AI model parameter acquisition method and device, storage medium and chip system

Abstract

The invention discloses an AI model parameter acquisition method and device, a storage medium and a chip system, and belongs to the technical field of communication. The method comprises the following steps: receiving n symbol sequences sent by network equipment; wherein when a symbol sequence is received, the symbol sequence is demodulated to obtain a first soft information sequence; generating a second soft information sequence with a preset length according to the first soft information sequence; and decoding the second soft information sequence according to side information related to an update parameter of a target NN layer in the target AI model to obtain a target coding block, the target coding block comprising the update parameter of the target NN layer, and the side information being determined according to an NN original parameter of the target AI model. According to the method, the side information related to the update parameter of the target NN layer is used as additional priori knowledge for decoding input, so that the decoding performance can be greatly improved.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a method, device, storage medium, and chip system for obtaining AI model parameters. Background Art

[0002] In wireless communication systems, artificial intelligence (AI) models can be used to enhance system performance. These AI models are typically trained in specific environments. However, as the actual usage environment changes, the deployed AI models may not adapt to the new environment, thus affecting system performance. To address this issue, network equipment can fine-tune the AI models to update their parameters and transmit the updated parameters to terminal devices. However, AI models typically include a large number of high-precision neural network (NN) parameters, which are easily affected by channel noise during wireless transmission, posing a significant challenge to decoding by terminal devices. Summary of the Invention

[0003] This application provides an AI model parameter acquisition method, device, storage medium, and chip system that can improve decoding performance. The technical solution is as follows: In the first aspect, a method for obtaining AI model parameters is provided. This method can be executed, for example, by a terminal device, or by a component configured in the terminal device (such as a circuit, chip, or chip system), or by a logic module or software that implements all or part of the terminal device's functions, although this application does not limit this. The following description uses a terminal device as an example.

[0004] The method includes: a terminal device receives n symbol sequences sent by a network device, where n is a positive integer. Each time the terminal device receives a symbol sequence, it demodulates the symbol sequence to obtain a first soft information sequence, and then generates a second soft information sequence of a preset length based on the first soft information sequence. Subsequently, the second soft information sequence is decoded based on side information related to the updated parameters of the target NN layer in the target AI model to obtain a target coding block. The target coding block includes the updated parameters of the target NN layer, and the side information is determined based on the original NN parameters of the target AI model.

[0005] Among them, side information is the core element of the Slepian-Wolf theory. It refers to the auxiliary information related to the source to be decoded that is already available at the decoding end. The decoding end can achieve efficient reconstruction by combining side information with the received data.

[0006] In this application, the terminal device can determine the side information related to the updated parameters of the target NN layer based on the original NN parameters of the target AI model. The terminal device uses the side information related to the updated parameters of the target NN layer as additional prior knowledge and performs joint decoding input with the soft information sequence obtained by demodulation, thereby greatly improving the decoding performance. In this way, even under poor channel conditions, decoding can be completed more accurately with a high decoding success rate, which helps to achieve robust transmission and accurate recovery of NN parameters.

[0007] In one possible implementation, before receiving the n symbol sequences sent by the network device, the terminal device may also receive control information sent by the network device. The control information includes a target AI model identifier, a target NN layer identifier, and a total number of blocks, where the total number of blocks is n. In this way, the terminal device can determine which NN layer parameters in which AI model need to be updated based on the control information.

[0008] In one possible implementation, after the terminal device receives the control information sent by the network device, it can also obtain the original NN parameters of the target AI model; input the binary quantization sequence of the original NN parameters of the target AI model and the parameters of the hidden Markov model into the FBA module to obtain the output data of the FBA module; based on the output data of the FBA module, estimate the soft information sequence of the binary quantization sequence of the NN update parameters to obtain the target soft information sequence, which includes side information related to the update parameters of the target NN layer.

[0009] The hidden Markov model is used to represent the correlation between the binary quantization sequence of the original NN parameters of the target AI model and the binary quantization sequence of the updated NN parameters of the target AI model.

[0010] The binary quantized sequence of the original NN parameters of the target AI model is an observable past sequence. The goal of the FBA module is to estimate the binary quantized sequence of the target AI model's NN update parameters using the hidden Markov model and the known observable sequence. The FBA module continuously corrects the parameters of the hidden Markov model during the iteration process. When the observable sequence samples are long enough, the impact of the error in the initial input state can be ignored. Based on this, the FBA module can output after reaching the preset number of iterations. Based on the FBA output data, the soft information sequence of the binary quantized sequence of the NN update parameters can be estimated more accurately.

[0011] In one possible implementation, before the terminal device receives the n symbol sequences sent by the network device, it can also obtain the soft information sequence corresponding to the target NN layer from the target soft information sequence, where the target soft information sequence is determined based on the original NN parameters of the target AI model; the soft information sequence corresponding to the target NN layer is divided into n soft information sequences, where the n soft information sequences are the side information. Accordingly, the terminal device decodes the second soft information sequence based on the side information related to the updated parameters of the target NN layer in the target AI model, and the operation of obtaining the target coding block can be: generating a fourth soft information sequence based on the second soft information sequence and the third soft information sequence, where the third soft information sequence is a soft information sequence among the n soft information sequences corresponding to the second soft information sequence; and inputting the fourth soft information sequence into the channel decoding module to obtain the target coding block output by the channel decoding module.

[0012] In the present application, joint decoding input can be performed based on the second soft information sequence obtained by demodulation and the third soft information sequence obtained by estimation, which can enhance decoding performance.

[0013] In one possible implementation, the operation of the terminal device generating the fourth soft information sequence based on the second soft information sequence and the third soft information sequence can be: adding the soft information of the first Q bits in the second soft information sequence to the soft information of the corresponding bits in the third soft information sequence one by one to obtain the soft information of the first Q bits in the fourth soft information sequence, where the first Q bits are punctured bits; and using the soft information of the remaining bits in the second soft information sequence except the soft information of the first Q bits as the soft information of the remaining bits in the fourth soft information sequence except the first Q bits.

[0014] In related art, the punctured bits are assumed to be unknown information, i.e., their decoding input is set to 0. This input setting loses some coded block information, resulting in a certain degradation in decoding performance. In this application, the decoding input of the punctured bits can be set based on the estimated third soft information sequence, thereby improving decoding performance.

[0015] In one possible implementation, the terminal device may generate a fourth soft information sequence based on the second soft information sequence and the third soft information sequence by: using the soft information of the first Q bits in the second soft information sequence as the soft information of the first Q bits in the fourth soft information sequence, where the first Q bits are punctured bits; adding the soft information of the Q+1th to Kth bits in the second soft information sequence to the soft information of corresponding bits in the third soft information sequence one by one to obtain the soft information of the Q+1th to Kth bits in the fourth soft information sequence, where the Q+1th to Kth bits are information bits; and using the soft information of the remaining bits in the second soft information sequence except the soft information of the first K bits as the soft information of the remaining bits in the fourth soft information sequence except the first K bits.

[0016] In the present application, for the Q+1th to Kth information bits, joint decoding input can be performed based on the second soft information sequence obtained by demodulation and the third soft information sequence obtained by estimation, thereby enhancing decoding performance.

[0017] In one possible implementation, the terminal device may generate a fourth soft information sequence based on the second soft information sequence and the third soft information sequence by: adding the soft information of the first Q bits in the second soft information sequence to the soft information of the corresponding bits in the third soft information sequence one by one to obtain the soft information of the first Q bits in the fourth soft information sequence, where the first Q bits are punctured bits; adding the soft information of the Q+1th to Kth bits in the second soft information sequence to the soft information of the corresponding bits in the third soft information sequence one by one to obtain the soft information of the Q+1th to Kth bits in the fourth soft information sequence, where the Q+1th to Kth bits are information bits; and using the soft information of the remaining bits in the second soft information sequence except the soft information of the first K bits as the soft information of the remaining bits in the fourth soft information sequence except the first K bits.

