Method of signal processing and receiver

Abstract

Provided are a signal processing method and a receiver. The method comprises: receiving, by the receiver, a wireless signal transmitted by a transmitter to obtain a received signal; inputting, by the receiver, the received signal into an artificial intelligence (AI) decoder to obtain a decoded signal; generating, by the receiver, a recovered signal of the received signal according to the decoded signal; and training, by the receiver, the AI decoder online according to a difference between the recovered signal and the received signal. Since the received signal and the recovered signal can both be obtained by the receiver, the method can implement an online training process of the AI decoder to improve the generalization capability of the AI decoder.

Description

Technical Field

[0001] This application relates to the field of communications, specifically to a signal processing method and a receiver. Background Technology

[0002] An autoencoder (AE) is an artificial intelligence (AI) model that uses input signals as training targets. The architecture of its AI decoder is naturally compatible with many architectures in communication systems. For example, an AI decoder can correspond to a receiver in a communication system.

[0003] Currently, to improve the accuracy of AI decoders, they need to be pre-trained on a pre-defined training set before being deployed to the receiver (i.e., online to the receiver), a process also known as "offline training." However, real-world communication systems are complex, and the training data in the training set cannot cover all situations, resulting in significant differences from the actual received signals. If the AI ​​decoder is trained offline solely on the training set, the generalization ability of the trained AI decoder will be poor. Summary of the Invention

[0004] This application provides a signal processing method and a receiver for online training of an AI decoder, which helps to improve the generalization ability of the AI ​​decoder.

[0005] In a first aspect, a signal processing method is provided, comprising: a receiver receiving a wireless signal transmitted by a transmitter to obtain a received signal; the receiver inputting the received signal into an artificial intelligence (AI) decoder for decoding to obtain a decoded signal; the receiver generating a recovered signal of the received signal based on the decoded signal; and the receiver performing online training on the AI ​​decoder based on the difference between the recovered signal and the received signal.

[0006] In a second aspect, a receiver is provided, comprising: a receiving unit for receiving a wireless signal transmitted by a transmitter to obtain a received signal; a processing unit for inputting the received signal into an artificial intelligence (AI) decoder for decoding to obtain a decoded signal; the processing unit for generating a recovered signal of the received signal based on the decoded signal; and the processing unit for performing online training on the AI ​​decoder based on the difference between the recovered signal and the received signal.

[0007] Thirdly, a receiver is provided, including a memory and a processor, the memory for storing a program and the processor for calling the program in the memory to perform the method as described in the first aspect.

[0008] Fourthly, an apparatus is provided, including a processor for calling a program from memory to perform the method described in the first aspect.

[0009] Fifthly, a chip is provided, including a processor for calling a program from memory, causing a device having the chip mounted to perform the method described in the first aspect.

[0010] A sixth aspect provides a computer-readable storage medium having a program stored thereon that causes a computer to perform the method described in the first aspect.

[0011] A seventh aspect provides a computer program product, including a program that causes a computer to perform the method described in the first aspect.

[0012] Eighthly, a computer program is provided that causes a computer to perform the method described in the first aspect.

[0013] The receiver uses an AI decoder to decode the received signal to obtain a decoded signal. Then, it generates a recovered signal of the received signal based on the decoded signal. Based on the difference between the received signal and the recovered signal, the AI ​​decoder is trained online. Since both the received signal and the recovered signal can be acquired by the receiver, the method of this application embodiment can realize the online training process of the AI ​​decoder, which is beneficial to improving the generalization ability of the AI ​​decoder. Attached Figure Description

[0014] Figure 1 This is a flowchart of the signal transmission in a wireless communication system to which the embodiments of this application are applicable.

[0015] Figure 2 This is a schematic diagram illustrating channel estimation and signal recovery applicable to embodiments of this application.

[0016] Figure 3 The present application describes the structure diagram of the neural network to which the embodiments are applicable.

[0017] Figure 4 This paper presents a structural diagram of a CNN to which the embodiments of this application are applicable.

[0018] Figure 5 This is a schematic diagram of the channel estimation process based on an AI decoder.

[0019] Figure 6 This is a schematic diagram of a CSI feedback system based on an autoencoder.

[0020] Figure 7 This is a schematic diagram of a receiver based on an AI decoder.

[0021] Figure 8 This is the wireless communication system 800 to which the embodiments of this application apply.

[0022] Figure 9 This is a flowchart of a signal processing method according to an embodiment of this application.

[0023] Figure 10 This is a schematic diagram of a signal processing method based on periodic online training according to an embodiment of this application.

[0024] Figure 11 This is a schematic diagram of a signal processing method based on non-continuous online training according to an embodiment of this application.

[0025] Figure 12 This is a schematic diagram of a signal processing method based on non-periodic online training according to an embodiment of this application.

[0026] Figure 13 This is a schematic diagram of the signal processing method in the channel estimation process according to an embodiment of this application.

[0027] Figure 14 This is a schematic diagram of the deployment of an autoencoder in a communication system.

[0028] Figure 15 This is a schematic diagram of the signal processing method in the CSI feedback process according to an embodiment of this application.

[0029] Figure 16 This is a schematic diagram of a signal processing method during data transmission according to an embodiment of this application.

[0030] Figure 17 This is a schematic structural diagram of the receiver according to an embodiment of this application.

[0031] Figure 18 This is a schematic structural diagram of a signal processing apparatus according to an embodiment of this application. Detailed Implementation

[0032] The technical solutions in this application will now be described with reference to the accompanying drawings. To facilitate understanding of this application, the following description will first refer to... Figures 1 to 7 This document introduces the terminology and communication processes involved in the embodiments of this application.

[0033] I. Signal Transmission Process in Wireless Communication Systems

[0034] Figure 1 This is a flowchart illustrating the signal transmission in a wireless communication system to which the embodiments of this application apply. For example... Figure 1 As shown, the signal transmission process in a wireless communication system can be roughly divided into: Figure 1 The various signal processing procedures S111 to S118 are shown. Figure 1 The signal processing steps shown, some or all, can be implemented using a separate AI model. For details, please refer to [link to relevant documentation]. Figures 5 to 7 Introduction.

[0035] In the channel coding process S111, the transmitter performs channel coding on the information to be transmitted to obtain the encoded bitstream. The information to be transmitted can be in the form of a bitstream.

[0036] In the modulation process S112, the code stream is modulated into modulation symbols.

[0037] In the pilot insertion process S113, pilot symbols are inserted into the above modulation symbols to form a signal to be transmitted. The pilot symbols can be used by the receiver for channel estimation and symbol detection.

[0038] In transmission signal S114, the aforementioned signal is carried on the channel and transmitted to the receiver. During transmission through the channel, noise is typically added to the signal.

[0039] In the channel estimation process S115, the receiver can perform channel estimation based on the pilot signal to obtain channel state information (CSI), and feed the CSI back to the transmitter through the feedback link so that the transmitter can adjust the channel coding, modulation, precoding and other methods.

[0040] In the symbol detection process S116, symbol detection is performed on the received modulation symbols to obtain the detection results.

[0041] In the demodulation process S117, the received modulation symbols are demodulated based on the detection results to obtain the code stream.

[0042] In the channel decoding process S118, the bitstream is decoded to obtain the recovered information, which can be in the form of a bitstream.

[0043] It should be understood that Figure 1 The signal processing procedures S111 to S118 shown are merely illustrative examples of common signal processing procedures in wireless communication systems. Wireless communication systems may also include signal processing procedures such as resource mapping, precoding, interference cancellation, and CSI measurement, all of which can be implemented using separate AI models. For the sake of brevity, these will not be elaborated further in this application.

