedback information includes at least one of the following: whether rollback occurs (i.e. using a traditional RACH method to obtain TA), the number of times rollback occurs, the difference between AI inference TA and actual TA, and the number of times AI inference fails.

[0442] The second device can obtain a large number of training samples for AI model training, and the training samples include second input information and at least one label.

[0443] In some implementations, the second input information at least includes the first input information used for AI model inference, i.e. the first input information is the same as the second input information, or the first input information is a subset of the second input information. The content included in the first input information is described in Embodiment One, which is not repeated here.

[0444] The content included in the AI model output information is described in Embodiment One, which is not repeated here.

[0445] The label generally refers to the identification or annotation of the true class or target value of the data sample. The label is a key component in supervised learning tasks, used to train machine learning models. The label of the sample is also called the true value (i.e. the target of model training). The quality and accuracy of the label are crucial to the performance of the model. The label of the sample can be manually annotated or machine annotated.

[0446] In some implementable manners, the label of the AI model training includes at least one of the following 10 items:

[0447] (1) whether the uplink adjustment information meeting the accuracy requirement is obtained

[0448] The uplink adjustment information output by the AI model may or may not meet the accuracy requirement.

[0449] (2) the number of obtained uplink adjustment information

[0450] In some scenarios, the AI model needs to output multiple uplink adjustment information, and therefore, the label of the training sample can be the number of obtained (i.e., output by the AI model) uplink adjustment information.

[0451] (3) the uplink adjustment information between the terminal and the network side device of the first cell

[0452] (4) the uplink adjustment information between the network side device of the first cell and the network side device of the second cell

[0453] The network side device of the first cell and the network side device of the second cell are the base station of the source cell and the base station of the target cell, or the base station of the primary cell and the base station of the secondary cell, or the base stations of two adjacent cells.

[0454] (5) the propagation delay (or called propagation delay) of uplink or downlink

[0455] The propagation delay can be the UL or DL propagation delay between the terminal and the target cell / scell.

[0456] (6) the delay difference between the uplink propagation delay of the terminal and the target cell and the uplink propagation delay of the terminal and the source cell

[0457] (7) the delay difference between the downlink propagation delay of the terminal and the target cell and the downlink propagation delay of the terminal and the source cell

[0458] (8) the delay difference between the uplink propagation delay of the terminal and the primary cell and the uplink propagation delay of the terminal and the secondary cell

[0459] (9) the delay difference between the downlink propagation delay of the terminal and the primary cell and the downlink propagation delay of the terminal and the secondary cell

[0460] (10) the true value of the uplink adjustment information output by the AI model.

[0461] In some implementable manners, the second device separately performs the AI model training, i.e., the terminal, the network side device or the server respectively performs the whole training of the model.

[0462] In some implementations, the second device jointly trains the AI model with the first device.

[0463] In some implementations, the second device jointly trains the AI model with the first device, including at least one of the following operations:

[0464] (1) The second device trains a base model of the AI model offline, and the first device fine-tunes the base model of the AI model to obtain a final model of the AI model.

[0465] The training of the base model of the AI model is usually completed offline, and the fine-tuning is usually completed online. The final model obtained by the first device through fine-tuning is a model used for inference. For example, a server trains a base model of an AI model offline, sends the base model to a terminal or a network-side device, and the terminal or the network-side device fine-tunes the base model of the AI model before performing AI inference to obtain a final model used for AI inference.

[0466] (2) The second device receives at least part of the second input information sent by the first device.

[0467] In this way, all or part of the second input information used for training by the second device is sent by the first device. The information sent by the first device can be information output by a first device model training. The information output by the first device model training is another model, which is different from the AI model described above.

[0468] For example, a terminal reports at least part of the output information of the model training to a network-side device or a server, and the network-side device or the server uses the information reported by the terminal (i.e., the output information of the terminal model training) as one of the input contents of its own model training (i.e., the AI model described above).

[0469] For another example, a network-side device sends at least part of the output information of the model training to a terminal or a server, and the terminal or the server uses the information sent by the network-side device (i.e., the output information of the network-side model training) as one of the input contents of its own model training (i.e., the AI model described above).

[0470] When the AI model training occurs or at least partially occurs at the terminal side or the server side, the terminal or the server can have multiple sets of trained AI models. The terminal or the server can report auxiliary information to the network-side device, and the network-side device determines and instructs which set of AI model to use for AI inference.

[0471] In some embodiments, when the second device is a terminal or a server, the method further comprises: sending, by the second device, second information to the network-side device, the second information being used to assist the network-side device in selecting an AI model from the plurality of candidate AI models; and receiving, by the second device, third information sent by the network-side device, the third information being used to indicate information of the AI model selected by the network-side device.

[0472] The plurality of candidate AI models differ in at least one of the following: training data sets, label information, input information, and output information.

[0473] In some embodiments, the second information comprises IDs of the plurality of candidate AI models and IDs of training data sets used by each candidate AI model for training. The third information comprises at least one of the following: an ID of the AI model selected by the network-side device, an ID of the training data set, and collection configuration information of the input information.

[0474] In this embodiment, the network-side device decides which AI model of the plurality of or multiple sets of candidate AI models to use for AI inference. After receiving the second information, the network-side device can learn the input information and the output information of each candidate AI model according to the ID of each candidate AI model. Therefore, the network-side device selects an AI model from the plurality of candidate AI models according to the training data sets, the input information, and the output information of the plurality of candidate AI models.

[0475] Optionally, the terminal or the server can also send the input information and the output information of each candidate model to the network-side device together with the ID of the training data set in the second information.

[0476] After selecting the AI model, the network-side device can send the ID of the selected AI model, the ID of the training data set, and the collection configuration information of the input information to the terminal or the server. The terminal or the server determines which AI model to use for inference according to the ID of the AI model, the ID of the training data set, and the collection configuration information of the input information indicated by the network-side device.

[0477] In some embodiments, the collection configuration information of the input information comprises a collection source (e.g., from which devices to collect) and / or a collection manner (how to collect the input information).

[0478] In some embodiments, the training trigger of the AI model comprises at least one of the following: a conditional trigger, a periodic trigger, or a semi-static trigger.

[0479] Periodic triggering refers to that the training of the AI model is periodic, and the second device can receive periodic configuration information sent by other devices, the periodic configuration information being used to periodically trigger the training of the AI model. For example, when the second device is a terminal or a server, the second device can receive periodic configuration information sent by a network-side device, and when the second device is a network-side device, the second device can receive periodic configuration information sent by other network-side devices or a server.

[0480] In some implementable manners, the periodic configuration information comprises at least one of the following:

[0481] (1) Start point of periodic model training

[0482] The start point of periodic model training can be a specific time point, which is carried in the periodic configuration information. Alternatively, the start point of periodic model training can also not be carried in the periodic configuration information, and the second device takes the time point at which the periodic configuration information is received as the start point of periodic model training.

[0483] (2) Interval of periodic model training

[0484] For example, model training is triggered at least once every interval N.

[0485] (3) Number of model training times or model training duration in a period

[0486] The second device triggers AI model training semi-statically based on semi-static configuration information sent by other devices. The semi-static configuration information comprises a start point of model training, a period, a continuous duration in the period, and the like. For example, when the second device is a terminal or a server, the second device can receive semi-static configuration information sent by a network-side device, and when the second device is a network-side device, the second device can receive semi-static configuration information sent by other network-side devices or a server.

[0487] In some implementable manners, other devices can trigger the sending and activation of semi-static configuration information based on specific conditions or events. When a specific event or condition occurs, other devices send semi-static configuration information to the second device, and when a specific event or condition occurs, other devices send activation information to the second device, which is used to activate the semi-static configuration information. The activation of semi-static configuration information can be understood as the activation of semi-static training of the model. After receiving the activation information, the second device activates the semi-static training.

[0488] In other implementable manners, other devices send semi-static configuration information to the second device through an RRC message, and then activate / deactivate semi-static training of the model through DCI or MAC-CE.

[0489] In some implementable manners, the triggering condition of the AI model training comprises at least one of the following:

[0490] (1) AI-related timer expires

[0491] The AI-related timer can be a newly defined specific timer for AI training, and the expiration of the specific timer can trigger the AI model training.

[0492] (2) TAG-related timer expires

[0493] (3) The terminal receives a change in TA value

[0494] The TA value received by the terminal can be the TA value carried in the TAC in the MAC CE

[0495] (4) The value of N changes TA

[0496] (5) The value of N changes TA_offset

[0497] (5) The terminal accesses a new cell, switches frequency points or frequency bands, switches operators or public land mobile networks (PLMN)

[0498] (6) AI model inference fails

[0499] (7) AI model inference fails N times in a row

[0500] (8) The number of times of AI model inference failure reaches a threshold value

[0501] (9) AI model inference is used

[0502] In the training triggering condition, the model training is triggered as long as the inference device uses AI model inference, and the inference device sends information to the training device after the AI model inference ends, triggering the training device to perform model training.

[0503] (10) The terminal moves to a new cell or TA or geographic location

[0504] (11) The terminal's moving speed changes significantly within a specified time period

[0505] (12) The terminal's RSRP measurement value changes or the change exceeds a certain threshold

[0506] (13) Model training-related timer expires for a period of time

[0507] (14) Continuous N times of model supervision occur or N times of model supervision occur ​​

[0508] When the inference device uses a certain AI model for N times of model supervision in succession or N times of model supervision occur, the inference device sends a message to the training device to trigger the training device to train the AI model.

[0509] (15) The external environment of the terminal changes

[0510] The change of the external environment in which the terminal is located includes a change of a channel environment, a change of an external weather state, a change of a location, etc., and the change of the external environment can be determined according to sensor measurement data.

[0511] (16) The speed or acceleration of the terminal exceeds a specified threshold

[0512] (17) The terminal or network location changes or changes exceed a specified threshold

[0513] (18) The terminal selected beam or spatial domain transmission filter changes

[0514] (19) The change of the first parameter of the terminal relative to a certain time point or within a certain time window exceeds a certain value.

[0515] The first parameter includes at least one of the following parameters: timing advance offset, timing pre-compensation, common timing advance offset, dedicated timing advance offset, frequency compensation, frequency offset, common frequency offset, dedicated frequency offset.

[0516] (20) The RACH-less technology is enabled

[0517] (21) The terminal is configured to use a short sequence for PRACH transmission

[0518] (22) The network side device configures PRACH resources dedicated to traditional RACH as a fallback mechanism, and the network side device configures an AI-based RACH mechanism

[0519] (23) The accuracy of the uplink adjustment information is less than or equal to a target threshold or less than or equal to a target threshold for more than a certain time length

[0520] (24) The accuracy of the uplink adjustment information calculated using a non-AI method is less than a certain threshold

[0521] (25) The number of transmission failures or consecutive failures of the uplink adjustment information calculated using a non-AI method exceeds a target threshold

[0522] (26) Default use in a specific scenario (such as in a cell-free scenario)

[0523] (27) Cell or TRP reselection

[0524] (28) cell or TRP switching

[0525] (29) conditions for cell switching

[0526] For example, the conditions for cell switching can include at least one of the following:

[0527] a measurement result (RSRP, RSRQ, RSSI, etc. of L1 or L3) of a source cell is greater than or less than a threshold;

[0528] a measurement result of a target cell (neighboring cell) is better than that of the source cell;

[0529] a measurement result of a target cell (neighboring cell) is greater than a threshold;

[0530] a measurement result of a source cell is less than a threshold 1 and a measurement result of a target cell (neighboring cell) is greater than a threshold 2;

[0531] a load of a source cell is greater than a threshold.