[0018] In the present application, not only can the decoding input of the punctured bits be set based on the estimated third soft information sequence, but also the joint decoding input of the Q+1th to Kth information bits can be performed based on the second soft information sequence obtained by demodulation and the estimated third soft information sequence, thereby significantly enhancing the decoding performance.

[0019] In a second aspect, a communication device is provided, which includes a processing module and a communication module. The communication module is used to receive n symbol sequences sent by a network device, where n is a positive integer. The processing module is used to demodulate a symbol sequence each time it is received to obtain a first soft information sequence; generate a second soft information sequence of a preset length based on the first soft information sequence; decode the second soft information sequence based on side information related to the update parameters of the target NN layer in the target AI model to obtain a target coding block, where the target coding block includes the update parameters of the target NN layer, and the side information is determined based on the original NN parameters of the target AI model.

[0020] The second aspect is the implementation on the device side corresponding to the first aspect. The explanation, supplement and description of the beneficial effects of the first aspect are also applicable to the second aspect and will not be repeated here.

[0021] In a third aspect, a communication device is provided, comprising a processor. The processor is coupled to a memory and can be configured to execute instructions or data in the memory to implement the method of any possible implementation of any of the above aspects. Optionally, the communication device further comprises a memory. Optionally, the communication device further comprises a communication interface, the processor being coupled to the communication interface.

[0022] In one implementation, the communication interface may be a transceiver, or an input / output interface.

[0023] In another implementation, the communication device is a chip configured in a terminal device. When the communication device is a chip configured in a terminal device, the communication interface may be an input / output interface.

[0024] In a fourth aspect, a computer program product is provided, which includes: a computer program (also referred to as code, or instructions), which, when executed, enables a computer to execute a method in any possible implementation of any of the above aspects.

[0025] In a fifth aspect, a computer-readable storage medium is provided, which stores a computer program (also referred to as code, or instructions), which, when executed on a computer, enables the computer to execute a method in any possible implementation of any of the above aspects.

[0026] In a sixth aspect, embodiments of the present application provide a chip system comprising one or more processors configured to retrieve and execute instructions stored in a memory, thereby executing the method of any possible implementation of any of the above aspects. The chip system may be composed of a chip or may include a chip and other discrete devices.

[0027] Among them, the chip system may include an input circuit or interface for sending information or data, and an output circuit or interface for receiving information or data.

[0028] In a seventh aspect, a communication system is provided, comprising the aforementioned terminal device and network device. Optionally, the communication system may further comprise other devices that communicate with the terminal device and / or the network device. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a schematic diagram of a communication system provided in an embodiment of the present application.

[0030] Figure 2 It is a schematic diagram of another communication system provided in an embodiment of the present application.

[0031] Figure 3 This is a flowchart of a method for obtaining AI model parameters provided in an embodiment of the present application.

[0032] Figure 4 This is a schematic diagram of a hidden Markov model provided in an embodiment of the present application.

[0033] Figure 5 This is a schematic diagram of data processing on a network device side provided in an embodiment of the present application.

[0034] Figure 6This is a schematic diagram of data processing on a terminal device side provided in an embodiment of the present application.

[0035] Figure 7 This is a schematic block diagram of a communication device provided in an embodiment of the present application.

[0036] Figure 8 This is a schematic block diagram of a communication device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0037] In the following description, specific details such as specific system structures and technologies are provided for illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details.

[0038] It should be understood that when used in the specification and appended claims of this application, the term "comprising" indicates the presence of the described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or their collections. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized.

[0039] It should be understood that the "one or more" mentioned in this application refers to one, two or more, and the "multiple" mentioned in this application refers to two or more. In the description of this application, unless otherwise specified, " / " means or, for example, A / B can mean A or B. The "and / or" in this article is merely a description of the association relationship of associated objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone.

[0040] To facilitate the clear description of the technical solutions of this application, the words "first" and "second" are used to distinguish between identical or similar items with substantially the same functions and effects. Those skilled in the art will understand that the words "first" and "second" do not limit the quantity or order of execution, and the words "first" and "second" do not necessarily mean different.

[0041] The phrases "one embodiment" or "some embodiments" described in this application mean that the specific features, structures, or characteristics described in the embodiment are included in one or more embodiments of the application. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," and "in other embodiments" that appear in different places in this application do not necessarily refer to the same embodiment, but rather mean "one or more but not all embodiments," unless otherwise specifically emphasized.

[0042] The embodiments of the present application can be applied to various communication systems, for example, a global system for mobile communications (GSM) system, a general packet radio service (GPRS) system, a wireless local area network (WLAN) system (such as a wireless fidelity (Wi-Fi) system), a long term evolution (LTE) system, an LTE frequency division duplex (FDD) system, an LTE time division duplex (TDD) system, a sidelink communication system, a universal mobile telecommunication system (UMTS), a world-wide interoperability for microwave access (WiMAX) communication system, a non-terrestrial network (NTN) communication system, a fourth generation (4G) mobile communication system, a fifth generation (5G) mobile communication system or a new radio access technology (NR) system, a sixth generation (6G) mobile communication system, etc. The 5G mobile communication system may include non-standalone (NSA) and / or standalone (SA) networking. It is understood that the embodiments of the present application may also be applied to future communication systems, and the embodiments of the present application are not limited thereto.

[0043] Figure 11 is a schematic diagram of a communication system 100 provided in an embodiment of the present application. The communication system 100 may include network devices such as Figure 1 The communication system 100 may also include terminal devices, such as Figure 1 The terminal device 120 is shown. The network device 110 and the terminal device 120 can communicate via a wireless link. Figure 1 The example shows one network device 110 and one terminal device 120. Optionally, the communication system 100 may also include multiple network devices and / or multiple terminal devices.

[0044] The network device in the embodiment of the present application may be a network-side device such as an access network device and a core network device.

[0045] Access network equipment is sometimes also referred to as an access node. Access network equipment has wireless transceiver capabilities and is used to communicate with terminal devices. For example, an access network device may be a base station, an evolved NodeB (eNodeB), a transmission reception point (TRP), a next-generation NodeB (gNB) in a 5G mobile communication system, an access network device or module in an open access network (ORAN) system, a satellite in an NTN communication system, a base station in a future mobile communication system, or an access point (AP) in a Wi-Fi system. An access network device may also be a module or unit that implements some of the functions of a base station, such as a macro base station, a micro base station, an indoor station, a relay node, or a donor node. The multiple access network devices in communication system 100 may be base stations of the same type or different types. The embodiments of this application do not limit the specific technology or device form factor employed by the access network devices.

[0046] Core network equipment performs functions such as data processing, session management, network interconnection, and operation administration and maintenance (OAM). Core network equipment can implement user access authentication, service bearer establishment, and data exchange with external networks. Through OAM functions, it can also complete operation and maintenance tasks such as network configuration monitoring, resource scheduling optimization, and fault repair. For example, core network equipment can be a mobility management entity (MME), serving gateway (SGW), or packet data network gateway (PGW) in a 4G mobile communication system, or an access and mobility management function (AMF), session management function (SMF), or user plane function (UPF) in a 5G mobile communication system. For example, core network equipment can also be an operation and maintenance management system, such as a network management subsystem (NMS) with integrated OAM functions, an automated operation and maintenance module, or an independent operation and maintenance server that works in conjunction with the core network. For example, a core network device may be a virtualized network element in a network function virtualization (NFV) architecture, such as a cloud-deployed network function module. Optionally, lightweight OAM capabilities may be integrated into the virtualized network element. A core network device may also be a new type of core network entity in a future communication system. Multiple core network devices in communication system 100 may be deployed in a centralized or distributed architecture, and each core network device may perform the same or different types of network functions. The embodiments of this application do not limit the specific technologies and device forms used by the core network devices.