[0044] II. Channel Estimation

[0045] Due to the complexity and time-varying nature of wireless channel environments, in wireless communication systems (such as the wireless communication systems described above), receivers need to recover the received signals based on channel estimation results. Figure 2 This is a schematic diagram illustrating channel estimation and signal recovery applicable to embodiments of this application.

[0046] like Figure 2 As shown, in step S210, in addition to transmitting data signals, the transmitter will also transmit a series of pilot signals known to the receiver, such as channel state information-reference signal (CSI-RS) and demodulation reference signal (DMRS), etc.

[0047] In step S211, the transmitter transmits the aforementioned data signal and pilot signal to the transmitter via the channel.

[0048] In step S212, after receiving the pilot signal, the receiver can perform channel estimation. In one possible implementation, the receiver can estimate the channel information of the channel transmitting the pilot signal based on a pre-stored pilot sequence and the received pilot sequence, using a channel estimation algorithm (e.g., least squares method (LS) channel estimation).

[0049] In step S213, the receiver can recover the channel information on the full-time frequency resources using an interpolation algorithm based on the channel information of the channel transmitting the pilot sequence, for use in subsequent CSI feedback or data recovery, etc.

[0050] III. CSI Feedback

[0051] In wireless communication systems, channel feature extraction and feedback are primarily achieved using codebook-based schemes. After the receiver performs channel estimation, it selects the precoding matrix that best matches the current channel from a pre-defined precoding codebook based on the estimation results and certain optimization criteria. The precoding matrix index (PMI) information is then fed back to the transmitter via an air interface feedback link, enabling the transmitter to perform precoding. In some implementations, the receiver can also feed back the measured channel quality indication (CQI) to the transmitter, allowing the transmitter to implement adaptive modulation and coding, etc.

[0052] IV. Self-encoder

[0053] An autoencoder is a neural network that uses the input signal as a training target. The architecture of its AI encoder and / or AI decoder is naturally compatible with many architectures in communication systems. For example, the AI ​​encoder and AI decoder can correspond to the transmitter and receiver in a wireless communication system, respectively. Alternatively, they can correspond to the channel compression and decompression modules in a CSI feedback process, respectively. Furthermore, the AI ​​decoder in an autoencoder can be used independently in the channel estimation process for the receiver to recover channel information. The following will combine... Figures 5 to 7 For the sake of brevity, the introduction will not be elaborated here.

[0054] Typically, an autoencoder can be trained on a training set before being deployed to a communication system. For example, when the autoencoder contains only an AI decoder, the AI ​​decoder can be trained on the training set. When the autoencoder contains an AI encoder, the AI ​​encoder can be trained on the training set. When the autoencoder contains both an AI encoder and an AI decoder, the AI ​​encoder and AI decoder can be trained jointly.

[0055] In some implementations, since AE is a neural network model that uses the input signal as the training target, the training objective of the autoencoder can be understood as optimizing the weights of the AI ​​encoder and AI decoder while minimizing the loss function, when the difference between the input and output of the autoencoder is represented by the loss function.

[0056] For example, an autoencoder containing an AI encoder f(·) and an AI decoder g(·) is denoted as g(f(·)). The original signal s is first encoded by the AI ​​encoder f(·), and the output of the AI ​​encoder f(·) is denoted as q = f(s). The encoded signal is then input into the AI ​​decoder g(·) for decoding, and the output of the AI ​​decoder g(·) is denoted as s` = g(q) = g(f(s)). During the joint training phase, min_ {g,f} l(s,g(f(s))) is used as the training objective to jointly train the AI ​​encoder f(·) and the AI ​​decoder g(·), where l(·) represents the loss function.

[0057] V. Neural Networks

[0058] In recent years, artificial intelligence research, represented by neural networks, has achieved remarkable results in many fields and will continue to play an important role in people's production and daily life for a long time to come. Common neural networks include convolutional neural networks (CNNs), recurrent neural networks (RNNs), and deep neural networks (DNNs).

[0059] The following text combines Figure 3 This application describes the neural network to which the embodiments are applicable. Figure 3 The neural network shown can be divided into three categories according to the position of different layers: input layer 310, hidden layer 320, and output layer 330. Generally speaking, the first layer is the input layer 310, the last layer is the output layer 330, and the intermediate layers between the first and last layers are hidden layers 320.

[0060] The input layer 310 is used to input data, which may be, for example, a received signal received by a receiver. The hidden layer 320 is used to process the input data, for example, to decompress the received signal. The output layer 330 is used to output the processed output data, for example, to output the decompressed signal.

[0061] like Figure 3 As shown, a neural network consists of multiple layers, each containing multiple neurons. The neurons between layers can be fully connected or partially connected. For connected neurons, the output of a neuron in one layer can serve as the input of a neuron in the next layer.

[0062] With the continuous development of neural network research, deep learning algorithms have been proposed in recent years. These algorithms introduce more hidden layers into neural networks, forming DNNs (Deep Neural Networks). More hidden layers allow DNNs to better depict complex situations in the real world. Theoretically, the more parameters a model has, the higher its complexity and the greater its "capacity," meaning it can accomplish more complex learning tasks. This type of neural network model is widely used in pattern recognition, signal processing, optimization, and anomaly detection.

[0063] CNN is a deep neural network with convolutional structures, and its structure is as follows: Figure 4 As shown, it may include an input layer 410, a convolutional layer 420, a pooling layer 430, a fully connected layer 440, and an output layer 450.

[0064] Each convolutional layer 420 can include many convolution operators, also known as kernels. Their function can be seen as a filter that extracts specific information from the input signal. A convolution operator can essentially be a weight matrix, which is usually predefined.

[0065] The weight values ​​in these weight matrices need to be obtained through extensive training in practical applications. The weight matrices formed by the weight values ​​obtained through training can extract information from the input signal, thereby helping the CNN to make correct predictions.

[0066] When a CNN has multiple convolutional layers, the initial convolutional layers tend to extract more general features, which can also be called low-level features. As the depth of the CNN increases, the features extracted by later convolutional layers become more and more complex.

[0067] Pooling layer 430: Because it is often necessary to reduce the number of training parameters, pooling layers are often introduced periodically after convolutional layers. For example, it could be... Figure 4 The diagram shows a convolutional layer followed by a pooling layer, or multiple convolutional layers followed by one or more pooling layers. In signal processing, the sole purpose of pooling layers is to reduce the spatial size of the extracted information.

[0068] After processing by convolutional layers 420 and pooling layers 430, the CNN is still insufficient to output the required information. As mentioned earlier, convolutional layers 420 and pooling layers 430 only extract features and reduce parameters introduced by the input data. However, to generate the final output information (e.g., the bitstream of the original information transmitted by the transmitter), the CNN still needs to utilize fully connected layers 440. Typically, fully connected layers 440 can include multiple hidden layers. The parameters contained in these hidden layers can be pre-trained based on training data relevant to a specific task type. For example, the task type could include decoding data signals received by a receiver, or it could include channel estimation based on pilot signals received by the receiver.

[0069] After the multiple hidden layers in the fully connected layer 440, which is the final layer of the entire CNN, is the output layer 450, used to output the result. Typically, this output layer 450 is equipped with a loss function (e.g., a loss function similar to the cross-entropy loss function for classification) to calculate the prediction error, or to evaluate the degree of difference between the output of the CNN model (also known as the predicted value) and the ideal result (also known as the true value).

