Methods performed by a receiver, wireless communication devices, and storage media

Abstract

Embodiments of the present disclosure provide a method performed by a receiver, a wireless communication device and a storage medium, relating to the technical field of wireless communication and the field of artificial intelligence. The method comprises: performing data fusion on at least one of a pilot signal, channel estimation and noise estimation obtained and a received signal to obtain first data; dividing the first data into a plurality of to-be-processed data blocks in at least one dimension of a time domain, a frequency domain and a space domain; and performing demodulation processing on the plurality of to-be-processed data blocks to obtain an original transmitted signal corresponding to the received signal. Optionally, the above method performed by the receiver can be executed using an artificial intelligence model.

Description

Technical Field

[0001] This disclosure relates to the fields of wireless communication technology and artificial intelligence, and more specifically, to methods performed by a receiver, wireless communication devices, and storage media. Background Technology

[0002] In wireless communication systems, signal transmission is subject to channel impairments such as multipath fading, Doppler shift, and noise interference, which can affect the system's bit error rate (BER) performance. To combat the distortion and interference caused by the wireless channel, the receiver typically needs to perform channel estimation and equalization. Channel estimation refers to the receiver estimating the characteristics of the wireless channel, such as channel fading and delay spread. Equalization refers to the receiver processing the received signal based on the channel estimation results to eliminate the effects of the channel and recover the original transmitted signal. However, the above processes are usually based on assumptions such as known statistical characteristics of the channel and slow channel changes. In real wireless communication environments, these assumptions often do not hold, leading to a degraded performance of the aforementioned channel estimation and equalization methods. Summary of the Invention

[0003] According to one aspect of this disclosure, a method performed by a receiver in a wireless communication system is provided, comprising: performing data fusion on at least one of an acquired pilot signal, a channel estimate, and a noise estimate, and a received signal to obtain first data; dividing the first data into multiple data blocks to be processed in at least one dimension of a time domain, a frequency domain, and a spatial domain; and demodulating the multiple data blocks to obtain an original transmitted signal corresponding to the received signal.

[0004] In an exemplary embodiment, the above-described method of dividing the first data into multiple data blocks to be processed in at least one dimension of the time domain, frequency domain, and spatial domain includes: performing data augmentation on the first data to obtain second data; and dividing the second data into multiple data blocks to be processed in at least one dimension of the time domain, frequency domain, and spatial domain.

[0005] In an exemplary embodiment, the above-described methods for performing data augmentation on the first data include at least one of the following: phase rotation; row position swapping; column position swapping; and mirror flipping.

[0006] In an exemplary embodiment, the method performed by the receiver further includes: fusing information features of at least two dimensions in the time domain, frequency domain, and spatial domain of each data block to be processed to obtain multiple data blocks after feature fusion; wherein, the demodulation processing of the multiple data blocks to be processed to obtain the original transmitted signal corresponding to the received signal includes: demodulating the multiple data blocks after feature fusion to obtain the original transmitted signal.

[0007] In an exemplary embodiment, the above-mentioned processing of the first data into multiple data blocks to be processed in at least one dimension of the time domain, frequency domain, and spatial domain includes: processing the first data into multiple intermediate data blocks based on the information features of the first data in the time domain, frequency domain, or spatial domain; and performing feature fusion on the information features of at least two dimensions of the time domain, frequency domain, and spatial domain of each intermediate data block to obtain multiple data blocks to be processed.

[0008] In an exemplary embodiment, the method performed by the receiver further includes preprocessing at least one of the pilot signal, channel estimation, and noise estimation, as well as the received signal.

[0009] In an exemplary embodiment, the above-described demodulation processing of multiple data blocks to be processed to obtain the original transmitted signal corresponding to the received signal includes: inputting multiple data blocks to be processed into an artificial intelligence (AI) network for demodulation processing, wherein the AI ​​network is used to output the original transmitted signal.

[0010] In an exemplary embodiment, the above-described input of multiple data blocks to be processed into an artificial intelligence (AI) network for demodulation processing includes: processing the multiple data blocks to be processed through the embedding layer and coding layer of the AI ​​network to obtain a frequency domain output signal; converting the frequency domain output signal into a first time domain output signal; converting the first time domain output signal into a second time domain output signal through the denoising layer of the AI ​​network; and performing weighted processing on the first time domain output signal and the second time domain output signal to obtain the original transmitted signal.

[0011] In an exemplary embodiment, the loss function for training the AI ​​network is obtained by weighting the frequency domain loss function corresponding to the frequency domain output signal and the time domain loss function corresponding to the second time domain output signal.

[0012] In an exemplary embodiment, the above-mentioned processing of multiple data blocks to be processed through the embedding layer and the coding layer of the AI ​​network to obtain a frequency domain output signal includes: jointly processing multiple data blocks to be processed through the embedding layer to obtain multiple data blocks to be encoded; and encoding multiple data blocks to be encoded through the coding layer to obtain a frequency domain output signal.

[0013] In an exemplary embodiment, the above-described encoding process of multiple data blocks to be encoded through an encoding layer to obtain a frequency domain output signal includes: encoding multiple data blocks to be encoded through at least one multilayer perceptron in the encoding layer corresponding to at least one dimension in the time domain, frequency domain, and spatial domain to obtain a frequency domain output signal.

[0014] In an exemplary embodiment, the method performed by the receiver further includes: acquiring at least one of pilot signals, channel estimation, and noise estimation based on a plurality of receiving antennas, as well as a received signal.

[0015] According to other aspects of this disclosure, a wireless communication device for performing the above-described method is also disclosed.

[0016] According to other aspects of this disclosure, a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described above, which is executed by a receiver.

[0017] According to embodiments of this disclosure, by fusing at least one of the acquired pilot signal, channel estimation, and noise estimation with the received signal, and by dividing the first data obtained through data fusion into multiple data blocks to be processed in at least one dimension of the time domain, frequency domain, and spatial domain, the original transmitted signal corresponding to the received signal can be obtained through further demodulation processing of these multiple data blocks to be processed, thus achieving demodulation of the received signal. This simplifies the signal processing flow and reduces the bit error rate, thereby improving the system's bit error rate performance and throughput. Attached Figure Description

[0018] Figure 1 An example wireless network according to an embodiment of this disclosure is shown.

[0019] Figure 2 An example base station according to an embodiment of the present disclosure is shown.

[0020] Figure 3 An example user equipment according to an embodiment of the present disclosure is shown.

[0021] Figure 4 A flowchart illustrating a method performed by a receiver in a wireless communication system according to an embodiment of the present disclosure is shown.

[0022] Figure 5 A flowchart illustrating another method performed by a receiver in a wireless communication system according to an embodiment of the present disclosure is shown.

[0023] Figure 6 A flowchart illustrating another method performed by a receiver in a wireless communication system according to an embodiment of the present disclosure is shown.

[0024] Figure 7 An exemplary block diagram according to an embodiment of the present disclosure is shown.

[0025] Figure 8 Another exemplary block diagram according to an embodiment of the present disclosure is shown.

[0026] Figure 9 Another exemplary block diagram according to an embodiment of the present disclosure is shown.

[0027] Figure 10 A schematic diagram illustrating the structure and processing of an exemplary AI network according to an embodiment of the present disclosure is shown.

[0028] Figure 11 It shows the basis Figure 10 The diagram shows the processing flow of the AI ​​network.