[0047] In the embodiments of the present application, the apparatus for implementing the functions of a network device may be a network device, or may be a device capable of supporting the network device in implementing the functions, such as a processor, circuit, chip, or chip system. The apparatus may be installed in the network device or connected to the network device for use. In the embodiments of the present application, the technical solution provided by the present application is described using the network device as an example.

[0048] The terminal device in the embodiments of the present application may be a wireless terminal device capable of receiving network device scheduling and instruction information. A wireless terminal device may be a device that provides voice and / or data connectivity to a user, a handheld device with wireless connectivity, or other processing device connected to a wireless modem. For example, the terminal device may communicate with one or more core networks or the Internet via a radio access network (RAN). The terminal device may also be referred to as a terminal, user equipment (UE), mobile station, or mobile terminal. The terminal device can be widely used in various scenarios, such as device-to-device (D2D), vehicle-to-everything (V2X) communication, machine-type communication (MTC), the Internet of Things (IoT), ultra-reliable low-latency communication (URLLC), virtual reality, augmented reality, industrial control, autonomous driving, telemedicine, smart grid, smart furniture, smart office, smart wearables, smart transportation, smart city, or satellite communication. The terminal device may be a mobile phone, tablet computer, computer with wireless transceiver function, wearable device, vehicle, aircraft (such as drone, helicopter, airplane), hot air balloon, ship, robot, robotic arm, or smart home device, etc. The embodiment of the present application does not limit the form of the terminal device.

[0049] In the embodiments of the present application, the apparatus for implementing the functions of a terminal device may be a terminal device, or may be a device capable of supporting the terminal device in implementing the functions, such as a processor, circuit, chip, or chip system. The apparatus may be installed in the terminal device or connected to the terminal device for use. In the embodiments of the present application, the technical solutions provided by the present application are described using the terminal device as an example.

[0050] The application scenarios involved in the embodiments of the present application are described below.

[0051] In wireless communication systems, AI models can be used to enhance system performance. For example, in 6G mobile communication systems, in application scenarios where communication and AI are deeply integrated, such as AI air interface or AI-enhanced mobility management, terminal devices will deploy a variety of specialized AI models to perform key inference tasks such as channel state information (CSI) prediction / compression, beamforming optimization, and resource scheduling and allocation. To ensure that these AI models achieve optimal system performance in specific wireless environments (such as specific channel models, interference conditions, and user distribution), the model training process requires pre-configuration of a series of key parameters and objectives, including: setting training channel environment parameters (such as a specific signal-to-noise ratio range), selecting an appropriate optimization objective (such as a specific loss function), and constructing a sufficiently representative training dataset (covering the typical characteristics and dynamic changes of the target scenario).

[0052] However, when the actual operating environment in which the AI model is deployed changes significantly, the AI model may not be able to adapt to the changed environment, resulting in a sharp degradation of model performance. For example, the CSI prediction model is trained under a set signal-to-noise ratio range, but is deployed in an environment with a very different actual signal-to-noise ratio, or the input data distribution processed during model inference is significantly offset from the data distribution learned during its training phase. For such situations where the inference environment and the training environment do not match, the current common solution is to fine-tune the model, that is, based on the original model parameters, a small number of samples are collected in the new environment for incremental training. However, due to the inherent resource constraints of the terminal device, such as insufficient computing power, limited storage space, and limited energy supply, the terminal device cannot independently complete local fine-tuning of large-scale model parameters. For this reason, Figure 2 As shown, the cloud server or edge server can fine-tune the model and then transmit the fine-tuned complete model or model increments to the base station. The base station can send the fine-tuned model parameters to the terminal device via a wireless downlink.

[0053] AI models typically include a large number of high-precision NN parameters, which are easily affected by channel noise during wireless transmission. If the NN parameters received by the terminal device are inaccurate, the model performance will be severely degraded. However, under low signal-to-noise ratio channel conditions, it is particularly difficult to ensure high-fidelity recovery of NN parameters. If the traditional automatic repeat request (ARQ) mechanism is relied upon to ensure the reliability of NN parameter transmission, huge signaling overhead and wireless resource consumption will be generated due to frequent retransmissions. This not only easily causes wireless link congestion, but also significantly extends the end-to-end delay of model updates, making it impossible for terminal devices to obtain updated AI models in a timely manner, ultimately hindering the effective adaptation of AI models in dynamic environments.

[0054] To this end, an embodiment of the present application provides an AI model parameter acquisition method, which provides a NN parameter decoding algorithm with side information. Specifically, the terminal device can extract side information for the fine-tuned NN parameters based on the correlation between the NN parameters before and after fine-tuning the AI model, and then perform decoding based on the side information to enhance decoding performance. In this way, the accuracy and robustness of NN parameter recovery under harsh wireless channel conditions such as low signal-to-noise ratio, high interference, and fast time variation can be significantly improved. Among them, side information is the core element of the Slepian-Wolf theory, which refers to auxiliary information related to the source to be decoded that is already available at the decoding end. The decoding end can achieve efficient reconstruction by combining the side information with the received data. In some embodiments, the embodiments of the present application can be applied to the 6G AI model life cycle management (LCM) framework.

[0055] Alternatively, the AI model described in the embodiments of the present application may also be referred to as a machine learning (ML) model. Alternatively, the decoding described in the embodiments of the present application may also be referred to as decoding.

[0056] The following describes the technical principles involved in the AI model parameter acquisition method provided in the embodiment of the present application: Based on the current model fine-tuning algorithm (including but not limited to low-rank adaptation (LoRA) fine-tuning algorithm, etc.), the NN original parameter matrix and the fine-tuned NN parameter matrix The relationship between can be modeled as: , It is the NN update parameters accumulated during model fine-tuning. Assume The binary quantization sequence is , The binary quantization sequence is The terminal device has an AI model deployed locally, so the terminal device has Based on this, a correlation transmission model with side information can be modeled according to the Slepian-Wolf theory to transform the original sequence Considered as target sequence The received signal after transmission through a correlation channel, the correlation transmission model can be expressed as: ,in is the correlation channel noise, and Correlation. Due to the correlation channel noise The distribution characteristics of are difficult to estimate accurately, so the embodiment of the present application is based on the correlation transmission model, according to the known original sequence Constructing an unknown target sequence The constructed side information is used as additional prior knowledge for decoding input, which can greatly enhance decoding performance. Compared with independent decoding algorithms, the embodiments of the present application can achieve lossless data transmission at a higher bit rate.

[0057] The following is a detailed description of the AI model parameter acquisition method provided in the embodiment of the present application in conjunction with the corresponding flowchart. It can be understood that the schematic flowchart provided in the embodiment of the present application mainly uses different devices (such as terminal devices, network devices) as the execution subjects of the interactive schematic to illustrate the AI model parameter acquisition method, but the embodiment of the present application does not limit the execution subjects of the interactive schematic. For example, the device (such as terminal device, network device) in the schematic flowchart can also be a chip, chip system, or processor that supports the device to implement the AI model parameter acquisition method, or it can be a logic module or software that can implement all or part of the functions of the device.

[0058] For a unified explanation here, in the interaction process of the embodiment of the present application, the message or signaling interaction involved can adopt the message or signaling in the standard, or it can be a newly introduced message or signaling, and the embodiment of the present application does not limit this.