[0070] To minimize the loss function, the CNN model needs to be trained. In some implementations, the backpropagation algorithm (BP) can be used to train the CNN model. The BP training process consists of forward propagation and backward propagation. In forward propagation (such as...) Figure 4 During the forward propagation (from 410 to 450), the input data is fed into each of the aforementioned layers of the CNN model, processed layer by layer, and then passed to the output layer. If the output result differs significantly from the ideal result, minimizing the aforementioned loss function is used as the optimization objective, and the process transitions to backpropagation (e.g., ...). Figure 4 The propagation from 450 to 410 is backpropagation. The partial derivatives of the optimization objective with respect to the weights of each neuron are calculated layer by layer, forming the gradient of the optimization objective with respect to the weight vector. This gradient serves as the basis for modifying the model weights. The CNN training process is completed during the weight modification process. When the aforementioned error reaches the desired value, the CNN training process ends.

[0071] It should be noted that, as Figure 4 The CNN shown is only an example of a convolutional neural network. In specific applications, convolutional neural networks can also exist in the form of other network models, and this application does not limit them.

[0072] The above text combined Figures 1 to 4 This paper introduces the signal transmission process, channel estimation, CSI feedback, and autoencoder of wireless communication systems. The following section combines these concepts with... Figures 5 to 7 This paper introduces the signal transmission process, channel estimation, and CSI feedback based on autoencoders.

[0073] VI. Channel Estimation Based on AI Decoder

[0074] Channel estimation based on AI decoders aims to achieve channel estimation by processing the pilot signals received by the receiver using an AI decoder. Figure 5 This illustrates the process of channel estimation based on an AI decoder. See also... Figure 5 The pilot signal received by receiver 500 is used as the input to AI decoder 510. Accordingly, AI decoder 510 processes the input pilot signal to output channel information. Furthermore, in some implementations, in addition to the pilot signal, other auxiliary information can be added to improve the accuracy of the channel information output by AI decoder. For example, the input to AI decoder 510 may include the original sequence of the pilot signal pre-stored by receiver 500, the energy level of the pilot signal received by receiver 500, the transmission delay during pilot signal transmission, or noise during pilot signal transmission.

[0075] VII. CSI Feedback Based on Autoencoder

[0076] The AI ​​encoder in the autoencoder can extract features from the received signal carrying CSI, and the AI ​​decoder in the autoencoder can restore the compressed feedback CSI from the transmitter at the receiver as much as possible. In this way, the communication overhead of feedback CSI can be saved without affecting the accuracy of CSI transmission.

[0077] Figure 6 A CSI feedback system based on an autoencoder is shown. For example... Figure 6 As shown, the entire feedback system includes an autoencoder, comprising an AI encoder 611 and an AI decoder 621. The AI ​​encoder 611 is deployed at the transmitter 610, and the AI ​​decoder 621 is deployed at the receiver 620. The transmitter 610 compresses and encodes the CSI to be transmitted using the AI ​​encoder 611 to obtain the compressed CSI. The compressed CSI is then fed back to the receiver 620 via the feedback link. The receiver 620 decodes the compressed CSI using the AI ​​decoder 621 to obtain the recovered CSI.

[0078] 8. Receiver based on AI decoder

[0079] The performance of a receiver can be improved by introducing an AI decoder into the receiver design and using the AI ​​decoder to perform internal signal processing (e.g., demodulation, decompression, etc.). Figure 7 This is a schematic diagram of a receiver based on an AI decoder. Figure 7 In the receiver 700 shown, the AI ​​decoder 710 takes the received signal as its input and outputs the decoded signal as its output.

[0080] As can be seen from the above introduction, modular communication system design based on autoencoders is a trend in the development of communication systems. It can make good use of the prior structure of traditional communication system models, and at the same time, it can flexibly adjust and train the AI ​​encoder and / or AI decoder in the autoencoder.

[0081] Currently, before the AI ​​decoder is deployed to the receiver, it is trained offline using a pre-set training set. However, the training effect is not ideal. Because real-world communication systems are complex, and the training data in the set cannot cover all scenarios, the trained AI decoder can only output a relatively accurate decoded signal when processing signals with similar characteristics to the training data. When the received signal differs significantly from the training data, the AI ​​decoder cannot decode the received signal accurately. In other words, the generalization ability of the offline-trained AI decoder is poor. The generalization ability of the AI ​​decoder describes its accuracy when decoding signals that differ significantly from the training data.

[0082] This application provides a signal processing scheme to realize the online training process of an AI decoder. To facilitate understanding of this application, the following will first combine... Figure 8 This application introduces the wireless communication system to which the embodiments are applicable, and then combines it with Figure 9 The method of the embodiments of this application is described.

[0083] Figure 8 This is the wireless communication system 800 to which this application embodiment applies. The wireless communication system 800 may include a network device 810. The network device 810 may be a device that communicates with a terminal device 820. The network device 810 may provide communication coverage for a specific geographical area and may communicate with the terminal device 820 located within that coverage area.

[0084] Figure 8 An exemplary embodiment shows a network device 810 and two terminal devices 820. Optionally, the wireless communication system 800 may include multiple network devices and each network device may include other numbers of terminal devices within its coverage area. This application embodiment does not limit this.

[0085] Optionally, the wireless communication system 800 may also include other network entities such as a network controller and a mobility management entity, which is not limited in this embodiment.

[0086] Alternatively, the terminal devices 820 can also communicate directly with each other. For example, two terminal devices 820 can communicate with each other through a device-to-device (D2D) link.

[0087] It should be understood that the technical solutions of the embodiments of this application can be applied to various communication systems, such as: 5th generation (5G) systems or new radio (NR), long term evolution (LTE) systems, LTE frequency division duplex (FDD) systems, LTE time division duplex (TDD) systems, etc. The technical solutions provided in this application can also be applied to future communication systems, such as 6th generation mobile communication systems, satellite communication systems, etc.

[0088] The terminal device in this application embodiment can also be referred to as user equipment (UE), access terminal, user unit, user station, mobile station, mobile station (MS), mobile terminal (MT), remote station, remote terminal, mobile device, user terminal, terminal, wireless communication device, user agent, or user device. The terminal device in this application embodiment can be a device that provides voice and / or data connectivity to a user, and can be used to connect people, objects, and machines, such as a handheld device with wireless connectivity, vehicle-mounted device, etc. The terminal devices in the embodiments of this application can be mobile phones, tablets, laptops, PDAs, mobile internet devices (MIDs), wearable devices, virtual reality (VR) devices, augmented reality (AR) devices, wireless terminals in industrial control, self-driving, remote medical surgery, smart grids, transportation safety, smart cities, and smart homes, etc. Optionally, the UE can act as a base station. For example, the UE can act as a scheduling entity, providing sidelink signals between UEs in V2X or D2D, etc. For example, cellular phones and cars communicate with each other using sidelink signals. Cellular phones and smart home devices communicate without relaying communication signals through a base station.

[0089] The network device in this application embodiment can be a device for communicating with a terminal device. This network device can also be called an access network device or a wireless access network device, such as a base station. In this application embodiment, the network device can refer to a radio access network (RAN) node (or device) that connects the terminal device to the wireless network. A base station can broadly encompass, or be replaced by, various names including: NodeB, evolved NodeB (eNB), next-generation NodeB (gNB), relay station, access point, transmitting and receiving point (TRP), transmitting point (TP), master MeNB, auxiliary SeNB, multi-mode radio (MSR) node, home base station, network controller, access node, wireless node, access point (AP), transmission node, transceiver node, baseband unit (BBU), remote radio unit (RRU), active antenna unit (AAU), remote radio head (RRH), central unit (CU), distributed unit (DU), positioning node, etc. A base station can be a macro base station, micro base station, relay node, donor node, or similar entities, or combinations thereof. A base station can also refer to a communication module, modem, or chip installed within the aforementioned equipment or apparatus. A base station can also be a mobile switching center, or a device that performs base station functions in device-to-device (D2D), vehicle-to-everything (V2X), and machine-to-machine (M2M) communications, a network-side device in a 6G network, or a device performing base station functions in future communication systems. Base stations can support networks using the same or different access technologies. The embodiments of this application do not limit the specific technologies or device forms used in the network equipment.

