First device and signal processing method

Abstract

The application provides a first device and a signal processing method. The signal processing method is applied to the first device and includes the following steps: obtaining target reverse sampling parameters corresponding to a diffusion model, the target reverse sampling parameters being related to target quality of service parameters of a first service, and the target quality of service parameters being used for indicating performance requirements of the first service; based on the target reverse sampling parameters, performing reverse de-noising inference on a first pilot signal received on a first channel by using the diffusion model to determine a second pilot signal obtained by de-noising the first pilot signal; determining channel state information of the first channel based on the second pilot signal; and demodulating data of the first service received on the first channel based on the channel state information.

Description

Technical Field

[0001] This application relates to, but is not limited to, the field of communication technology, and in particular to a first device and a signal processing method. Background Technology

[0002] With the continuous advancement of sixth-generation (6G) mobile communication technology, millimeter-wave communication, with its extremely wide available spectrum resources, has become a key technology for achieving ultra-high-speed transmission. To overcome the severe path loss in the millimeter-wave band, communication systems are typically equipped with massive MIMO antenna arrays (MIMOs), utilizing beamforming technology to improve coverage and link reliability. However, the performance of millimeter-wave MIMOs is highly dependent on high-precision channel state information. Due to the high dimensionality, rapid time-varying nature, and susceptibility to obstruction of millimeter-wave channels, coupled with the massive number of parameters inherent in MIMOs, accurate channel state information cannot be obtained at the receiving end. Summary of the Invention

[0003] This application provides at least one first device and a signal processing method.

[0004] The technical solution of this application embodiment is implemented as follows: This application provides a first device, which includes a first transceiver, and A first processor, coupled to a first transceiver; the first processor is configured to: Obtain the target backsampling parameters corresponding to the diffusion model. The target backsampling parameters are related to the target quality of service parameters of the first service. The target quality of service parameters are used to indicate the performance requirements of the first service. Based on the target reverse sampling parameters, the diffusion model is used to perform reverse denoising inference on the first pilot signal received on the first channel to determine the second pilot signal after denoising the first pilot signal. Based on the second pilot signal, the channel state information of the first channel is determined; Based on channel state information, the data of the first service received on the first channel is demodulated.

[0005] This application provides a signal processing method, including: Obtain the target backsampling parameters corresponding to the diffusion model. The target backsampling parameters are related to the target quality of service parameters of the first service. The target quality of service parameters are used to indicate the performance requirements of the first service. Based on the target reverse sampling parameters, the diffusion model is used to perform reverse denoising inference on the first pilot signal received on the first channel to determine the second pilot signal after denoising the first pilot signal. Based on the second pilot signal, the channel state information of the first channel is determined; Based on channel state information, the data of the first service received on the first channel is demodulated.

[0006] In this embodiment, the target backsampling parameters corresponding to the diffusion model are obtained. These parameters are related to the target quality of service (QoS) parameters of the first service, which indicate the performance requirements of the first service. Based on the target backsampling parameters, the diffusion model is used to perform back-denoising inference on the first pilot signal received on the first channel to determine the second pilot signal after denoising. Based on the second pilot signal, the channel state information of the first channel is determined. Based on the channel state information, the data of the first service received on the first channel is demodulated. Thus, by modeling channel estimation as a back-denoising inference process of the diffusion model and establishing a correspondence between QoS parameters and backsampling parameters, the first device can dynamically select target backsampling parameters matching the target QoS parameters for back-denoising inference, improving the accuracy of channel estimation. Furthermore, it can achieve an adaptive balance between computational overhead, processing latency, and energy consumption while ensuring the accuracy of channel state information. Attached Figure Description

[0007] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with this application and, together with the specification, serve to explain the technical solutions of this application.

[0008] Figure 1 A schematic flowchart of an optional signal processing method provided in an embodiment of this application; Figure 2 A schematic flowchart of an optional signal processing method provided in an embodiment of this application; Figure 3 A schematic diagram illustrating the processing mechanism of a diffusion model provided in an embodiment of this application; Figure 4 A schematic flowchart of an optional signal processing method provided in an embodiment of this application; Figure 5 A schematic diagram illustrating the implementation flow of a model training algorithm for a signal processing method based on a diffusion model, provided in an embodiment of this application; Figure 6 A schematic diagram illustrating the implementation flow of a model inference algorithm for a signal processing method based on a diffusion model, provided in an embodiment of this application; Figure 7 A schematic diagram of the architecture of a diffusion model signal processing method based on service quality awareness provided in an embodiment of this application; Figure 8 A schematic diagram of an architecture for adaptive channel estimation based on internal cross-layer interaction on the terminal side, provided for an embodiment of this application; Figure 9 A schematic diagram of a cooperative channel estimation architecture based on network-side signaling control provided for an embodiment of this application; Figure 10 A schematic diagram illustrating the implementation process of a cooperative signal processing method based on network-side signaling control, provided in an embodiment of this application; Figure 11 This is a schematic diagram of an optional structure of the device provided in an embodiment of this application. Detailed Implementation

[0009] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application are further described in detail below with reference to the accompanying drawings and embodiments. The described embodiments should not be regarded as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0010] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0011] It should be noted that the terms "first, second, and third" used in the embodiments of this application are merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first, second, and third" can be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0012] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which the embodiments of this application pertain. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0013] This application provides a first device and a signal processing method.

[0014] In a first aspect, the signal processing method provided in the embodiments of this application is applied to a first device, such as... Figure 1 As shown, it includes: S101. Obtain the target backsampling parameters corresponding to the diffusion model. The target backsampling parameters are related to the target service quality parameters of the first service. The target service quality parameters are used to indicate the performance requirements of the first service. S102. Based on the target reverse sampling parameters, use the diffusion model to perform reverse denoising inference on the first pilot signal received on the first channel to determine the second pilot signal after denoising the first pilot signal. S103. Based on the second pilot signal, determine the channel state information of the first channel; S104. Based on channel state information, demodulate the data of the first service received on the first channel.

[0015] Secondly, the signal processing method provided in this application embodiment is applied to a communication system including a first device and a second device, such as... Figure 2 As shown, it includes: S201. The second device sends a first pilot signal and data of the first service to the first device on the first channel; S202. The first device obtains the target backsampling parameters corresponding to the diffusion model. The target backsampling parameters are related to the target quality of service parameters of the first service. The target quality of service parameters are used to indicate the performance requirements of the first service. S203. The first device, based on the target reverse sampling parameters, uses a diffusion model to perform reverse denoising inference on the first pilot signal received on the first channel to determine the second pilot signal after denoising the first pilot signal. S204. The first device determines the channel state information of the first channel based on the second pilot signal; S205. The first device demodulates the data of the first service received on the first channel based on the channel state information.

[0016] Below, on Figure 1 or Figure 2 The signal processing method shown is described.

[0017] Here, the first device can be a network device or a terminal device. In one example, the network device can be a base station.

[0018] The first channel is a data transmission channel between the first device and the second device. In some embodiments, when the first device is a terminal device, the first channel is a downlink channel; when the first device is a network device, the first channel is an uplink channel.

[0019] In this embodiment, the first channel can be a millimeter-wave channel or a low-frequency band channel. A millimeter-wave channel refers to a communication channel that uses the millimeter-wave frequency band for wireless signal transmission, while a low-frequency band channel refers to a communication channel that uses electromagnetic wave frequencies below the millimeter-wave frequency band for wireless signal transmission.

[0020] Millimeter wave band refers to the electromagnetic wave band with a frequency range of 30 GHz to 300 GHz (corresponding to wavelengths of 1 mm to 10 mm).

[0021] In this embodiment of the application, when performing downlink channel estimation, the first device is a terminal device, the second device is a network device, the first device is a terminal located within the signal coverage area of ​​the second device, and the first channel is the downlink channel between the first device and the second device; when performing uplink channel estimation, the first device is a network device, the second device is a terminal device, the second device is a terminal located within the signal coverage area of ​​the first device, and the first channel is the uplink channel between the first device and the second device.

[0022] Diffusion models are a class of generative models that include a forward diffusion process and a backsampling process (also known as a backward denoising inference process). By simulating the gradual addition and denoising of data, diffusion models can model and reconstruct complex data distributions. They utilize a learnable backward generation chain to gradually "restore" noisy samples into structured data.

[0023] In this embodiment, the reverse denoising inference of the first pilot signal received on the first channel is performed using a diffusion model. This can be understood as using a neural network model that implements or employs the diffusion model to perform reverse denoising inference of the first pilot signal received on the first channel. This neural network model can also be called a diffusion neural network or diffusion network. This neural network model can be used to implement the reverse denoising inference process in the diffusion model.

[0024] In this embodiment, a neural network model implementing a diffusion model is used to progressively remove noise from the received noisy first pilot signal to obtain a denoised second pilot signal. Based on the second pilot signal, the channel state information of the first channel is estimated. The pilot signal can also be referred to as a reference signal. The first pilot signal is a pilot signal transmitted from the second device to the first device based on the first channel; the second pilot signal is a reconstructed signal obtained by progressively denoising the first pilot signal using a diffusion model.

[0025] In this embodiment of the application, the first pilot signal can be understood as the object of inverse denoising inference using the diffusion model, and the second pilot signal can be understood as the denoised reconstructed signal obtained by inverse iterative sampling of the first pilot signal using the diffusion model.

[0026] In this embodiment of the application, the first device is equipped with a pre-trained neural network model.

[0027] In this embodiment, the second device sends a first pilot signal and data of a first service to the first device via a first channel; the first device receives the first pilot signal and data of the first service via the first channel; using a diffusion model and target backsampling parameters, it performs reverse denoising inference on the first pilot signal step by step to determine the second pilot signal after denoising. The target backsampling parameters refer to the sampling parameters used by the first device during the reverse denoising inference process using the diffusion model. The first service can be a communication service.

[0028] In some embodiments, the second device sends the first pilot signal and the data of the first service together to the first device, or the second device may send the first pilot signal and the data of the first service separately to the first device.