[0059] The interpretation of some technical terms in the embodiments of this application can refer to the interpretation in the 3rd Generation Partnership Project (3GPP) standard protocol. It should be understood that the technical terms in the embodiments of this application are for illustrative purposes only and are not limiting. As technology evolves, technical terms will also change. If the technical meaning is the same, other technical terms should also apply to the embodiments of this application.

[0060] Understandably, the following Figure 3 The terminal device described in the embodiment may be Figure 1 Any terminal device described in the embodiment may also be a device in the terminal device (such as a processor, chip, or chip system, etc.). Figure 3 The network device described in the embodiment may be Figure 1 Any network device described in the implementation manner (such as an access network device or a core network device) may also be a device in the network device (such as a processor, a chip, or a chip system, etc.).

[0061] Figure 3 This is a flow chart of a method for obtaining AI model parameters provided by an embodiment of the present application. Figure 3 As shown, the AI model parameter acquisition method may include the following steps: Step 301: The network device obtains the NN update parameters of the target AI model.

[0062] The target AI model is the AI model that needs to be updated. The NN update parameters are used to update the target AI model. The NN update parameters may be some or all of the NN parameters in the target AI model.

[0063] The NN update parameters may include one or more update parameters in one or more NN layers in the target AI model. The update parameters of a NN layer are used to update the NN layer.

[0064] The network device can be an access network device or a core network device, and this embodiment of the present application does not limit this.

[0065] The NN update parameters can be generated by the network device or received by the network device from other devices. For example, the network device is a base station. The base station can receive the updated target AI model sent by the core network device (such as a cloud server or edge server, etc.), and the base station can determine the NN update parameters of the target AI model based on the target AI model before the update and the target AI model after the update. Alternatively, the base station can receive the NN update parameters of the target AI model sent by the core network device. Of course, this is not limited to this, and the NN update parameters can also be obtained by the network device through other means, and the embodiments of the present application are not limited to this.

[0066] Step 302: For the update parameters of any one NN layer (hereinafter referred to as the target NN layer) in the NN update parameters, the network device obtains a binary quantized sequence of the update parameters of the target NN layer.

[0067] It should be noted that, if the NN update parameters include only update parameters for one NN layer, the network device may send the update parameters for that NN layer to the terminal device. If the NN update parameters include update parameters for multiple NN layers, the network device may send the update parameters for each of the multiple NN layers to the terminal device in sequence.

[0068] For the update parameters of a specific NN layer (i.e., the target NN layer) that currently needs to be transmitted, the network device may first flatten and quantize the update parameters of the target NN layer to obtain a corresponding binary quantization sequence, and then transmit the binary quantization sequence in blocks. For example, the binary quantization sequence may be encapsulated into a single transport block (TB).

[0069] Optionally, before sending the binary quantized sequence of updated parameters of the target NN layer in blocks, the network device may first send control information to the terminal device based on the binary quantized sequence. The control information may include a target AI model identifier (ID), a target NN layer identifier, and the total number of blocks. After receiving the control information, the terminal device can obtain the target AI model identifier, the target NN layer identifier, and the total number of blocks.

[0070] The target AI model identifier is used to identify the target AI model. The target NN layer identifier is used to identify the target NN layer. The total number of blocks is the number of blocks into which the binary quantized sequence will be divided when it is sent. It is assumed here that the total number of blocks is n, where n is a positive integer.

[0071] That is to say, each time the network device needs to send an updated parameter of the NN layer to the terminal device, it can first send the control information related to the NN layer to the terminal device, so that the terminal device can subsequently accurately decode the NN parameters according to the control information.

[0072] For example, the network device may carry the control information in the AI model update control signaling and send it to the terminal device. Of course, the network device may also carry the control information in other messages or signaling and send it to the terminal device, which is not limited in this embodiment of the present application.

[0073] Step 303: The network device divides the binary quantization sequence of the update parameters of the target NN layer into n blocks.

[0074] Optionally, the network device may segment the binary quantization sequence sequentially according to a first preset length to obtain n blocks, where the n blocks are in order. The length of each of the n blocks is the first preset length. The first preset length may be preset, and the unit of the first preset length is bits. Here, it is assumed that the first preset length is L.

[0075] It is easy to understand that one or more blocks of a first preset length can be sequentially intercepted from the binary quantization sequence. In some cases, when the length of the remaining bits in the binary quantization sequence (assuming it is r) is insufficient (i.e., less than the first preset length), all the remaining bits are taken as the valid part of the nth block (i.e., the last block) among the n blocks, and Lr bits are padded after the valid part, which can usually be padded with 0 or padded according to the protocol. Based on this, it is possible that the n blocks are all directly intercepted from the binary quantization sequence; or, the first n-1 blocks among the n blocks are all directly intercepted from the binary quantization sequence, the 1st to rth bits of the nth block among the n blocks are intercepted from the binary quantization sequence, and the r+1th to Lth bits of the nth block are padded.

[0076] Step 304: The network device performs channel coding on each of the n blocks to obtain n first coded blocks.

[0077] The length of each of the n first coding blocks is a second preset length. The second preset length can be preset, and the unit of the second preset length can be bits. It is assumed here that the second preset length is K.

[0078] A first coded block is obtained by adding parity bits to a block. The first coded block is a systematic code. That is, the first L bits in the first coded block are information bits, and bits (L+1) through (K) in the first coded block are parity bits. The information bits include the updated parameters of the NN layer. The parity bits are redundant data generated by the channel coding algorithm.

[0079] Optionally, for any one block among the n blocks, the network device may input the block into a channel coding module, and the channel coding module may output a first coding block.

[0080] The channel coding module is used to perform channel coding. For example, the channel coding module may be a module that uses a low-density parity check code (LDPC) coding algorithm, i.e., an LDPC coding module; or, the channel coding module may be a module that uses a Turbo coding algorithm, i.e., a Turbo coding module. Of course, the channel coding module may also be a module that uses other channel coding algorithms, which is not limited in this embodiment of the present application.

[0081] Step 305: The network device performs a puncture operation on each of the n first coding blocks to obtain n second coding blocks.

[0082] The length of each of the n second coding blocks is a third preset length. The third preset length can be preset, and the unit of the third preset length can be bits. Here, it is assumed that the third preset length is 1.

[0083] Puncturing is used to remove some information bits from the first coded block, while retaining all parity bits. Here, we assume that puncturing is used to remove the first Q bits from the first coded block to match the current channel transmission rate. In this case, Q < L, and I = KQ.

[0084] In some implementations, the above-mentioned blocks, the first coding block, and the second coding block may all be referred to as code blocks (CB).

[0085] Step 306: The network device modulates each of the n second coding blocks respectively to obtain n symbol sequences.

[0086] The n symbol sequences are all complex-valued symbols.

[0087] Optionally, the network device may send downlink control information (DCI) to the terminal device, where the DCI may include the time-frequency position of each second coding block in the multiple second coding blocks.

[0088] Step 307: The network device sends the n symbol sequences to the terminal device.

[0089] Step 308: The terminal device receives n symbol sequences sent by the network device, wherein each time a symbol sequence is received, the symbol sequence is demodulated to obtain a first soft information sequence.

[0090] The terminal device can separate the symbol sequences corresponding to each second coding block from the received symbol sequence according to the time-frequency positions of each second coding block indicated by the DCI sent by the network device, and can determine the block index of each symbol sequence.

[0091] This soft information can be a log-likelihood ratio (LLR), where the LLR can be an estimated value. The LLR is used to quantify the confidence level of a bit's reliability. An LLR greater than 0 indicates a higher probability of judging the bit as "0," while an LLR less than 0 indicates a higher probability of judging the bit as "1." A larger |LLR| indicates a higher confidence level, while a smaller |LLR| indicates a greater influence of noise and a higher likelihood of error. An LLR of 0 indicates neutral confidence, representing "no information."