[0090] Base stations can be fixed or mobile. For example, a helicopter or drone can be configured to act as a mobile base station, and one or more cells can move depending on the location of the mobile base station. In other examples, a helicopter or drone can be configured as a device to communicate with another base station.

[0091] In some deployments, the network device in this application embodiment may refer to a CU or a DU, or the network device may include both a CU and a DU. The gNB may also include an AAU.

[0092] Network devices and terminal devices can be deployed on land, including indoors or outdoors, handheld or vehicle-mounted; they can also be deployed on water; and they can also be deployed in the air on airplanes, balloons, and satellites. This application does not limit the scenario in which the network devices and terminal devices are located.

[0093] It should be understood that the communication equipment involved in this application can be a network device or a terminal device. For example, the first communication device is a network device and the second communication device is a terminal device. Alternatively, the first communication device may be a terminal device and the second communication device may be a network device. Or, both the first and second communication devices may be network devices or both may be terminal devices.

[0094] It should also be understood that all or part of the functions of the communication device in this application can also be implemented by software functions running on hardware, or by virtualization functions instantiated on a platform (e.g., a cloud platform).

[0095] Figure 9 This is a flowchart of a signal processing method according to an embodiment of this application. Figure 9 The method shown includes steps S910 to S940. It should be understood that... Figure 9 The receiver shown can be one of the terminal devices described above, and correspondingly, the transmitter can also be one of the terminal devices described above. In this case, the receiver and transmitter can communicate via a D2D link. Of course, when the receiver is a terminal device, the transmitter can also be a network device.

[0096] In step S910, the receiver receives the wireless signal transmitted by the transmitter and obtains the received signal.

[0097] In step S920, the receiver inputs the received signal into the AI ​​decoder for decoding to obtain the decoded signal.

[0098] In step S930, the receiver generates a recovered signal of the received signal based on the decoded signal.

[0099] In the embodiments of this application, the method for generating the recovered signal is not limited. In some embodiments, the recovered signal can be obtained by performing simple signal processing on the decoded signal. In other embodiments, the decoded signal can be input into an AI encoder to generate a recovered signal of the received signal. The following will combine... Figures 14 to 16 For the sake of brevity, a detailed introduction will not be elaborated here.

[0100] In step S940, the receiver performs online training on the AI ​​decoder based on the difference between the recovered signal and the received signal. The difference between the recovered signal and the received signal can be represented by the error between them.

[0101] In some implementations, step S940 may include the receiver updating the weights of the AI ​​decoder by minimizing the difference between the recovered signal and the received signal as an optimization objective, or optimizing the weights of the AI ​​decoder to train the AI ​​decoder online.

[0102] When the AI ​​decoder described above is the neural network model, its weights can be understood as the weights used by each layer of the neural network model when processing the input data. For example, see... Figure 4 The weights of the AI ​​decoder can include the weights used by the input layer 410, convolutional layer 420, pooling layer 430, fully connected layer 440, and output layer 450 to process the input of each layer.

[0103] The aforementioned update of weights in the AI ​​decoder can be understood as updating all weights of the AI ​​decoder, or updating some weights of the AI ​​decoder.

[0104] The difference between the recovered signal and the received signal can be represented by the error between them. In some implementations, the error between the recovered signal and the received signal can be calculated using a loss function; in other words, the loss function represents the error between the recovered signal and the received signal.

[0105] Typically, to minimize the loss function, the AI ​​decoder needs to be trained online. For example, a backpropagation (BP) algorithm can be used to train the AI ​​decoder. The BP training process can consist of forward propagation and backward propagation. Taking the AI ​​decoder as an example... Figure 4 The model shown is used as an example to illustrate the forward propagation (e.g.) Figure 4 During the forward propagation (from 410 to 450), the received signal is input into each layer of the AI ​​decoder model, processed layer by layer, and then transmitted to the output layer. The output layer outputs the decoded signal accordingly, and the receiver obtains the recovered signal based on the decoded signal. If the difference between the received signal and the recovered signal is significant, minimizing the loss function is used as the optimization objective, and the process transitions to backpropagation (e.g., ...). Figure 4The propagation from 450 to 410 is backpropagation. The partial derivatives of the optimization target with respect to the weights of each neuron are calculated layer by layer, forming the gradient of the optimization target with respect to the weight vector. This gradient serves as the basis for modifying the weights of the AI ​​decoder. The online training process of the AI ​​decoder is completed during the weight modification process. When the aforementioned error reaches the desired value, the online training process of the AI ​​decoder ends. Of course, other known training principles can also be used for the online training process of the AI ​​decoder in this application, and this embodiment does not limit this.

[0106] In this embodiment, the receiver uses an AI decoder to decode the received signal, obtaining a decoded signal. Then, it generates a recovered signal based on the decoded signal. Based on the difference between the received signal and the recovered signal, the AI ​​decoder is trained online. Since both the received signal and the recovered signal can be acquired by the receiver, the method in this embodiment can realize the online training process of the AI ​​decoder, which is beneficial for improving the generalization ability of the AI ​​decoder. Alternatively, after the AI ​​decoder is online, it is trained online using the actual received signal. In this way, even if there is a significant difference between the received signal and the training data, the online-trained AI decoder can still achieve high accuracy when decoding the received signal, which is beneficial for improving the generalization ability of the AI ​​decoder.

[0107] Compared to the approach of training the AI ​​decoder based on the difference between the original signal to be transmitted by the transmitter and the received signal received by the receiver, the solution of this application avoids the problem that the original signal cannot be accurately known by the receiver due to the distributed deployment of the receiver and transmitter, which would prevent the AI ​​decoder from being trained online on the receiver.

[0108] The online training time can be configured by the network device for the receiver. In some implementations, the network device can send online training indication information to the receiver, whereby the online training indication information is used to indicate the time for the AI ​​decoder to perform online training (for ease of understanding, the time for the AI ​​decoder to perform online training will be referred to as the "online training time" below). Of course, the receiver can also perform online training according to the online training time specified in the communication protocol. This application does not specifically limit this aspect.

[0109] The online training instructions described above can indicate the duration of online training in various ways. For example, the instructions can indicate the online training period, instructing the receiver to periodically perform online training for the AI ​​decoder. Alternatively, the instructions can directly indicate the start and / or end times of the online training.

[0110] In some implementations, the above-mentioned online training instruction information can be transmitted via radio resource control (RRC) signaling. In other implementations, the above-mentioned online training instruction information can also be transmitted via the control channel. This application does not limit this aspect.

[0111] Based on the different instructions used in online training, the online training configuration methods for AI decoders can be divided into periodic online training, non-continuous online training, and non-periodic online training. The following section combines... Figures 10 to 12 The above three online training configuration methods are introduced.

[0112] Online training configuration method one: periodic online training. This means the receiver periodically trains the AI ​​decoder online based on a configured online training period.

[0113] Since online training requires the actual signal transmitted between the transmitter and receiver (i.e., the received signal mentioned above), to simplify the configuration of the online training cycle, in some implementations, the online training cycle can include multiple transmission cycles of the actual transmitted signal. Accordingly, the receiver can perform online training in any of these multiple transmission cycles. Typically, to ensure the accuracy of the AI ​​decoder, online training can be performed in the first transmission cycle of the multiple transmission cycles. This way, the trained AI decoder can be used for decoding in the subsequent transmission cycles.