[0029] Figure 12 It shows Figure 10 The diagram shows the processing flow of the coding layer in the AI ​​network.

[0030] Figure 13 It shows Figure 10 The diagram shows the processing flow of the denoising layer in the AI ​​network.

[0031] Figure 14 A schematic diagram of the processing procedure of another AI network according to an embodiment of the present disclosure is shown.

[0032] Figure 15 A schematic diagram of model migration according to an embodiment of the present disclosure is shown.

[0033] Figure 16a A schematic flowchart illustrating the process of generating samples according to an embodiment of the present disclosure is shown.

[0034] Figure 16b A schematic diagram illustrating the construction of a sample according to an embodiment of the present disclosure is shown.

[0035] Figure 17 An exemplary structural block diagram of a wireless communication device applicable to this disclosure is shown.

[0036] Figure 18 An exemplary structural block diagram of an electronic device applicable to this disclosure is shown. Detailed Implementation

[0037] The following description, with reference to the accompanying drawings, is provided to aid in a thorough understanding of the various embodiments of this disclosure as defined by the claims and their equivalents. This description includes various specific details to aid understanding but should be considered exemplary only. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the various embodiments described herein without departing from the scope and spirit of this disclosure. Furthermore, for clarity and brevity, descriptions of well-known functions and structures may be omitted.

[0038] The terms and wording used in the following description and claims are not limited to their dictionary meanings, but are merely used by the inventors to enable a clear and consistent understanding of this disclosure. Therefore, it will be apparent to those skilled in the art that the following description of various embodiments of this disclosure is for illustrative purposes only and not for limiting the purpose of this disclosure as defined in the appended claims and their equivalents.

[0039] It should be understood that the singular forms of “a,” “an,” and “the” can also include plural references unless the context clearly indicates otherwise. Thus, for example, the reference to “component surface” includes referring to one or more such surfaces. When we say that an element is “connected” or “coupled” to another element, the element can be directly connected or coupled to the other element, or it can mean that the element and the other element are connected through an intermediate element. Furthermore, the use of “connected” or “coupled” herein can include wireless connections or wireless couplings.

[0040] The terms “comprising” or “may include” refer to the presence of a corresponding disclosed function, operation, or component that may be used in the various embodiments of this disclosure, rather than limiting the presence of one or more additional functions, operations, or features. Furthermore, the terms “comprising” or “having” may be interpreted as indicating certain characteristics, numbers, steps, operations, constituent elements, components, or combinations thereof, but should not be construed as excluding the possibility of the presence of one or more other characteristics, numbers, steps, operations, constituent elements, components, or combinations thereof.

[0041] The term "or" as used in the various embodiments of this disclosure includes any of the listed terms and all combinations thereof. For example, "A or B" may include A, may include B, or may include both A and B. When describing multiple (two or more) items, if the relationship between the multiple items is not explicitly defined, the multiple items may refer to one, more, or all of the multiple items. For example, the description "parameter A includes A1, A2, A3" can be implemented as parameter A includes A1 or A2 or A3, or it can be implemented as parameter A includes at least two of the three items A1, A2, and A3.

[0042] Unless otherwise defined, all terms used in this disclosure (including technical or scientific terms) have the same meaning as understood by one of those skilled in the art as described herein. Common terms as defined in dictionaries are to be interpreted as having a meaning consistent with the context in the relevant technical field and should not be interpreted ideally or overly formally unless expressly defined in this disclosure.

[0043] The following Figures 1 to 3 Various embodiments of this disclosure implemented in wireless communication systems are described. Figures 1 to 3The description does not imply any physical or architectural limitations on the ways in which different embodiments can be implemented. Different embodiments of this disclosure can be implemented in any suitably arranged communication system.

[0044] Figure 1 An example wireless network according to an embodiment of this disclosure is shown. Figure 1 The embodiments of the wireless network shown are for illustrative purposes only. Other embodiments of the wireless network 100 may be used without departing from the scope of this disclosure.

[0045] like Figure 1 As shown, the wireless network includes a base station (next generation nodeB, gNB or gNodeB) 101, gNB 102, and gNB 103. gNB 101 communicates with gNB 102 and gNB 103. gNB 101 also communicates with at least one network 130 such as the Internet, a proprietary Internet Protocol (IP) network, or other data networks.

[0046] gNB 102 provides wireless broadband access to network 130 to multiple first user equipments (UEs) within coverage area 120 of gNB 102. The multiple first UEs include UE 111, which may be located in a small business (SB); UE 112, which may be located in an enterprise (E); UE 113, which may be located in a WiFi hotspot (HS); UE 114, which may be located in a first residence (R1); UE 115, which may be located in a second residence (R2); and UE 116, which may be a mobile device (M) such as a cellular phone, wireless laptop, or wireless personal digital assistant (PDA). gNB 103 provides wireless broadband access to network 130 to multiple second UEs within coverage area 125 of gNB 103. The multiple second UEs include UE 115 and UE 116, and subscriber stations (SS, such as UEs) 117, 118, and 119. In some embodiments, one or more of gNBs 101-103 may communicate with each other and UEs 111-116 using existing wireless communication technologies, and one or more of UEs 111-119 may communicate directly with each other (e.g., UEs 117-119) using other existing or proposed wireless communication technologies.

[0047] Depending on the network type, the term "base station" or "BS" can refer to any component (or set of components) configured to provide wireless access to a network, such as a transmit point (TP), transmit-receive point (TRP), enhanced (or "evolved") base station (eNodeB or eNB), 5G base station (gNB), macro cell, femtocell, wireless fidelity (WiFi) access point (AP), or other wirelessly capable device. A base station can provide wireless access according to one or more wireless communication protocols, such as 3GPP 5G new radio (NR), Long Term Evolution (LTE), LTE-Advanced (LTE-A), high-speed packet access (HSPA), Wi-Fi 802.11a / b / g / n / ac, etc. For convenience, various names for base station type devices and functions may be used interchangeably in this patent document to refer to network infrastructure components that provide wireless access to remote terminals. Furthermore, depending on the network type, the term "User Equipment" (UE) can refer to any component such as a mobile station (MS), user station (SS), remote terminal, wireless terminal, receiving point, or user device. For convenience, various names for user equipment type devices and functions may be used interchangeably in this patent document to refer to remote wireless devices that wirelessly access the BS regardless of whether the UE is a mobile device (such as a mobile phone or smartphone) or is generally considered a fixed device (such as a desktop computer or vending machine).

[0048] The dashed lines indicate the approximate extent of coverage areas 120 and 125, which are shown as roughly circular for illustrative and explanatory purposes only. It should be clearly understood that coverage areas such as 120 and 125 associated with the gNB can have other shapes, including irregular shapes, depending on the configuration of the gNB and variations in the wireless environment associated with natural and man-made obstacles.

[0049] As described in more detail below, one or more of UEs 111-119 include circuitry, programming, or a combination thereof. In some embodiments, one or more of gNBs 101-103 include circuitry, programming, or a combination thereof.

[0050] although Figure 1 An example of a wireless network is shown, but more can be found on... Figure 1Various modifications can be made. For example, wireless network 100 can include any number of gNBs and any number of UEs in any suitable arrangement. Furthermore, gNB 101 can communicate directly with any number of UEs and provide those UEs with wireless broadband access to network 130. Similarly, each gNB 102-103 can communicate directly with network 130 and provide UEs with direct wireless broadband access to network 130. Additionally, gNBs 101, 102, and / or 103 can provide access to other or additional external networks, such as external telephone networks or other types of data networks.