[0029] In some embodiments, the target backsampling parameters may include at least one of the following: target sampling step size and target sampling number. The target sampling step size can be understood as the sampling step size related to the target Quality of Service (QoS) parameters, which is the span of denoising iterations in the backsampling inference phase of the diffusion model. A larger sampling step size results in faster backsampling inference speed, but relatively lower accuracy. The target sampling number can be understood as the number of sampling steps related to the target QoS parameters, which is the number of denoising iterations in the backsampling inference phase of the diffusion model. More sampling steps result in higher accuracy of the generated channel matrix, but higher computational latency and energy consumption; fewer target sampling steps result in faster backsampling inference speed, but relatively lower accuracy.

[0030] In one example, the target backsampling parameters may include the target sampling step size and the target sampling number.

[0031] In one example, the target backsampling parameter may include a target sampling step size. In this case, the number of sampling steps used by the diffusion model can be a predefined value or determined based on the initial sampling step size and the target sampling step size. The product of the target sampling step size and the number of sampling steps can be the initial sampling step size.

[0032] In one example, the target backsampling parameter may include the target number of sampling steps. In this case, the sampling step size used by the diffusion model can be a predefined value or determined based on the initial sampling step size and the target number of sampling steps. The product of the target number of sampling steps and the sampling step size can be the initial sampling step size.

[0033] Here, the initial sampling step size can be determined based on the noise of the first pilot signal.

[0034] In some embodiments, the target backsampling parameters can be determined based on the target quality of service (QoS) parameters of the first service.

[0035] The target quality of service parameter can be understood as the quality of service parameter of the first service. The service parameter is a set of parameters that indicate the performance requirements of the service. In the embodiments of this application, the quality of service parameter refers to the indicators related to channel estimation accuracy and latency. In some embodiments, the quality of service parameter may include at least one of the following: quality of service parameter identifier, packet delay budget (PDB), block error rate target (BLER Target), and service priority (PriorityLevel).

[0036] In one example, the target sampling step size and the target number of sampling steps are determined based on the target quality of service parameters.

[0037] In one example, the target sampling step size is determined based on the target quality of service parameters; the target number of sampling steps is determined based on the target sampling step size and the starting step number.

[0038] In one example, the target number of sampling steps is determined based on the target quality of service parameters; the target sampling step size is determined based on the target number of sampling steps and the initial number of steps. The initial number of steps is determined based on the noise power of the first pilot signal. The product of the target sampling step size and the target number of sampling steps is the initial number of steps.

[0039] In some embodiments, the target backsampling parameters may be determined by the first device or indicated by the second device.

[0040] In one example, the first device is the terminal device, and the target backsampling parameters are determined by the terminal device.

[0041] In one example, the first device is a terminal device, the second device is a network device, and the target backsampling parameters are indicated to the first device (i.e., the terminal device) by the second device (i.e., the network device).

[0042] In one example, the first device is a network device, and the target backsampling parameters are determined by the network device.

[0043] The embodiments of this application may include the following scenarios: Scenario 1: The first device is a terminal device. The first device determines the target inverse sampling parameters, performs inverse denoising inference based on the inverse sampling parameters to obtain the second pilot signal, and performs channel estimation and demodulation on the data transmitted on the first channel based on the second pilot signal.

[0044] Scenario 2: The first device is a network device. The first device determines the target inverse sampling parameters, performs inverse denoising inference based on the inverse sampling parameters to obtain the second pilot signal, and performs channel estimation and demodulation on the data transmitted on the first channel based on the second pilot signal.

[0045] Scenario 3: The first device is a terminal device, and the second device is a network device. The second device determines the target back sampling parameters and instructs the target back sampling parameters to the first device. The first device performs back denoising inference based on the received back sampling parameters to obtain the second pilot signal, and performs channel estimation and demodulation of the data transmitted on the first channel based on the second pilot signal.

[0046] In this embodiment of the application, the first device determines the starting number of inverse denoising inference using the diffusion model based on the noise power of the first pilot signal, uses the first pilot signal as the initial input of the inverse denoising inference process of the diffusion model, and starts from the starting number of steps, performs inverse denoising inference on the first pilot signal step by step according to the target sampling step size and target sampling step number specified by the target inverse sampling parameters to obtain the second pilot signal.

[0047] In this embodiment, the first device determines the channel state information (CSI) of the first channel based on the denoised second pilot signal, and demodulates the data of the first service received on the first channel based on the CSI. The CSI describes the effect of the channel on the signals transmitted on that channel.

[0048] The first device can perform inverse transformation and least squares processing on the second pilot signal to obtain the estimated channel matrix, and determine the channel state information based on the channel matrix.

[0049] In this embodiment of the application, if the first device successfully demodulates the data of the first service received on the first channel, a positive acknowledgment (ACK) is generated; if the first device fails to demodulate the data of the first service received on the first channel, a negative acknowledgment (NACK) is generated; the block error rate is calculated based on the feedback positive acknowledgment / negative acknowledgment; and the target backsampling parameter corresponding to the target quality of service parameter is updated based on the block error rate.

[0050] In this embodiment of the application, the diffusion model can be implemented as follows: Figure 3 As shown, it includes the following stages: Stage 1: Forward diffusion process; Noise is gradually injected into the original data sample, causing the original data sample to completely degenerate into a random noise distribution. In one example, the injected noise can be Gaussian noise.

[0051] Phase Two: Learning the Reverse Process; Train a neural network model (e.g., a time-dependent U-Net structure) so that the neural network model can learn the conditional probability distribution in each forward diffusion step, that is, predict and remove the noise component in each forward diffusion step.

[0052] Phase 3: Reverse denoising inference process (also known as reverse iterative sampling process).

[0053] During the inference phase, starting from the noisy signal (corresponding to the noisy first pilot signal received by the first device in this application), the trained neural network model is used to perform iterative denoising. By gradually removing the noise from the noisy signal, a denoised signal (corresponding to the second pilot signal in this application) that is close to the original data sample is obtained.

[0054] The offline training of the diffusion model can include the following steps: loading the original signal from the dataset, gradually injecting noise into the original signal at each time step, so that the original signal gradually approaches a state of pure noise; constructing a neural network model to predict the noise component in the signal at each step, with the training objective of the neural network model being to minimize the mean square error between the predicted noise and the actual noise, until the neural network gradually learns the ability to recover and reconstruct the signal from the noisy signal; updating the parameters of the neural network model using the gradient descent method until the neural network model can accurately estimate the noise component in the signal at each step. After training, the neural network model can be used for the reverse denoising inference process.

[0055] The offline model training function described above for training neural networks can be deployed on network operations, administration and maintenance (OAM) systems, over-the-top (OTT) servers, or network devices (such as core networks or base stations).

[0056] In some embodiments, the above model training can be implemented by an OAM system, an OTT server, or a network device.

[0057] In some embodiments, for downlink data transmission, the first device is a terminal device, which is the receiving end, and the network device is the sending end. The network device sends the trained neural network to the terminal device. For uplink data transmission, the first device is a network device, which can perform neural network model training, or the OAM system or OTT server can perform neural network model training and send the trained neural network model to the network device.

[0058] In this embodiment, the target backsampling parameters corresponding to the diffusion model are obtained. These parameters are related to the target quality of service (QoS) parameters of the first service, which indicate the performance requirements of the first service. Based on the target backsampling parameters, the diffusion model is used to perform back-denoising inference on the first pilot signal received on the first channel to determine the second pilot signal after denoising. Based on the second pilot signal, the channel state information of the first channel is determined. Based on the channel state information, the data of the first service received on the first channel is demodulated. Thus, by modeling channel estimation as a back-denoising inference process of the diffusion model and establishing a correspondence between QoS parameters and backsampling parameters, the first device can dynamically select target backsampling parameters matching the target QoS parameters for back-denoising inference, improving the accuracy of channel estimation. Furthermore, it can achieve an adaptive balance between computational overhead, processing latency, and energy consumption while ensuring the accuracy of channel state information.

[0059] The method for determining the target backsampling parameters in the embodiments of this application is described below. In the embodiments of this application, the target backsampling parameters can be determined by a first device or by a second device.

[0060] In some embodiments, the target backsampling parameters are determined based on the target quality of service parameters and a first relationship; the first relationship is the mapping relationship between the quality of service parameters and the backsampling parameters.

[0061] In this embodiment of the application, the first relationship includes a mapping relationship between one or more sets of service quality parameters and backsampling parameters. Based on the target service quality parameters, the first relationship can be queried to determine the target backsampling parameters corresponding to the target service quality parameters in the first relationship.

[0062] In one example, the first relationship may include at least one set of mappings between quality of service parameters and backsampling parameters: The first group: Service quality parameters characterizing the first service quality requirement, the first backsampling parameters; The second group consists of service quality parameters that characterize the second service quality requirement, and second backsampling parameters.

[0063] In this embodiment of the application, the reverse sampling parameters include the sampling step size and / or the number of sampling steps.

[0064] In one example, the quality of service parameter indicates that the first service is a low-latency communication service and the latency budget is lower than a preset latency threshold. The sampling step size is the first step size, and the number of sampling steps is the first step number. The quality of service parameter indicates that the first service is a high-priority service and the signal-to-noise ratio reported by the physical layer is lower than a preset signal-to-noise ratio threshold. The sampling step size is the first step size, and the number of sampling steps is the first step number. Wherein, the first step size is greater than the second step size, and the first step number is less than the second step number.

[0065] In this embodiment of the application, the first relationship may be determined by the first device or indicated by the second device.

[0066] In this embodiment of the application, the first device pre-sets multiple sets of service quality parameters and backsampling parameters to obtain a first relationship.

[0067] In one example, the first device is a terminal device, which determines the target backsampling parameters based on the target quality of service parameters and the first relationship.

[0068] In one example, the second device is a network device. The second device determines the target backsampling parameters based on the target quality of service parameters and the first relationship, and instructs the target backsampling parameters to the first device.

[0069] In one example, the first device is a network device, which determines the target backsampling parameters based on the target quality of service parameters and the first relationship.

[0070] In this embodiment of the application, the target quality of service parameters are directly mapped to the target backsampling parameters through the first relationship, which can improve the efficiency of determining the target backsampling parameters.

[0071] In some embodiments, the target backsampling parameters are determined based on the target quality of service parameters using a parameter prediction model.

[0072] In this embodiment, the parameter prediction model is a pre-trained decision network used to predict the target backsampling parameters. The input to the parameter prediction model is the target quality of service parameter, and the output is the target backsampling parameter.

[0073] In this embodiment of the application, the first device or the second device inputs the target quality of service parameters into the parameter prediction model to obtain the output target backsampling parameters.