[0114] As mentioned above, the receiver can perform online training in any of the multiple transmission cycles. That is, the online training cycles include a first type and a second type of transmission cycle. In the first type of transmission cycle, the receiver can perform online training on the AI ​​decoder and use the trained AI decoder to decode the received signal in the current transmission cycle. In the second type of transmission cycle, the receiver can directly use the AI ​​decoder to decode the received signal without needing to perform online training on the AI ​​decoder. Typically, to improve the accuracy of online training, the second type of transmission cycle can also be used to collect received signals as training data for online training.

[0115] It should be noted that the arrangement of the first type of transmission period and the second type of transmission period in the online training period can be pre-configured based on the communication protocol. Of course, it can also be indicated by the network device through signaling, and this application embodiment does not limit this.

[0116] Furthermore, this application does not specifically limit the number of transmission cycles included in the aforementioned online training cycle. In some implementations, the online training cycle may contain fewer transmission cycles, resulting in a higher online training frequency. This configuration of the online training cycle is suitable for communication scenarios with frequent changes, such as high-speed rail. In other embodiments, the online training cycle may contain more transmission cycles, resulting in a lower online training frequency, which helps reduce the computational resources required for online training. This configuration of the online training cycle is suitable for communication scenarios with less frequent changes, such as office buildings.

[0117] The following text is incomplete and cannot be translated. Figure 10 Taking the periodic online training shown as an example, the signal processing method of this application embodiment is introduced. Figure 10 As shown, the online training cycle r1 includes 4 transmission cycles t1, and the first transmission cycle of the 4 transmission cycles t1 is the first type of transmission cycle, and the other 3 transmission cycles are the second type of transmission cycles.

[0118] In the first type of transmission cycle, the receiver needs to train the AI ​​decoder online based on the training data, and then use the online trained AI decoder to decode the received signal received in the current transmission cycle.

[0119] In the second type of transmission cycle, the receiver can directly use the AI ​​decoder trained online as described above to decode the received signal in the current transmission cycle. Simultaneously, the receiver can collect the received signal in the current transmission cycle as training data for subsequent online training.

[0120] Online training configuration method two: Non-continuous online training. That is, based on periodic online training, the start and end times of online training can be indicated through online training instruction information to improve the flexibility of configuring online training.

[0121] In this training method, periodic online training can be found in the introduction to online training configuration method one above. For the sake of brevity, the following text focuses on the method of indicating the start and end times of online training.

[0122] The online training instruction information can indicate the start time of online training either explicitly or implicitly. For example, the online training instruction information can directly carry the start time. Alternatively, the receiver can use the time domain unit where the online training instruction information is transmitted as the starting point, offset by a preset time domain unit to obtain the offset time domain unit, which is the start time. The time domain unit can be, for example, a time slot or subframe.

[0123] Accordingly, the online training instructions can indicate the end time of the online training either explicitly or implicitly. For example, the online training instructions can directly carry the end time. Alternatively, the online training instructions can indicate the end time by specifying the total number of online training cycles.

[0124] The following text is incomplete and cannot be translated. Figure 11 Taking the non-continuous online training shown as an example, the signal processing method of this application embodiment is introduced. Figure 11 As shown, assume that the online training indication information indicates the start time of online training is the i-th transmission period t. i Given that the total number of online training cycles is 3, and each online training cycle r2 includes 2 transmission cycles, the start time of the online training cycle r2 is the i-th transmission cycle t. i The start time is the online training period r2, and the end time is the (i+6)th transmission period t. i+6 The end time.

[0125] The first transmission cycle in each online training cycle r2 is a first-type transmission cycle. In the first-type transmission cycle, the receiver needs to train the AI ​​decoder online based on the training data, and then use the online trained AI decoder to decode the received signal received in the current transmission cycle.

[0126] The second transmission cycle in each online training cycle r2 is a second type of transmission cycle. In this second type of transmission cycle, the receiver can directly use the AI ​​decoder trained online to decode the received signal in the current transmission cycle. Simultaneously, the receiver can collect the received signal in the current transmission cycle as training data for subsequent online training.

[0127] The third online training configuration method is non-periodic online training. This means that each online training session requires the use of online training instructions.

[0128] In some implementations, the start and end times of online training can be included in the online training instruction information, and the receiver performs online training within the time period specified by the start and end times. In other implementations, only the start time of online training can be included in the online training instruction information, and the receiver performs one online training session at the start time after receiving the online training instruction information.

[0129] The online training instruction information can indicate the start time of the online training in a visible or implicit manner. For example, the online training instruction information can directly carry the start time. Alternatively, the receiver can use the time slot containing the online training instruction information as the starting point, offset by a preset time slot to obtain the offset time slot, which then becomes the start time.

[0130] The online training instruction information can indicate the end time of the online training in a visible or implicit manner. For example, the online training instruction information can directly carry the end time. Alternatively, the receiver can use the time slot where the online training instruction information is transmitted as the starting point, offset by a preset time slot to obtain the offset time slot, which then becomes the end time.

[0131] The following text is incomplete and cannot be translated. Figure 12 Taking the non-periodic online training shown as an example, the signal processing method of this application embodiment is introduced. Figure 12 As shown, assume that the online training instruction information 1 indicates that during the transmission period t i Online training is performed, and the online training instruction information 2 indicates that during the transmission period t... i+4 Conduct online training.

[0132] After the receiver receives the online training instruction information 1, during the transmission period t i The receiver needs to train the AI ​​decoder online based on the training data, and then use the trained AI decoder to decode the received signal in the current transmission cycle.

[0133] During the transmission period t i+1 up to transmission period t i+3 The receiver directly uses the AI ​​decoder trained online to decode the received signal in the current transmission cycle.

[0134] After the receiver receives the online training instruction information 2, during the transmission period t i+4 The receiver continues to train the AI ​​decoder online based on the training data, and then uses the trained AI decoder to decode the received signal in the current transmission cycle.

[0135] In the online training process described above, the receiver can collect received signals as training data for online training. However, in some cases, the amount of received signals collected by the receiver is insufficient for online training of the AI ​​decoder. In such cases, some or all of the training data in the receiver's pre-stored training set can be selected for online training. In some implementations, the pre-stored training set can be the dataset used for offline training of the AI ​​decoder.

[0136] It should be noted that the fewer training data in the additional training set and the more received signals collected by the receiver, the higher the generalization performance of the AI ​​decoder after online training.

[0137] In some implementations, the size of the training data used for online training can be predetermined. In other embodiments, the size of the training data used for online training can also be indicated by the network device. For example, the network device can indicate the size of the training data used in the first type of transmission cycle through online training indication information. As another example, the network device can indicate the size of the training data used in the first type of transmission cycle through other information.

[0138] The size of the training data used in the aforementioned online training can be understood as the total amount of all training data used in each online training session. In some implementations, if the number of received signals collected by the receiver is insufficient for online training of the AI ​​decoder, the size of the training data used in the aforementioned online training includes the sum of the number of received signals collected by the receiver and the number of training data selected from the training set. In other implementations, if the number of received signals collected by the receiver is sufficient, the size of the training data used in the aforementioned online training may only include the number of received signals used as training data.

[0139] It should be noted that the above-mentioned scheme for indicating the size of training data by network devices can be used in combination with any of the three online configuration methods described above. For example, when used in combination with online training configuration method one, the above-mentioned online training indication information may include first indication information indicating the online training cycle for the AI ​​decoder to perform online training, and the above-mentioned online training indication information may also include second indication information indicating the size of the training data used in the first type of transmission cycle.