[0092] The length of the first soft information sequence is the same as the length of the second coding block, that is, both are the third preset length.

[0093] Optionally, the terminal device may demodulate the symbol sequence using the following formula to obtain a first soft information sequence.

[0094]

[0095] in, For this symbol sequence The corresponding second coding block The k-th LLR in . Represents the symbol sequence The sign value of the i-th bit in . is the channel noise variance, which determines the quality of the received signal.

[0096] Step 309: The terminal device generates a second soft information sequence according to the first soft information sequence.

[0097] The length of the second soft information sequence is the same as the length of the first coding block, that is, both are a second preset length. The first L bits of the second soft information sequence are soft information of the information bits, and the L+1th to Kth bits of the second soft information sequence are soft information of the check bits.

[0098] Optionally, the terminal device may add Q bits to the head of the first soft information sequence and set all Q bits to 0 to obtain a second soft information sequence. In this case, the first Q bits (i.e., the punctured bits in the information bits) in the second soft information sequence are 0, and bits (Q+1) to (K) in the second soft information sequence constitute the first soft information sequence.

[0099] Step 310: The terminal device decodes the second soft information sequence according to the side information related to the update parameters of the target NN layer in the AI model to obtain a target coding block, where the target coding block includes the update parameters of the target NN layer.

[0100] The side information related to the updated parameters of the target NN layer is determined based on the original NN parameters of the target AI model. The side information related to the updated parameters of the target NN layer is the side information specific to the updated parameters. In this case, performing side information-assisted joint decoding on the second soft information sequence can accurately recover the updated parameters of the target NN layer.

[0101] An AI model is deployed in the terminal device, so the terminal device has the original NN parameters of the target AI model. Based on this, the terminal device can determine the side information related to the update parameters of the target NN layer. In the embodiment of the present application, the side information related to the update parameters of the target NN layer is used as additional prior knowledge for decoding input, which can greatly improve the decoding performance. Moreover, since the decoding gain of the embodiment of the present application is higher and there is a certain performance redundancy, the decoding can be completed more accurately even under poor channel conditions, and the decoding success rate is high. In this way, robust transmission and accurate recovery of NN parameters can be achieved. In addition, due to the improvement in decoding performance, the transmitter can adopt a more aggressive coding strategy to further improve transmission efficiency.

[0102] In some embodiments, before receiving the n symbol sequences, the terminal device receives control information sent by the network device. Based on the target AI model identifier, target NN layer identifier, and total number of blocks in the control information, the terminal device can learn the update parameters of the target NN layer in the target AI model to be received next, and can also learn that the update parameters correspond to the n symbol sequences.

[0103] In this case, the terminal device can obtain the original NN parameters of the target AI model. The binary quantization sequence of the original NN parameters of the target AI model and the parameters of the hidden Markov model are input into the forward-backward algorithm (FBA) module to obtain the output data of the FBA module. Afterwards, based on the output data of the FBA module, the soft information sequence of the binary quantization sequence of the NN update parameters of the target AI model is estimated to obtain the target soft information sequence. The target soft information sequence includes side information related to the update parameters of the target NN layer. That is, the side information related to the update parameters of the target NN layer is the soft information sequence of the binary quantization sequence of the update parameters of the target NN layer, that is, the soft information sequence corresponding to the target NN layer. The soft information sequence of a binary quantization sequence includes the soft information of each bit in the binary quantization sequence.

[0104] The hidden Markov model is used to represent the correlation between the binary quantization sequence of the original NN parameters of the target AI model and the binary quantization sequence of the updated NN parameters of the target AI model.

[0105] For example, Figure 4 As shown, the embodiment of the present application converts the binary quantization sequence of the original NN parameters of the target AI model into Binary quantized sequence of NN update parameters of the target AI model The correlation between them is modeled as a hidden Markov model, Figure 4 The binary hidden Markov model is taken as an example. is an observable sequence, is the predicted sequence.

[0106] Assumptions and The length of each is M, and the state of its mth bit is , then the initial input of the FBA module is: .in, are the initial parameters of the hidden Markov model. is the initial hidden state probability; is the state transition probability matrix, represents the corresponding state transition probability; is the observation probability matrix, Indicates that in the corresponding hidden state The probability of being 0 or 1.

[0107] As the binary quantized sequence of the original NN parameters, it is the observable past sequence. The goal of the FBA module is to use the hidden Markov model and the known , estimated The FBA module will continuously modify the parameters of the hidden Markov model during the iteration process. , when the observable sequence sample When the time is long enough, the impact of the error in the initial input state can be ignored. Based on this, the FBA module can output the following after reaching the preset number of iterations: .in, express The transition probability in the corresponding state, express The probability of express probability.

[0108] Therefore, it can be determined by the following formula The LLR of the mth position in , where LLR is an estimated value.

[0109]

[0110] in, Indicates that given the current hidden state Under the premise of satisfying the joint probability of the following two events at the same time: 1. The system remains in the same state: the hidden state at the next moment is still (Right now ); 2. Generate a specific bit: the observation bit output at the current moment . Indicates that given the current hidden state Under the premise of satisfying the joint probability of the following two events at the same time: 1. The system remains in the same state: the hidden state at the next moment is still (Right now ); 2. Generate a specific bit: the observation bit output at the current moment . express probability. express probability.

[0111] Through the above method, it can be estimated that LLR sequence, that is, the target soft information sequence is obtained.

[0112] It should be noted that there may be one or more NN layer parameters in the target AI model that need to be updated. In this case, before the network device sends a symbol sequence corresponding to the updated parameters of an NN layer to the terminal device each time, it may first send control information carrying the target AI model identifier, the NN layer identifier, and the total number of blocks corresponding to the updated parameters of the NN layer to the terminal device. In an embodiment of the present application, the terminal device does not need to calculate the target soft information sequence each time it receives control information, but can calculate the target soft information sequence when it first receives control information related to the target AI model update, and does not need to calculate the target soft information sequence when it subsequently receives control information related to the target AI model update, thereby saving computing resources.

[0113] In this case, before receiving the symbol sequence corresponding to the update parameters of any NN layer in the target AI model, the terminal device has already obtained the target soft information sequence based on the original NN parameters of the target AI model. The target soft information sequence includes the soft information sequence of the binary quantized sequence of the update parameters of each NN layer in one or more NN layers that have update requirements in the target AI model, that is, the target soft information sequence includes the soft information sequence corresponding to each NN layer in the one or more NN layers. In this case, the side information related to the update parameters of the target NN layer can be the soft information sequence corresponding to the target NN layer.

[0114] It should be noted that since the control information received by the terminal device includes the target AI model identifier, the target NN layer identifier, and the total number of blocks (i.e., n), after receiving the control information and obtaining the target soft information sequence, the terminal device can also obtain the soft information sequence corresponding to the target NN layer from the target soft information sequence and divide the soft information sequence corresponding to the target NN layer into n soft information sequences. In this way, the terminal device can subsequently perform decoding processing based on these n soft information sequences.

[0115] The terminal device may segment the soft information sequence corresponding to the target NN layer in sequence according to the first preset length to obtain n soft information sequences, where the n soft information sequences exist in order. In this case, the length of each of the n soft information sequences is the first preset length, i.e., L.

[0116] It is easy to understand that one or more soft information sequences of a first preset length can be sequentially extracted from the soft information sequence corresponding to the target NN layer. In some cases, when the length of the remaining bits (assuming it is r) in the soft information sequence corresponding to the target NN layer is insufficient (i.e., less than the first preset length), all remaining bits are used as the valid portion of the nth soft information sequence among the n soft information sequences. This valid portion is then padded with Lr bits, typically zeros. Based on this, the n soft information sequences may all be directly extracted from the soft information sequence corresponding to the target NN layer; alternatively, the first n-1 soft information sequences among the n soft information sequences may all be directly extracted from the soft information sequence corresponding to the target NN layer, bits 1 to r of the nth soft information sequence among the n soft information sequences are extracted from the soft information sequence corresponding to the target NN layer, and bits r+1 to L of the nth soft information sequence are padded.