[0532] (30) terminal triggers access to a secondary cell

[0533] (31) terminal moves to a specific location

[0534] (32) uplink or downlink data arrives

[0535] (33) a specific type of service arrives

[0536] (34) receiving at least one of the second input information

[0537] It should be noted that some of the conditions for triggering AI model training are the same as some of the conditions for triggering AI model inference and model supervision described above. For details, please refer to the description of the foregoing related content, which will not be repeated here.

[0538] In some implementations, when the triggering condition for AI model training is met, the second device determines to train the AI model.

[0539] In other implementations, the second device receives training indication information sent by the network side device, the training indication information being used to indicate training of the AI model, and the second device determines to train the AI model according to the training indication information and the triggering condition for AI model training, or the second device determines to train the AI model according to the training indication information, the triggering condition for AI model training, and AI capability information of the second device.

[0540] The network-side device (e.g., a base station) can send the training indication information through a TA MAC CE command. When the second device determines to train the AI model according to the training indication information, the trigger condition of the AI model training, and the AI capability information of the second device, the second device can determine whether to train the AI model according to the training indication information and the AI capability information of the second device when the trigger condition of the AI model training is met.

[0541] The training indication information can be sent by the network-side device together with the trigger condition of the AI model training, or can be sent by the network-side device through a separate message. The AI capability information of the second device includes whether the second device has the capability to train the AI model.

[0542] Correspondingly, when the second device determines to train the AI model according to the training indication information, the trigger condition of the AI model training, and the AI capability information of the second device, the second device determines that the AI model training needs to meet the following three conditions at the same time: the trigger condition of the AI model training is met, the second device receives the training indication information, and the second device has the capability to train the AI model. If any one of the three conditions is not met, the second device determines not to train the AI model.

[0543] In some implementations, the training condition of the AI model inference is met refers to at least one of the training conditions is met.

[0544] Optionally, the network-side device also configures a training end condition of the AI model, and the second device can determine to end the training of the AI model according to the training end condition of the AI model configured by the network-side device when training the AI model.

[0545] In some implementations, the training end condition of the AI model includes at least one of the following six conditions:

[0546] (1) The loss function of the AI model meets a predefined value (or referred to as an index)

[0547] The loss function of the AI model can be at least one of the following: mean square error or normalized mean square error of a predicted value and an actual value, and mean absolute error of the predicted value and the actual value. Correspondingly, the loss function of the AI model meeting the predefined value can be that the error of the AI model is less than a predefined threshold value.

[0548] (2) The number of training reaches a predefined number

[0549] (3) The number of iterations of fine-tuning reaches a predefined number

[0550] (4) At least one information in the label used for training of the model is greater than or equal to or less than a threshold value

[0551] (5) the length of the model training is greater than or equal to a predefined length

[0552] For example, if the length of the AI model training is greater than or equal to a target length, the second device considers that the training is complete.

[0553] (6) the energy consumption of the model training is greater than or equal to a predefined value

[0554] For example, if the energy consumption of the AI model training is greater than or equal to a target value, the second device considers that the training is complete.

[0555] In this embodiment, the second device obtains a training sample, the training sample including second input information and a label, and the second device uses the training sample to perform model training to obtain an AI model, the AI model outputting uplink adjustment information, the uplink adjustment information being used to adjust an uplink transmission state of a terminal, the uplink adjustment information including at least one of the following information: TA-related information and frequency-related information, wherein the second input information at least includes first input information used for AI model inference. By training the AI model, the uplink adjustment information can be predicted using the AI model, and the AI model has higher accuracy in predicting the uplink adjustment information, which can meet the accuracy requirements of various scenarios.

[0556] Embodiment Four

[0557] Embodiments One to Three describe the training, inference, and supervision of AI models, and before the terminal, network-side device, or server performs AI model training, inference, or supervision, the related capabilities (also referred to as AI capability information or AI-related capability information) of the terminal, network-side device, and server need to be defined, and a method for determining the related capabilities is given.

[0558] In this embodiment, at least one of the following AI capability information is defined:

[0559] The capability of the terminal to support (or have) AI model training, i.e., the capability of training an AI model used for uplink adjustment information determination;

[0560] The capability of the terminal to support AI inference using an AI model, i.e., the capability of using the AI model for uplink adjustment information determination;

[0561] The capability of the terminal to support reporting one or more types of assistance information for AI model inference;

[0562] The capability of the network-side device (such as a base station, TRP, or core network device) to support AI model training, i.e., the capability of the AI model used for uplink adjustment information determination;

[0563] The network-side device (such as a base station, a TRP, or a core network device) supports an AI inference capability using an AI model, that is, a capability of using the AI model for uplink adjustment information determination;

[0564] The network-side device (such as a base station, a TRP, or a core network device) supports a capability of indicating one or more types of assistance information for AI model inference;

[0565] The server supports an AI model training capability, that is, a capability of training an AI model for uplink adjustment information determination;

[0566] The server supports an AI inference capability using an AI model, that is, a capability of using the AI model for TA-related information determination

[0567] The server supports a capability of indicating one or more types of assistance information for AI model inference.

[0568] When the terminal, the network-side device, or the server is a first device, the first device is configured to use an AI model for inference and to supervise the AI model. Correspondingly, the first device sends, to a third device, indication information of AI capability information of the first device, the indication information being used to indicate AI-related capabilities that the first device possesses or does not possess (or whether the first device supports or does not support the AI-related capabilities), and the AI capability information of the first device including at least one of the following: a capability of the first device of using an AI model for AI inference; a capability of the first device of sending first assistance information used for the AI inference; and a capability of the first device of supervising the AI model.

[0569] The third device can be a network-side device, and the third device needs to acquire the AI capability information of the first device and make some decisions or send configuration information based on the AI capability information of the first device. When the first device is a terminal or a server, the terminal or the server can send its own AI capability information to the network-side device, so that the network-side device makes some decisions or sends configuration information based on the AI capability information of the terminal or the server.

[0570] For example, when the network-side device determines that the terminal or the server possesses an AI inference capability, the third device sends, to the terminal, first configuration information of an AI model, and can also send, to the terminal or the server, activation indication information or deactivation indication information, the activation indication information being used to indicate activation of AI inference of the AI model, and the deactivation indication information being used to deactivate the AI model.

[0571] For another example, when the network-side device determines that the terminal or the server possesses a capability of supervising an AI model, the third device can send, to the terminal or the server, supervision configuration information of the AI model.

[0572] When the terminal, the network side device or the server is the second device, the second device is configured to train the AI model. Correspondingly, the second device sends the indication information of the AI capability information of the second device to the third device, the indication information is configured to indicate the AI related capability possessed or not possessed (also can be described as whether supported) by the second device, the AI capability information of the second device includes at least one of the following: the second device possesses or does not possess the capability of training the AI model; the second device possesses or does not possess the capability of sending the third auxiliary information, the third auxiliary information is configured to train the AI model.

[0573] The third device needs to obtain the AI capability information of the second device, and makes some decisions or issues configuration information based on the AI capability information of the second device. When the second device is a terminal or a server, the terminal or the server can send its own AI capability information to the network side device, so that the network side device makes some decisions or issues configuration information based on the AI capability information of the terminal or the server.

[0574] For example, when the network side device determines that the terminal or the server possesses the capability information of training the AI model, the third device sends the trigger condition, the periodic configuration information or the semi-static configuration information of the AI model training to the terminal or the server, the trigger condition, the periodic configuration information or the semi-static configuration information of the AI model training is configured to trigger the training of the AI model.

[0575] In some implementable manners, the third device is configured to determine the AI capability information of the first device or the second device based on at least one of the following four items:

[0576] (1) Device type

[0577] In this way, different AI related capabilities are introduced for different types of devices (the first device or the second device). Taking the terminal as an example, the terminal type can be reduced capability (RedCap) type and Internet of Things (IoT). RedCap is a 5G lightweight terminal type, which achieves the goals of cost saving, size reduction, power consumption reduction and service life extension by simplifying device capability and reducing device complexity.

[0578] In some implementable manners, different types of devices can have different AI model training or inference capabilities, including but not limited to the following capabilities:

[0579] (1.1) For different types of devices, the input information for model training / inference is different

[0580] For example, the input information for model training for a device with weak capability (e.g., RedCap type terminal) should be less, and the input information for model training for a device with strong capability can be more.

[0581] (1.2) For different types of devices, the labels for model training are different

[0582] (1.3) For different types of devices, the execution mode of model training is different

[0583] For example, for a terminal with weak capability, it can be considered that only the network side device performs model training, or the terminal performs a small part of joint model training (for example, model training involving user privacy data can be performed at the terminal).

[0584] (1.4) For different types of devices, the AI model for model training is different

[0585] For example, a device with weak capability may not be able to apply a too complex AI model.

[0586] (2) Network type

[0587] In this way, different AI related capabilities are introduced for different network types. The network type can be TNT type or Terrestrial Network (TN).

[0588] (3) One or more reference signals

[0589] The AI capability information of the device can be indicated by some specific reference signals, and when the third device receives the specific reference signal, the AI capability information of the device sending the specific reference signal can be determined.

[0590] For example, a specific resource (such as a specific RO or preamble) of PRACH indicates that the terminal has the capability of AI model training or AI model inference.

[0591] (4) Indication information of AI capability information sent by the first device or the second device

[0592] In this way, the third device determines the AI capability information of the first device or the second device through the explicit indication information of the AI capability information sent by the first device or the second device.

[0593] Exemplarily, the indication information of the AI capability of the first device comprises at least one of the following: first capability indication information, second capability indication information, and third capability indication information. The first capability indication information is used to indicate whether the first device has the capability of using an AI model for AI inference; the second capability indication information is used to indicate whether the first device has the capability of sending first auxiliary information; and the third capability indication information is used to indicate whether the first device has the capability of model supervision on the AI model.

[0594] Exemplarily, the indication information of the AI capability of the second device comprises at least one of the following: fourth capability indication information and fifth capability indication information. The fourth capability indication information is used to indicate whether the second device has the capability of training an AI model; and the fifth capability indication information is used to indicate whether the second device has the capability of sending third auxiliary information.

[0595] In some implementation manners, the first device or the second device can send the indication information of its own AI capability to the third device in the following ways:

[0596] (1) Control information sent by the first device or the second device

[0597] The indication information of the AI capability information of the first device or the second device can be carried in control information, which can be uplink control information or downlink control information. For example, the terminal indicates the indication information of its own AI capability information through uplink control information, which can be physical layer control information (such as UCI information reported by the terminal to the network).

[0598] (2) RRC signaling sent by the first device or the second device

[0599] The indication information of the AI capability information of the first device or the second device can be carried in RRC signaling.

[0600] (3) Interface message between the first device or the second device and the third device

[0601] The indication information of the AI capability information of the first device or the second device can be carried in the interface message, which can include a specific interface message between the terminal and the server, a specific interface message between the terminal and the network side device, or a specific interface message between the server and the network side device.

[0602] The interface message can be a message related to a specific AI model or a message related to all AI models.

[0603] The above embodiments one to four illustrate the training, inference and supervision of the AI model. The following embodiments five to eight illustrate the execution process of AI model inference in some practical application scenarios.

[0604] Embodiment five

[0605] FIG. 8 is a signaling flowchart of a RACH-less cell switching process provided by the embodiment five of the present application. In the present embodiment, the terminal obtains the TA related information between the terminal and the network side device of the cell based on AI model inference. As shown in FIG. 8, the method provided by the present embodiment includes the following steps.

[0606] S401, the terminal accesses cell 1.

[0607] In the present embodiment, the terminal can access cell 1 by using a traditional random access process. Optionally, the terminal also sends indication information of its AI capability information to the base station of cell 1, which is used to indicate the AI capability information of the terminal, and the AI capability information at least includes the capability of the terminal to use AI model for AI inference.

[0608] S402, the terminal obtains the AI model and the first configuration information of the AI model from the base station of cell 1.