[0074] In this embodiment of the application, the parameter prediction model can be trained by a first device and deployed to the first device, or trained by a second device and deployed to the second device, or trained by a second device and sent to the first device, and the first device deploys the parameter prediction model to the first device.

[0075] In one example, where the first device is a terminal device, the first device uses a parameter prediction model to determine the target backsampling parameters based on the target quality of service parameters.

[0076] In one example, where the first device is a terminal device and the second device is a network device, the second device uses a parameter prediction model to determine the target backsampling parameters based on the target quality of service parameters, and sends the target backsampling parameters to the first device.

[0077] In one example, where the first device is a network device, the first device uses a parametric prediction model to determine the target backsampling parameters based on the target quality of service parameters.

[0078] In this embodiment of the application, the target backsampling parameters can be predicted more accurately based on the target quality of service parameters using a parameter prediction model.

[0079] In some embodiments, the target reverse sampling parameters are determined based on the sampling mode of the first service; the sampling mode of the first service is determined based on the target quality of service parameters, and the sampling mode includes one of the following: fast sampling mode, low-power sampling mode, and robust sampling mode.

[0080] In this embodiment, the sampling mode is used to describe the performance of backsampling, and different sampling modes can be adapted to different quality of service parameters.

[0081] In this embodiment, the pre-set sampling modes include at least one of the following: fast sampling mode, low-power sampling mode, and robust sampling mode; each sampling mode includes a sampling step size and a number of sampling steps. In one example, the sampling step size corresponding to the fast sampling mode is 20, and the number of sampling steps is 5.

[0082] In this embodiment, a sampling mode for a first service that is compatible with the target quality of service (QoS) parameter is determined from a plurality of pre-set sampling modes, and a target reverse sampling parameter is determined based on the sampling mode of the first service. For example, if the target QoS parameter indicates that the first service is a low-latency service, the sampling mode for the video with the target QoS parameter can be determined to be a fast sampling mode. Therefore, the sampling mode for the first service is determined to be a fast sampling mode, and the target reverse sampling parameter is the reverse sampling parameter corresponding to the fast sampling mode.

[0083] In some embodiments, when the first device is a terminal device, the first device determines the sampling mode of the first service based on the target quality of service parameters; and determines the target backsampling parameters based on the sampling mode of the first service.

[0084] In some embodiments, when the first device is a terminal device and the second device is a network device, the second device determines the sampling mode of the first service based on the target quality of service parameters; determines the target backsampling parameters based on the sampling mode of the first service, and instructs the target backsampling parameters to the first device.

[0085] In some embodiments, when the first device is a network device, the first device determines the sampling mode of the first service based on the target quality of service parameters; and determines the target backsampling parameters based on the sampling mode of the first service.

[0086] In some embodiments, the target backsampling parameters are determined based on the target quality of service parameters and the channel quality of the first channel.

[0087] In this embodiment of the application, channel quality can be described by a Channel Quality Indicator (CQI).

[0088] In some embodiments, when the first device is a terminal device, the terminal device feeds back the channel quality of the first channel to the network device through the uplink channel.

[0089] In some embodiments, when the first device is a network device, the network device estimates the channel quality of the first channel.

[0090] In this embodiment, a mapping relationship can be established between quality of service parameters, channel quality, and backsampling parameters. Based on the target quality of service parameters, the channel quality of the first channel, and the mapping relationship between the quality of service parameters, channel quality, and backsampling parameters, the target backsampling parameters can be determined.

[0091] In this embodiment of the application, the same quality of service parameter can correspond to different backsampling parameters depending on the channel quality.

[0092] In one example, the quality of service parameter is a first quality of service parameter, the channel quality of the first channel is a first value, and the corresponding backsampling parameter is a first sampling parameter; the quality of service parameter is a second quality of service parameter, the channel quality of the first channel is a first value, and the corresponding backsampling parameter is a second sampling parameter.

[0093] In the embodiments of this application, the same quality of service parameter may correspond to one or more reverse sampling parameters depending on the channel quality, and this application does not limit this.

[0094] In this embodiment, service quality parameters and channel quality are considered comprehensively, so that the target backsampling parameters can not only meet the requirements of the target service quality parameters, but also adapt to the current channel quality, thereby improving the accuracy of subsequent backsampling inference based on the target backsampling parameters.

[0095] In some embodiments, the first channel is a downlink channel; the target quality of service parameters are determined based on the target service transmission channel identifier and the second relationship of the data of the first service, wherein the second relationship is a mapping relationship between the service transmission channel identifier and the quality of service parameters.

[0096] In this embodiment, the second relationship includes a mapping relationship between one or more sets of service transmission channel identifiers and service quality parameters. The second relationship can also be referred to as a service quality parameter mapping table. In some embodiments, the second relationship can be pre-set by the first device, or it can be set by a network device and sent to the first device.

[0097] In this embodiment of the application, the first channel is a downlink channel, the first device is a terminal device, the second relationship is stored inside the first device, and the first device queries the second relationship based on the target service transmission channel identifier of the data of the first service to determine the target quality of service parameter corresponding to the target service transmission channel identifier in the second relationship.

[0098] In this embodiment of the application, when the first device determines the target quality of service parameters, the target service transmission channel identifier may be indicated by the second device.

[0099] In one possible embodiment, where the first device is a terminal device and the second device is a network device, when scheduling downlink resources, the network device determines the target service transmission channel identifier corresponding to the downlink resource, adds an indication field of the target service transmission channel identifier to the scheduling information sent to the terminal device, so that the terminal device can obtain the target service transmission channel identifier of the data of the first service, and the terminal device determines the target quality of service parameters based on the target service transmission channel identifier.

[0100] In some embodiments, the service transmission channel identifier includes at least one of the following: logical channel identifier; service flow identifier; data radio bearer identifier.

[0101] In this embodiment of the application, the second relationship may be a mapping relationship between one or more of the logical channel identifier, service flow identifier, and data radio bearer identifier, and between service quality parameters.

[0102] In one example, the second relationship could be a mapping between logical channel identifiers and quality of service parameters.

[0103] In one example, the second relationship could be a mapping between logical channel identifiers, traffic flow identifiers, and quality of service parameters.

[0104] In this embodiment, the first device determines the target quality of service parameters based on the target service transmission channel identifier and a second relationship of the data of the first service. The second relationship is a mapping relationship between the service transmission channel identifier and the quality of service parameters. In this way, the first device can quickly obtain the target quality of service parameters.

[0105] In some embodiments, the first device is a network device and the first channel is an uplink channel; or, the first device is a terminal device and the first channel is a downlink channel. The target service quality parameters are determined based on the service type of the first service.

[0106] In this embodiment of the application, the target quality of service parameter is determined by the network device based on the service type of the first service when scheduling resources.

[0107] In this embodiment, the network device possesses global service awareness capabilities, enabling it to obtain target quality of service parameters matching the service type of the first service when scheduling resources. The target quality of service parameters can be determined at the MAC layer or RRC layer of the network device based on the service type of the first service.

[0108] In some embodiments, the first device is a network device, the second device is a terminal device, and the first channel is an uplink channel; the terminal device sends a first pilot signal and data of a first service to the network device; the network device receives the first pilot signal and data of the first service sent by the terminal device; the network device determines a target quality of service parameter based on the service type of the first service; determines a target backsampling parameter based on the target quality of service parameter; performs back denoising inference on the first pilot signal received on the first channel using a diffusion model based on the target backsampling parameter to determine a second pilot signal after denoising the first pilot signal; determines the channel state information of the first channel based on the second pilot signal; and demodulates the data of the first service received on the first channel based on the channel state information.

[0109] In some embodiments, the first device is a terminal device, the second device is a network device, and the first channel is a downlink channel; the network device sends a first pilot signal and data of a first service to the terminal device; the network device determines a target quality of service parameter based on the service type of the first service; determines a target backsampling parameter based on the target quality of service parameter; and sends backsampling parameter indication information to the terminal device based on the target backsampling parameter; the terminal device receives the backsampling parameter indication information, decodes the backsampling parameter indication information, and obtains the target backsampling parameter; the terminal device uses a diffusion model to perform backsampling inference on the first pilot signal received on the first channel based on the target backsampling parameter to determine a second pilot signal after denoising the first pilot signal; determines the channel state information of the first channel based on the second pilot signal; and demodulates the data of the first service received on the first channel based on the channel state information.

[0110] In some embodiments, the first channel is an uplink channel, and the first device is a network device; and / or, the first channel is a downlink channel, and the first device is a terminal device; based on Figure 1 The signal processing method shown in this application embodiment further includes: determining the target backsampling parameters corresponding to the diffusion model based on the target quality of service parameters of the first service.

[0111] In this embodiment of the application, when the first device is a network device and the second device is a terminal device, the first channel is an uplink channel, which is estimated by the first device. Based on the target quality of service parameters of the first service, the first device determines the target backsampling parameters corresponding to the diffusion model, and based on the target backsampling parameters, uses the diffusion model to perform back denoising inference on the first pilot signal sent by the second device to determine the second pilot signal after denoising the first pilot signal. Based on the second pilot signal, the channel state information of the first channel is determined. Based on the channel state information, the data of the first service received on the first channel is demodulated.

[0112] In this embodiment, when the first device is a terminal device and the second device is a network device, the first channel is a downlink channel, which is estimated by the first device. The first device obtains the target backsampling parameters corresponding to the diffusion model. The target backsampling parameters are related to the target quality of service parameters of the first service, which are used to indicate the performance requirements of the first service. Based on the target backsampling parameters, the diffusion model is used to perform inverse denoising inference on the first pilot signal received on the first channel to determine the second pilot signal after denoising the first pilot signal. Based on the second pilot signal, the channel state information of the first channel is determined. Based on the channel state information, the data of the first service received on the first channel is demodulated.

[0113] In some embodiments, based on Figure 1 The signal processing method shown in this application embodiment further includes: determining the target backsampling parameters corresponding to the diffusion model based on the target quality of service parameters at the Radio Resource Control (RRC) layer or the Media Access Control (MAC) layer; and sending the target backsampling parameters to the physical layer of the first device through the RRC layer or the MAC layer.