[0117] In this case, the side information related to the updated parameters of the target NN layer can be the n soft information sequences. The operation of step 309 can be: generating a fourth soft information sequence based on the second soft information sequence and the third soft information sequence, where the third soft information sequence is a soft information sequence among the n soft information sequences corresponding to the second soft information sequence; and inputting the fourth soft information sequence into a channel decoding module to obtain a target coding block output by the channel decoding module.

[0118] The n soft information sequences correspond one-to-one to the n symbol sequences corresponding to the updated parameters of the target NN layer. Since the n soft information sequences exist in an orderly manner, the terminal device can obtain a corresponding soft information sequence from the n soft information sequences as the third soft information sequence based on a block index of a symbol sequence corresponding to the second soft information sequence.

[0119] The channel decoding module is used to perform channel decoding. For example, the channel decoding module may be an LDPC decoding module, a Turbo decoding module, or the like, although this embodiment of the present application is not limited thereto. For example, the LDPC decoding module may employ a belief propagation (BP) decoding algorithm. The Turbo decoding module may employ a maximum a posteriori probability iterative decoding algorithm.

[0120] It should be noted that the channel coding module should correspond to the channel decoding module. That is, when the channel coding module in the network device is an LDPC coding module, the channel decoding module in the terminal device can be an LDPC decoding module; when the channel coding module in the network device is a Turbo coding module, the channel decoding module in the terminal device can be a Turbo decoding module.

[0121] The length of the fourth soft information sequence is the same as the length of the second soft information sequence. The operation of the terminal device generating the fourth soft information sequence based on the second soft information sequence and the third soft information sequence may include any one of the following three methods: In the first method, the terminal device adds the soft information of the first Q bits in the second soft information sequence to the soft information of the corresponding bits in the third soft information sequence one by one to obtain the soft information of the first Q bits in the fourth soft information sequence, where the first Q bits are punctured bits; and uses the soft information of the remaining bits in the second soft information sequence except the soft information of the first Q bits as the soft information of the remaining bits in the fourth soft information sequence except the first Q bits.

[0122] It should be noted that in related art, the punctured bits are assumed to be unknown information, i.e., their decoding input is set to 0. This input setting loses some coded block information, resulting in a certain degradation in decoding performance. In the embodiment of the present application, the decoding input of the punctured bits can be set based on the estimated third soft information sequence, thereby improving decoding performance.

[0123] In a second manner, the terminal device uses the soft information of the first Q bits in the second soft information sequence as the soft information of the first Q bits in the fourth soft information sequence, where the first Q bits are punctured bits; adds the soft information of the Q+1th to Kth bits in the second soft information sequence to the soft information of the corresponding bits in the third soft information sequence one by one, to obtain the soft information of the Q+1th to Kth bits in the fourth soft information sequence, where the Q+1th to Kth bits are information bits; and uses the soft information of the remaining bits in the second soft information sequence except the soft information of the first K bits as the soft information of the remaining bits in the fourth soft information sequence except the first K bits.

[0124] In the embodiment of the present application, for the Q+1th to Kth information bits, joint decoding input can be performed based on the second soft information sequence obtained by demodulation and the third soft information sequence obtained by estimation, thereby enhancing decoding performance.

[0125] In a third manner, the terminal device adds the soft information of the first Q bits in the second soft information sequence to the soft information of the corresponding bits in the third soft information sequence one by one to obtain the soft information of the first Q bits in the fourth soft information sequence, where the first Q bits are punctured bits; adds the soft information of the Q+1th to Kth bits in the second soft information sequence to the soft information of the corresponding bits in the third soft information sequence one by one to obtain the soft information of the Q+1th to Kth bits in the fourth soft information sequence, where the Q+1th to Kth bits are information bits; and uses the soft information of the remaining bits in the second soft information sequence except the soft information of the first K bits as the soft information of the remaining bits in the fourth soft information sequence except the first K bits.

[0126] In an embodiment of the present application, not only can the decoding input of the punctured bits be set based on the estimated third soft information sequence, but also the joint decoding input of the Q+1th to Kth information bits can be performed based on the demodulated second soft information sequence and the estimated third soft information sequence, thereby significantly enhancing the decoding performance.

[0127] For example, the operation of obtaining the fourth soft information sequence in the third method can be expressed by the following formula:

[0128] It should be noted that the channel decoding module generally pre-sets a maximum number of iterations. If the decoded information bits and check bits meet the requirements before the maximum number of iterations is reached, the decoding result (i.e., the coded block) can be output. If the decoded information bits and check bits still do not meet the requirements after the maximum number of iterations is reached, the decoding will fail. The embodiments of the present application improve the decoding success rate by introducing side information for joint decoding input. Moreover, compared to related technologies, the prior knowledge added in the embodiments of the present application can accelerate convergence, thereby reducing the number of iterations and improving decoding speed.

[0129] In an embodiment of the present application, the network device may send n symbol sequences corresponding to the update parameters of the target NN layer; the terminal device may decode each received symbol sequence to obtain a target coding block, and ultimately obtain n target coding blocks corresponding one-to-one to the n symbol sequences, and determine the update parameters of the target NN layer based on the n target coding blocks. In addition, the network device may also sequentially send to the terminal device symbol sequences corresponding to the update parameters of each of one or more NN layers that need to be updated in the target AI model; the terminal device may thereby determine the update parameters of each of the one or more NN layers, and ultimately obtain the NN update parameters of the target AI model, and the terminal device may thereby update the target AI module.

[0130] For ease of understanding, the following Figure 5 and Figure 6 The following is an example of how to transfer AI model parameters: like Figure 5 As shown, on the network device side, the update parameters of the NN layer in the AI model are flattened and quantized to obtain a binary quantization sequence. The binary quantization sequence is then divided into n blocks. Each of the n blocks is channel-coded to obtain n first coding blocks. Each of the n first coding blocks is punctured to obtain n second coding blocks. Each of the n second coding blocks is modulated to obtain n symbol sequences, which are then sent to the terminal device.

[0131] like Figure 6 As shown, after receiving one of the n symbol sequences, the terminal device demodulates the symbol sequence to obtain an LLR sequence. This LLR sequence is then jointly decoded with an LLR sequence associated with updated NN parameters estimated from the original NN parameters to obtain a decoding result. Each decoding result is processed to obtain n blocks. Based on these n blocks and the AI model ID, NN layer ID, and total number of blocks in the AI model update control signaling sent by the network device, the NN layer in the AI model that can be updated is determined, and the parameters of the NN layer are updated based on the n blocks.

[0132] In an embodiment of the present application, the terminal device can determine the side information related to the updated parameters of the NN layer based on the original NN parameters of the AI model. The terminal device can use the side information related to the updated parameters of the NN layer as additional prior knowledge and perform joint decoding input with the soft information sequence obtained by demodulation, thereby greatly improving the decoding performance. In this way, even under poor channel conditions, decoding can be completed more accurately with a high decoding success rate, thereby helping to achieve robust transmission and accurate recovery of NN parameters.

[0133] It should be understood that Figures 1 to 6 The flowcharts or scenario diagrams shown are only for ease of understanding and are not intended to limit the embodiments of the present application to the examples shown in the diagrams. In fact, those skilled in the art will Figures 1 to 6 The examples in can be equivalently transformed to obtain more implementation methods.