[0609] The AI model is trained by the base station of cell 1 and sent to the terminal. It can be understood that during the process of the terminal obtaining the AI model and the first configuration information from the base station of cell 1, the terminal and the base station of cell 1 may need to interact once or more times.

[0610] For example, the terminal sends a request message of the AI model to the base station of cell 1, and the base station of cell 1 sends the AI model to the terminal according to the request message. Optionally, after receiving the request message of the AI model, the base station of cell 1 also determines the AI capability information of the terminal, and if the AI capability information of the terminal is that the terminal has the capability to use the AI model for AI inference, the base station of cell 1 determines to send the AI model to the terminal.

[0611] In one implementation manner, the base station of cell 1 sends the first configuration information to the terminal at the same time when sending the AI model to the terminal, that is, the base station of cell 1 carries the AI model and the first configuration information in a message and sends them to the terminal. The first configuration information includes the trigger condition of the AI model inference. The specific content of the first configuration information is referred to the related description of the embodiment one, and the like, which will not be described here.

[0612] In another implementation manner, the base station of cell 1 sends the AI model and the first configuration information to the terminal based on different messages.

[0613] S403, the terminal triggers handover with the cell 2 when the terminal moves and the handover condition is met.

[0614] The cell 2 is a target cell or Scell to which the terminal needs to be handed over, and the handover condition is described in the foregoing embodiments.

[0615] S404, the terminal uses the AI model to perform AI inference to obtain TA-related information between the terminal and the base station of the cell 2.

[0616] Before the terminal uses the AI model to perform AI inference, the terminal needs to determine whether to perform AI inference according to the trigger condition of AI model inference, or determine whether to perform AI inference according to the trigger condition of AI model inference, AI capability information of the terminal, and activation indication information sent by the base station.

[0617] When it is determined that AI inference can be performed, the terminal obtains first input information, uses the first input information as input of the AI model, and infers to obtain TA-related information between the terminal and the base station of the cell 2. The TA-related information can be an absolute TA value between the terminal and the cell. The first input information includes at least one of the following:

[0618] Signal strength information, signal quality information, path loss-related information, distance information, propagation delay of downlink signals, RTT, actual transmission power, and the like of the downlink signals between the terminal and the base station of the cell 1 (source cell or Pcell of the terminal).

[0619] Signal strength information, signal quality information, path loss-related information, distance information, propagation delay of downlink signals, RTT, actual transmission power, and the like of the downlink signals between the terminal and the base station of the cell 2 (target cell or Scell of the terminal).

[0620] S405, the terminal uses RACH-less technology to switch to the cell 2.

[0621] In the handover process, the terminal uses the AI model to infer the TA value between the terminal and the target cell / Scell, and the base station of the target cell / Scell does not need to configure or indicate the TA through the RACH channel. Therefore, RACH-less technology can be used for handover, and the terminal and the base station do not need to transmit and receive RACH in the RACH-less handover process, thereby saving the energy consumption of RACH transmission and reception.

[0622] S406, the terminal sends uplink information to the base station of the cell 2 using the TA-related information output by the AI model.

[0623] Specifically, the terminal adjusts the time of sending uplink information using the TA-related information output by the AI model.

[0624] In this embodiment, the terminal predicts the TA related information between the terminal and the target cell / Scell based on the AI model sent by the base station during the cell switching process, and communicates with the target cell / Scell based on the inferred TA value of the terminal. This method does not require sending RACH signals during the switching process, has smaller switching delay, and does not require the terminal and the base station to transmit and receive RACH, thereby saving the energy consumption of RACH transmission and reception.

[0625] Embodiment Six

[0626] FIG. 9 is a signaling flowchart based on the RACH-less cell switching process provided in Embodiment Six of the present application. In this embodiment, the base station infers the TA related information between the base station and a specific terminal based on an AI model. As shown in FIG. 9, the method provided in this embodiment includes the following steps.

[0627] S501, the terminal accesses cell 1.

[0628] In this embodiment, the terminal can access cell 1 using the traditional random access process.

[0629] S502, the base station of cell 2 performs model training to obtain an AI model.

[0630] S503, the terminal moves and meets the switching condition, and the terminal triggers switching with cell 2.

[0631] The cell 2 is the target cell or Scell to which the terminal needs to switch, and the switching condition is described in the foregoing embodiments.

[0632] S504, the base station of cell 2 uses the AI model to perform AI inference to obtain the TA related information between cell 2 and the terminal.

[0633] The AI model used by the base station of cell 2 is an AI model trained by the base station itself. Optionally, the base station of cell 2 also obtains first configuration information of the AI model, and the first configuration information includes the triggering condition of AI inference.

[0634] Before using the AI model to perform AI inference, the base station of cell 2 also needs to determine whether to perform AI inference according to the triggering condition of AI inference, or to determine whether to perform AI inference according to the triggering condition of AI inference and its own AI capability information.

[0635] In a case where it is determined that AI inference can be performed, the base station acquires first input information, takes the first input information as an input of an AI model, and infers TA-related information between the base station of cell 2 and the terminal, which can be an absolute TA value between the base station of cell 2 and the terminal. The first input information includes at least one of the following:

[0636] Signal strength information, signal quality information, path loss-related information, distance information, propagation delay of an uplink signal, RTT, actual transmission power, and the like of an uplink signal between the base station of cell 1 (a source cell or Pcell of the terminal) and the terminal.

[0637] Signal strength information, signal quality information, path loss-related information, distance information, propagation delay of an uplink signal, RTT, actual transmission power, and the like of an uplink signal between the base station of cell 2 (a target cell or Scell of the terminal) and the terminal.

[0638] S505, the base station of cell 2 sends the TA-related information to the terminal.

[0639] The base station sends the TA-related information inferred by the AI model to the terminal through downlink information, which can be a downlink channel or a downlink signal. For example, the downlink signal can be an SSB.

[0640] S506, the terminal switches to cell 2 using the RACH-less technology.

[0641] S507, the terminal sends uplink information to the base station of cell 2 using the received TA-related information.

[0642] Specifically, the terminal adjusts the time of sending uplink information using the TA-related information output by the AI model.

[0643] In this embodiment, during the cell switching process, the base station of the target cell / Scell infers the TA-related information between the base station and the specific terminal based on the AI model, and indicates the terminal in the downlink information. No RACH signal needs to be sent during the switching process, the time delay of switching is smaller, and the terminal and the base station do not need to transmit and receive RACH, which saves the energy consumption of RACH transmission and reception.

[0644] Embodiment Seven

[0645] FIG. 10 is a signaling flowchart of a communication method in a Cell-free scenario according to an embodiment of the present application. The communication method according to the embodiment is applied in a Cell-free scenario. In the Cell-free scenario, due to the large difference in geographical positions of the distributedly deployed TRPs, the coverage ranges of different TRPs are different. How to quickly and accurately obtain the TA used for transmission between the terminal and the plurality of TRPs is a problem to be solved. In the embodiment, the macro base station of the Cell-free cell obtains the TA related information between the plurality of TRPs in the coverage range of the macro base station and the specific terminal based on AI model inference. As shown in FIG. 10, the method provided by the embodiment includes the following steps.

[0646] S601, the macro base station of the Cell-free cell trains to obtain an AI model.

[0647] S602, after the terminal is powered on, the terminal selects one or more TRPs in the Cell-free cell.

[0648] In the Cell-free cell, the terminal can select one or more TRPs to implement high-speed data transmission. Before the terminal and the one or more TRPs implement high-speed data transmission, the TA related information between the terminal and the one or more TRPs needs to be obtained.

[0649] S603, the macro base station of the Cell-free cell uses the AI model to perform AI inference to obtain the TA related information between the terminal and the one or more TRPs.

[0650] The macro base station obtains first input information of the AI model, takes the first input information as the input of the AI model, and can infer the TA related information between all TRPs in the coverage range of the macro base station and the terminal. The TA related information can be an absolute TA value.

[0651] The first input information can include at least one of the following: actual transmission power of the one or more TRPs, geographical position information of the one or more TRPs, multipath information of a channel between the one or more TRPs and the terminal, and the like.

[0652] S603, the macro base station of the Cell-free cell sends the TA related information corresponding to the one or more TRPs to the terminal.

[0653] The macro base station carries the TA related information corresponding to the one or more TRPs in the downlink information and sends it to the terminal. The downlink signal can be a downlink channel or a downlink signal, for example, an SSB. The macro base station can carry the inferred TA related information corresponding to the one or more TRPs in the SSB and send it to the terminal. The TA related information corresponding to each TRP can be different.

[0654] S604, the terminal sends uplink signals to the corresponding TRP using the TA-related information corresponding to the one or more TRPs.

[0655] When the terminal communicates with different TRPs, the terminal uses the TA-related information corresponding to the TRP for uplink transmission, thereby realizing high-speed data transmission between the terminal and multiple TRPs.

[0656] In this embodiment, the macro base station of the Cell-free cell infers the TA-related information between the one or more TRPs and the specific terminal within the coverage of the cell or within the coverage of the macro base station based on the AI model trained by itself, and indicates the TA-related information between the multiple TRPs and the specific terminal to the terminal. This method can obtain the TA-related information between the multiple TRPs and the specific terminal within the coverage of the macro base station at one time through the AI model.

[0657] Embodiment Eight

[0658] FIG. 11 is a signaling flowchart of a communication method in a Cell-free scenario according to an embodiment of the present application. The difference between this embodiment and embodiment seven is that in this embodiment, the terminal infers the TA-related information between the multiple TRPs and the terminal in the Cell-free cell based on the AI model. As shown in FIG. 11, the method provided in this embodiment includes the following steps.

[0659] S701, the macro base station of the Cell-free cell performs model training to obtain an AI model.

[0660] S702, the terminal accesses the Cell-free cell.

[0661] The terminal accesses the Cell-free cell by interacting with the macro base station of the Cell-free cell.

[0662] S703, the terminal obtains the AI model and first configuration information of the AI model from the macro base station of the Cell-free cell.

[0663] The AI model can be requested by the terminal to the macro base station, or can be actively sent by the macro base station to the terminal. For example, after the terminal accesses the Cell-free cell, the macro base station determines that the terminal supports AI inference using the AI model, and then sends the AI model and the first configuration information of the AI model to the terminal.

[0664] S704, the terminal moves or detects that uplink data arrives.

[0665] Terminal movement or detection of uplink traffic arrival will trigger uplink transmission. In a cell-free cell, a terminal can communicate with multiple TRPs simultaneously to achieve high-speed data transmission. When the terminal moves, TRP switching is triggered. At this time, the TA related information between the terminal and multiple TRPs needs to be obtained.

[0666] S705, the terminal uses the AI model to perform AI inference to obtain TA related information between the terminal and one or more TRPs.

[0667] The terminal obtains first input information of the AI model, and uses the first input information as input of the AI model. The TA related information between the one or more TRPs and the terminal can be inferred, and the TA related information can be an absolute TA value. The AI model can predict the TA related information between each of the multiple TRPs and the terminal.

[0668] The first input information can include at least one of the following: actual transmission power of the one or more TRPs, geographical position information of the one or more TRPs, multipath information of a channel between the one or more TRPs and the terminal, etc.

[0669] S706, the terminal uses the TA related information corresponding to the one or more TRPs to send uplink signals to the corresponding TRPs.

[0670] In this embodiment, the terminal uses the TA related information between the one or more TRPs corresponding to the AI model to communicate with the corresponding TRPs. The method can obtain the TA related information between the multiple TRPs and the terminal at one time through the AI model.

[0671] Embodiment Nine

[0672] FIG. 12 is a signaling flowchart of a communication method provided by Embodiment Nine of the present application. In this embodiment, a base station infers a CommonTA based on an AI model and a predefined reference point, and broadcasts the CommonTA to all terminals in a cell. The terminal uses the CommonTA for uplink transmission, and continuously updates the CommonTA according to the uplink signal. As shown in FIG. 12, the method provided by this embodiment includes the following steps.