[0114] In this embodiment, the first device can be a terminal device or a network device. The RRC layer or MAC layer of the first device determines target backsampling parameters based on target quality of service parameters. The RRC layer or MAC layer of the first device sends the target backsampling parameters to the physical layer of the first device through its internal interface. Based on the target backsampling parameters, the physical layer of the first device performs inverse denoising inference on the first pilot signal received on the first channel using a diffusion model to determine the second pilot signal. Based on the second pilot signal, the physical layer of the first device determines the channel state information of the first channel. Based on the channel state information, the data of the first service received on the first channel is demodulated. In some embodiments, the internal interface can be an internal application programming interface (API), shared memory, or hardware registers.

[0115] In this embodiment, the target backsampling parameters are written into the shared memory area, hardware control register, or inter-layer primitive interface through the RRC layer or MAC layer of the first device, so that the physical layer of the first device obtains the target backsampling parameters from the shared memory area, hardware control register, or inter-layer primitive interface before performing backsampling inference on the first pilot signal.

[0116] In some embodiments, the first channel is a downlink channel, and the first device is a terminal device; based on Figure 1 The signal processing method shown in the embodiments of this application further includes: The system receives backsampling parameter indication information sent by the network device. The backsampling parameter indication information is used to indicate the target backsampling parameters.

[0117] This application provides a signal processing method, wherein the first channel is a downlink channel and the first device is a terminal device; for example... Figure 4 As shown, it includes: S401. The network device sends reverse sampling parameter indication information to the terminal device; S402. The terminal device receives the reverse sampling parameter indication information sent by the network device. The reverse sampling parameter indication information is used to indicate the target reverse sampling parameters.

[0118] In this embodiment of the application, when the first device is a terminal device and the second device is a network device, the second device sends back sampling parameter indication information to the first device; the first device receives the back sampling parameter indication information sent by the second device and determines the target back sampling parameter based on the back sampling parameter indication information.

[0119] In this embodiment of the application, the first device includes a first communication module, and the second device includes a second communication module; the second communication module generates backsampling parameter indication information based on the target backsampling parameters and sends the backsampling parameter indication information to the first communication module; the first communication module receives the backsampling parameter indication information sent by the second communication module, decodes the backsampling parameter indication information, and determines the target backsampling parameters.

[0120] In this embodiment, the second device determines a target quality of service parameter based on the service type of the first service; determines a target reverse sampling parameter based on the target quality of service parameter, and generates reverse sampling parameter indication information based on the target reverse sampling parameter; the second device sends the reverse sampling parameter indication information to the first device; the first device receives the sampling parameter indication information, decodes the sampling parameter indication information to obtain the target reverse sampling parameter, and, based on the target reverse sampling parameter, uses a diffusion model to perform reverse denoising inference on the first pilot signal received on the first channel to determine the second pilot signal after denoising the first pilot signal.

[0121] In this embodiment, the timing relationship between the first device receiving the reverse sampling parameter indication information sent by the second device and receiving the first pilot signal and the data of the first service is not limited; in one example, the first device may receive the reverse sampling parameter indication information first; in another example, the first device may receive the first pilot signal and the data of the first service first.

[0122] In this embodiment of the application, the target backsampling parameters can be determined at the MAC layer or RRC layer of the network device.

[0123] In this embodiment of the application, the first device obtains a target backsampling parameter that better matches the target quality of service parameter by receiving backsampling parameter indication information, thereby improving the accuracy of backsampling inference on the first pilot signal received on the first channel.

[0124] In some embodiments, based on Figure 4 The signal processing method shown above, in step S402, the first device receives the reverse sampling parameter indication information sent by the network device, including: receiving downlink control information (DCI) sent by the network device, wherein the DCI includes the reverse sampling parameter indication information.

[0125] Accordingly, in step S401 above, the network device sends reverse sampling parameter indication information to the first device, including: the network device sends downlink control information (DCI) to the first device.

[0126] In this embodiment of the application, the network device carries back sampling parameter indication information in the downlink control information (DCI) sent to the terminal device, so that the terminal device receives the DCI sent by the network device, obtains the back sampling parameter indication information by decoding the DCI, and determines the target back sampling parameter based on the back sampling parameter indication information.

[0127] In this embodiment of the application, the second communication module of the network device generates the DCI and sends the DCI to the first communication module of the terminal device through the second communication module; the first communication module of the terminal device receives the DCI and parses the DCI to obtain the reverse sampling parameter indication information.

[0128] In this embodiment of the application, the network device can send DCI to the terminal device through a millimeter wave channel or a low-frequency band channel.

[0129] In this embodiment of the application, the target backsampling parameters of the diffusion model are dynamically indicated by the DCI signaling sent by the network device, so that the terminal device can dynamically adjust the target backsampling parameters of the diffusion model, thereby balancing the accuracy and inference efficiency of the back denoising inference process.

[0130] In some embodiments, the backsampling indication information includes a target configuration identifier, based on Figure 4 The signal processing method shown in this application embodiment further includes: receiving a candidate configuration set sent by a network device, the candidate configuration set including at least one set of backsampling parameters and configuration identifiers of each set of backsampling parameters, the candidate configuration set being used to determine target backsampling parameters with the target configuration identifier.

[0131] Accordingly, the second device is a network device, and the second device sends a candidate configuration set to the first device.

[0132] In this embodiment, the candidate configuration set consists of at least one set of backsampling parameters and configuration identifiers for each set of backsampling parameters pre-set by the second device. In some embodiments, the candidate configuration set is set at the RRC layer or MAC layer of the first device.

[0133] In this embodiment of the application, the network device determines a set of backsampling parameters and configuration identifiers of the backsampling parameters that best match the target quality of service parameters from the candidate configuration set; the terminal device determines the configuration identifier of the backsampling parameters in the set of backsampling parameters and configuration identifiers of the backsampling parameters as the target configuration identifier.

[0134] In this embodiment, the second device sends a candidate configuration set to the first device; the first device receives the candidate configuration set sent by the second device and stores the candidate configuration set in a local database; the second device determines a target configuration identifier, generates backsampling indication information based on the target configuration identifier, and sends the backsampling indication information to the first device; the first device receives the backsampling indication information, parses the backsampling indication information, and obtains the target configuration identifier; the first device determines the target backsampling parameters corresponding to the target configuration identifier from the candidate configuration set based on the target configuration identifier.

[0135] In this embodiment of the application, the candidate configuration set includes at least one set of backsampling parameters and configuration identifiers for each set of backsampling parameters; the first device determines the target backsampling parameters from the candidate configuration set based on the target configuration identifier.

[0136] In one example, the candidate configuration set includes the following sets of backsampling parameters and configuration identifiers for each set of backsampling parameters: The configuration identifier is set to 0, and the reverse sampling parameter is the first sampling parameter; The configuration identifier is set to 1, and the reverse sampling parameter is the second sampling parameter; The configuration identifier is 2, and the reverse sampling parameter is the third sampling parameter.

[0137] The first device receives a target configuration identifier of 1 sent by the second device. Based on the target configuration identifier and the candidate configuration set, the target backsampling parameter is determined as the second sampling parameter.

[0138] In this embodiment of the application, the candidate configuration set includes at least one candidate configuration subset, and each candidate configuration subset includes at least one set of reverse sampling parameters corresponding to a service transmission channel identifier and configuration identifiers of each set of reverse sampling parameters; the first device first determines the target candidate configuration subset corresponding to the target service transmission channel identifier from the candidate configuration set according to the target service transmission channel identifier; and then determines the target reverse sampling parameters corresponding to the target configuration identifier from the target candidate configuration subset based on the target configuration identifier.

[0139] In one example, the candidate configuration set includes a first candidate configuration subset, a second candidate configuration subset, and a third candidate configuration subset.

[0140] The first candidate configuration subset includes: The first service transmission channel identifier is configured as 0, and the reverse sampling parameter is the first sampling parameter; The first service transmission channel identifier is configured as 1, and the reverse sampling parameter is the second sampling parameter.

[0141] The second candidate configuration subset includes: The second service transmission channel identifier is configured as 0, and the reverse sampling parameter is the third sampling parameter. The second service transmission channel identifier is configured as 1, and the reverse sampling parameter is the fourth sampling parameter.

[0142] The third candidate configuration subset includes: The third service transmission channel identifier is configured as 0, and the reverse sampling parameter is the fifth sampling parameter. The third service transmission channel identifier is configured as 1, and the reverse sampling parameter is the sixth sampling parameter.

[0143] The first device determines the target service transmission channel identifier as the first service transmission channel identifier, and determines the target candidate configuration subset corresponding to the target service transmission channel identifier as the first candidate configuration subset; the first device determines the target configuration identifier as 1; and determines the target reverse sampling parameter as the second sampling parameter based on the target configuration identifier and the candidate configuration subset.

[0144] In some embodiments, receiving the candidate configuration set sent by the network device includes receiving an RRC message sent by the network device, wherein the RRC message includes the candidate configuration set.

[0145] Accordingly, the second device is a network device, and the second device sends an RRC message to the first device. The RRC message includes a candidate configuration set.

[0146] In this embodiment of the application, the second device carries a candidate configuration set in the RRC message sent to the first device; the first device receives the RRC message sent by the second device and obtains the candidate configuration set by parsing the RRC message.

[0147] In this embodiment of the application, the second communication module of the second device generates an RRC message based on the candidate configuration set and sends the RRC message to the first communication module of the first device; the first communication module of the first device receives the RRC message and obtains the candidate configuration set by decoding the RRC message.

[0148] In this embodiment of the application, the network device can send RRC messages to the terminal device through a millimeter wave channel or a low-frequency band channel.

[0149] The following describes the application of the embodiments of this application in a real-world scenario.

[0150] With the rapid development of generative artificial intelligence (AI), its application potential in the field of wireless communication is being gradually explored. In particular, generative models such as diffusion models have shown significant advantages in handling noise and reconstructing structured signals. This has brought new ideas to signal processing methods in related technologies, especially in complex, low signal-to-noise ratio scenarios, where the prior modeling capabilities of generative models can be used to improve reconstruction accuracy.