[0140] The above text combined Figures 10 to 12 The signal processing method of the embodiments of this application is introduced below, in conjunction with Figure 13 Taking the application of AI decoders in channel estimation as an example, this paper introduces signal processing methods. It should be noted that the online training process described below can be combined with any of the online training configurations mentioned above. For the sake of brevity, further details will not be elaborated upon below.

[0141] During the channel estimation process, the received signal includes the pilot signal transmitted by the transmitter. Step S920 includes the receiver inputting the received signal into the AI ​​decoder for decoding to perform channel estimation and obtain a decoded signal. The decoded signal includes the estimated first channel information. Step S930 includes the receiver processing the pilot signal stored in the receiver according to the first channel information and the first random noise to obtain a recovered signal.

[0142] Typically, the pilot signal transmitted by the transmitter can be configured to be the same as the pilot signal stored in the receiver, so that the receiver can perform channel estimation based on the comparison between the received signal and the stored pilot signal, and obtain the estimated channel information.

[0143] The aforementioned first random noise may be pre-stored by the receiver or pre-generated by the receiver; this application embodiment does not limit this.

[0144] To improve the accuracy of the channel information estimated by the AI ​​decoder, in some implementations, the pilot signal stored in the receiver can be input into the AI ​​decoder along with the received signal. Of course, it is also possible to input only the received signal into the AI ​​decoder; this application does not limit this approach.

[0145] Furthermore, due to the difficulty in acquiring noise data from actual communication environments, during the process of obtaining the recovered signal, the receiver can process the pilot signal based on various pre-stored random noises to obtain multiple recovered signals. Then, the recovered signal most similar to the received signal is selected from these recovered signals. Based on the differences between the similar recovered signal and the received signal, the AI ​​decoder is trained online.

[0146] The following text combines Figure 13 This paper introduces a signal processing method in the channel estimation process according to embodiments of this application.

[0147] like Figure 13 As shown, transmitter 1310 transmits the pilot signal p through channel H. During transmission, due to the presence of noise N in the actual channel, receiver 1320 collects the received signal Y. p It can be represented as Y p =H*p+N.

[0148] Receiver 1320 will collect the received signal Y p The AI ​​decoder 1321 is input to obtain channel information H'. Then, the receiver 1320 processes the pilot signal p stored in the receiver 1320 based on the channel information H' and random noise N' to obtain the recovered signal Y. p ', that is, Y p '=H'*p+N'. Finally, receiver 1320 based on the recovered signal Y p 'With received signal Y p The differences between them were used to train the AI ​​decoder 1321 online.

[0149] As mentioned earlier, autoencoders can also include AI encoders. For autoencoders that include both AI encoders and AI decoders, a joint training approach is typically used to train both the AI ​​encoder and the AI ​​decoder. That is, the autoencoder formed by the AI ​​encoder and the AI ​​decoder is jointly trained based on the difference between the input signal of the AI ​​encoder and the output signal of the AI ​​decoder.

[0150] Similarly, after going live, the AI ​​encoder and AI decoder are deployed in the receiver and transmitter, respectively. Due to the distributed deployment of the receiver and transmitter, the input of the AI ​​encoder cannot be obtained at the receiver, resulting in the aforementioned joint training having to be conducted offline based on the training set. Figure 14 As shown, the AI ​​encoder 1411 is deployed at the transmitter 1410 to encode the original signal s to obtain the encoded signal q. The transmitter 1410 transmits the encoded signal q to the receiver 1420 via a wireless link. Accordingly, the receiver 1420 inputs the received encoded signal q into the AI ​​decoder 1421 for decoding to obtain the decoded signal s'. During online training, it is necessary to base the training on the difference between the original signal s and the decoded signal s'. However, the original signal s cannot be accurately transmitted through the communication link between the transmitter 1410 and the receiver 1420. Therefore, it is not possible to jointly train the AI ​​encoder 1411 and the AI ​​decoder 1421 in an online training manner. Joint training can be understood as treating the AI ​​encoder and AI decoder as a whole and updating the weights in the AI ​​encoder and AI decoder.

[0151] To improve the generalization ability of the autoencoder, this application also provides a signal processing method for jointly training an AI encoder and an AI decoder in an online training manner. By configuring an AI encoder in the receiver, the configured AI encoder and the receiver's original AI decoder can be jointly trained, that is, the AI ​​encoder and AI decoder are jointly trained based on the difference between the input of the AI ​​decoder and the output of the AI ​​encoder. In other words, step S930 above includes: the receiver inputs the decoded signal into the AI ​​encoder to generate a recovered signal.

[0152] In some implementations, to improve the accuracy of joint training, the AI ​​encoder configured in the receiver can have the same model structure and / or the same model parameters, such as the same model weights, as the AI ​​encoder in the transmitter. Of course, if the goal of joint training is primarily to improve the generalization performance of the AI ​​decoder, a simpler AI encoder can be configured in the receiver. Alternatively, the parameters of the AI ​​encoder can be frozen during joint training, and only the parameters of the AI ​​decoder can be trained.

[0153] If we want to improve the generalization performance of the AI ​​encoder in the transmitter and the AI ​​decoder in the receiver by using joint training, we can train the AI ​​encoder and AI decoder in the receiver online, and then feed back the model parameters of the trained AI encoder in the receiver to the transmitter after the joint training is completed.

[0154] In some implementations, the training objective (or optimization objective) of the receiver's autoencoder is represented by a loss function as: min_ {g’} l`(q,f'(g'(q))), where l`(·) represents the loss function; q represents the received signal received by the receiver; g'(·) represents the AI ​​decoder in the receiver; and f'(·) represents the AI ​​encoder in the receiver.

[0155] The following text combines Figure 15 and Figure 16 Taking the application of an autoencoder formed by an AI encoder and AI decoder in the CSI feedback and data transmission processes as an example, this paper introduces the online training process of the autoencoder using a joint training method. It should be noted that the online training process described below can be combined with any of the online training configurations mentioned above. For the sake of brevity, further details will not be elaborated upon below.

[0156] During the CSI feedback process, the received signal includes the CSI transmitted by the transmitter, which is a compressed CSI. Step S920 includes the receiver inputting the CSI into the AI ​​decoder for decoding to obtain a decoded signal, which contains the recovered channel information. Step S930 includes the receiver inputting the decoded CSI into the AI ​​encoder for encoding to recover the CSI and generate a recovery signal, which contains the recovered CSI.

[0157] Figure 15 This is a schematic diagram of the signal processing method during the CSI feedback process according to an embodiment of this application. Figure 15 As shown, transmitter 1510 inputs the CSI to be transmitted into AI encoder 1511 for encoding, obtaining the encoded CSI (also known as "compressed CSI"). Transmitter 1510 then transmits the compressed CSI to receiver 1520. Correspondingly, receiver 1520 inputs the received signal containing the compressed CSI into AI decoder 1521 for decoding, obtaining the decoded CSI. Receiver 1520 then inputs the decoded CSI into AI encoder 1522 for encoding, obtaining the recovered CSI. Based on the difference between the recovered CSI and the compressed CSI, receiver 1520 performs joint training on AI encoder 1522 and AI decoder 1521.

[0158] During data transmission, the received signal includes a first data signal transmitted by the transmitter. Step S920 includes the receiver inputting the first data signal into an AI decoder for decoding to recover the data transmitted by the transmitter and obtain a decoded signal containing the recovered data. The receiver inputs the decoded signal into an AI encoder to generate a recovered signal, including: the receiver inputting the recovered data into the AI ​​encoder to obtain a second data signal; and the receiver processing the second data signal based on the estimated second channel information and second random noise to obtain the recovered signal.