[0673] S801, the base station performs model training to obtain an AI model.

[0674] S802, the base station determines to use the AI model to perform AI inference, and determines a CommonTA between the reference point and the base station according to the predefined reference point information and the AI model.

[0675] The base station can determine to use the AI model for AI inference when the trigger condition of the AI model inference is met, or can determine to use the AI model for AI inference when the trigger condition of the AI model inference is met and the activation indication information sent by the other network side device is received.

[0676] The predefined reference point information (or referred to as reference point configuration information) can be position information of the reference point, which can be a position coordinate in the cell or a region in the cell.

[0677] The base station takes the reference point information as the first input information of the AI model, and the AI model performs inference based on the reference point information to obtain the CommonTA between the reference point and the base station.

[0678] S803, the base station sends the CommonTA to a group of terminals or specific terminals.

[0679] The base station sends the CommonTA to a group of terminals or specific terminals in the form of broadcast, multicast or unicast. For example, the base station broadcasts the CommonTA to all terminals within the cell coverage or to terminals within a certain distance range from the reference point through the common PDCCH channel.

[0680] S804, the terminal receiving the CommonTA transmits uplink signals using the CommonTA.

[0681] S805, the base station measures the uplink signals transmitted by the terminal and calculates an updated TA.

[0682] The updated TA can be an absolute TA value or a relative TA value, wherein the absolute TA value refers to a new TA value obtained by adding an offset value to the CommonTA, or an absolute value of a new TA. The relative TA value is an offset value relative to the CommonTA. In this step, the base station calculates the updated TA using a non-AI method.

[0683] S806, the base station sends the updated TA to the corresponding terminal.

[0684] It can be understood that the updated TA calculated by the base station based on the uplink signals transmitted by different terminals is different, and therefore the base station needs to send the updated TA to the corresponding terminal.

[0685] S807, the terminal transmits uplink signals according to the updated TA.

[0686] For the group of terminals that have received the CommonTA, some terminals can have received the updated TA, and all terminals can have received the updated TA. For the terminals that have received the updated TA, the terminals perform uplink signal transmission according to the updated TA.

[0687] When the updated TA is an absolute TA value, the terminal directly uses the updated TA value to replace the CommonTA for subsequent uplink transmission. When the updated TA is a relative TA value, the terminal superimposes the relative TA value on the CommonTA to obtain a final TA value, and uses the final TA value for subsequent uplink transmission.

[0688] In this embodiment, the base station infers a CommonTA based on the AI model and the predefined reference point, and sends the CommonTA to a group or specific terminals. The terminals that receive the CommonTA use the CommonTA for uplink transmission. The base station continuously updates the CommonTA according to the uplink signal. The CommonTA predicted by the AI model has higher accuracy, thereby improving the accuracy and reliability of uplink transmission.

[0689] It should be noted that the method for determining uplink adjustment information based on the AI model is not limited to the scenarios described in Embodiments 5 to 9. The method for determining uplink adjustment information based on the AI model can be used in any scenario that requires determination of uplink adjustment information. For example, it can also be applied to the NTN scenario. The AI model infers based on information of the satellite (such as ephemeris information, or information such as the reference position of the cell and the moving trajectory).

[0690] Embodiment 10

[0691] The communication method provided in the embodiments of the present application can be executed by a communication device. In the embodiments of the present application, the communication device executes the communication method as an example, and the communication device provided in the embodiments of the present application is described.

[0692] The communication device provided in the embodiments of the present application can be a first device or a component in the first device, such as a chip. The first device can be a terminal, a network side device, or a server, etc. For example, the terminal can include but is not limited to the types of the terminal 11 listed above, the network side device can include but is not limited to the types of the network side device 12 listed above, and the server can include but is not limited to the types of the server 13 listed above, which are not limited in the embodiments of the present application.

[0693] The communication apparatus includes a receiving module, a sending module and a processing module. The receiving module, the sending module and the processing module can be implemented by software or hardware. When implemented by hardware, the processing module can be implemented by a processor, which can include a general-purpose processor, a special-purpose processor, a central processing unit (CPU), a microprocessor, a digital signal processor (DSP), an artificial intelligent (AI) processor, a graphics processing unit (GPU), an application specific integrated circuit (ASIC), a network processor (NP), a field programmable gate array (FPGA) or other programmable logic device, a gate circuit, a transistor, a discrete hardware component, etc. The receiving module and the sending module can be implemented by a communication interface, which can include a transceiver, a pin, a circuit, a bus, a radio frequency unit, etc.

[0694] Specifically, referring to FIG. 13, which is a schematic block diagram of a communication apparatus provided by an embodiment of the present application, when the communication apparatus 900 is a first device or a component in the first device, the communication apparatus 900 includes:

[0695] The processing module 901 is configured to obtain first input information.

[0696] The processing module 901 is further configured to obtain uplink adjustment information according to an artificial intelligent (AI) model and the first input information, the uplink adjustment information being used to adjust an uplink transmission state of a terminal, and the uplink adjustment information including at least one of the following:

[0697] Uplink adjustment information of a specific terminal;

[0698] Uplink adjustment information of a group of terminals;

[0699] Uplink adjustment information in a specific area;

[0700] Uplink adjustment information of at least one beam direction;

[0701] Uplink adjustment information of a specific reference point;

[0702] Uplink adjustment information corresponding to a specific path loss range or a specific reference signal received power (RSRP) range;

[0703] uplink adjustment information corresponding to a specific sequence or sequence format;

[0704] common uplink adjustment information;

[0705] uplink adjustment information between the specific terminal and at least one network-side device;

[0706] uplink adjustment information of other terminals within the terminal group;

[0707] a group of candidate uplink adjustment information and / or a probability value associated with the group of candidate uplink adjustment information.

[0708] In some implementations, the first input information includes at least one of the following:

[0709] signal strength information between the terminal and the network-side device of the first cell;

[0710] signal quality information between the terminal and the network-side device of the first cell;

[0711] path loss information;

[0712] distance information;

[0713] absolute TA value between the terminal and the network-side device of the first cell;

[0714] relative TA value between the terminal and the network-side device of the first cell;

[0715] identification of a transmission reference point (TRP);

[0716] identification of a TRP group;

[0717] cell identification of the first cell;

[0718] group identification of a cell group in which the first cell is located;

[0719] group identification of a timing advance group (TAG) in which the first cell is located;

[0720] tracking area (TA) identification in which the first cell is located;

[0721] radio access network notification area (RNA) identification in which the first cell is located;

[0722] frequency domain information in which the first cell operates;

[0723] reference signal index, beam index, or beam direction transmitted between the terminal and the network-side device of the first cell;

[0724] transmission delay between the terminal and the network-side device of the first cell;

[0725] round trip time (RTT) between the terminal and the network-side device of the first cell;

[0726] transmission power information of a network side device of the first cell;

[0727] location information of a network side device of the first cell;

[0728] location information or distribution information of the terminal;

[0729] a moving direction and / or a moving speed of the terminal, the satellite or the network side device of the first cell;

[0730] energy consumption status and / or power status of the terminal, the satellite or the network side device of the first cell;

[0731] operator information and network type information supported by the terminal;

[0732] antenna orientation information of the terminal, the satellite or the network side device of the first cell;

[0733] energy consumption status and / or power status of the first cell;

[0734] perception information of the terminal;

[0735] scenario information of a network in which the terminal is located;

[0736] environment information in which the terminal is located;

[0737] ephemeris information;

[0738] reference position or moving trajectory information of a cell in an NTN scenario;

[0739] multipath information of a channel;

[0740] time information;

[0741] one or more of the following information used currently or within a certain time window: timing advance offset, timing pre-compensation, common timing advance offset, dedicated timing advance offset, frequency compensation, frequency offset, common frequency offset and dedicated frequency offset;

[0742] The first cell includes one of the following: a source cell, a target cell, a currently camped cell, a currently accessed cell, a candidate cell, a primary cell or a secondary cell of the terminal.

[0743] In some implementations, the uplink adjustment information includes at least one of the following information:

[0744] timing advance TA related information, the TA related information including any one of the following: absolute TA value, relative TA value, timing advance offset, timing pre-compensation, common timing advance offset, dedicated timing advance offset;

[0745] Frequency related information, including at least one of: frequency error, frequency compensation, frequency offset, common frequency offset, specific frequency offset.

[0746] In some implementations, the first device is a terminal device, and the apparatus 900 further includes:

[0747] The receiving module 902 is configured to receive at least part of the first input information sent by the network-side device.

[0748] In some implementations, the first device is a terminal device, and after the processing module 901 obtains the uplink adjustment information according to the AI model and the first input information, the processing module 901 is further configured to:

[0749] According to the uplink adjustment information, at least one of the following operations is performed:

[0750] Report first information to a training device, the first information being used for updating the AI model, and the first information including at least one of the following information: input information of this AI inference, output information of this AI inference, and current state information of the terminal;

[0751] Fall back to a determination process of uplink adjustment information in a non-AI manner;

[0752] Trigger updating of the AI model;

[0753] Trigger change of input information of the AI model;

[0754] Trigger supervision of the AI model.

[0755] In some implementations, the first device is a network-side device, and the apparatus 900 further includes:

[0756] The receiving module 902 is configured to receive at least part of the first input information sent by the terminal.

[0757] In some implementations, the first device is a server, and the apparatus 900 further includes:

[0758] The receiving module 902 is configured to receive at least part of the first input information sent by the terminal and / or the network-side device.

[0759] In some implementations, the processing module 901 is further configured to:

[0760] Obtain first configuration information of the AI model;

[0761] The first configuration information includes at least one of the following:

[0762] An identifier of the AI model used for AI inference;

[0763] application scope of the AI model inference;

[0764] period of the AI model inference;

[0765] validity duration of the AI model inference;

[0766] trigger condition of the AI model inference;

[0767] configuration parameter of the AI model, the configuration parameter of the AI model comprising a type of first input information and / or a type of output information of the AI model;

[0768] whether joint inference of multiple devices is supported.

[0769] In some implementations, the processing module 901 is further configured to:

[0770] obtain activation indication information or deactivation indication information;

[0771] determine to use the AI model for AI inference according to the activation indication information and the trigger condition of the AI model inference;

[0772] or, determine not to use the AI model for AI inference according to the deactivation indication information.

[0773] In some implementations, the processing module 901 is further configured to:

[0774] determine to use the AI model for inference according to the trigger condition of the AI model inference.

[0775] In some implementations, the trigger condition of the AI model inference comprises at least one of the following:

[0776] AI-related timer timeout;

[0777] RACH-less technology is enabled;

[0778] the terminal is configured to use a short sequence for transmission of a physical random access channel (PRACH);

[0779] the network-side device configures PRACH resources dedicated to a legacy RACH as a fallback mechanism, and the network-side device configures an AI-based RACH mechanism;

[0780] accuracy of uplink adjustment information is less than or equal to a target threshold or less than or equal to a target threshold for more than a certain duration;

[0781] accuracy of uplink adjustment information calculated using a non-AI method is less than a certain threshold;

[0782] The number of failures or the number of consecutive failures of signal transmission using the uplink adjustment information calculated in a non-AI manner exceeds a target threshold;

[0783] Default use in specific scenarios;

[0784] Cell or TRP reselection;

[0785] Cell or TRP switching;

[0786] Satisfy the conditions of cell switching;

[0787] Terminal triggers access to a secondary cell;

[0788] Terminal moves to a specific location;

[0789] Uplink or downlink data arrives;

[0790] A specific type of service arrives;

[0791] Receiving at least one information in the first input information;

[0792] Terminal speed or acceleration exceeds a set threshold;

[0793] Terminal or network location changes or changes exceed a set threshold;

[0794] The terminal selected beam or spatial transmission filter changes;

[0795] The change of the first parameter of the terminal relative to a certain time point or within a certain time window exceeds a certain value, and the first parameter includes at least one of the following parameters: timing advance offset, timing pre-compensation, common timing advance offset, dedicated timing advance offset, frequency compensation, frequency offset, common frequency offset, and dedicated frequency offset.