[0151] Meanwhile, with the continuous advancement of sixth-generation (6G) mobile communication technology, millimeter-wave (mmWave) communication (high-frequency band and large bandwidth) has become a key technology for achieving ultra-high-speed transmission due to its extremely wide available spectrum resources. To overcome the severe path loss in the millimeter-wave band, communication systems are typically equipped with massive MIMO (Multi-Input Multiple-Output) antenna arrays, utilizing beamforming technology to improve coverage and link reliability. However, the performance of millimeter-wave massive MIMO systems is highly dependent on high-precision channel state information. Due to the high dimensionality, rapid time-varying nature, and susceptibility to obstruction of millimeter-wave channels, coupled with the massive number of parameters brought by massive MIMO antenna arrays, obtaining accurate channel state information at the receiver presents a severe challenge, especially in scenarios with low signal-to-noise ratio (SNR) or limited pilot overhead. For the millimeter-wave channel estimation problem, related technologies are mainly divided into linear estimation algorithms and deep learning-based methods, but both suffer from the following technical bottlenecks in practical applications: (1) Performance limitations of linear estimation methods: Linear estimation methods (such as LMMSE) rely on accurate channel statistical characteristics, which are difficult to obtain in real time in the rapidly changing millimeter-wave environment, and involve high-dimensional matrix inversion, resulting in extremely high computational complexity. Although compressed sensing (CS) based methods (such as OMP and AMP) utilize the angular domain sparsity of millimeter-wave channels, noise will severely damage the sparse structure of the signal under low signal-to-noise ratio (Low SNR) conditions, leading to a sharp decrease in reconstruction probability, which makes it difficult to meet the high reliability requirements of 6G systems.

[0152] (2) Limitations of deep learning methods in related technologies: End-to-end deep learning methods in related technologies typically require retraining the model for different SNRs and lack adaptability to channel conditions.

[0153] Generative AI, particularly diffusion models, has been introduced into the field of channel estimation to address noise problems due to its superior distribution modeling capabilities and noise resistance. However, diffusion model-based channel estimation schemes in related technologies suffer from significant rigidity: Fixed sampling strategy: Most works employ a fixed maximum diffusion step count T (e.g., a fixed 50 or 100 steps) in the reverse generation stage. This mechanism cannot be adjusted based on real-time received signal quality (e.g., SNR). Inefficient resource utilization: In high SNR scenarios, the channel is relatively easy to recover, and excessive sampling steps lead to wasted computational resources and unnecessary inference delays; while in low SNR scenarios, a fixed number of steps may affect accuracy due to incomplete denoising.

[0154] (3) Lack of Quality of Service (QoS) Awareness Cross-Layer Mechanism: This is a key pain point faced by related technologies. Channel estimation schemes in related technologies typically only minimize the mean square error (MSE) as a single optimization objective, ignoring the differentiated QoS requirements of various vertical industry applications in 6G networks. This leads to a severe disconnect between physical layer processing and upper-layer service requirements, for example: Ultra-Reliable Low-Latency Communication (URLLC): This type of service is extremely sensitive to air interface latency (such as industrial control). In related technologies, if the diffusion model uses complex long-step sampling, its huge iterative computation will cause the inference time to exceed the latency budget, resulting in communication interruption.

[0155] Massive Machine-Type Communications (mMTC): These services (such as IoT sensors) are limited by battery capacity and are extremely sensitive to power consumption. The high-precision backsampling process involves a large number of neural network operations. If the number of sampling steps is not dynamically reduced according to actual needs, it will lead to a sharp increase in the power consumption of terminal devices and shorten the life of the devices.

[0156] Based on the above description, the relevant technologies lack an adaptive mechanism for transceiver linkage, making it impossible for the receiver to dynamically adjust the inference strategy of the channel estimation model according to the current channel conditions and the specific QoS requirements of the service (such as latency priority, power consumption priority, or accuracy priority). To address this problem, this application proposes a service demand-aware millimeter-wave system signal processing method based on a diffusion model. By introducing a multi-dimensional step-matching mechanism, a mapping relationship is established between channel quality, service requirements, and model inference parameters, achieving an adaptive balance between computational overhead, processing latency, and energy consumption while ensuring reconstruction accuracy.

[0157] This application provides the following two solutions in its embodiments: Option 1: The terminal device queries the locally stored QoS parameter mapping table (corresponding to the second relationship in the aforementioned embodiments) based on the Logical Channel Identity (LCID), QoS Stream ID, or Data Radio Bearer Identity (DRB ID) of the currently transmitted data to determine the service quality requirements of the current service. Combining these service quality requirements, the terminal device determines the configuration parameters of the diffusion model (corresponding to the target backsampling parameters in the aforementioned embodiments, such as the target sampling steps) through preset policy rules (corresponding to the first relationship in the aforementioned embodiments) or an AI decision network (corresponding to the parameter prediction model in the aforementioned embodiments). and target sampling step size (or the sampling mode of the diffusion model); based on the configuration parameters of the read diffusion model, perform a reverse diffusion sampling process and channel estimation on the received downlink pilot signal (such as DMRS or CSI-RS).

[0158] Option 2: The terminal device receives a semi-static candidate set configuration of the diffusion model sent by the network device (corresponding to the candidate configuration set in the previous embodiment); the terminal device receives dynamic parameter adjustment information of the diffusion model sent by the network device (corresponding to the backsampling parameter indication information in the previous embodiment); the terminal device selects the configuration parameters of the diffusion model from the semi-static candidate set according to the dynamic parameter adjustment information; and performs a backsampling sampling process and channel estimation on the received downlink pilot signal (such as DMRS or CSI-RS) according to the read configuration parameters of the diffusion model.

[0159] In this embodiment, channel estimation is a fundamental and core component of a wireless communication system. Its goal is to acquire or evaluate the propagation characteristics of a signal during transmission between the transmitter and receiver, i.e., channel state information (CSI). Accurate CSI is a prerequisite for implementing communication functions such as coherent detection, beamforming, resource allocation, and interference management. Channel estimation typically includes the following main steps: Step S11: The transmitter sends a known pilot sequence.

[0160] Step S12: The receiver receives the pilot signal. The received signal is the version of the pilot sequence after actual channel fading and additional noise pollution.

[0161] Step S13: The receiver uses the received signal and the known pilot sequence to calculate and recover the channel coefficients using a specific estimation algorithm (such as the least squares (LS) method or the linear least mean square error (LMMSE) algorithm).

[0162] Step S14: The recovered channel state information is used to guide subsequent signal processing procedures, such as channel equalization, demodulation, or feedback to the transmitter for precoding.

[0163] In this embodiment, diffusion models are a type of generative model that has emerged in recent years. They model and reconstruct complex data distributions by simulating the gradual addition and removal of noise from data. Figure 3 As shown, a learnable reverse generation chain is used to progressively "reconstruct" noisy samples into structured data. Diffusion models typically include the following steps: Step S21: Forward diffusion process.

[0164] To the original data sample ( Gaussian noise is gradually added to the data. This process is fixed and mathematically easy to follow, eventually causing the original data to completely degenerate into a random noise distribution. The noise addition process is described in formula (1): (1); in, It is the first Step-by-step noisy data, It is random Gaussian noise. , , It is the variance of the forward process.

[0165] Step S22: Learn the reverse process. Train a neural network (e.g., a time-dependent U-Net structure) to learn the conditional probability distribution in each diffusion step, i.e., predict and remove the current state. The noise component in the equation. The training loss function is given by equation (2): (2); in, It is a neural network.

[0166] Step S23: Iterative sampling generation.

[0167] During the inference phase, from randomly sampled pure Gaussian noise Initially, a trained neural network is used for iterative noise reduction. The model refines the data step by step through hundreds to thousands of steps, eventually generating a new sample with a distribution similar to the original data. The sampling formula is shown in formula (3): (3); in, This refers to the standard normal distribution Sampling random noise, and Same dimensions. This refers to the noise standard deviation at time step t-1.

[0168] The terms used in the embodiments of this application are explained as follows: Quality of Service (QoS) attributes: A set of parameters indicating the performance requirements of communication services. In the embodiments of this application, it refers to indicators related to channel estimation accuracy and latency, including but not limited to the 5G QoS identifier (5QI), packet delay budget (PDB), block error rate target (BLER Target), and service priority level.

[0169] Diffusion model: A probabilistic model based on generative artificial intelligence, comprising a forward diffusion process and a backsampling process. In the embodiments of this application, it refers to a neural network model used by the physical layer to reconstruct channel state information (CSI) from noisy received signals.

[0170] Backsampling step size: The span of denoising iterations in the inference phase (backward process) of the diffusion model. The larger the step size, the faster the inference speed, but the relatively lower the accuracy.

[0171] Backsampling steps: The number of denoising iterations performed by the diffusion model during the inference phase (backward process). More steps result in a more accurate channel matrix, but also higher computational latency and energy consumption; fewer steps result in faster inference speed, but relatively lower accuracy.

[0172] Logical Channel: The interface between the MAC layer and the RLC layer, used to distinguish different types of service data streams. The MAC layer identifies the QoS attributes of the service by recognizing the Logical Channel Identifier (LCID).

[0173] Configuration signaling: Control information sent by the network device to the terminal device to instruct the terminal to adjust the channel estimation strategy. In the embodiments of this application, it includes semi-static Radio Resource Control (RRC) signaling or Media Access Layer Control Unit (MACCE) and dynamic Downlink Control Information (DCI).

[0174] Candidate configuration set: A predefined set of diffusion model parameter combinations, which is usually pre-configured by RRC signaling.

[0175] Dynamic indication information: The index field carried in DCI is used to activate a specific target configuration from the candidate configuration set.

[0176] The channel estimation architecture based on the diffusion model proposed in this application includes a transmitter (Tx) and a receiver (Rx), and a pre-trained AI model is deployed on the receiver side. It mainly includes the following two stages: Phase 1: Offline Model Training Phase (Offline Phase) includes the following steps: Step S31: Gradually add noise to the original signal to form a diffusion sequence; Load the raw signal from the dataset. At each time step To the original signal Gaussian noise is injected into the signal to gradually bring it closer to a pure noise state, as shown in formula (4): (4); in, , This is a preset noise variance sequence.

[0177] Step S32: Learn the inverse denoising mapping; Build a neural network To predict in the first Noise components in step signal The training objective of the model is to minimize the mean squared error between the predicted noise and the actual noise. See formula (5): (5); By optimizing the objective function, the neural network gradually learns the ability to recover a clean signal from a noisy signal.

[0178] Step S33: Model optimization and convergence.