[0134] Combined with the above Figures 1 to 6 , describes in detail the AI model parameter acquisition method provided by the embodiment of this application. Figures 7 and 8 The device embodiments of the present application are described in detail. It should be understood that the communication device of the embodiment of the present application can execute the various AI model parameter acquisition methods of the aforementioned embodiment of the present application, that is, the specific working processes of the following various products can refer to the corresponding processes in the aforementioned method embodiments.

[0135] In each of the above embodiments, the terminal device may perform some or all of the steps in each embodiment; the network device may perform some or all of the steps in each embodiment. These steps or operations are merely examples, and the embodiments of the present application may also perform other operations or variations of various operations. In addition, the various steps may be performed in a different order as presented in the embodiments, and it is possible that not all of the operations in the embodiments of the present application need to be performed. The size of the sequence number of each step does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0136] Figure 77 is a schematic block diagram of a communication device 700 provided in an embodiment of the present application. Figure 7 As shown, communication device 700 may include a communication module 720. Communication module 720 can implement corresponding communication functions, which may be internal communication functions of communication device 700 or communication functions between communication device 700 and other devices. Optionally, communication module 720 may also be referred to as a communication interface or a transceiver module. Optionally, communication device 700 also includes a processing module 710. Processing module 710 can implement corresponding processing functions.

[0137] Optionally, the communication device 700 further includes a storage module, which can be used to store instructions and / or data; the processing module 710 can read the instructions and / or data in the storage module to enable the communication device 700 to implement the aforementioned method embodiment.

[0138] In one possible design, communication device 700 may correspond to the terminal device in the above method embodiments, or a component configured in the terminal device (such as a circuit, chip, or chip system). Communication device 700 can be used to execute the steps or processes performed by the terminal device in any of the above method embodiments.

[0139] For example, the communication module 720 is configured to receive n symbol sequences sent by a network device, where n is a positive integer; The processing module 710 is used to demodulate a symbol sequence each time it is received to obtain a first soft information sequence; generate a second soft information sequence of a preset length based on the first soft information sequence; decode the second soft information sequence based on side information related to the update parameters of the target NN layer in the target AI model to obtain a target coding block, where the target coding block includes the update parameters of the target NN layer, and the side information is determined based on the original NN parameters of the target AI model.

[0140] For example, the communication module 720 is used to receive control information sent by a network device, where the control information includes a target AI model identifier, a target NN layer identifier, and a total number of blocks, where the total number of blocks is n.

[0141] For example, the processing module 710 is used to obtain the original NN parameters of the target AI model; the binary quantization sequence of the original NN parameters of the target AI model and the parameters of the hidden Markov model are input into the FBA module to obtain the output data of the FBA module, and the hidden Markov model is used to represent the correlation between the binary quantization sequence of the original NN parameters of the target AI model and the binary quantization sequence of the NN update parameters of the target AI model; based on the output data of the FBA module, the soft information sequence of the binary quantization sequence of the NN update parameters is estimated to obtain the target soft information sequence, and the target soft information sequence includes the side information.

[0142] For example, the processing module 710 is used to obtain the soft information sequence corresponding to the target NN layer from the target soft information sequence, where the target soft information sequence is determined based on the original NN parameters of the target AI model; the soft information sequence corresponding to the target NN layer is divided into n soft information sequences, and the n soft information sequences are the side information.

[0143] For example, the processing module 710 is configured to generate a fourth soft information sequence based on the second soft information sequence and the third soft information sequence, where the third soft information sequence is a soft information sequence among the n soft information sequences corresponding to the second soft information sequence; and input the fourth soft information sequence into the channel decoding module to obtain a target coding block output by the channel decoding module.

[0144] For example, the processing module 710 is configured to add the soft information of the first Q bits in the second soft information sequence to the soft information of the corresponding bits in the third soft information sequence one by one to obtain the soft information of the first Q bits in the fourth soft information sequence, where the first Q bits are punctured bits; and use the soft information of the remaining bits in the second soft information sequence except the soft information of the first Q bits as the soft information of the remaining bits in the fourth soft information sequence except the first Q bits.

[0145] For example, the processing module 710 is configured to use the soft information of the first Q bits in the second soft information sequence as the soft information of the first Q bits in the fourth soft information sequence, where the first Q bits are punctured bits; add the soft information of the Q+1th to Kth bits in the second soft information sequence to the soft information of the corresponding bits in the third soft information sequence one by one to obtain the soft information of the Q+1th to Kth bits in the fourth soft information sequence, where the Q+1th to Kth bits are information bits; and use the soft information of the remaining bits in the second soft information sequence except the soft information of the first K bits as the soft information of the remaining bits in the fourth soft information sequence except the first K bits.

[0146] For example, the processing module 710 is configured to add the soft information of the first Q bits in the second soft information sequence to the soft information of the corresponding bits in the third soft information sequence one by one to obtain the soft information of the first Q bits in the fourth soft information sequence, where the first Q bits are punctured bits; add the soft information of the Q+1th to Kth bits in the second soft information sequence to the soft information of the corresponding bits in the third soft information sequence one by one to obtain the soft information of the Q+1th to Kth bits in the fourth soft information sequence, where the Q+1th to Kth bits are information bits; and use the soft information of the remaining bits in the second soft information sequence except the soft information of the first K bits as the soft information of the remaining bits in the fourth soft information sequence except the first K bits.

[0147] The above is only an example, and for detailed steps or processes, please refer to the description of the aforementioned embodiments.

[0148] Figure 88 is a schematic block diagram of a communication device 800 provided in an embodiment of the present application. Communication device 800 may be a chip, chip system, or processor, etc., that implements the above-described method in a terminal device or network device. Communication device 800 may be used to implement the method described in the above-described method embodiment. For details, please refer to the description of the above-described method embodiment.

[0149] like Figure 8 As shown, communication device 800 may include one or more processors 810, which may also be referred to as processing units or processing modules, and may implement certain control functions. Processor 810 may be a general-purpose processor or a dedicated processor, such as a baseband processor or a central processing unit. The baseband processor may be used to process communication protocols and communication data, while the central processing unit may be used to control communication device 800 (e.g., base station, baseband chip, user, user chip), execute software programs, and process data in the software programs.

[0150] In an optional design, the processor 810 may also store instructions and / or data, which can be executed by the processor 810 to enable the communication device 800 to perform the method described in the above method embodiment.

[0151] In another optional design, the communication device 800 may include a communication interface 820 for implementing receiving and transmitting functions. For example, the communication interface 820 may be a transceiver circuit, an interface, an interface circuit, or a transceiver. The transceiver circuit, interface, interface circuit, or transceiver for implementing the receiving and transmitting functions may be separate or integrated. The transceiver circuit, interface, interface circuit, or transceiver may be used for reading and writing code / data, or the transceiver circuit, interface, interface circuit, or transceiver may be used for transmitting or delivering signals.

[0152] Optionally, the communication device 800 may include one or more memories 830, which may store instructions that can be executed on the processor 810, causing the communication device 800 to perform the method described in the above method embodiment. Optionally, the memory 830 may also store data. Optionally, the processor 810 may also store instructions and / or data. The processor 810 and memory 830 may be provided separately or integrated together.

[0153] It should be understood that, in one possible design, each step in the method embodiment provided in the present application can be completed by an integrated logic circuit of the hardware in the processor or by instructions in the form of software. The steps of the method disclosed in conjunction with the embodiments of the present application can be directly embodied as being executed by a hardware processor, or can be executed by a combination of hardware and software modules in the processor. The software module can be located in a storage medium mature in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in a memory, and the processor reads the information in the memory and completes the steps of the above method in conjunction with its hardware. To avoid repetition, it will not be described in detail here.