[0051] Figure 2 An example base station according to an embodiment of the present disclosure is shown. Figure 2 The embodiment of gNB 102 shown is for illustrative purposes only, and Figure 1 gNBs 101 and 103 can have the same or similar configurations. However, gNBs come in a variety of configurations, and Figure 2 This disclosure is not intended to limit the scope to any particular implementation of gNB.

[0052] like Figure 2 As shown, gNB 102 includes multiple antennas 200a-200n, multiple radio frequency (RF) transceivers 201a-201n, transmit (TX) processing circuitry 203, and receive (RX) processing circuitry 204. gNB 102 also includes a controller / processor 205, a memory 206, and a backhaul or network interface 207.

[0053] RF transceivers 201a-201n receive incoming RF signals from antennas 200a-200n, such as signals transmitted by the UE in network 100. RF transceivers 201a-201n down-convert the incoming RF signals to generate intermediate frequency (IF) or baseband signals. The IF or baseband signal is sent to RX processing circuitry 204, which generates a processed baseband signal by filtering, decoding, and / or digitizing the baseband or IF signal. RX processing circuitry 204 sends the processed baseband signal to controller / processor 205 for further processing.

[0054] The TX processing circuit 203 receives analog or digital data (such as voice data, web data, email, or interactive video game data) from the controller / processor 205. The TX processing circuit 203 encodes, multiplexes, and / or digitizes the outgoing baseband data to generate a processed baseband or IF signal. RF transceivers 201a-201n receive the processed baseband or IF signal from the TX processing circuit 203 and up-convert the baseband or IF signal into an RF signal transmitted via antennas 201a-201n.

[0055] The controller / processor 205 may include one or more processors or other processing devices that control the overall operation of the gNB 102. For example, the controller / processor 205 may control the reception of forward channel signals and the transmission of reverse channel signals by the RF transceivers 201a-201n, the RX processing circuit 204, and the TX processing circuit 203, according to known principles. The controller / processor 205 may also support additional functions, such as more advanced wireless communication functions.

[0056] For example, the controller / processor 205 can support beamforming or directional routing operations, where signals emitted from multiple antennas 200a-200n are weighted differently to effectively redirect the emitted signals in the desired direction. Any of a variety of other functions can be supported in the gNB 102 via the controller / processor 205.

[0057] The controller / processor 205 is also capable of executing programs and other processes located in memory 206, such as the operating system (OS). The controller / processor 205 can move data into or out of memory 206 as needed by the executing process.

[0058] The controller / processor 205 is also connected to a backhaul or network interface 207. The backhaul or network interface 207 allows the gNB 102 to communicate with other devices or systems via a backhaul connection or over a network. Interface 207 can support communication via any suitable wired or wireless connection(s). For example, when the gNB 102 is implemented as part of a cellular communication system (such as a cellular communication system supporting 5G, LTE, or LTE-A), interface 207 can allow the gNB 102 to communicate with other gNBs via a wired or wireless backhaul connection. When the gNB 102 is implemented as an access point, interface 207 can allow the gNB 102 to communicate via a wired or wireless local area network or via a wired or wireless connection to a larger network (such as the Internet). Interface 207 includes any suitable structure that supports communication via wired or wireless connections such as Ethernet or RF transceivers.

[0059] Memory 206 is connected to controller / processor 205. A portion of memory 206 may include random access memory (RAM), and another portion of memory 206 may include flash memory or other read-only memory (ROM).

[0060] although Figure 2 An example of gNB 102 is shown, but it is possible to see more. Figure 2 Various changes can be made. For example, gNB 102 can include any number of Figure 2 Each component is shown in the diagram. As a specific example, an access point may include multiple interfaces 207, and the controller / processor 205 may support routing functionality to route data between different network addresses. As another specific example, although shown as a single instance of TX processing circuitry 203 and a single instance of RX processing circuitry 204, gNB102 may include multiple instances of each (such as one per RF transceiver). For example, Figure 2 The various components can be combined, further subdivided, or omitted, and additional components can be added as needed.

[0061] Figure 3 An example user equipment according to an embodiment of the present disclosure is shown. Figure 3 The embodiment of UE 116 shown is for illustrative purposes only, and Figure 1 UEs 111-115 and 117-119 can have the same or similar configurations. However, UEs appear in multiple configurations, and Figure 3 This disclosure is not intended to limit the scope to any particular implementation of the UE.

[0062] like Figure 3 As shown, UE 116 includes an antenna 301, a radio frequency (RF) transceiver 302, a TX processing circuit 303, a microphone 304, and a receive (RX) processing circuit 305. UE 116 also includes a speaker 306, a controller or processor 307, an input / output (I / O) interface (IF) 308, an input device 309, a touchscreen display 310, and memory 311. Memory 311 includes an OS 312 and one or more applications 313.

[0063] RF transceiver 302 receives incoming RF signals transmitted by gNB of network 100 from antenna 301. RF transceiver 302 down-converts the incoming RF signals to generate an IF or baseband signal. The IF or baseband signal is sent to RX processing circuitry 305, which generates a processed baseband signal by filtering, decoding, and / or digitizing the baseband or IF signal. RX processing circuitry 305 sends the processed baseband signal to speaker 306 (e.g., for voice data) or processor 307 for further processing (e.g., for web browsing data).

[0064] TX processing circuit 303 receives analog or digital voice data from microphone 304 or other outgoing baseband data (such as web data, email, or interactive video game data) from processor 307. TX processing circuit 303 encodes, multiplexes, and / or digitizes the outgoing baseband data to generate a processed baseband or IF signal. RF transceiver 302 receives the processed baseband or IF signal from TX processing circuit 303 and up-converts the baseband or IF signal into an RF signal transmitted via antenna 301.

[0065] Processor 307 may include one or more processors or other processing devices and executes OS 312 stored in memory 311 to control the overall operation of UE 116. For example, processor 307 may control the reception of forward channel signals and the transmission of reverse channel signals by RF transceiver 302, RX processing circuitry 305, and TX processing circuitry 303 according to known principles. In some embodiments, processor 307 includes at least one microprocessor or microcontroller.

[0066] Processor 307 is also capable of executing other processes and programs located in memory 311, such as processes for CSI reporting on the uplink channel. Processor 307 can move data into or out of memory 311 as needed for executing processes. In some embodiments, processor 307 is configured to execute application 313 based on OS 312 or in response to signals received from gNB or operator. Processor 307 is also coupled to I / O interface 308, which provides UE 116 with the ability to connect to other devices such as laptops and laptops. I / O interface 308 is the communication path between these accessories and processor 307.

[0067] The processor 307 is also connected to the touchscreen display 310. The user of the UE 116 can use the touchscreen display 310 to input data into the UE 116. The touchscreen display 310 can be a liquid crystal display, a light-emitting diode display, or other display capable of rendering text and / or at least limited graphics such as those from a website.

[0068] Memory 311 is connected to processor 307. A portion of memory 311 may include RAM, and another portion of memory 311 may include flash memory or other ROM.