[0796] In some implementations, the processing module 901 is further configured to:

[0797] Obtain supervision configuration information of the AI model;

[0798] Supervise the AI model according to the supervision configuration information.

[0799] In some implementations, the supervision configuration information includes at least one of the following:

[0800] Identification of the AI model that needs to be supervised;

[0801] Period of model supervision;

[0802] Number of model supervision;

[0803] Duration of model supervision;

[0804] a window-related information of model supervision;

[0805] a trigger condition of model supervision;

[0806] an index of model supervision;

[0807] a label of model supervision.

[0808] In some implementations, the label of model supervision has the same or different value as a label in the AI model training process.

[0809] In some implementations, the trigger condition of model supervision includes at least one of the following:

[0810] an inference result of the AI model does not meet accuracy requirements;

[0811] at least one of inference indexes of the AI model does not meet requirements;

[0812] the index of model supervision does not meet requirements, the index of model supervision including an error between a predicted value and a real value of the AI model, and / or a performance index of the network;

[0813] a supervision timer of the AI model is timed out;

[0814] the terminal switches to a new cell, TRP or beam;

[0815] after the AI model inference fails;

[0816] the AI model continuously fails N times of inference;

[0817] a number of times of AI model inference failure reaches a threshold value;

[0818] using the AI model for inference;

[0819] the terminal moves to a new cell, tracking area or geographic location;

[0820] a moving speed of the terminal significantly changes within a specified time length;

[0821] an external environment of an inference device of the AI model changes.

[0822] In some implementations, the inference index of the AI model includes at least one of the following:

[0823] a complexity of the AI model;

[0824] a time delay of the AI model inference;

[0825] a success rate of the AI model inference;

[0826] Reliability of an AI model inference result.

[0827] In some implementations, the apparatus further includes:

[0828] The sending module 903 is configured to send, to a third device, indication information of AI capability information of the first device.

[0829] The AI capability information of the first device includes at least one of the following:

[0830] The first device has or does not have the capability of using the AI model for AI inference;

[0831] The first device has or does not have the capability of sending first auxiliary information used for the AI inference;

[0832] The first device has or does not have the capability of model supervision on the AI model.

[0833] In some implementations, the third device determines the AI capability information of the first device based on at least one of the following: device type, network type, one or more reference signals, and the indication information of the AI capability information of the first device.

[0834] In some implementations, the indication information of the AI capability of the first device includes at least one of the following:

[0835] First capability indication information for indicating that the first device has or does not have the capability of using the AI model for AI inference;

[0836] Second capability indication information for indicating that the first device has or does not have the capability of sending the first auxiliary information;

[0837] Third capability indication information for indicating that the first device has or does not have the capability of model supervision on the AI model.

[0838] In some implementations, the first input information is the same as second input information used when the AI model is trained, or the first input information is a subset of the second input information.

[0839] The communication apparatus 900 obtains first input information, obtains uplink adjustment information according to the AI model and the first input information, the uplink adjustment information being used to adjust an uplink transmission state of a terminal, sends the uplink adjustment information obtained by AI model inference to a corresponding terminal, and the terminal adjusts the uplink transmission state according to the uplink adjustment information. The accuracy of the uplink adjustment information obtained by AI model inference is high, thereby ensuring the accuracy and reliability of data transmission.

[0840] The communication apparatus 900 provided in the embodiments can implement each process performed by the first device in the methods described in the embodiments one to nine, and achieve the same technical effects. To avoid repetition, details are not described herein.

[0841] Embodiment eleven

[0842] The AI model training method provided in the embodiments can be performed by an AI model training apparatus. In the embodiments, the AI model training method is taken as an example to illustrate the AI model training apparatus provided in the embodiments.

[0843] The AI model training apparatus provided in the embodiments can be a second device or a component in the second device, for example, a chip. The second device can be a terminal, a network side device, a server, or the like. For example, the terminal can include, but is not limited to, the types of the terminal 11 listed above, the network side device can include, but is not limited to, the types of the network side device 12 listed above, and the server can include, but is not limited to, the types of the server 13 listed above. The embodiments are not limited in this regard.

[0844] The AI model training apparatus includes a receiving module, a sending module, and a processing module. The receiving module, the sending module, and the processing module can be implemented by software or hardware. When implemented by hardware, the processing module can be implemented by a processor. For example, the processor can include a general processor, a special-purpose processor, or the like, such as a CPU, a microprocessor, a DSP, an AI processor, a GPU, an ASIC, an NP, an FPGA, or other programmable logic devices, gate circuits, transistors, discrete hardware components, or the like. The receiving module and the sending module can be implemented by a communication interface, which can include one or more of a transceiver, a pin, a circuit, a bus, a radio frequency unit, or the like.

[0845] Specifically, referring to FIG. 14, which is a schematic block diagram of the AI model training apparatus provided in the embodiments, when the AI model training apparatus 1000 is a second device or a component in the second device, the AI model training apparatus 1000 includes:

[0846] The processing module 1001 is configured to obtain a training sample, the training sample including second input information and at least one label.

[0847] The processing module 1001 is further configured to perform model training on the training sample to obtain an AI model used in Embodiment Ten, the AI model outputting uplink adjustment information, the uplink adjustment information being used to adjust an uplink transmission state of a terminal, wherein the second input information at least includes first input information used for inference of the AI model, and the uplink adjustment information includes at least one of the following:

[0848] uplink adjustment information of a specific terminal;

[0849] uplink adjustment information of a group of terminals;

[0850] uplink adjustment information in a specific area;

[0851] uplink adjustment information of at least one beam direction;

[0852] uplink adjustment information of a specific reference point;

[0853] uplink adjustment information corresponding to a specific path loss range or a specific reference signal received power (RSRP) range;

[0854] uplink adjustment information corresponding to a specific sequence or sequence format;

[0855] common uplink adjustment information;

[0856] uplink adjustment information between a specific terminal and at least one network side device;

[0857] uplink adjustment information of other terminals in a terminal group;

[0858] a group of candidate uplink adjustment information and / or a probability value associated with the group of candidate uplink adjustment information.

[0859] In some implementations, the label of the training sample includes at least one of the following:

[0860] uplink adjustment information meeting an accuracy requirement;

[0861] a number of obtained uplink adjustment information;

[0862] uplink adjustment information between a terminal and a network side device of a first cell;

[0863] uplink adjustment information between a network side device of a first cell and a network side device of a second cell;

[0864] a propagation delay of uplink or downlink;

[0865] a delay difference between an uplink propagation delay between a terminal and a target cell and an uplink propagation delay between the terminal and a source cell;

[0866] a difference in a downlink propagation delay between the terminal and the target cell and a downlink propagation delay between the terminal and the source cell;

[0867] a difference in an uplink propagation delay between the terminal and the primary cell and an uplink propagation delay between the terminal and the secondary cell;

[0868] a difference in a downlink propagation delay between the terminal and the primary cell and a downlink propagation delay between the terminal and the secondary cell;

[0869] a real value of uplink adjustment information output by the AI model.

[0870] In some implementations, the second device jointly trains the AI model with the first device.

[0871] In some implementations, when the second device is a terminal or a server, the apparatus 1000 further includes:

[0872] The sending module 1003 is configured to send second information to a network side device, the second information being used to assist the network side device in selecting an AI model from a plurality of candidate AI models;

[0873] The receiving module 1002 is configured to receive third information sent by the network side device, the third information being used to indicate information of the AI model selected by the network side device.

[0874] In some implementations, at least one of the following information of the plurality of candidate AI models is different: a training data set, label information, input information, and output information.

[0875] In some implementations, the second information includes IDs of the plurality of candidate AI models and IDs of training data sets used for training of each candidate AI model;

[0876] The third information includes at least one of the following: an ID of the AI model selected by the network side device, an ID of the training data set, and collection configuration information of the input information.

[0877] In some implementations, the trigger type of the AI model training includes at least one of the following: a conditional trigger, a periodic trigger, or a semi-static trigger.

[0878] In some implementations, the trigger condition of the AI model training includes at least one of the following:

[0879] timeout of an AI related timer;

[0880] timeout of a timing advance group (TAG) related timer;

[0881] the terminal receives a change in a TA value;

[0882] N TA a value of the first parameter has changed;

[0883] N TA_offset a value of the first parameter has changed;

[0884] the terminal accesses a new cell, switches a frequency point or a frequency band, switches an operator or a public land mobile network (PLMN);

[0885] AI model inference fails;

[0886] AI model continuous inference fails N times;

[0887] the number of times of AI model inference failure reaches a threshold value;

[0888] AI model inference is used;

[0889] the terminal moves to a new cell or a tracking area or a geographic location;

[0890] a moving speed of the terminal changes significantly within a specified time length;

[0891] a reference signal received power (RSRP) measurement value of the terminal changes or a change amount exceeds a certain threshold value;

[0892] a model training related timer times out for a period of time;

[0893] continuous N times of model supervision occurs or N times of model supervision occurs;

[0894] an external environment of the terminal changes;

[0895] a speed or an acceleration of the terminal exceeds a specified threshold value;

[0896] a terminal or network location changes or a change exceeds a specified threshold value;

[0897] a terminal selected beam or a spatial domain transmission filter changes;

[0898] a first parameter of the terminal changes relative to a certain time point or within a certain time window, the first parameter including at least one of the following parameters: a timing advance offset, a timing pre-compensation, a common timing advance offset, a dedicated timing advance offset, a frequency compensation, a frequency offset, a common frequency offset, a dedicated frequency offset;

[0899] RACH-less technology is enabled;

[0900] the terminal is configured to use a short sequence for PRACH transmission;

[0901] The network-side device configures PRACH resources specially used for traditional RACH as a fallback mechanism, and the network-side device configures an AI-based RACH mechanism;

[0902] The accuracy of the uplink adjustment information is less than or equal to a target threshold or is less than or equal to a target threshold for more than a certain time length;

[0903] The accuracy of the uplink adjustment information calculated using a non-AI manner is less than a specific threshold;

[0904] The number of transmission failures or the number of consecutive failures of the uplink adjustment information calculated using a non-AI manner exceeds a target threshold;

[0905] Default use in a specific scenario;

[0906] Cell or TRP reselection;

[0907] Cell or TRP switching;

[0908] Satisfy the conditions of cell switching;

[0909] Terminal triggers access to a secondary cell;

[0910] The terminal moves to a specific location;

[0911] Uplink or downlink data arrives;

[0912] A specific type of service arrives;

[0913] Receiving at least one information in the second input information.

[0914] In some implementations, the apparatus further includes:

[0915] The receiving module 1002 is configured to receive training indication information sent by a network-side device;

[0916] The processing module 1001 is further configured to:

[0917] When the trigger condition of the AI model training is met, determine to train the AI model according to the training indication information and the AI capability information of the first device;

[0918] Or, when the trigger condition of the AI model training is met, determine to train the AI model.

[0919] In some implementations, the training end condition of the AI model includes at least one of the following:

[0920] The loss function of the AI model meets a predefined value;

[0921] The number of training times reaches a predefined number of times;

[0922] The number of iterations of fine-tuning reaches a predefined number;

[0923] At least one piece of information in the label for training of the model is greater than or equal to or less than a threshold value;

[0924] The length of model training is greater than or equal to a predefined length;

[0925] The energy consumption of model training is greater than or equal to a predefined value.