[0159] The aforementioned second random noise may be pre-stored by the receiver or pre-generated by the receiver; this application embodiment does not limit this.

[0160] In some implementations, a pilot signal can be inserted into the first data signal, allowing the receiver to perform channel estimation based on the pilot signal, thereby improving the similarity between the recovered signal and the received signal. Alternatively, channel estimation can be omitted, and the second data signal can be directly input into the AI ​​model to obtain the recovered signal; this application does not limit this approach.

[0161] Figure 16 This is a schematic diagram of a signal processing method during data transmission according to an embodiment of this application. Figure 16 As shown, transmitter 1610 inputs the data to be transmitted into AI encoder 1611 for encoding to obtain a first data signal. Transmitter 1620 transmits the first data signal to receiver 1620 via wireless channel H, and during transmission, noise N is superimposed on the first data signal to form a received signal Y. Correspondingly, after receiving the received signal Y, receiver 1620 inputs the received signal Y into AI decoder 1621 for decoding to obtain the recovered data. Receiver 1620 then inputs the recovered data into AI encoder 1622 for encoding to obtain a second data signal. Receiver 1620 processes the second data signal based on the random noise N' and the estimated channel information H' to obtain the recovered signal Y'. Based on the difference between the recovered signal Y' and the received signal Y, receiver 1620 performs joint training on AI encoder 1622 and AI decoder 1621 in receiver 1620.

[0162] Of course, during data transmission, only the AI ​​decoder can be used without the AI ​​encoder. In this case, the signal processing flow performed by the transmitter can be similar to that of traditional signal processing, such as encoding and modulating the signal to be transmitted. Correspondingly, the receiver can also perform signal processing on the decoded signal output by the AI ​​decoder in a similar manner to traditional signal processing, such as encoding and modulating the signal to be transmitted. In this case, it can be understood as only training the AI ​​decoder online. For details, please refer to the online training process of the AI ​​decoder above, which will not be repeated here for the sake of brevity.

[0163] The above text combined Figures 1 to 16 The method embodiments of this application are described in detail below, in conjunction with... Figures 17 to 18 The present application provides a detailed description of the apparatus embodiments. It should be understood that the descriptions of the method embodiments correspond to the descriptions of the apparatus embodiments; therefore, any parts not described in detail can be found in the foregoing method embodiments.

[0164] Figure 17 This is a schematic structural diagram of the receiver according to an embodiment of this application. Figure 17 The receiver 1700 shown includes a receiving unit 1710 and a processing unit 1720.

[0165] The receiving unit 1710 can be used to receive the wireless signals transmitted by the transmitter and obtain the received signals.

[0166] The processing unit 1720 can be used to input the received signal into the AI ​​decoder to decode the received signal to obtain a decoded signal; generate a recovered signal of the received signal based on the decoded signal; and perform online training on the AI ​​decoder based on the difference between the recovered signal and the received signal.

[0167] Optionally, the receiving unit 1710 can also be used to receive online training instruction information sent by the network device, the online training instruction information being used to indicate the time for the AI ​​decoder to perform the online training.

[0168] Optionally, the online training instruction information includes first instruction information, which is used to instruct the AI ​​decoder to perform an online training cycle. The online training cycle includes multiple transmission cycles, including a first type of transmission cycle and a second type of transmission cycle. The first type of transmission cycle is used to perform online training on the AI ​​decoder, and the second type of transmission cycle is used to collect training data for online training.

[0169] Optionally, the online training indication information further includes second indication information, which indicates the size of the training data used in the first type of transmission cycle.

[0170] Optionally, the online training time includes the start time and / or end time of the online training.

[0171] Optionally, the received signal is the received signal of the pilot signal transmitted by the transmitter. The processing unit 1720 can also be used to input the received signal into the AI ​​decoder for decoding, perform channel estimation, and obtain the decoded signal, which includes the estimated first channel information; and to process the pilot signal stored in the receiver according to the first channel information and the first random noise to obtain the recovered signal, wherein the pilot signal transmitted by the transmitter is the same as the pilot signal stored in the receiver.

[0172] Optionally, the processing unit can also be used to input the pilot signal stored in the receiver and the received signal into the AI ​​decoder to perform channel estimation and obtain the decoded signal.

[0173] Optionally, the processing unit 1720 may be specifically used to input the decoded signal into the AI ​​encoder to generate the recovery signal.

[0174] Optionally, the received signal includes the compressed CSI transmitted by the transmitter. The CSI is a compressed CSI. The processing unit 1720 can also be used to input the CSI into the AI ​​decoder for decoding to obtain the decoded signal, which includes the recovered channel information; and to input the recovered channel information into the AI ​​encoder for encoding to recover the CSI and generate the recovered signal, which includes the recovered CSI.

[0175] Optionally, the received signal includes a first data signal transmitted by the transmitter. The processing unit 1720 can also be used to input the first data signal into the AI ​​decoder for decoding, recover the data transmitted by the transmitter, and obtain the decoded signal, the decoded signal containing the recovered data; input the recovered data into the AI ​​encoder to obtain a second data signal; and process the second data signal according to the estimated second channel information and second random noise to obtain the recovered signal.

[0176] Optionally, the processing unit 1720 can also be used to minimize the difference as an optimization objective, update the weights of the AI ​​decoder, and perform the online training on the AI ​​decoder.

[0177] Optionally, the difference between the recovered signal and the received signal is represented by the error between the recovered signal and the received signal.

[0178] Figure 18 This is a schematic structural diagram of a signal processing apparatus according to an embodiment of this application. Figure 18 The dashed lines indicate that the unit or module is optional. The device 1800 can be used to implement the methods described in the above method embodiments. The device 1800 can be a chip, a terminal device, or a network device.

[0179] Apparatus 1800 may include one or more processors 1810. The processor 1810 may support apparatus 1800 in implementing the methods described in the preceding method embodiments. The processor 1810 may be a general-purpose processor or a special-purpose processor. For example, the processor may be a central processing unit (CPU). Alternatively, the processor may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0180] The apparatus 1800 may further include one or more memories 1820. The memories 1820 store a program that can be executed by the processor 1810, causing the processor 1810 to perform the methods described in the preceding method embodiments. The memories 1820 may be independent of the processor 1810 or integrated within the processor 1810.

[0181] The device 1800 may also include a transceiver 1830. The processor 1810 can communicate with other devices or chips via the transceiver 1830. For example, the processor 1810 can send and receive data with other devices or chips via the transceiver 1830.

[0182] This application also provides a computer-readable storage medium for storing a program. This computer-readable storage medium can be applied to a terminal or network device provided in this application, and the program causes a computer to execute the methods performed by the terminal or network device in various embodiments of this application.

[0183] This application also provides a computer program product. The computer program product includes a program. The computer program product can be applied to a terminal or network device provided in this application embodiment, and the program causes a computer to execute the methods performed by the terminal or network device in various embodiments of this application.

[0184] This application also provides a computer program. This computer program can be applied to the terminal or network device provided in this application, and the computer program causes the computer to execute the methods performed by the terminal or network device in various embodiments of this application.

[0185] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0186] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0187] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0188] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0189] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0190] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can read or a data storage device such as a server or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., digital video discs, DVDs) or semiconductor media (e.g., solid-state disks, SSDs), etc.