[0179] The model parameters are updated using gradient descent until the neural network can accurately estimate the noise component in each diffusion step. After training, the neural network can be used for the backpropagation process.

[0180] The aforementioned offline model training function can be deployed on network OAM systems, OTT servers, or network nodes (core network or base stations). This offline model training function progressively updates model parameters by gradually adding noise to the original signal and learning an inverse denoising mapping. For downlink data transmission, the terminal device acts as the receiver, and the base station (network device) acts as the transmitter; the offline model training function sends the trained model to the terminal device. For uplink output transmission, the offline model training function sends the trained neural network model to the base station.

[0181] The online channel estimation phase (online phase - inference phase) includes the following steps: Step S41: Step matching; The receiver determines the noise power of the received signal (the signal actually received after passing through the channel). Calculate the initial number of steps for back diffusion using the following formula. See formula (6): (6); Step S42: Backsampling; From the received signal (corresponding to the first) Starting from the state of step 0, the trained neural network is used for iterative denoising to gradually restore the pure pilot response to step 0. .from Start iterative denoising until t=0.

[0182] (7); Step S43: Channel recovery.

[0183] The denoised signal is subjected to inverse transform and least squares processing to obtain the estimated channel matrix. .

[0184] In this embodiment of the application, the model training algorithm flow of the signal processing method based on the diffusion model is as follows: Figure 5 As shown, the steps include S501 to S506 as follows: Step S501: Training begins; Step S502: Load data from the dataset; Step S503: The neural network outputs the predicted noise value; Step S504: Calculate the loss function; Step S505: Has the loss function converged? If yes, proceed to step S506; if no, proceed to step S502. Step S506: Training complete.

[0185] In this embodiment of the application, the model inference algorithm flow of the signal processing method based on the diffusion model is as follows: Figure 6 As shown, the steps include S601 to S608 as follows: Step S601: Reasoning begins; Step S602: Receive the signal and preprocess it; Step S603: Match the number of steps to obtain t=ts; Step S604: Perform sampling according to the sampling formula; Step S605: Calculate t = t - 1; Step S606: Is t less than 0? If yes, proceed to step S607; if no, proceed to step S604. Step S607: Signal post-processing; Step S608: Reasoning complete.

[0186] The architecture of the service quality-aware diffusion model signal processing method provided in this application embodiment is as follows: Figure 7 As shown, it includes a control plane 701, an execution plane 702, and an interaction plane 703. The functions of each plane are defined as follows: Control Plane 701: Used for service-aware global scheduling and policy decisions, enabling channel estimation strategies to adapt to current service requirements. The control plane mainly includes the following functions: Quality of Service (QoS) Awareness 711: Acquires contextual information associated with the data stream to be transmitted or received, including logical channel priority, 5QI, and latency sensitivity (QoS attributes). For example, for each QoS flow, the network device distributes the QoS parameters of that flow to the terminal device through the NAS layer. Each QoS flow is mapped to a corresponding Data Radio Bearer (DRB), and each DRB is configured with corresponding scheduling information, such as logical channel priority.

[0187] Strategy Mapping 712: Maintain the "Service Quality Parameters - Configuration Parameters" mapping table or run a lightweight decision network (corresponding to the parameter prediction model in the aforementioned embodiments) to determine the optimal working mode of the diffusion model (such as the number of sampling steps, number of sampling times, etc.) based on the service quality parameters.

[0188] Command 713: Generates configuration instructions for the diffusion model and sends them to the execution entity via the interaction plane.

[0189] Execution Plane 702: Deployed at the physical layer, it is used for specific signal processing and AI model inference, enabling channel estimation tasks to be completed under given resource constraints. The execution plane mainly includes the following functions: Signal reception and preprocessing 721: Receives wireless air interface signals, extracts pilot signals (such as CSI-RS or DMRS) and performs preliminary noise estimation.

[0190] Adaptive channel estimation 722: Receives configuration instructions from the control plane and dynamically adjusts the number of iterations of the reverse process of the diffusion model (e.g., from T steps to T' steps).

[0191] Interaction Plane 703: Used for information exchange between the control plane and the execution plane. Depending on the deployment mode, this plane includes two mechanisms: internal interface interaction and air interface signaling interaction. Internal interface interaction: Supports writing parameters such as sampling steps and noise scheduling table index from the MAC layer to the physical layer via internal API, shared memory, or hardware registers.

[0192] Air interface signaling interaction: RRC signaling is used to configure a predefined list of model parameters, and DCI signaling is used to activate a specific target configuration from a set of candidate configurations.

[0193] In this application embodiment, the above architecture has two implementation schemes: Scheme 1: internal cross-layer optimization, the MAC layer sends "model configuration instructions" to the physical layer processing unit through the internal interface; Scheme 2: network-side signaling control: the network side controls through RRC configuration and DCI dynamic triggering.

[0194] For Scheme 1, in one possible implementation, this architecture is applied to downlink channel estimation on the terminal device side. The control entity is the MAC layer entity or RRC layer entity of the terminal device; the execution entity is the physical layer of the terminal device. When the terminal device receives downlink data, the MAC layer or RRC layer, based on the decoded downlink control information or logical channel configuration, for example, obtains the service quality parameters or scheduling priority corresponding to the data through LCID or QFI, thereby identifying the current service requirement (e.g., low-latency service), and then instructs the physical layer to adopt a "fast sampling mode" (e.g., step size) through an internal interface. (t=20, sampling times N=5); the physical layer configures the diffusion model accordingly and quickly outputs channel state information to meet service requirements. In another possible implementation, this architecture is applied to uplink channel estimation on the network side. The control entity is the MAC layer entity or RRC layer entity of the base station; the execution entity is the physical layer entity of the base station. When the network device schedules uplink resources, the MAC layer knows the service type of the scheduled user and determines the service quality parameters corresponding to that service type. Based on the service quality parameters, the MAC layer or RRC layer determines that the optimal sampling model of the extended model is the "low-power sampling mode" and instructs the physical layer to adopt the "low-power sampling mode" (e.g., reduce the number of iteration steps), thereby significantly reducing the computing load and energy consumption on the network side when processing channel estimation for a large number of concurrent users.

[0195] For Scheme 2, one possible implementation is that the control entity resides on the network side, the execution entity resides on the terminal side, and the interaction is achieved through air interface signaling. The network device determines the sampling mode to be used by the terminal device (e.g., fast sampling mode or low-power sampling mode of the diffusion model when using backsampling with a diffusion model) or the sampling step size and sampling frequency of the diffusion model used by the terminal device, based on the service quality parameters of the service to be scheduled (such as QoS flow or DRB) and uplink channel quality feedback. The network device sends indication information to the terminal via air interface signaling. The terminal device parses the indication field in the signaling, retrieves the corresponding parameters from the candidate configuration set, and adjusts the backsampling parameters of the diffusion model for inference.

[0196] In this embodiment of the application, the above architecture supports diverse policy mapping mechanisms: Rule-based mapping: The control entity directly maps service quality parameters to sampling parameters of the diffusion model according to a pre-defined rule lookup table.

[0197] AI-based mapping: The control entity uses a lightweight decision network to predict the optimal sampling parameters by combining service quality requirements with real-time physical layer parameters (such as SNR and Doppler shift).

[0198] This application describes in detail a diffusion model signal processing method based on Quality of Service (QoS) awareness applied to the user equipment (UE) side. The control entity and execution entity are integrated within the same UE, achieving information interaction and control through a cross-layer interface within the device. The terminal-side architecture configuration for adaptive channel estimation based on internal cross-layer interaction is as follows: Figure 8 As shown, the terminal device is logically divided into a context-aware module, a policy generation module, an internal interface unit, and a computing module: Context Awareness Module 801: Deployed at the MAC or RRC layer of the terminal device, the Context Awareness Module 801 maintains and retrieves the mapping relationship between service transmission channels and quality of service parameters (this mapping relationship can be configured by the network or pre-configured by the terminal device), thereby providing contextual basis for subsequent policy formulation. Internally, this Context Awareness Module stores a pre-configured user equipment context or quality of service parameter mapping table. The quality of service parameter mapping table records the specific quality of service parameters corresponding to each logical channel identifier (LCID), QoS flow ID, or DRB ID, such as 5G QoS identifier (5QI), packet delay budget, block error rate target, and service priority. During terminal device operation, this Context Awareness Module receives scheduling information from the scheduler, identifies the logical channel identifier, QoS flow ID, or DRB ID of the data stream to be processed within the current transmission time interval, and queries the service flow mapping table accordingly to extract the quality of service parameters for the current service, then passes the quality of service parameters as output to downstream modules.

[0199] Policy Generation Module 802: Also deployed at the MAC or RRC layer, this module follows the context-aware module and serves as the decision-making core for cross-layer control. It translates abstract quality of service (QoS) parameters from upstream inputs into specific, physically-layer-executable configuration parameters for the diffusion model. This module incorporates a step-count matching mechanism, enabling it to determine the optimal configuration parameters for the diffusion model under current conditions based on rule-based table lookups or intelligent neural network predictions, according to the received QoS parameters (and optionally combined with channel state statistics fed back from the physical layer). Finally, the module generates a model configuration instruction containing specific configuration parameters (such as the number of backsampling iterations and sampling step size) and sends it to the internal interface.

[0200] Internal Interface 803: Located at the boundary between the MAC / RRC layer and the PHY layer, this internal interface establishes an information transmission channel between the control plane and the execution plane, enabling configuration commands to be transmitted from the MAC / RRC layer to the PHY layer in real time and accurately. Depending on the specific hardware implementation, this internal interface can be configured as a shared memory area, a hardware control register, or an inter-layer primitive interface. This interface receives model configuration commands output by the policy generation module and caches or passes them through to the physical layer, allowing the physical layer to obtain the latest configuration parameters before initiating diffusion model inference.

[0201] Calculation Module 804: Deployed at the physical layer, the calculation module acts as the execution entity for signal processing. It contains a pre-trained diffusion model neural network, capable of reconstructing channel state information from noisy data. This module simultaneously receives model configuration instructions from the interface and noisy received signals from the RF front-end. During processing, the module first initializes the inference parameters of the diffusion model according to the model configuration instructions. Then, it uses the received signal as the initial input for the inverse denoising inference process of the diffusion model, strictly adhering to the specified sampling step size and number of sampling steps to perform the inverse inference denoising process, gradually removing noise from the received signal and restoring the signal structure. Finally, it outputs a high-precision channel estimation result for subsequent signal demodulation or beamforming.