[0069] although Figure 3 An example of UE 116 is shown, but it is possible to modify it. Figure 3 Make various changes. For example, Figure 3 The various components can be combined, further subdivided, or omitted, and additional components can be added as needed. As a specific example, processor 307 can be divided into multiple processors, such as one or more central processing units (CPUs) and one or more graphics processing units (GPUs). Moreover, although... Figure 3 The UE 116 is shown configured as a mobile phone or smartphone, but the UE can be configured to operate as other types of mobile or fixed devices.

[0070] At least some of the functions of the device or electronic device provided in this disclosure embodiment can be implemented by an AI model, such as implementing at least one module of a plurality of modules of the device or electronic device by an AI model. AI-related functions can be executed by non-volatile memory, volatile memory, and a processor.

[0071] The processor may include one or more processors. In this case, the one or more processors may be general-purpose processors, such as central processing unit (CPU), application processor (AP), etc., or pure graphics processing unit, such as graphics processing unit (GPU), vision processing unit (VPU), and / or AI-specific processors, such as neural processing unit (NPU).

[0072] The one or more processors control the processing of input data based on predefined operating rules or AI models stored in non-volatile memory and volatile memory. These predefined operating rules or AI models are provided through training or learning.

[0073] Here, "providing through learning" refers to obtaining predefined operating rules or an AI model with desired characteristics by applying a learning algorithm to multiple learning datasets. This learning can be performed within the device or electronic device itself, in which the AI ​​is executed according to the embodiment, and / or can be implemented via a separate server / system.

[0074] AI models can contain multiple neural network layers. Each layer has multiple weight values, and each layer performs neural network computations by calculating the input data of that layer (such as the computation results of the previous layer and / or the input data of the AI ​​model) and the multiple weight values ​​of the current layer. Examples of neural networks include, but are not limited to, convolutional neural networks (CNNs), deep neural networks (DNNs), recurrent neural networks (RNNs), restricted Boltzmann machines (RBMs), deep belief networks (DBNs), bidirectional recurrent deep neural networks (BRDNNs), generative adversarial networks (GANs), and deep Q-networks.

[0075] A learning algorithm is a method of training a predetermined target device (e.g., a robot) using multiple learning data sets to enable, allow, or control the target device to make determinations or predictions. Examples of such learning algorithms include, but are not limited to, supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning.

[0076] The methods provided in this disclosure may relate to one or more fields in the technical fields of speech, language, image, video, or data intelligence.

[0077] Optionally, in the context of speech or language, according to this disclosure, in a method performed by an electronic device, a method for recognizing a user's speech and interpreting the user's intent may receive a speech signal as an analog signal via a speech signal acquisition device (e.g., a microphone) and convert the speech portion into computer-readable text using an Automatic Speech Recognition (ASR) model. The user's utterance intent can be obtained by interpreting the converted text using a Natural Language Understanding (NLU) model. The ASR model or NLU model can be an artificial intelligence model. The artificial intelligence model can be processed by a dedicated artificial intelligence processor designed in a hardware architecture specified for processing the artificial intelligence model. The artificial intelligence model can be obtained through training. Here, "obtained through training" means obtaining a predefined operating rule or artificial intelligence model configured to perform desired features (or purposes) by training a basic artificial intelligence model with multiple training data using a training algorithm. Language understanding is a technique for recognizing and applying / processing human language / text, including, for example, natural language processing, machine translation, dialogue systems, question answering, or speech recognition / synthesis.

[0078] Optionally, in the realm of images or videos, according to this disclosure, in methods executed in electronic devices, the method for identifying objects can obtain output data for identifying images or image features in images by using image data as input data for an artificial intelligence model. The artificial intelligence model can be obtained through training. Here, "obtained through training" means obtaining a predefined operating rule or artificial intelligence model configured to perform desired features (or purposes) by training a basic artificial intelligence model with multiple training data using a training algorithm. The methods of this disclosure can relate to the field of visual understanding in artificial intelligence technology, which is a technology for recognizing and processing things like human vision, and includes, for example, object recognition, object tracking, image retrieval, human recognition, scene recognition, 3D reconstruction / localization, or image enhancement.

[0079] Optionally, in the context of data intelligence processing, according to this disclosure, in methods executed in electronic devices, the method for reasoning or predicting object categories can use an artificial intelligence model to recommend / perform category recognition by using feature data. The processor of the electronic device can perform preprocessing operations on the data to transform it into a form suitable for use as input to an artificial intelligence model. The artificial intelligence model can be obtained through training. Here, "obtained through training" means obtaining a predefined operating rule or artificial intelligence model configured to perform desired features (or purposes) by training a basic artificial intelligence model with multiple training data using a training algorithm. Reasoning and prediction are techniques for logical reasoning and prediction based on determined information, including, for example, knowledge-based reasoning, optimization prediction, preference-based planning, or recommendation.

[0080] Optionally, in the field of signal processing in wireless communication, according to this disclosure, in methods executed in communication devices (such as terminals and / or base stations), the method for reasoning or predicting constellation point symbols can use an artificial intelligence model to recommend / perform symbol detection by using information features. The processor of the communication device can perform preprocessing operations on the data to transform it into a form suitable for use as input to the artificial intelligence model. The artificial intelligence model can be obtained through training. Here, "obtained through training" means obtaining a predefined operating rule or artificial intelligence model configured to perform desired features (or purposes) by training a basic artificial intelligence model with multiple training data using a training algorithm. Reasoning and prediction are techniques for logical reasoning and prediction based on determined information, including, for example, knowledge-based reasoning, optimization prediction, preference-based planning, or recommendation.

[0081] In wireless communication systems, signal transmission is subject to channel impairments such as multipath fading, Doppler shift, and noise interference, which can affect the system's bit error rate performance. To combat the distortion and interference caused by wireless channels, one approach is to use non-artificial intelligence (AI) based solutions. With the widespread application of AI technology in the field of wireless communication, another approach is to use AI-based solutions.

[0082] In non-AI-based solutions, the receiver needs to perform channel estimation and equalization. Channel estimation can begin using pilot signals: the channel impulse response (CIR) or channel frequency response (CFR) is calculated based on the received pilot signals. Common algorithms include, but are not limited to, the least squares (LS) method and the minimum mean squared error (MMSE) method. Then, channel interpolation is performed at the data symbol positions, and channel equalization is performed using the interpolated channel, i.e., compensating for data symbols using the estimated channel information. Specific equalizers include MMSE equalizers and zero-forcing (ZF) equalizers. This approach suffers from high bit error rate and throughput loss.

[0083] In AI-based solutions, given AI's powerful learning and adaptive capabilities, it can learn the complex characteristics of the channel from large amounts of data, thereby achieving more accurate channel estimation and equalization. Specifically, channel estimation, equalization, and decoding can be integrated into a single network, such as using a Recurrent Neural Network (RNN), a Deep Neural Network (DNN), a Residual Network (ResNet), or a Convolutional Neural Network (CNN). While this eliminates the need for explicit channel estimation and reduces error propagation caused by channel estimation to some extent, it requires collecting a large amount of data for training, resulting in high complexity and limited gain in receiver bit error rate performance.

[0084] Therefore, a solution is needed to improve the system's bit error rate performance and throughput.

[0085] The following description of several optional embodiments illustrates the technical solutions of this disclosure and the technical effects produced by these solutions. It should be noted that the following embodiments can be referenced, learned from, or combined with each other. Identical terms, similar features, and similar implementation steps in different embodiments will not be repeated.