[0154] In one implementation, the communication device 800 may correspond to the terminal device in the above-mentioned method embodiment and may be used to execute the various steps and / or processes performed by the terminal device in the above-mentioned method embodiment. The processor 810 may be used to execute instructions stored in the memory 830, and when the processor 810 executes the instructions stored in the memory, the processor 810 is used to execute the various steps and / or processes of the above-mentioned method embodiment corresponding to the terminal device.

[0155] In another implementation, the communication device 800 may correspond to the network device in the above-mentioned method embodiment and may be used to execute the various steps and / or processes performed by the network device in the above-mentioned method embodiment. The processor 810 may be used to execute instructions stored in the memory 830, and when the processor 810 executes the instructions stored in the memory, the processor 810 is used to execute the various steps and / or processes of the above-mentioned method embodiment corresponding to the network device.

[0156] It should be understood that the processing device may be one or more chips. For example, the processing device may be a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on chip (SoC), a central processor unit (CPU), a network processor (NP), a digital signal processor (DSP), a microcontroller unit (MCU), a programmable logic device (PLD), or other integrated chips.

[0157] It is understood that the memory in the embodiments of the present application may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. The non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), and direct RAM bus RAM (DR RAM). It should be noted that the memory of the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0158] Based on the methods provided in the embodiments of the present application, the present application also provides a chip system, which includes one or more processors configured to retrieve and execute instructions stored in a memory, thereby executing the methods of the embodiments of the present application. The chip system can be composed of a chip or can include a chip and other discrete devices.

[0159] Among them, the chip system may include an input circuit or interface for sending information or data, and an output circuit or interface for receiving information or data.

[0160] According to the method provided in the embodiment of the present application, the present application also provides a communication system, which includes the aforementioned network device and terminal device.

[0161] According to the method provided in the embodiments of the present application, the present application also provides a computer program product, which includes: computer program code, which, when running on a computer, enables the computer to execute the various steps or processes executed by the network device and terminal device in any of the aforementioned method embodiments.

[0162] According to the method provided in the embodiments of the present application, the present application also provides a computer-readable storage medium, which stores program code. When the program code runs on a computer, the computer executes the various steps or processes performed by the network device and terminal device in any of the aforementioned method embodiments.

[0163] The computer-readable storage medium may be the aforementioned volatile memory or non-volatile memory, or may include both volatile memory and non-volatile memory.

[0164] In the embodiments of this application, each term and English abbreviation is provided for convenience of description and shall not constitute any limitation to this application. This application does not exclude the possibility of defining other terms that can achieve the same or similar functions in existing or future agreements.

[0165] In the above embodiments, all or part of the embodiments may be implemented using software, hardware, firmware, or any combination thereof. When implemented using software, all or part of the embodiments may be implemented in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer instructions are loaded and executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part.

[0166] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0167] It should be understood that in the various embodiments of the present application, the size of the serial number of each process does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0168] In short, the above description is only an optional embodiment of the technical solution of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of this application shall be included in the scope of protection of this application.

Claims

1. A method for obtaining AI model parameters, characterized in that: Applied to a terminal device, the method includes: Receiving n symbol sequences sent by a network device, where n is a positive integer; Whenever a symbol sequence is received, the symbol sequence is demodulated to obtain a first soft information sequence; generating a second soft information sequence of a preset length according to the first soft information sequence; According to the side information related to the update parameters of the target neural network NN layer in the target artificial intelligence AI model, the second soft information sequence is decoded to obtain a target coding block, where the target coding block includes the update parameters of the target NN layer, and the side information is determined based on the original NN parameters of the target AI model.

2. The method according to claim 1, wherein Before receiving the n symbol sequences sent by the network device, the method further includes: Receive control information sent by the network device, where the control information includes a target AI model identifier, a target NN layer identifier, and a total number of blocks, where the total number of blocks is n.

3. The method according to claim 2, wherein After receiving the control information sent by the network device, the method further includes: Obtaining the original NN parameters of the target AI model; Input the binary quantized sequence of the original NN parameters of the target AI model and the parameters of the hidden Markov model into the forward-backward algorithm FBA module to obtain the output data of the FBA module, wherein the hidden Markov model is used to represent the correlation between the binary quantized sequence of the original NN parameters of the target AI model and the binary quantized sequence of the updated NN parameters of the target AI model; According to the output data of the FBA module, a soft information sequence of the binary quantization sequence of the NN update parameter is estimated to obtain a target soft information sequence, where the target soft information sequence includes the side information.

4. The method according to any one of claims 1 to 3, characterized in that Before receiving the n symbol sequences sent by the network device, the method further includes: Obtaining a soft information sequence corresponding to the target NN layer from a target soft information sequence, where the target soft information sequence is determined based on the original NN parameters of the target AI model; Divide the soft information sequence corresponding to the target NN layer into n soft information sequences, where the n soft information sequences are the side information; The decoding of the second soft information sequence according to the side information related to the update parameters of the target neural network NN layer in the target artificial intelligence AI model to obtain the target coding block includes: generating a fourth soft information sequence according to the second soft information sequence and the third soft information sequence, wherein the third soft information sequence is a soft information sequence corresponding to the second soft information sequence among the n soft information sequences; The fourth soft information sequence is input into a channel decoding module to obtain a target coding block output by the channel decoding module.

5. The method according to claim 4, wherein Generating a fourth soft information sequence according to the second soft information sequence and the third soft information sequence includes: Adding the soft information of the first Q bits in the second soft information sequence to the soft information of the corresponding bits in the third soft information sequence one by one to obtain the soft information of the first Q bits in the fourth soft information sequence, where the first Q bits are punctured bits; The soft information of the remaining bits except the first Q bits in the second soft information sequence is used as the soft information of the remaining bits except the first Q bits in the fourth soft information sequence.

6. The method according to claim 4, wherein Generating a fourth soft information sequence according to the second soft information sequence and the third soft information sequence includes: Using the soft information of the first Q bits in the second soft information sequence as the soft information of the first Q bits in the fourth soft information sequence, where the first Q bits are punctured bits; Adding the soft information of the Q+1th to Kth bits in the second soft information sequence to the soft information of the corresponding bits in the third soft information sequence one by one to obtain the soft information of the Q+1th to Kth bits in the fourth soft information sequence, where the Q+1th to Kth bits are information bits; The soft information of the remaining bits except the first K bits in the second soft information sequence is used as the soft information of the remaining bits except the first K bits in the fourth soft information sequence.

7. The method according to claim 4, wherein Generating a fourth soft information sequence according to the second soft information sequence and the third soft information sequence includes: Adding the soft information of the first Q bits in the second soft information sequence to the soft information of the corresponding bits in the third soft information sequence one by one to obtain the soft information of the first Q bits in the fourth soft information sequence, where the first Q bits are punctured bits; Adding the soft information of the Q+1th to Kth bits in the second soft information sequence to the soft information of the corresponding bits in the third soft information sequence one by one to obtain the soft information of the Q+1th to Kth bits in the fourth soft information sequence, where the Q+1th to Kth bits are information bits; The soft information of the remaining bits except the first K bits in the second soft information sequence is used as the soft information of the remaining bits except the first K bits in the fourth soft information sequence.

8. A communication device, characterized in that: The communication device includes at least one processor coupled to a memory, wherein the memory stores a program or instruction. The processor executes the program or instruction so that the communication device is configured to perform the method according to any one of claims 1 to 7.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program or instructions, which, when executed, causes a computer to perform the method according to any one of claims 1 to 7.

10. A chip system, characterized in that: The chip system includes one or more processors, and the one or more processors are used to call and execute instructions stored in the memory from the memory, so that the method according to any one of claims 1 to 7 is executed.

Images