[0202] In this embodiment of the application, a complete adaptive downlink channel estimation and data reception process includes the following steps: Step S51: Service quality perception and strategy decision-making; The terminal device queries the locally stored service quality parameter mapping table based on the LCID, QoS stream ID, or DRB ID of the currently transmitted data to determine the service quality requirements of the current service. The control module (including the aforementioned context-aware module and policy generation module) determines the configuration parameters of the diffusion model based on the current QoS requirements and through preset policy rules or parameter prediction models. For example, the configuration parameters include, but are not limited to, the following possible scenarios: Scenario A (low latency priority): If the LCID corresponds to a low-latency communication service (e.g., 5QI=1), and the latency budget (PDB) is lower than a preset threshold (e.g., 10ms), the control module decides to adopt the "fast sampling mode" and sets the sampling step size. Number of samplings Scenario B (High Reliability Priority): If the LCID corresponds to a high-priority service, and the signal-to-noise ratio (SNR) reported by the physical layer is lower than a preset threshold (indicating a poor channel environment), the control module decides to adopt "robust mode" and sets the sampling step size. Number of samplings To maximize the noise reduction effect.

[0203] Step S52: Issuance of cross-layer instructions; The control module writes the determined configuration parameters to the specified register address of the internal interface unit, or sends them to the computing module through inter-layer configuration primitives.

[0204] Step S53: Adaptive channel estimation; Before initiating diffusion model inference, the computation module (physical layer) reads configuration parameters from the internal interface unit. The computation module then uses the read sampling step number N and sampling step size... The module t performs a back-diffusion sampling process on the received downlink pilot signal (such as DMRS or CSI-RS). Specifically, the computation module controls the neural network to perform N iterations of denoising to obtain the denoised reconstructed signal, and generates reconstructed channel state information based on the reconstructed signal.

[0205] Step S54: Data demodulation and feedback.

[0206] The computation module uses the generated channel state information to perform equalization and demodulation on the Physical Downlink Shared Channel (PDSCH). If demodulation is successful, the physical layer generates an ACK; if demodulation fails, it generates a NACK. The MAC layer can use the ACK / NACK feedback to calculate the block error rate (BLER) and dynamically update the preset policy rules accordingly (e.g., if the BLER increases, the number of steps N is appropriately increased in the next decision).

[0207] This application describes in detail a signal processing method based on network-side signaling control, where the control entity is deployed on the network device and the execution entity is deployed on the terminal device. The network device, based on global scheduling information and service quality parameters, instructs the terminal device to adjust the configuration parameters of its local diffusion model via radio air interface signaling (interaction plane). The architecture configuration of cooperative channel estimation based on network-side signaling control is as follows: Figure 9 As shown, this architecture configuration includes network devices and terminal devices: Network devices include: Control module 901: Corresponds to the MAC layer scheduler and RRC control unit of the network device. The control module has global service awareness capabilities, can know the service quality parameters (such as 5QI, latency budget) of the data to be scheduled, and maintain the "QoS parameter-model configuration" policy mapping logic.

[0208] The second communication module 902 is used to generate and send configuration signaling, including semi-static Radio Resource Control (RRC) signaling, MAC CE, and dynamic downlink control information (DCI).

[0209] User Equipment (UE) includes: First Communication Module 903: Used to decode signaling from network devices and parse configuration signaling for channel estimation.

[0210] Computing Module 904: Corresponds to the terminal physical layer. It internally deploys a diffusion model that can switch between different sampling parameters based on instructions from the network side.

[0211] In this embodiment of the application, the specific process of cooperative channel estimation based on network-side signaling control can be as follows: Figure 10 As shown, it includes the following steps: Step S1001: Semi-static candidate set configuration; During the connection establishment or reconfiguration phase, the network device's control module sends an RRC configuration message to the terminal device. This message contains a newly added AI channel estimation configuration list (AI-ChannelEst-ConfigList) (corresponding to the candidate configuration set in the aforementioned embodiments). The AI ​​channel estimation configuration list defines M groups of candidate configurations, for example: Configuration ID 0: sampling step size 10, Number of samplings (Suitable for low-latency scenarios). Configuration ID 1: Sampling step size 5. Number of samplings (Applicable to general scenarios). Configuration ID 2: Sampling step size 1. Number of samplings (Applicable to high reliability / weak coverage scenarios). After receiving this message, the terminal device stores the above candidate configuration set in its local context database.

[0212] Step S1002: Dynamic scheduling and strategy decision-making; In each scheduling slot, the network device's control module makes scheduling decisions for the downlink data to be transmitted. The control module identifies that the data stream currently scheduled to a specific terminal (e.g., UE_A) belongs to a low-latency service, and combines this with the current channel quality (e.g., the current Channel Quality Indicator (CQI) is good). Based on the combination of "service + channel quality," the control module determines which parameters the terminal device's diffusion model should use (e.g., the control module selects the target configuration identifier as configuration ID 0).

[0213] In this embodiment, the same quality of service parameter can correspond to different configuration parameters depending on the channel quality. For example, when the channel quality is good, the diffusion model uses a first sampling parameter; when the channel quality is poor, the diffusion model uses a second sampling parameter.

[0214] Step S1003: Dynamic signaling indication; The network device's communication module generates downlink control information (DCI). The DCI (e.g., DCI Format 1_1) includes an "AI Model Indicator Field" (corresponding to the target configuration identifier in the aforementioned embodiment). The network device sets the value of this indicator field to 00 (corresponding to configuration ID 0) and sends the DCI to the terminal device via the Physical Downlink Control Channel (PDCCH).

[0215] Step S1004: Terminal-side parsing and configuration; The terminal device's communication module decodes the DCI and extracts the value of the "AI model indicator domain". Based on this value, the terminal device queries the locally stored candidate configuration set to determine the diffusion model parameters for the current time slot (e.g., if the extracted "AI model indicator domain" value is 00, then the sampling step size is selected). t=10, sampling times N=5).

[0216] Step S1005: Adaptive inference execution.

[0217] The computing module (physical layer) of the terminal device initializes the diffusion model according to the configuration parameters determined in step S1004. The computing module uses the diffusion model to perform back-sampling denoising on the pilot signal of the current time slot to generate channel state information (CSI).

[0218] In this application, an adaptive estimation framework based on a step-matching mechanism is proposed to address the limitation of fixed sampling steps in diffusion models in related technologies. By modeling channel estimation as the reverse process of the diffusion model and introducing a step-matching mechanism, the diffusion process can dynamically adjust the starting point of reverse sampling according to the signal-to-noise ratio (SNR) of the received signal, thereby improving the robustness and efficiency of the diffusion model under different noise environments.

[0219] In this embodiment, by utilizing the progressive denoising characteristics of the diffusion model, under low SNR conditions (e.g., -10dB to 0dB), the normalized mean square error (NMSE) is superior to LMMSE, CS algorithm and direct mapping deep learning method in related technologies.

[0220] In this embodiment, a channel estimation method for a terminal device is provided. The terminal device acquires the Quality of Service (QoS) parameters associated with the data to be processed (data corresponding to the first service in the aforementioned embodiment); based on the QoS parameters, it determines the inference configuration parameters of the diffusion model (corresponding to the target backsampling parameters in the aforementioned embodiment), the inference configuration parameters including at least the number of iteration steps and the step size of the backsampling process; based on the inference configuration parameters, it performs back denoising inference on the pilot observation data (corresponding to the first pilot signal in the aforementioned embodiment) using a pre-trained diffusion model to generate a reconstructed signal (corresponding to the second pilot signal in the aforementioned embodiment); and determines the channel state information based on the reconstructed signal.

[0221] In some embodiments, the number of backsampling start time steps is determined. This includes: calculating the time step that minimizes the Kullback-Leibler (KL) divergence between the signal distribution during the forward diffusion process and the actual received signal distribution; specifically calculated using formula (6).

[0222] In some embodiments, obtaining the Quality of Service (QoS) attribute associated with the data stream to be processed includes: parsing scheduling information at the Media Access Control (MAC) layer of the terminal device to identify the Logical Channel Identifier (LCID) of the current data stream; and querying a preset context mapping table (corresponding to the second relationship in the aforementioned embodiments) based on the Logical Channel Identifier to determine the QoS attribute corresponding to the data stream.

[0223] In some embodiments, determining the inference configuration parameters of the diffusion model includes: the MAC layer of the terminal device generating a model configuration instruction based on QoS parameters; the MAC layer sending the model configuration instruction to the physical (PHY) layer through an internal interface; the internal interface includes shared memory, hardware control registers, or inter-layer primitive interfaces.

[0224] In some embodiments, obtaining the Quality of Service (QoS) parameters associated with the data stream to be processed includes: receiving configuration signaling from a network device; parsing the configuration signaling to obtain channel estimation policy indication information carried therein, and using the indication information as a control parameter characterizing the QoS parameters.

[0225] In some embodiments, the configuration signaling includes a first signaling and a second signaling; determining the inference configuration parameters of the diffusion model includes: receiving the first signaling, which is used to configure a candidate configuration set containing multiple sets of candidate parameters; receiving the second signaling, which contains a target configuration index; and retrieving the corresponding parameters from the candidate configuration set as inference configuration parameters according to the target configuration index.

[0226] In some embodiments, the first signaling is Radio Resource Control (RRC) signaling, and the second signaling is Downlink Control Information (DCI).

[0227] In some embodiments, the inference configuration parameters of the diffusion model are determined based on the QoS parameters, including any of the following methods: Method 1: Query a preset mapping rule table, and when the QoS parameters indicate low latency services, select a first iteration step number less than a preset threshold; when the QoS parameters indicate high reliability services, select a second iteration step number greater than a preset threshold; Method 2: Input the QoS attributes and the currently measured channel state statistics into the parameter decision neural network to obtain the recommended sampling parameters output by the neural network.

[0228] In some embodiments, the above-described inverse denoising inference using a pre-trained diffusion model on pilot observation data includes: using the pilot observation data as the initial state of the reverse diffusion process. Based on a determined number of iterations N, a denoising network and sampling equation are used to perform N iterations to gradually remove noise components.