[0086] Figure 4 A flowchart illustrating a method performed by a receiver in a wireless communication system according to an embodiment of the present disclosure is shown, such as... Figure 4 As shown, the process 400 of the method executed by the receiver may include the following steps: Step 401: Perform data fusion on at least one of the acquired pilot signal, channel estimation, and noise estimation, as well as the received signal, to obtain the first data.

[0087] This step aims to fuse at least one of the pilot signal, channel estimation, and noise estimation with the received signal (Rx) by the aforementioned execution entity to obtain fused first data. The specific form of the first data is not limited in this embodiment; it depends on the specific circumstances. In one implementation of this embodiment, the obtained pilot signal, channel estimation, and noise estimation, along with the received signal, can be represented in matrix form. When fusing the pilot signal, channel estimation, noise estimation, and the received signal, a splicing fusion method or a stacking fusion method can be used. The splicing fusion method refers to understanding the matrix in the current dimension (rows or columns, etc.), i.e., without creating new dimensions, the number of dimensions remains unchanged, but the length of a certain dimension increases. The stacking fusion method refers to combining matrices in new dimensions to generate a higher-dimensional matrix, i.e., increasing the number of dimensions (e.g., adding an axis, from two-dimensional to three-dimensional). For example, the first data can include, but is not limited to, data in tensor form, where tensors are high-dimensional extensions of vectors and matrices.

[0088] The pilot signal is used by the receiver to estimate channel quality, perform channel estimation and tracking to compensate for the channel's impact on the data signal, thereby accurately demodulating the received data and recovering the original transmitted signal. In some optional implementations of this disclosure, the pilot signal includes, but is not limited to, a demodulation reference signal (DMRS), such as... Figure 6 Specific examples are provided.

[0089] Channel estimation (CE) can estimate the channel at the location of the pilot signal (i.e., the pilot position). For example, taking the pilot signal as DMRS, due to frequency-selective fading of the channel, the channel coefficients (such as phase and amplitude) of DMRS are usually continuously changing. Therefore, the Hilbert-Huang Transform (HHT) can be used to perform denoising channel estimation on DMRS. In some optional implementations of this disclosure, the channel estimation can be based on the LS algorithm, i.e., it can be LS CE, such as... Figure 6 Specific examples are provided.

[0090] Noise estimation can be implemented based on a reference signal such as DMRS. In some alternative implementations disclosed herein, noise estimation can also be expressed as noise and interference estimation (N&I), such as... Figure 6 Specific examples are provided. Specifically, noise estimation of the pilot positions can be performed first (e.g., ...). Figure 6 The shaded area in the lower right corner is then used to calculate the noise estimate at the pilot position. The average value of this estimate is then assigned to the noise estimate corresponding to the location of the data symbol (e.g., ...). Figure 6 The time-domain averaged noise is indicated by the dotted portion shown in the lower right corner, thus obtaining the final noise estimate.

[0091] Step 402: Divide the first data into multiple data blocks to be processed in at least one dimension of the time domain, frequency domain, and spatial domain.

[0092] In the embodiments of this disclosure, the communication dimension involved may include at least one of the time domain, frequency domain and spatial domain, and the block processing of the first data obtained by data fusion can be implemented in one or more of these dimensions to obtain a fine granularity for recovering the original transmitted signal, i.e., each data block to be processed.

[0093] Step 403: Demodulate multiple data blocks to obtain the original transmitted signal corresponding to the received signal.

[0094] According to the method provided in this disclosure, data fusion is performed on at least one of the acquired pilot signal, channel estimation, and noise estimation, as well as the received signal. The first data obtained through data fusion is then divided into multiple data blocks to be processed in at least one dimension (time domain, frequency domain, and spatial domain). Further demodulation processing of these multiple data blocks to be processed yields the original transmitted signal corresponding to the received signal, thus achieving demodulation of the received signal. This simplifies the signal processing flow and reduces the bit error rate, thereby improving the system's bit error rate performance and throughput.

[0095] Figure 5 A flowchart illustrating another method performed by a receiver in a wireless communication system according to an embodiment of the present disclosure is shown, such as... Figure 5 As shown, the process 500 of the method executed by the receiver may include the following steps: Step 501: Perform data fusion on at least one of the acquired pilot signal, channel estimation, and noise estimation, as well as the received signal, to obtain the first data.

[0096] This step 501 is similar to... Figure 4 The steps shown in step 401 are the same; for the same parts, please refer to the corresponding parts of the previous embodiment, which will not be repeated here. Further, step 402 in the above embodiment can be specifically executed as steps 502 and 503 as follows: Step 502: Perform data augmentation on the first data to obtain the second data.

[0097] In this step, in order to solve the problem of limited data in wireless communication, data augmentation can be used to expand the data, so as not to collect a large amount of data. Then, based on the data augmented data, block processing is performed, as shown in step 503 below, so as to obtain the original transmitted signal more accurately.

[0098] In some optional implementations of the embodiments of this disclosure, the methods for performing data augmentation processing on the first data include, but are not limited to, at least one of the following: phase rotation; row position swapping; column position swapping; and mirror flipping.

[0099] For example, the received signal can be represented as y, the channel estimate as H, the original transmitted signal as X, and the noise estimate as n. Then the transmit / receive relationship can be represented as:

[0100] Furthermore, the dataset can be expanded using one of the following data augmentation methods: In data augmentation method 1, phase rotation is performed simultaneously on the received signal and the channel estimation. If the received signal is rotated and the channel estimation is combined while keeping the transmitted signal unchanged, then the new transmit / receive relationship can be expressed as:

[0101] In data augmentation method 2, the received signal and the original transmitted signal are simultaneously phase rotated. If we combine the rotated received signal and transmitted signal while keeping the channel estimation unchanged, then the new transmit / receive relationship can be expressed as:

[0102] In data augmentation method 3, the row or column positions of the received and transmitted signals are simultaneously swapped. This swapping can be represented by a permutation matrix P, where each row or column contains only one 1 element and all other elements are 0. The new transmit / receive relationship can then be represented as:

[0103] In this data augmentation method 3, the aforementioned permutation matrix P includes, but is not limited to, a cyclic shift matrix.

[0104] In data augmentation method 4, the received signal, channel estimation, and transmitted signal are simultaneously mirrored and flipped, and the flipped value is... H X The new send / receive relationship can then be represented as:

[0105] In some alternative implementations of the embodiments of this disclosure, for multi-layer signal input, exemplarily, such as in a multi-user multiple-input multiple-output (MU-MIMO) communication system, M single-antenna users can transmit signals to a gNB with NR receiving antennas on the same physical resource. The received signal on that physical resource (e.g., a resource element (RE) in each frequency domain) can be represented as y, and the channel estimate for the m-th user can be represented as... The original transmitted signal of the m-th user is represented as Where m is the user number, and the noise estimation is expressed as... Then the sending and receiving relationship can be represented as:

[0106] Furthermore, the dataset can be expanded using one of the following data augmentation methods: In data augmentation method 1', phase rotation is performed simultaneously on the received signal and the channel estimation. If the received signal is rotated and the channel estimation is combined while keeping the transmitted signal unchanged, then the new transmit / receive relationship can be expressed as:

[0107] In data augmentation method 2', the received signal and the original transmitted signal are simultaneously phase rotated. If we combine the rotated received signal and transmitted signal while keeping the channel estimation unchanged, then the new transmit / receive relationship can be expressed as:

[0108] In data augmentation method 3', the row or column positions of the received and transmitted signals are simultaneously swapped. This swapping can be represented by a permutation matrix P, where each row or column contains only one 1 element and all others are 0. This cyclic shifting along the input time domain expands the diversity of the time domain. The new transmit / receive relationship can then be represented as:

[0109] In this data augmentation method 3', the aforementioned permutation matrix P includes, but is not limited to, a cyclic shift matrix.