[0926] In some implementations, the apparatus 1000 further includes:

[0927] The sending module 1003 is configured to send, to a third device, indication information of AI capability information of the second device;

[0928] The AI capability information of the second device includes at least one of the following:

[0929] The second device has or does not have the capability of training the AI model;

[0930] The second device has or does not have the capability of sending third auxiliary information for training of the AI model.

[0931] In some implementations, the third device is configured to determine the AI capability information of the second device by at least one of the following: device type, network type, one or more reference signals, and indication information of the AI capability information of the second device.

[0932] In some implementations, the indication information of the AI capability of the second device includes at least one of the following:

[0933] Fourth capability indication information for indicating that the second device has or does not have the capability of training the AI model;

[0934] Fifth capability indication information for indicating that the second device has or does not have the capability of sending the third auxiliary information.

[0935] The training apparatus 1000 of the AI model obtains a training sample, the training sample including second input information and a label, and performs model training using the training sample to obtain an AI model, the AI model outputting uplink adjustment information, the uplink adjustment information being used to adjust an uplink transmission state of a terminal, and the second input information including at least first input information used for inference of the AI model. Through training of the AI model, the AI model can be used to predict the uplink adjustment information, and the AI model has higher accuracy in predicting the uplink adjustment information, thereby meeting the accuracy requirements of various scenarios.

[0936] The training device 1000 of the AI model provided in the embodiments can implement each process performed by the second device in the methods of the first to ninth embodiments and achieve the same technical effects. To avoid repetition, details are not described herein.

[0937] As shown in FIG. 15, the embodiments of the present application also provide a communication device 1100, which includes a processor 1101 and a memory 1102, and the memory 1102 stores programs or instructions executable on the processor 1101. For example, when the communication device 1100 is the first device, the programs or instructions are executed by the processor 1101 to implement each step of the first device in the above method embodiments and achieve the same technical effects. When the communication device 1100 is the second device, the programs or instructions are executed by the processor 1101 to implement each step of the second device in the above method embodiments and achieve the same technical effects. To avoid repetition, details are not described herein.

[0938] The embodiments of the present application also provide a terminal, which includes a processor and a communication interface, the communication interface and the processor are coupled, and the processor is configured to run programs or instructions to implement the steps in the method embodiments shown in FIGS. 5-12. The terminal embodiments correspond to the above terminal-side method embodiments, and each implementation process and implementation manner of the above method embodiments can be applied to the terminal embodiments and achieve the same technical effects. The terminal can be the communication device shown in FIG. 13 or the training device of the AI model shown in FIG. 14. Specifically, FIG. 16 is a schematic diagram of the hardware structure of a terminal for implementing the embodiments of the present application.

[0939] The terminal 1200 includes, but is not limited to, at least some of the radio frequency unit 1201, the network module 1202, the audio output unit 1203, the input unit 1204, the sensor 1205, the display unit 1206, the user input unit 1207, the interface unit 1208, the memory 1209, and the processor 1210.

[0940] Those skilled in the art can understand that the terminal 1200 can also include a power supply (such as a battery) for supplying power to each component, and the power supply can be logically connected to the processor 1210 through a power management system, so as to realize functions such as management of charging, discharging, and power consumption management through the power management system. The terminal structure shown in FIG. 16 does not constitute a limitation on the terminal, and the terminal can include more or fewer components than shown, or combine certain components, or different component arrangements, which are not described herein.

[0941] It should be understood that in the embodiments of the present application, the input unit 1204 can include a graphics processor 12041 and a microphone 12042, and the graphics processor 12041 processes image data of a still picture or a video obtained by an image capture device (such as a camera) in a video capture mode or an image capture mode. The display unit 1206 can include a display panel 12061, which can be configured in the form of a liquid crystal display, an organic light-emitting diode, etc. The user input unit 1207 includes at least one of a touch panel 12071 and other input devices 12072. The touch panel 12071 is also called a touch screen. The touch panel 12071 can include two parts of a touch detection device and a touch controller. The other input devices 12072 can include, but are not limited to, a physical keyboard, function keys (such as volume control keys, on-off keys, etc.), a trackball, a mouse, a joystick, and the like, which will not be described here.

[0942] In the embodiments of the present application, after the radio frequency unit 1201 receives the downlink data from the network side device, it can be transmitted to the processor 1210 for processing. In addition, the radio frequency unit 1201 can send uplink data to the network side device. Generally, the radio frequency unit 1201 includes, but is not limited to, an antenna, an amplifier, a transceiver, a coupler, a low noise amplifier, a duplexer, etc.

[0943] The memory 1209 can be used to store software programs or instructions and various data. The memory 1209 can mainly include a first storage area storing programs or instructions and a second storage area storing data, wherein the first storage area can store an operating system, application programs or instructions required by at least one function (such as a sound playing function, an image playing function, etc.), and the like. In addition, the memory 1209 can include a volatile memory or a non-volatile memory. The non-volatile memory can be a Read-Only Memory (ROM), a Programmable ROM (PROM), an Erasable PROM (EPROM), an Electrically EPROM (EEPROM), or a flash memory. The volatile memory can be a Random Access Memory (RAM), a Static RAM (SRAM), a Dynamic RAM (DRAM), a Synchronous DRAM (SDRAM), a Double Data Rate SDRAM (DDR SDRAM), an Enhanced SDRAM (ESDRAM), a Synch link DRAM (SLDRAM), and a Direct Rambus RAM (DRRAM). The memory 1209 in the embodiments of the present application includes but is not limited to these and any other suitable types of memory.

[0944] The processor 1210 can include one or more processing units; optionally, the processor 1210 integrates an application processor and a modem processor, wherein the application processor mainly processes operations related to an operating system, a user interface, and an application program, and the modem processor mainly processes wireless communication signals, such as a baseband processor. It can be understood that the above-mentioned modem processor can also not be integrated into the processor 1210.

[0945] The processor 1210 is configured to obtain first input information, obtain uplink adjustment information according to an AI model and the first input information, and the uplink adjustment information is used to adjust an uplink transmission state of a terminal, and the uplink adjustment information includes at least one of the following:

[0946] The uplink adjustment information of a specific terminal;

[0947] The uplink adjustment information of a group of terminals;

[0948] uplink adjustment information in a specific region;

[0949] uplink adjustment information of at least one beam direction;

[0950] uplink adjustment information of a specific reference point;

[0951] uplink adjustment information corresponding to a specific path loss range or a specific reference signal receiving power (RSRP) range;

[0952] uplink adjustment information corresponding to a specific sequence or sequence format;

[0953] common uplink adjustment information;

[0954] uplink adjustment information between a specific terminal and at least one network side device;

[0955] uplink adjustment information of other terminals in a terminal group;

[0956] a group of candidate uplink adjustment information and / or a probability value associated with the group of candidate uplink adjustment information.

[0957] The terminal obtains first input information, obtains uplink adjustment information according to an AI model and the first input information, and adjusts the uplink transmission state of the terminal according to the uplink adjustment information. The accuracy of the uplink adjustment information obtained by the AI model is high, thereby ensuring the accuracy and reliability of data transmission.

[0958] Alternatively, the processor 1210 is configured to: obtain training samples, wherein the training samples include second input information and at least one label; and perform model training to obtain an AI model using the training samples, wherein the AI model outputs uplink adjustment information, and the uplink adjustment information is used to adjust the uplink transmission state of the terminal, and the second input information at least includes first input information used for inference of the AI model, and the uplink adjustment information includes at least one of the following:

[0959] uplink adjustment information of a specific terminal;

[0960] uplink adjustment information of a group of terminals;

[0961] uplink adjustment information in a specific region;

[0962] uplink adjustment information of at least one beam direction;

[0963] uplink adjustment information of a specific reference point;

[0964] uplink adjustment information corresponding to a specific path loss range or a specific reference signal receiving power (RSRP) range;

[0965] uplink adjustment information corresponding to a specific sequence or sequence format;

[0966] common uplink adjustment information;

[0967] uplink adjustment information between a specific terminal and at least one network side device;

[0968] uplink adjustment information of other terminals in a terminal group;

[0969] a set of candidate uplink adjustment information and / or a probability value associated with the set of candidate uplink adjustment information.

[0970] The terminal obtains a training sample including second input information and a label, uses the training sample to perform model training to obtain an AI model, and the AI model outputs uplink adjustment information used to adjust the uplink transmission state of the terminal, and the second input information at least includes first input information used for inference of the AI model. Through training of the AI model, the AI model can be used to predict the uplink adjustment information, and the AI model has higher accuracy in predicting the uplink adjustment information, and can meet the accuracy requirements of various scenarios.

[0971] It can be understood that the implementation processes of the implementation manners mentioned in the embodiment can refer to the related descriptions of the method embodiments one to nine, and achieve the same or corresponding technical effects. To avoid repetition, they will not be described here again.

[0972] The embodiment of the application also provides a network side device, which comprises a processor and a communication interface, the communication interface and the processor are coupled, and the processor is used to run programs or instructions to realize the steps of the method embodiments shown in FIGS. 5-12. The network side device embodiment corresponds to the network side device method embodiment described above, and each implementation process and implementation manner of the above method embodiments can be applied to the network side device embodiment, and the same technical effects can be achieved.

[0973] Specifically, the embodiment of the application also provides a network side device, which can be the communication device shown in FIG. 13 or the training device of the AI model shown in FIG. 14. As shown in FIG. 17, the network side device 1300 comprises an antenna 1301, a radio frequency device 1302, a baseband device 1303, a processor 1304 and a memory 1305. The antenna 1301 is connected with the radio frequency device 1302. In the uplink direction, the radio frequency device 1302 receives information through the antenna 1301, and sends the received information to the baseband device 1303 for processing. In the downlink direction, the baseband device 1303 processes the information to be sent and sends it to the radio frequency device 1302, and the radio frequency device 1302 processes the received information and sends it out through the antenna 1301.

[0974] The method performed by the network side device in the above embodiments can be implemented in the baseband device 1303, which includes a baseband processor.

[0975] The baseband device 1303 may, for example, include at least one baseband board on which a plurality of chips are disposed, as shown in FIG. 17, one of which is a baseband processor, for example, connected to the memory 1305 through a bus interface to invoke a program in the memory 1305 to perform the network device operations shown in the above method embodiments.

[0976] The network side device may, for example, further include a network interface 1306, which is a Common Public Radio Interface (CPRI), for example.

[0977] Specifically, the network side device 1300 of the embodiments of the present application further includes instructions or programs stored in the memory 1305 and executable on the processor 1304, and the processor 1304 invokes the instructions or programs in the memory 1305 to perform the method performed by each module shown in FIG. 13 or FIG. 14 and achieve the same technical effects. To avoid repetition, details are not described here.

[0978] The embodiments of the present application also provide a readable storage medium having programs or instructions stored thereon, which are executed by a processor to implement each process of the method embodiments shown in FIGS. 5-12 above and achieve the same technical effects. To avoid repetition, details are not described here.

[0979] The processor is the processor in the terminal described in the above embodiments. The readable storage medium includes a computer readable storage medium, such as a computer readable only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc. In some examples, the readable storage medium can be a non-transitory readable storage medium.

[0980] The embodiments of the present application further provide a chip including a processor and a communication interface, the communication interface being coupled to the processor, and the processor being configured to execute programs or instructions to implement each process of the method embodiments shown in FIGS. 5-12 above and achieve the same technical effects. To avoid repetition, details are not described here.

[0981] It should be understood that the chip mentioned in the embodiments of the present application can also be referred to as a system chip, a system on chip, a chip system or a system on chip, etc.

[0982] The embodiment of the present application further provides a computer program / product stored in a storage medium, which is executed by at least one processor to implement each process of the method embodiments shown in FIG. 5 to FIG. 12, and can achieve the same technical effects. To avoid repetition, details are not described herein.