[0229] This application provides a channel estimation apparatus, comprising: a communication module for receiving wireless signals; a memory for storing computer programs and pre-trained diffusion model parameters; and a processor coupled to the communication module and the memory for executing the computer program to implement the channel estimation method described above.

[0230] In some embodiments, the channel estimation device terminal equipment (UE) described above has a processor that includes a logically separate control module and a computing module; the control module is used to execute MAC layer protocols, obtain QoS parameters and generate configuration instructions; the computing module is used to execute physical layer protocols and run a diffusion model according to the instructions.

[0231] This application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the channel estimation method described above.

[0232] In this embodiment, a transceiver linkage strategy based on the quality of service parameters of the transmitting end is constructed. It not only depends on the physical layer channel state, but also further configures the denoising strategy of the diffusion model according to the quality of service parameters of the transmitting end, so as to better match the network requirements, reduce unnecessary waste of computing resources, and achieve a balance between performance and overhead.

[0233] Fourthly, to implement the above signal processing method, an embodiment of this application provides a device 1100 (a first device or a second device), such as... Figure 11As shown, it may include at least one processor 1101 and at least one transceiver 1102 coupled to at least one processor 1101. The transceiver 1102 may include at least one separate receiving circuitry and a transmitting circuitry, or at least one integrated receiving circuitry and a transmitting circuitry. The at least one processor 701 may be a central processing unit (CPU), a microprocessor (MPU), a digital signal processor (DSP), or a field-programmable gate array (FPGA), etc.

[0234] According to some embodiments of this application, when device 1100 is a first device, the first device includes a first transceiver; and A first processor, coupled to a first transceiver; the first processor is configured to: Obtain the target backsampling parameters corresponding to the diffusion model. The target backsampling parameters are related to the target quality of service parameters of the first service. The target quality of service parameters are used to indicate the performance requirements of the first service. Based on the target reverse sampling parameters, the diffusion model is used to perform reverse denoising inference on the first pilot signal received on the first channel to determine the second pilot signal after denoising the first pilot signal. Based on the second pilot signal, the channel state information of the first channel is determined; Based on channel state information, the data of the first service received on the first channel is demodulated.

[0235] In some embodiments, the target backsampling parameters are determined based on the target quality of service parameters and a first relationship; the first relationship is the mapping relationship between the quality of service parameters and the backsampling parameters.

[0236] In some embodiments, the target backsampling parameters are determined based on the target quality of service parameters using a parameter prediction model.

[0237] In some embodiments, the target reverse sampling parameters are determined based on the sampling mode of the first service; the sampling mode of the first service is determined based on the target quality of service parameters, and the sampling mode includes one of the following: fast sampling mode, low-power sampling mode, and robust sampling mode.

[0238] In some embodiments, the target backsampling parameters are determined based on the target quality of service parameters and the channel quality of the first channel.

[0239] In some embodiments, the target backsampling parameters include one of the following: target sampling step size and target sampling number.

[0240] In some embodiments, the first channel is a downlink channel; The target quality of service parameters are determined based on the target service transmission channel identifier and the second relationship of the data of the first service. The second relationship is the mapping relationship between the service transmission channel identifier and the quality of service parameters.

[0241] In some embodiments, the service transport channel identifier includes at least one of the following: Logical channel identifier; Business flow identifier; Data wireless bearer identifier.

[0242] In some embodiments, the first device is a network device and the first channel is an uplink channel; or, the first device is a terminal device and the first channel is a downlink channel; the target quality of service parameter is determined based on the service type of the first service.

[0243] In some embodiments, the target quality of service parameters include at least one of the following: Service quality parameters, packet latency budget, error rate target, and service priority.

[0244] The first channel is the uplink channel, and the first device is a network device; and / or... The first channel is a downlink channel, the first device is a terminal device; the first processor is configured to: Based on the target service quality parameters of the first service, the target backsampling parameters corresponding to the diffusion model are determined.

[0245] In some embodiments, the first processor is configured to: At the Radio Resource Control (RRC) layer or the Media Access Control (MAC) layer, the target backsampling parameters corresponding to the diffusion model are determined based on the target Quality of Service (QoS) parameters. The target backsampling parameters are sent to the physical layer of the first device via the RRC layer or MAC layer.

[0246] In some embodiments, the first channel is a downlink channel, and the first device is a terminal device; the first processor is configured to: Receive backsampling parameter indication information sent by the network device. The backsampling parameter indication information is used to indicate the target backsampling parameters.

[0247] In some embodiments, the first processor is configured to: Receive downlink control information (DCI) sent by the network device. The DCI includes reverse sampling parameter indication information.

[0248] In some embodiments, the backsampling indication information includes a target configuration identifier, and the first processor is configured to: The network device receives a candidate configuration set, which includes at least one set of backsampling parameters and configuration identifiers for each set of backsampling parameters. The candidate configuration set is used to determine the target backsampling parameters with the target configuration identifier.

[0249] In some embodiments, the first processor is configured to: Receive RRC messages sent by network devices. The RRC messages include a set of candidate configurations.

[0250] According to some embodiments of this application, when device 1100 is a second device, the second device includes a second transceiver; and A second processor, coupled to a second transceiver; the second processor is configured to: The first pilot signal and the data of the first service are transmitted to the first device on the first channel.

[0251] In some embodiments, the first channel is a downlink channel, the first device is a terminal device, the second device is a network device, and the second processor is configured to: Send back sampling parameter indication information to the first device. The back sampling parameter indication information is used to indicate the target back sampling parameters.

[0252] In some embodiments, the second processor is configured to: Send downlink control information (DCI) to the first device. The DCI includes inverse sampling parameter indication information.

[0253] In some embodiments, the second processor is configured to: A candidate configuration set is sent to the first device. The candidate configuration set includes at least one set of backsampling parameters and configuration identifiers for each set of backsampling parameters. The candidate configuration set is used to determine the target backsampling parameters with the target configuration identifier.

[0254] In some embodiments, the second processor is configured to: Send an RRC message to the first device. The RRC message includes a set of candidate configurations.

[0255] The descriptions of the above device embodiments are similar to those of the above method embodiments, and have similar beneficial effects. For technical details not disclosed in the device embodiments of this application, please refer to the descriptions of the method embodiments of this application for understanding.

[0256] It should be noted that, in the embodiments of this application, if the above-mentioned wireless communication method is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of this application, or the part that contributes to the related technology, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, mobile hard drives, read-only memory (ROM), magnetic disks, or optical disks. Thus, the embodiments of this application are not limited to any specific hardware and software combination.

[0257] Fifthly, to implement the above-mentioned wireless communication method, embodiments of this application provide an electronic device, including a memory and a processor. The memory stores a computer program that can run on the processor, and when the processor executes the program, it implements the steps in the wireless communication method provided in the above embodiments.

[0258] Sixthly, embodiments of this application provide a storage medium, namely a computer-readable storage medium, on which a computer program is stored, which, when executed by a processor, implements the steps in the wireless communication method provided in the above embodiments.

[0259] It should be noted that the descriptions of the storage medium and device embodiments above are similar to the descriptions of the method embodiments above, and have similar beneficial effects. For technical details not disclosed in the storage medium and device embodiments of this application, please refer to the descriptions of the method embodiments of this application for understanding.

[0260] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in some embodiments" appearing throughout the specification do not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential 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 this application. The sequence numbers of the above-described embodiments are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0261] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0262] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and 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. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0263] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of this embodiment according to actual needs.

[0264] In addition, each functional unit in the various embodiments of this application can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0265] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as mobile storage devices, read-only memory (ROM), magnetic disks, or optical disks.

[0266] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to related technologies, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0267] The above are merely embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A first device, the first device comprising a first transceiver, and A first processor, coupled to the first transceiver; the first processor is configured to: Obtain the target backsampling parameters corresponding to the diffusion model. The target backsampling parameters are related to the target quality of service parameters of the first service. The target quality of service parameters are used to indicate the performance requirements of the first service. Based on the target reverse sampling parameters, the diffusion model is used to perform reverse denoising inference on the first pilot signal received on the first channel to determine the second pilot signal after denoising the first pilot signal. Based on the second pilot signal, the channel state information of the first channel is determined; Based on the channel state information, the data of the first service received on the first channel is demodulated.

2. The first device according to claim 1, wherein the target backsampling parameter is determined based on the target quality of service parameter and a first relationship; the first relationship is a mapping relationship between the quality of service parameter and the backsampling parameter.

3. The first device according to claim 1, wherein the target backsampling parameters are determined based on the target quality of service parameters using a parameter prediction model.

4. The first device according to claim 1, wherein the target reverse sampling parameter is determined based on the target quality of service parameter and the channel quality of the first channel, and the target reverse sampling parameter includes one of the following: target sampling step size and target sampling number.

5. The first device according to claim 1, wherein the first channel is a downlink channel; The target quality of service parameters are determined based on the target service transmission channel identifier and a second relationship of the data of the first service. The second relationship is the mapping relationship between the service transmission channel identifier and the quality of service parameters.

6. The first device according to claim 1, wherein the first device is a network device and the first channel is an uplink channel; or, the first device is a terminal device and the first channel is a downlink channel; The target quality of service parameters are determined based on the service type of the first service.

7. The first device according to claim 1, wherein the target quality of service parameter includes at least one of the following: Service quality parameters, packet latency budget, error rate target, and service priority.

8. The first device according to claim 1, wherein the first channel is an uplink channel, and the first device is a network device; and / or, The first channel is a downlink channel, and the first device is a terminal device; the first processor is configured to: Based on the target quality of service parameters of the first service, the target backsampling parameters corresponding to the diffusion model are determined.

9. The first device according to claim 1, wherein the first channel is a downlink channel, and the first device is the terminal device; the first processor is configured to: The system receives backsampling parameter indication information sent by the network device, the backsampling parameter indication information being used to indicate the target backsampling parameters.

10. A signal processing method, comprising: Obtain the target backsampling parameters corresponding to the diffusion model. The target backsampling parameters are related to the target quality of service parameters of the first service. The target quality of service parameters are used to indicate the performance requirements of the first service. Based on the target reverse sampling parameters, the diffusion model is used to perform reverse denoising inference on the first pilot signal received on the first channel to determine the second pilot signal after denoising the first pilot signal. Based on the second pilot signal, the channel state information of the first channel is determined; Based on the channel state information, the data of the first service received on the first channel is demodulated.

Images