[0110] In data augmentation method 4', the received signal, channel estimation, and transmitted signal are simultaneously mirrored and flipped, and the flipped value is... H X The new send / receive relationship can then be represented as:

[0111] Step 503: Divide the second data into multiple data blocks to be processed in at least one dimension of the time domain, frequency domain, and spatial domain.

[0112] This step 503 is similar to... Figure 4 The steps 403 shown are basically the same. For the same or similar parts, please refer to the corresponding parts of the previous embodiment, which will not be repeated here. Furthermore, step 403 in the above embodiment can be specifically executed as the following step 504: Step 504: Input multiple data blocks to be processed into the artificial intelligence (AI) network for demodulation processing to obtain the original transmitted signal corresponding to the received signal. The AI ​​network is used to output the original transmitted signal.

[0113] In this embodiment, an AI-based receiver algorithm can be used to input multiple data blocks to be processed, obtained by processing at least one of the acquired pilot signal, channel estimation, and noise estimation, and the received signal as described above, into a pre-trained AI network. Without the need for channel interpolation of the data signal position or explicit channel equalization using the interpolated channel, the original transmitted signal corresponding to the received signal can be obtained, thus achieving efficient implicit channel equalization and demodulation of the data signal.

[0114] It should be noted that the AI ​​network in this embodiment can be trained based on sample received signals, sample pilot signals, sample channel estimation, and sample noise estimation, using the original transmitted signals as labels. Furthermore, to ensure the richness of the sample dataset, data augmentation processing can be performed on the aforementioned sample data. This generates a large amount of high-quality sample data, effectively compensating for the problems of model overfitting and insufficient robustness caused by insufficient data. The specific training process is not specifically limited here.

[0115] exist Figure 5 Based on the corresponding embodiments, in some optional implementations of the embodiments of this disclosure, the AI ​​network in step 504 above may at least sequentially include an embedding layer, an encoding layer, and a denoising layer. The process of processing the multiple data blocks to be processed based on this AI network may specifically include: (1) Multiple data blocks to be processed are processed through the embedding layer and the coding layer of the AI ​​network to obtain the frequency domain output signal; specifically, multiple data blocks to be processed can be jointly processed through the embedding layer to obtain multiple data blocks to be encoded; and multiple data blocks to be encoded can be encoded through the coding layer to obtain the frequency domain output signal.

[0116] (2) Convert the frequency domain output signal into the first time domain output signal. For example, the frequency domain output signal can be converted into the time domain output signal by means of the inverse discrete Fourier transform (IDFT).

[0117] (3) The first time domain output signal is converted into the second time domain output signal through the denoising layer of the AI ​​network.

[0118] (4) Weight the first time domain output signal and the second time domain output signal to obtain the original transmission signal.

[0119] In this embodiment, by transforming the time-frequency and frequency-domain data, the feature capture capability in the time and frequency domains is enhanced, the noise impact in the frequency and time domains can be effectively reduced, and the diversity characteristics of the received signal can be achieved through weighting.

[0120] In some optional implementations of the embodiments of this disclosure, the loss function for training the AI ​​network is obtained by weighting the frequency domain loss function corresponding to the frequency domain output signal and the time domain loss function corresponding to the second time domain output signal. Thus, in this embodiment, when training the AI ​​network for demodulating multiple data to be processed based on at least one of the acquired pilot signal, channel estimation, and noise estimation, and the received signal, a dual-supervised network mode in both the time and frequency domains is adopted. By designing multi-level loss functions, the information in both the time and frequency domains can be enhanced, further improving the neural network's ability to learn the relationships between multiple signal domains, thereby enhancing the signal decoding performance and reducing the bit error rate.

[0121] In some optional implementations of the embodiments of this disclosure, the step of encoding multiple data blocks to be encoded through an encoding layer to obtain a frequency domain output signal can be specifically executed as follows: encoding multiple data blocks to be encoded using at least one multilayer perceptron (MLP) in the encoding layer corresponding to at least one dimension in the time domain, frequency domain, and spatial domain to obtain a frequency domain output signal. This enables effective extraction of multi-domain or multi-dimensional features.

[0122] exist Figure 4 or Figure 5 Based on the corresponding embodiments, in some optional implementations of the embodiments of this disclosure, the method executed by the receiver may further include the following steps: acquiring at least one of pilot signals, channel estimation, and noise estimation, as well as the received signal, based on multiple receiving antennas. In this embodiment, the method executed by the receiver can be applied to multi-antenna systems, such as 8-antenna, 16-antenna, etc., and can also be applied to massively multi-antenna systems, such as 64-antenna, 128-antenna systems, etc.

[0123] exist Figure 4 or Figure 5 Based on the corresponding embodiments, in some optional implementations of the embodiments of this disclosure, the method executed by the receiver may further include the following steps: At least one of the pilot signal, channel estimation, and noise estimation, as well as the received signal, is preprocessed. This preprocessing, also known as pre-processing, can be performed in some optional implementations of this disclosure. Methods such as normalization and denoising can be used to preprocess or pre-process the pilot signal, channel estimation, and noise estimation, as well as the received signal, thereby ensuring the reliability of the signal processing.

[0124] like Figure 6In a specific example, during the preprocessing operation of process 601, the received signal, DMRS, LS CE, and noise and interference estimates can be normalized to obtain normalized DMRS, LS CE, and noise and interference estimates.

[0125] Furthermore, in Figure 5 Based on the corresponding embodiments, in some optional implementations of the embodiments of this disclosure, the step 502 of performing data augmentation processing on the first data can be specifically executed as follows: performing data fusion based on at least one of the preprocessed pilot signal, channel estimation and noise estimation and the received signal to obtain the first data, and performing data augmentation processing.

[0126] In some alternative implementations of the embodiments of this disclosure, at least one of the acquired pilot signal, channel estimation and noise estimation and the received signal can be fused to obtain first data; then the first data can be preprocessed, and then the preprocessed first data can be augmented.

[0127] like Figure 6 In a specific example, in the data augmentation operation of process 602, data augmentation processing can be performed on the DMRS, LS CE, noise and interference estimation, and received signal after the preprocessing or preprocessing operations (such as normalization, denoising, etc.) in process 601.

[0128] exist Figure 4 or Figure 5 Based on the corresponding embodiments, in some optional implementations of the embodiments of this disclosure, the method executed by the receiver may further include the following: feature fusion of information features in at least two dimensions of the time domain, frequency domain, and spatial domain of each data block to be processed to obtain multiple feature-fused data blocks. Furthermore, step 403 or 503 may be specifically executed as follows: demodulation processing of the multiple feature-fused data blocks to obtain the original transmitted signal.

[0129] exist Figure 4 or Figure 5 Based on the corresponding embodiments, in some optional implementations of the embodiments of this disclosure, the process of block processing in step 402 or step 503 above can be specifically executed as follows: Based on the information features of the first or second data in the time domain, frequency domain, or spatial domain, the first or second data is divided into multiple intermediate data blocks; the information features of at least two dimensions in the time domain, frequency domain, and spatial domain of each intermediate data block are fused to obtain multiple data blocks to be processed.