[0983] The embodiment of the present application further provides a communication system, including a terminal and a network side device. The terminal can be used to execute the steps of the communication method described above. The network side device can be used to execute the steps of the training method of the AI model described above. Alternatively, the terminal can be used to execute the steps of the training method of the AI model described above. The network side device can be used to execute the steps of the communication method described above.

[0984] It should be noted that in this document, the terms "comprising", "containing", or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device that includes a list of elements not only includes those elements, but also includes other elements not explicitly listed, or inherent to such a process, method, article, or device. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article, or device that includes the element. In addition, it should be pointed out that the scope of the methods and devices in the embodiments of the present application is not limited to performing functions in the order shown or discussed, but can also include performing functions in a substantially simultaneous manner or in reverse order, for example, the described method can be performed in an order different from that described, and various steps can also be added, omitted or combined. In addition, the features described with reference to certain examples can be combined in other examples.

[0985] From the above description of the embodiments, those skilled in the art can clearly understand that the above-mentioned method embodiments can be realized by means of computer software product and general hardware platform, of course, it can also be realized by hardware. The computer software product is stored in a storage medium (such as ROM, RAM, magnetic disc, optical disc, etc.), including a plurality of instructions, used to make the terminal or network side device execute the method described in each embodiment of the present application.

[0986] The embodiments of the present application are described above in combination with the drawings, but the present application is not limited to the above-mentioned specific embodiments, which are only illustrative and not limiting. Those skilled in the art can make many forms of embodiments under the inspiration of the present application without departing from the scope of the present application and the protection scope of the claims.

Claims

1. A communication method, wherein, Comprising: A first device obtains first input information; The first device obtains uplink adjustment information according to an artificial intelligence AI model and the first input information, the uplink adjustment information being used to adjust an uplink transmission state of a terminal, and the uplink adjustment information including at least one of the following: Uplink adjustment information of a specific terminal; Uplink adjustment information of a group of terminals; Uplink adjustment information in a specific area; Uplink adjustment information of at least one beam direction; Uplink adjustment information of a specific reference point; Uplink adjustment information corresponding to a specific path loss range or a specific reference signal receiving power RSRP range; Uplink adjustment information corresponding to a specific sequence or sequence format; Common uplink adjustment information; Uplink adjustment information between a specific terminal and at least one network side device; Uplink adjustment information of other terminals in a terminal group; A group of candidate uplink adjustment information and / or a probability value associated with the group of candidate uplink adjustment information.

2. The method of claim 1, wherein, The first input information includes at least one of the following: Signal strength information between the terminal and the network side device of the first cell; Signal quality information between the terminal and the network side device of the first cell; Path loss information; Distance information; Absolute TA value between the terminal and the network side device of the first cell; Relative TA value between the terminal and the network side device of the first cell; Identification of a transmission reference point TRP; Identification of a TRP group; Cell identification of the first cell; Group identification of a cell group in which the first cell is located; Group identification of a timing advance group TAG in which the first cell is located; Tracking area TA identification in which the first cell is located; Radio access network notification area RNA identification in which the first cell is located; Frequency domain information in which the first cell operates; Reference signal index, beam index or beam direction transmitted between the terminal and the network side device of the first cell; Transmission delay between the terminal and the network side device of the first cell; Round trip time RTT between the terminal and the network side device of the first cell; Transmission power information of the network side device of the first cell; Position information of the network side device of the first cell; Position information or distribution information of the terminal; Moving direction and / or moving speed of the terminal, satellite or network side device of the first cell; Energy consumption status and / or power status of the terminal; Operator information and network type information supported by the terminal; Antenna orientation information of the terminal, satellite or network side device of the first cell; Energy consumption status and / or power status of the first cell; Perception information of the terminal; Scenario information of the network in which the terminal is located; Environment information in which the terminal is located; Ephemeris information; Reference position or moving trajectory information of a cell in an NTN scenario; Multipath information of a channel; Time information; One or more of the following information used currently or within a certain time window: timing advance offset, timing pre-compensation, common timing advance offset, specific timing advance offset, frequency compensation, frequency offset, common frequency offset and specific frequency offset; The first cell includes a source cell, a target cell, a current camped cell, a current accessed cell, a candidate cell, a primary cell or a secondary cell of the terminal.

3. The method of claim 1, wherein, The uplink adjustment information includes at least one of the following information: Timing advance TA related information, the TA related information comprising any of: an absolute TA value, a relative TA value, a timing advance offset, a timing pre-compensation, a common timing advance offset, a dedicated timing advance offset; Frequency related information, the frequency related information comprising at least one of: a frequency error, a frequency compensation, a frequency offset, a common frequency offset, a dedicated frequency offset.

4. The method of any one of claims 1-3, wherein, The first device is a terminal, and the first device obtains first input information, comprising: The terminal receives at least part of the information of the first input information sent by the network side device.

5. The method of any one of claims 1-3, wherein, The first device is a terminal, and after the first device obtains uplink adjustment information according to the AI model and the first input information, the method further comprises: According to the uplink adjustment information, at least one of the following operations is performed: Reporting first information to a training device, the first information being used for updating the AI model, the first information comprising at least one of the following information: input information of this AI inference, output information of this AI inference, current state information of the terminal; Falling back to a determination process of uplink adjustment information in a non-AI manner; Triggering updating of the AI model; Triggering change of input information of the AI model; Triggering supervision of the AI model.

6. The method of any one of claims 1-3, wherein, The first device is a network side device, and the first device obtains first input information, comprising: The network side device receives at least part of the information of the first input information sent by the terminal.

7. The method of any one of claims 1-3, wherein, The first device is a server, and the first device obtains first input information, comprising: The server receives at least part of the information of the first input information sent by the terminal and / or the network side device.

8. The method of any one of claims 1-7, wherein, The method further comprises: The first device obtains first configuration information of the AI model; The first configuration information comprises at least one of the following: An identifier of the AI model used for AI inference; An application range of AI model inference; A period of AI model inference; An effective duration of AI model inference; A trigger condition of AI model inference; Configuration parameters of the AI model, the configuration parameters of the AI model comprising types of first input information and / or types of output information of the AI model; Whether to support joint inference of multiple devices.

9. The method of claim 8, wherein, The method further comprises: The first device obtains activation indication information or deactivation indication information; The first device determines to use the AI model for AI inference according to the activation indication information and the trigger condition of the AI model inference; Or, the first device determines not to use the AI model for AI inference according to the deactivation indication information.

10. The method of claim 8, wherein, The method further comprises: The first device determines to use the AI model for inference according to the trigger condition of the AI model inference.

11. The method of any one of claims 8-10, wherein, The trigger condition of the AI model inference comprises at least one of the following: An AI related timer timeout; A RACH-less technology is enabled; A terminal is configured to use a short sequence for transmission of a PRACH; A network side device configures PRACH resources dedicated to a traditional RACH as a fallback mechanism, and the network side device configures an AI based RACH mechanism; The accuracy of the uplink adjustment information is less than or equal to a target threshold or less than or equal to the target threshold for more than a certain time length; The accuracy of the uplink adjustment information calculated using a non-AI method is less than a certain threshold; The number of failures or the number of consecutive failures of signal transmission using the uplink adjustment information calculated using a non-AI method exceeds a target threshold; Default use in a specific scenario; Cell or TRP reselection; Cell or TRP switching; Satisfying the conditions for cell switching; Terminal triggering access to a secondary cell; Terminal moving to a specific location; Arrival of uplink or downlink data; Arrival of a specific type of service; Receiving at least one piece of information in the first input information; Terminal speed or acceleration exceeding a set threshold; Terminal or network location changes or changes exceeding a set threshold; Change in the beam or spatial transmission filter selected by the terminal; Change in the first parameter of the terminal relative to a certain time point or within a certain time window, the first parameter including at least one of the following parameters: timing advance offset, timing pre-compensation, common timing advance offset, dedicated timing advance offset, frequency compensation, frequency offset, common frequency offset, and dedicated frequency offset.

12. The method of any one of claims 1-11, wherein, The method further includes: The first device obtains supervision configuration information of the AI model; The first device supervises the AI model according to the supervision configuration information.

13. The method of claim 12, wherein, The supervision configuration information includes at least one of the following: Identification of the AI model that needs to be supervised; Period of model supervision; Number of model supervision; Duration of model supervision; Window-related information of model supervision; Triggering condition of model supervision; Index of model supervision; Label of model supervision.

14. The method of claim 13, wherein, The value of the label of the model supervision is the same as or different from the value of the label in the AI model training process.

15. The method of claim 13, wherein, The triggering condition of the model supervision includes at least one of the following: The inference result of the AI model does not meet the accuracy requirement; At least one of the inference indicators of the AI model does not meet the requirement; The index of the model supervision does not meet the requirement, including the error between the predicted value and the true value of the AI model, and / or the performance indicator of the network; The supervision timer of the AI model is timed out; The terminal switches to a new cell, TRP, or beam; After the AI model inference fails; The AI model has failed N times in a row; The number of AI model inference failures reaches a threshold value; Using the AI model for inference; The terminal moves to a new cell, tracking area, or geographic location; The moving speed of the terminal changes significantly within a specified time period; The external environment of the inference device of the AI model has changed.

16. The method of claim 15, wherein, The inference indicator of the AI model includes at least one of the following: Complexity of the AI model; Latency of the AI model inference; Success rate of the AI model inference; Reliability of the AI model inference result.

17. The method of any one of claims 1-16, wherein, The method further includes: The first device sends indication information of the AI capability information of the first device to a third device; The AI capability information of the first device includes at least one of the following: The first device has or does not have the capability to use the AI model for AI inference; The first device has or does not have the capability of sending first auxiliary information used for the AI inference; The first device has or does not have the capability of model supervision on the AI model.

18. The method of claim 17, wherein, The third device determines the AI capability information of the first device based on at least one of the following: device type, network type, one or more reference signals, and indication information of the AI capability information of the first device.

19. The method of claim 17, wherein, The indication information of the AI capability of the first device includes at least one of the following: First capability indication information, used for indicating that the first device has or does not have the capability of using the AI model for AI inference; Second capability indication information, used for indicating that the first device has or does not have the capability of sending the first auxiliary information; Third capability indication information, used for indicating that the first device has or does not have the capability of model supervision on the AI model.

20. The method of any one of claims 1-19, wherein, The first input information is the same as second input information used when the AI model is trained, or the first input information is a subset of the second input information.

21. A method of training an AI model, wherein, Comprise: The second device obtains training samples, and the training samples include second input information and at least one label; The second device performs model training on the training samples to obtain the AI model of any one of claims 1-20, and the AI model outputs uplink adjustment information used for adjusting the uplink transmission state of a terminal, wherein the second input information at least includes first input information used for AI model inference, and the uplink adjustment information includes at least one of the following: Uplink adjustment information of a specific terminal; Uplink adjustment information of a group of terminals; Uplink adjustment information in a specific area; Uplink adjustment information of at least one beam direction; Uplink adjustment information of a specific reference point; Uplink adjustment information corresponding to a specific path loss range or a specific reference signal receiving power (RSRP) range; Uplink adjustment information corresponding to a specific sequence or sequence format; Common uplink adjustment information; Uplink adjustment information between a specific terminal and at least one network side device; Uplink adjustment information of other terminals in a terminal group; A group of candidate uplink adjustment information and / or a probability value associated with the group of candidate uplink adjustment information.