[0191] The above description is merely a specific embodiment 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 signal processing method, characterized in that, include: The receiver receives the wireless signal transmitted by the transmitter and obtains the received signal, which is the received signal of the pilot signal transmitted by the transmitter. The receiver inputs the received signal into the artificial intelligence (AI) decoder for decoding to obtain a decoded signal. The receiver inputs the received signal into the AI ​​decoder for decoding and performs channel estimation to obtain the decoded signal. The decoded signal contains the estimated first channel information. The receiver generates a recovered signal of the received signal based on the decoded signal. The receiver processes the pilot signal stored in the receiver based on the first channel information and the first random noise to obtain the recovered signal. The pilot signal transmitted by the transmitter is the same as the pilot signal stored in the receiver. The receiver performs online training on the AI ​​decoder based on the difference between the recovered signal and the received signal.

2. The method as described in claim 1, characterized in that, The method further includes: The receiver receives online training instruction information sent by the network device, which is used to instruct the AI ​​decoder to perform the online training for a specified period.

3. The method as described in claim 2, characterized in that, The online training instruction information includes first instruction information, which is used to instruct the AI ​​decoder to perform the online training cycle. The online training cycle includes multiple transmission cycles, including a first type of transmission cycle and a second type of transmission cycle. The first type of transmission cycle is used to perform online training on the AI ​​decoder, and the second type of transmission cycle is used to collect training data for online training.

4. The method as described in claim 3, characterized in that, The online training instruction information also includes second instruction information, which indicates the size of the training data used in the first type of transmission cycle.

5. The method according to any one of claims 2-4, characterized in that, The online training time includes the start time and / or end time of the online training.

6. The method as described in claim 1, characterized in that, The receiver inputs the received signal into the AI ​​decoder for decoding to perform channel estimation and obtain the decoded signal, including: The receiver inputs the pilot signal stored in the receiver and the received signal into the AI ​​decoder to perform channel estimation and obtain the decoded signal.

7. The method as described in claim 1, characterized in that, The receiver generates a recovered signal of the received signal based on the decoded signal, including: The receiver inputs the decoded signal into the AI ​​encoder to generate the recovered signal.

8. The method as described in claim 7, characterized in that, The received signal includes Channel State Information (CSI) transmitted by the transmitter, and the CSI is a compressed version of CSI. The receiver inputs the received signal into the AI ​​decoder for decoding to obtain a decoded signal, including: The receiver inputs the CSI into the AI ​​decoder for decoding to obtain the decoded signal, which contains the recovered channel information. The receiver inputs the decoded signal into the AI ​​encoder to generate the recovered signal, including: The receiver inputs the decoded CSI into the AI ​​encoder for encoding, recovers the CSI, and generates the recovered signal, which contains the recovered CSI.

9. The method as described in claim 7, characterized in that, The received signal includes the first data signal transmitted by the transmitter. The receiver inputs the received signal into the AI ​​decoder for decoding to obtain a decoded signal, including: The receiver inputs the first data signal into the AI ​​decoder for decoding, recovers the data transmitted by the transmitter, and obtains the decoded signal, which contains the recovered data. The receiver inputs the decoded signal into the AI ​​encoder to generate the recovered signal, including: The receiver inputs the recovered data into the AI ​​encoder to obtain a second data signal; The receiver processes the second data signal based on the estimated second channel information and second random noise to obtain the recovered signal.

10. The method as described in claim 1, characterized in that, The receiver performs online training on the AI ​​decoder based on the difference between the recovered signal and the received signal, including: The receiver uses minimizing the difference as the optimization objective to update the weights of the AI ​​decoder and perform online training on the AI ​​decoder.

11. The method as described in claim 1, characterized in that, The difference between the recovered signal and the received signal is represented by the error between the recovered signal and the received signal.

12. A receiver, characterized in that, include: The receiving unit is used to receive the wireless signal transmitted by the transmitter and obtain the received signal, wherein the received signal is the received signal of the pilot signal transmitted by the transmitter; The processing unit is configured to input the received signal into an artificial intelligence (AI) decoder for decoding to obtain a decoded signal, wherein the received signal is input into the AI ​​decoder for decoding and channel estimation is performed to obtain the decoded signal, and the decoded signal contains the estimated first channel information; The processing unit is configured to generate a recovered signal of the received signal based on the decoded signal, wherein the pilot signal stored in the receiver is processed based on the first channel information and the first random noise to obtain the recovered signal, wherein the pilot signal transmitted by the transmitter is the same as the pilot signal stored in the receiver; The processing unit is used to perform online training on the AI ​​decoder based on the difference between the recovered signal and the received signal.

13. The receiver as claimed in claim 12, characterized in that, The receiving unit is further configured to: The system receives online training instruction information sent by a network device, which is used to instruct the AI ​​decoder to perform the online training for a specified period.

14. The receiver as claimed in claim 13, characterized in that, The online training instruction information includes first instruction information, which is used to instruct the AI ​​decoder to perform the online training cycle. The online training cycle includes multiple transmission cycles, including a first type of transmission cycle and a second type of transmission cycle. The first type of transmission cycle is used to perform online training on the AI ​​decoder, and the second type of transmission cycle is used to collect training data for online training.

15. The receiver as claimed in claim 14, characterized in that, The online training instruction information also includes second instruction information, which indicates the size of the training data used in the first type of transmission cycle.

16. The receiver as claimed in any one of claims 13-15, characterized in that, The online training time includes the start time and / or end time of the online training.

17. The receiver as claimed in claim 12, characterized in that, The processing unit is further configured to: The pilot signal stored in the receiver and the received signal are input into the AI ​​decoder to perform channel estimation and obtain the decoded signal.

18. The receiver as claimed in claim 12, characterized in that, The processing unit is further configured to input the decoded signal into the AI ​​encoder to generate the recovery signal.

19. The receiver as claimed in claim 18, characterized in that, The received signal includes Channel State Information (CSI) transmitted by the transmitter, wherein the CSI is a compressed version of the CSI. The processing unit is further configured to: The CSI is input to the AI ​​decoder for decoding to obtain the decoded signal, which contains the recovered channel information; and The recovered channel information is input into the AI ​​encoder for encoding to recover the CSI and generate the recovered signal, which contains the recovered CSI.

20. The receiver as claimed in claim 18, characterized in that, The received signal includes the first data signal transmitted by the transmitter, and the processing unit is further configured to: The first data signal is input into the AI ​​decoder for decoding to recover the data transmitted by the transmitter, thereby obtaining the decoded signal, which contains the recovered data. The recovered data is input into the AI ​​encoder to obtain the second data signal; as well as The second data signal is processed based on the estimated second channel information and the second random noise to obtain the recovered signal.

21. The receiver as claimed in claim 12, characterized in that, The processing unit is further configured to: Minimizing the difference is used as the optimization objective to update the weights of the AI ​​decoder, and the AI ​​decoder is then trained online.

22. The receiver as claimed in claim 12, characterized in that, The difference between the recovered signal and the received signal is represented by the error between the recovered signal and the received signal.

23. A receiver, characterized in that, It includes a memory and a processor, the memory being used to store a program, and the processor being used to invoke the program in the memory to perform the method as described in any one of claims 1-11.

24. An apparatus, characterized in that, Includes a processor for calling a program from memory to perform the method as described in any one of claims 1-11.

25. A chip, characterized in that, Includes a processor for calling a program from memory, causing a device on which the chip is mounted to perform the method as described in any one of claims 1-11.

26. A computer-readable storage medium, characterized in that, It contains a program that causes a computer to perform the method as described in any one of claims 1-11.

27. A computer program product, characterized in that, Includes a program that causes a computer to perform the method as described in any one of claims 1-11.

Images