[0130] In this embodiment, feature fusion can be performed using information features from multiple domains (at least two of the time domain, frequency domain, and spatial domain), and multi-dimensional or multi-domain data blocks (or signal blocks) can be obtained based on this. That is, by combining and dividing information features from at least two domains in time, frequency, and space, multi-domain features are fused so that the original transmitted signal can be obtained after processing by the AI ​​network. In other words, the relationship between the original transmitted signal and the received signal can be accurately mapped, thereby achieving high throughput and low bit error rate symbol detection.

[0131] like Figure 6 A specific example is found in the time-frequency spatial partitioning operation of process 603. Specifically, the receiving antenna (spatial domain), frequency (frequency domain), and time (time domain) dimensions can be stacked into a three-dimensional matrix (e.g., Figure 7 As shown in Part A), it is then divided into blocks, and the length, width, and height of each block can be arbitrarily configured, thereby obtaining local features under different windows (such as, but not limited to, time-frequency spatial domain relationships). After block division, the features of each block are processed by the AI ​​network in process 604 to combine the blocks obtained in process 603 (such as... Figure 7 The B portion shown in the diagram serves as the input to the AI ​​network, and the predicted original transmitted signal can be output in process 605.

[0132] In some optional implementations of the embodiments of this disclosure, the process of dividing the first data into multiple data blocks to be processed in at least one dimension of the time domain, frequency domain, and spatial domain (such as step 402 or step 503 above) can be integrated into the AI ​​network according to specific needs, and the AI ​​network can implement the block division function. Alternatively, it can be implemented outside the AI ​​network using other methods or other neural networks, which are not specifically limited here.

[0133] The following description uses the uplink wireless communication process as an example. It should be noted that the embodiments of this disclosure are also applicable to the downlink wireless communication process. In this specific implementation, the wireless communication system includes one transmitting antenna and four receiving antennas, i.e., K=4. In each subframe's received signal (it should be noted that the received signal can also be acquired at other granularities such as time slots and resource blocks), there are 120 frequency domain subcarriers F, 14 time domain symbols T, and DMRS is placed at positions 3 and 10. The signals on each receiving antenna are arranged according to the corresponding positions in the subframe as follows: Receive antenna 1: [Y_real, Y_imag, DMRS_real, DMRS_imag, H_real, H_imag,Noise]_ant1 Receiver antenna 2: [Y_real, Y_imag, DMRS_real, DMRS_imag, H_real, H_imag, Noise] _ant2 Receiver antenna 3: [Y_real, Y_imag, DMRS_real, DMRS_imag, H_real, H_imag, Noise] _ant3 Receiver antenna 4: [Y_real, Y_imag, DMRS_real, DMRS_imag, H_real, H_imag, Noise] _ant4 As described above, Y represents the received signal, H represents the channel estimation, and Noise represents the noise estimation. real represents the real part, and image represents the imaginary part. The total dimension of the matrix is ​​K. F T 7. Specifically, it can be a four-dimensional tensor with spatial, temporal, and frequency domains, as well as features. Here, 7 refers to the number of subframe signal features, including the aforementioned Y_real, Y_imag, DMRS_real, DMRS_imag, H_real, H_imag, and Noise.

[0134] As described above, considering multi-layer signal input, where Y represents the received signal, H represents channel estimation, and Noise represents noise estimation, real represents the real part, and image represents the imaginary part, the total dimension of the matrix is ​​K. F T S can specifically be a four-dimensional tensor with spatial, temporal, and frequency domains, as well as features. Where S = 2 + L 5 refers to the number of subframe signal features, where L is the number of multi-layer signals, and the feature dimensions include: Y_real, Y_imag, and the real and imaginary parts of each DMRS symbol, i.e., DMRS_real. L、DMRS_imag L, the real and imaginary components of the channel estimation at each layer, i.e., H_real L and H_imag Noise estimation for L and each layer L.

[0135] In this specific implementation, time-frequency spatial partitioning can be considered as a combination of K, F, and T. In one example, each antenna can be considered as a unit, and the entire F... Let T be a plane, considering the time and frequency domain relationships across the entire antenna, i.e., divided into K F... The plane of T, such as Figure 8 As shown.

[0136] In another example, the receiving antenna and the frequency domain can also be considered as a plane, taking into account the combination of the spatial and frequency domains, i.e., divided into T K... F plane, such as Figure 9 As shown.

[0137] In another example, the receiving antenna can be combined with the frequency and time domains to form a cube, comprehensively considering the combination of the spatial, frequency, and time domains. Each cube can have different dimensions (i.e., length, width, and height) or the same dimensions (e.g., ...). Figure 7 (as shown in Part B).

[0138] Furthermore, the result after time-frequency spatial partitioning can be input into the AI ​​network to obtain the corresponding output. The structure and processing of the AI ​​network are as follows: Figure 10 As shown, the AI ​​network includes an embedding layer, an encoding layer, and a denoising layer, specifically: Step 1: The embedding layer concatenates blocks based on time-frequency spatial division. The length and size of each block can vary and be adjusted arbitrarily. The embedding layer can also set the window size (or the size of the coding field of view) to perform joint processing or masking processing on multiple blocks.

[0139] Step 2: The encoding layer encodes the output of the embedding layer. The encoding method can incorporate attention mechanisms, convolution, channel mixing, etc. For example, an attention mechanism can be introduced into this encoding layer to perform attention calculations on each block / face, or channel mixing can be performed on each block / face.

[0140] In this specific implementation, a domain hybrid network is provided to improve the feature fusion capability of the received signal. By comprehensively considering complexity and performance, the data is segmented, and the signal tensor is separated into multiple planes. For example... Figure 11 As shown, specifically, it can include: (1) First, the resources of each slice (or each surface), i.e., spatial and frequency domains, can be obtained by slicing along the time domain, and each slice can be connected to the MLP module marked as spatial and frequency domain; then the input is transposed, and the time domain features are processed by the channel mixing MLP i.e., the time domain MLP module. (2) Then, the resources of each slice (or each surface), i.e., spatial and time domains, can be obtained by slicing along the frequency domain, and each slice can be connected to the MLP module marked as spatial and time domains; then the input is transposed, and the frequency domain features are processed by the channel mixing MLP i.e., the frequency domain MLP module. (3) Finally, the resources of each slice (or each surface), i.e., frequency and time domains, can be obtained by slicing along the spatial domain, and each slice can be connected to the MLP module marked as frequency and time domains; then the input is transposed, and the spatial domain features are processed by the channel mixing MLP i.e., the spatial domain MLP module. For example, the above Figure 8In the corresponding example, there are K planes, each connected to a corresponding time-domain and frequency-domain multilayer perceptron (MLP) in both the time and frequency domains, and then spatial mixing is performed, as shown below. Figure 12 As shown.

[0141] Step 3: After encoding in the coding layer to obtain the frequency domain output, the inverse Fourier transform is performed to convert it to the time domain, resulting in the time domain output 1.