22. The method of claim 21, wherein, The label of the training sample includes at least one of the following: Whether the obtained uplink adjustment information meets the accuracy requirement; The number of obtained uplink adjustment information; Uplink adjustment information between the terminal and the network side device of the first cell; Uplink adjustment information between the network side device of the first cell and the network side device of the second cell; Propagation delay of uplink or downlink; Delay difference between the uplink propagation delay between the terminal and the target cell and the uplink propagation delay between the terminal and the source cell; Delay difference between the downlink propagation delay between the terminal and the target cell and the downlink propagation delay between the terminal and the source cell; Delay difference between the uplink propagation delay between the terminal and the primary cell and the uplink propagation delay between the terminal and the secondary cell; Delay difference between the downlink propagation delay between the terminal and the primary cell and the downlink propagation delay between the terminal and the secondary cell; Actual value of the uplink adjustment information output by the AI model.

23. The method of claim 21 or 22, wherein, The second device jointly trains with the first device to obtain the AI model.

24. The method of claim 23, wherein, When the second device is a terminal or a server, the method further comprises: The second device sends second information to a network side device, the second information being used to assist the network side device to select an AI model from a plurality of candidate AI models; The second device receives third information sent by the network side device, the third information being used to indicate information of the AI model selected by the network side device.

25. The method of claim 24, wherein, At least one of the following information of the plurality of candidate AI models is different: training data set, label information, input information and output information.

26. The method of claim 25, wherein, The second information comprises IDs of the plurality of candidate AI models and IDs of training data sets used for training of each candidate AI model; The third information comprises at least one of the following: ID of the AI model selected by the network side device, ID of the training data set, and collection configuration information of the input information.

27. The method of any one of claims 21-26, wherein, The trigger type of the AI model training comprises at least one of the following: conditional trigger, periodic trigger or semi-static trigger.

28. The method of claim 27, wherein, The trigger condition of the AI model training comprises at least one of the following: AI related timer timeout; Timing advance group (TAG) related timer timeout; The terminal receives a change in TA value; N TA the value of the parameter has changed; N TA_offset the value of the parameter has changed; The terminal accesses a new cell, switches frequency or frequency band, switches operator or public land mobile network (PLMN); AI model inference fails; AI model continuous inference fails N times; The number of AI model inference failures reaches a threshold value; AI model inference is used; The terminal moves to a new cell or tracking area or geographic location; The terminal's moving speed changes significantly within a specified time period; The terminal's reference signal received power (RSRP) measurement value changes or the change exceeds a certain threshold value; Model training related timer timeout for a period of time; Continuous N times of model supervision occurs or N times of model supervision occurs; The terminal's external environment changes; The terminal's speed or acceleration exceeds a specified threshold value; The terminal or network location changes or changes exceed a specified threshold value; The terminal's selected beam or spatial transmission filter changes; The terminal's first parameter changes relative to a certain time point or within a certain time window, the first parameter comprising at least one of the following parameters: timing advance offset, timing pre-compensation, common timing advance offset, dedicated timing advance offset, frequency compensation, frequency offset, common frequency offset, dedicated frequency offset; RACH-less technology is enabled; The terminal is configured to use a short sequence for PRACH transmission; The network side device configures PRACH resources dedicated to traditional RACH as a fallback mechanism, and the network side device configures an AI-based RACH mechanism; The accuracy of uplink adjustment information is less than or equal to a target threshold or less than or equal to a target threshold for a certain period of time; The accuracy of uplink adjustment information calculated using a non-AI method is less than a certain threshold value; The number of transmission failures or consecutive failures of uplink adjustment information calculated using a non-AI method exceeds a target threshold; Default use in a specific scenario; Cell or TRP reselection; Cell or TRP switching; Conditions for cell switching are met; Terminal triggers access to a secondary cell; Terminal moves to a specific location; Arrival of uplink or downlink data; Arrival of a specific type of service; Receiving at least one of the second input information.

29. The method of claim 28, wherein, The method further comprises: The second device receives training instruction information sent by the network side device; When the trigger condition of the AI model training is met, the second device determines to train the AI model according to the training instruction information and the AI capability information of the first device; Or, when the trigger condition of the AI model training is met, the second device determines to train the AI model.

30. The method of any one of claims 21-29, wherein, The training end condition of the AI model comprises at least one of the following: The loss function of the AI model meets a predefined value; The number of training reaches a predefined number; The number of iterations of fine-tuning reaches a predefined number; At least one information in the label for model training is greater than or equal to or less than a threshold value; The duration of model training is greater than or equal to a predefined duration; The energy consumption of model training is greater than or equal to a predefined value.

31. The method of any one of claims 21-30, wherein, The method further comprises: The second device sends the indication information of the AI capability information of the second device to a third device; The AI capability information of the second device comprises at least one of the following: The second device has or does not have the ability to train the AI model; The second device has or does not have the ability to send third auxiliary information for training of the AI model.

32. The method of claim 31, wherein, The third device is configured to determine the AI capability information of the second device by at least one of the following: device type, network type, one or more reference signals, and indication information of the AI capability information of the second device.

33. The method of claim 31, wherein, The indication information of the AI capability of the second device comprises at least one of the following: Fourth capability indication information, used to indicate that the second device has or does not have the ability to train the AI model; Fifth capability indication information, used to indicate that the second device has or does not have the ability to send the third auxiliary information.

34. A communications device, wherein, Comprise: The processing module is configured to obtain first input information; The processing module is further configured to obtain uplink adjustment information according to an artificial intelligence (AI) model and the first input information, the uplink adjustment information being used to adjust an uplink transmission state of a terminal, and the uplink adjustment information comprising at least one of the following: Uplink adjustment information of a specific terminal; Uplink adjustment information of a group of terminals; Uplink adjustment information in a specific area; Uplink adjustment information of at least one beam direction; Uplink adjustment information of a specific reference point; Uplink adjustment information corresponding to a specific path loss range or a specific reference signal receiving power (RSRP) range; Uplink adjustment information corresponding to a specific sequence or sequence format; Common uplink adjustment information; Uplink adjustment information between a specific terminal and at least one network side device; Uplink adjustment information of other terminals in a terminal group; A group of candidate uplink adjustment information and / or a probability value associated with the group of candidate uplink adjustment information.

35. The apparatus of claim 34, wherein, The first input information comprises at least one of the following: Signal strength information between a terminal and a network side device of a first cell; signal quality information between the terminal and a network-side device of the first cell; path loss information; distance information; absolute TA value between the terminal and the network-side device of the first cell; relative TA value between the terminal and the network-side device of the first cell; identification of a transmission reference point (TRP); identification of a TRP group; cell identification of the first cell; group identification of a cell group in which the first cell is located; group identification of a timing advance group (TAG) in which the first cell is located; tracking area (TA) identification in which the first cell is located; radio access network notification area (RNA) identification in which the first cell is located; frequency domain information in which the first cell operates; reference signal index, beam index or beam direction transmitted between the terminal and the network-side device of the first cell; transmission delay between the terminal and the network-side device of the first cell; round trip time (RTT) between the terminal and the network-side device of the first cell; transmission power information of the network-side device of the first cell; location information of the network-side device of the first cell; location information or distribution information of the terminal; moving direction and / or moving speed of the terminal, satellite or network-side device of the first cell; energy consumption status and / or power status of the terminal; operator information and network type information supported by the terminal; antenna orientation information of the terminal, satellite or network-side device of the first cell; energy consumption status and / or power status of the first cell; perception information of the terminal; scenario information of a network in which the terminal is located; environment information in which the terminal is located; ephemeris information; reference position or moving trajectory information of a cell in an NTN scenario; multipath information of a channel; time information; one or more of the following information used currently or within a certain time window: timing advance offset, timing pre-compensation, common timing advance offset, dedicated timing advance offset, frequency compensation, frequency offset, common frequency offset and dedicated frequency offset. The first cell includes one of the following: a source cell, a target cell, a current camping cell, a current access cell, a candidate cell, a primary cell or a secondary cell of the terminal.

36. The apparatus of claim 34, wherein, The uplink adjustment information includes at least one of the following information: timing advance (TA) related information, the TA related information including any one of the following: absolute TA value, relative TA value, timing advance offset, timing pre-compensation, common timing advance offset, dedicated timing advance offset; frequency related information, the frequency related information including at least one of the following: frequency error, frequency compensation, frequency offset, common frequency offset and dedicated frequency offset.

37. The apparatus of any one of claims 34-36, wherein, The processing module is further configured to: obtain first configuration information of the AI model; The first configuration information includes at least one of the following: identification of the AI model used for AI inference; application range of AI model inference; period of AI model inference; validity duration of AI model inference; triggering condition of AI model inference; configuration parameters of the AI model, the configuration parameters of the AI model including types of first input information and / or types of output information of the AI model; whether to support joint inference of multiple devices.

38. The apparatus of claim 37, wherein, The processing module is further configured to: obtain activation indication information or deactivation indication information; According to the activation indication information and the trigger condition of the AI model inference, it is determined to use the AI model for AI inference. Alternatively, according to the deactivation indication information, it is determined not to use the AI model for AI inference.

39. The device of claim 37, wherein, The processing module is further configured to: According to the trigger condition of the AI model inference, it is determined to use the AI model for inference.

40. The apparatus of any one of claims 34-39, wherein, The processing module is further configured to: Obtain supervision configuration information of the AI model; Supervise the AI model according to the supervision configuration information.

41. The apparatus of claim 40, wherein, The supervision configuration information includes at least one of: An identifier of an AI model that needs to be supervised; A period of model supervision; A number of model supervision; A time length of model supervision; Window-related information of model supervision; A trigger condition of model supervision; An index of model supervision; A label of model supervision.

42. An apparatus for training an AI model, wherein, Comprise: A processing module configured to obtain a training sample, the training sample comprising second input information and at least one label; The processing module is further configured to: use the training sample to perform model training to obtain the AI model of any one of claims 1-20, the AI model outputting uplink adjustment information, the uplink adjustment information being used to adjust the uplink transmission state of a terminal, wherein the second input information at least includes first input information used by the AI model for inference, and the uplink adjustment information includes at least one of: Uplink adjustment information of a specific terminal; Uplink adjustment information of a group of terminals; Uplink adjustment information in a specific area; Uplink adjustment information of at least one beam direction; Uplink adjustment information of a specific reference point; Uplink adjustment information corresponding to a specific path loss range or a specific reference signal received power (RSRP) range; Uplink adjustment information corresponding to a specific sequence or sequence format; Common uplink adjustment information; Uplink adjustment information between a specific terminal and at least one network side device; Uplink adjustment information of other terminals in a terminal group; A group of candidate uplink adjustment information and / or a probability value associated with the group of candidate uplink adjustment information.

43. The device of claim 42, wherein, The label of the training sample includes at least one of: Whether the uplink adjustment information meets the accuracy requirement; The number of obtained uplink adjustment information; Uplink adjustment information between the terminal and the network side device of the first cell; Uplink adjustment information between the network side device of the first cell and the network side device of the second cell; Propagation delay of uplink or downlink; Delay difference between uplink propagation delay between the terminal and the target cell and uplink propagation delay between the terminal and the source cell; Delay difference between downlink propagation delay between the terminal and the target cell and downlink propagation delay between the terminal and the source cell; Delay difference between uplink propagation delay between the terminal and the primary cell and uplink propagation delay between the terminal and the secondary cell; Delay difference between downlink propagation delay between the terminal and the primary cell and downlink propagation delay between the terminal and the secondary cell; True value of the uplink adjustment information output by the AI model.

44. A communications device, comprising: A processor and a memory, the memory stores programs or instructions executable on the processor, and the programs or instructions are executed by the processor to implement the steps of the communication method according to any one of claims 1 to 20.

45. A communications device, comprising: comprising a processor and a memory storing a program or instructions executable on the processor, the program or instructions, when executed by the processor, implement the steps of the training method of the AI model according to any one of claims 21 to 33.

46. A readable storage medium, wherein, The program or instructions are stored on the readable storage medium, and when executed by the processor, implement the communication method according to any one of claims 1 to 20, or implement the steps of the training method of the AI model according to any one of claims 21 to 33.

Images