[0142] Step 4: The denoising layer further denoises the signal in the time domain. Specifically, it can employ techniques such as wavelet transform or AI networks to obtain the time-domain output. Wavelet transform is a mathematical tool for signal analysis and processing. By decomposing the signal into wavelet basis functions of different scales (frequency) and locations (time), it enables localized analysis of the signal in both the time and frequency domains. The denoising layer can be wavelet transform, an AI sub-network, a combination of both, or other time-domain denoising methods. For example, it can first perform wavelet transform to denoise the noise on each antenna, and then weight the signals from multiple antennas before further denoising via an AI sub-network, such as... Figure 13 As shown.

[0143] Step 5: Weight and fuse time-domain output 1 and time-domain output 2 to obtain the final time-domain signal, such as... Figure 10 The weighted output is shown in the figure. The time-domain weighted output can be further Fourier transformed to the frequency domain to obtain the frequency-domain output signal, depending on whether the constellation point symbols in the system are in the frequency domain or the time domain.

[0144] For example, the receiver can use the frequency domain output, transform it to the time domain using IDFT to obtain the time domain output r1, then pass it through a denoising layer to obtain the time domain output r2, and finally the output of the AI ​​network is a weighted sum of the two. The weighting method can be as follows: ,in, .

[0145] In yet another example, such as Figure 14The diagram illustrates another AI network processing procedure. This AI network uses a serial structure. First, the input to the AI ​​network is segmented. Optionally, it's segmented along the spatial domain into time-frequency domain slices. Then, similar to the processing of an MLP mixer, these time-frequency domain slices are mixed, followed by spatial channel mixing. Next, along the frequency domain, the result obtained from the aforementioned spatial channel mixing is segmented into time-space domain slices, which are then mixed again, followed by frequency channel mixing. To further enhance the relationship between domains, the result of the aforementioned frequency channel mixing can also be segmented along the time domain into frequency-spatial domain slices, which are then mixed again, followed by time channel mixing. Finally, after a fully connected layer, the predicted constellation points are output. Here, each slice can be understood as a two-dimensional segmentation, and the aforementioned block segmentation can be understood as a three-dimensional segmentation.

[0146] Furthermore, the loss function of this AI network can be designed as a combination of time-domain and frequency-domain components. That is, the frequency-domain signal is transformed into the time domain after undergoing a discrete inverse Fourier transform, and then connected to the sub-network (such as...). Figure 10 The denoising layer shown in the figure implements a multi-level loss function (such as...). Figure 10 The combination shown (where the frequency domain loss function and the time domain loss function are weighted to obtain a weighted loss function) can reduce noise and interference.

[0147] In the embodiments of this disclosure, by transforming the time-frequency and frequency-domain data, the learning and feature capture capabilities in the time and frequency domains are enhanced, the noise impact in the frequency and time domains is effectively reduced, and the diversity characteristics of the received signal are achieved through a weighted approach.

[0148] Furthermore, the loss function of this AI network can be designed as a loss function based on the Hamming distance between the predicted transmit constellation points and the tag transmit constellation points, or a function related to the Hamming distance. This loss function can be expressed as:

[0149] Where B represents the batch size. Indicates the marker to send constellation points, Indicates the predicted constellation point for transmission. To predict the Hamming distance between the transmitting constellation points and the tag transmitting constellation points, f() represents the operation function for processing the Hamming distance, which can be an exponential function or a linear function, etc. This loss function can more directly reflect the complexity of the constellation points.

[0150] Furthermore, in this embodiment, the LayerNorm operation in the aforementioned AI network can be replaced by a Dynamic Hyperbolic Tangent (DyT) mechanism instead of the static distance-based LayerNorm mechanism. This mechanism eliminates LayerNorm's dependence on fixed mean and variance, extracting local statistical features of the input sequence through a lightweight convolutional network. This lightweight convolutional network captures the distribution characteristics of neighboring tokens (the smallest unit of text processed by AI) using a sliding window approach, generating a dynamic offset Δμ and scaling factor that need to be normalized based on the dimension. The DyT mechanism can better adapt to different data distributions, improve the model's generalization ability, and be more flexible in handling sequences of different lengths. Additionally, it can reduce training instability or increase training speed.

[0151] In this embodiment of the disclosure, the DyT mechanism can employ a learnable dynamic slope parameter to adaptively adjust the curvature threshold of the tanh function tanh(·) during activation, the expression of which is: DyT(X) = γ tanh(αX)+β Here, X represents the input tensor; α represents the learning parameters that can be scaled differently according to the varying input range; γ and β represent the learnable channel vector parameters, the same as those used in all normalization layers, which enable the output to scale back to an arbitrary range. This DyT mechanism can achieve differentiated nonlinear mappings between hierarchical features based on input features. By maintaining the inherent boundary constraints of the activation function while performing local feature sensitivity adjustments, DyT effectively alleviates gradient path redundancy caused by the separation of normalization and activation operations.

[0152] Furthermore, to enhance the adaptability of AI models to different modulation schemes and / or channels, transfer learning can be used to train the AI ​​model for new scenarios. Specifically, a pre-trained model can be used, freezing some or all of the MLP layers used for time-frequency spatial domain processing, and training only the unfrozen MLP layers and output layer for the new scenario. This allows for high-quality model training with a small number of samples. Specific applicable scenarios include, but are not limited to: (1) Transfer learning between different modulation methods can promote knowledge transfer between different digital modulation formats (such as amplitude, phase, frequency modulation, etc.).

[0153] (2) Transfer learning between different channel scenarios can achieve model generalization in different propagation environments; among them, different channel scenarios can be distinguished by parameters such as multipath fading, Doppler shift and noise interference.

[0154] (3) Transfer learning between different modulation schemes and different channel scenarios to be applicable to scenarios where modulation schemes and channel characteristics are switched at the same time.

[0155] Figure 15 This embodiment of the present disclosure provides a model transfer illustration, which gives an example of model transfer of the base model in other propagation environments that are different from its corresponding propagation environment. Through this model transfer, the convergence speed can be significantly accelerated while maintaining detection accuracy.

[0156] Reference Figure 15 The base model obtained after initial training is a model under a 16-quadrature amplitude modulation (16QAM) tapped delay line A (TDL-A) channel. In one example, it can be used to freeze certain layers of the supervised fine-tuning (SFT) mixer (such as...). Figure 15 In the case of the frozen layer shown, only the remaining layers of the SFT mixer and the fully connected layers (as shown) are trained. Figure 15 The unfrozen layers shown in the figure yield the transfer model 1 for the 16QAM Clustered Delay Line A (CDL-A) channel. In another example, all layers of the SFT mixer can be frozen, and only the fully connected layers (such as...) can be trained. Figure 15 The non-frozen layer shown in the figure can be used to obtain the migration model 2 under the Quadrature Phase Shift Keying (QPSK) TDL-A channel.

[0157] Furthermore, for the online AI model detection mechanism, periodically transmitted preset signals (such as codewords known at the transmitting and receiving ends) are used as training samples to fine-tune the online AI model, ensuring the stability of the receiver's operation. Alternatively, when a significant drop in throughput occurs, preset signals can be transmitted as training samples to fine-tune the online AI model.

[0158] Furthermore, this disclosure also provides a performance optimization mechanism that integrates model diagnosis and updating. The implementation process of this mechanism may include, but is not limited to, three stages: a diagnosis stage; a data acquisition stage; and a model tuning stage.

[0159] In some optional implementations of the embodiments of this disclosure, during the diagnostic phase, the system addresses the performance degradation problem of the AI ​​wireless receiver through an adaptive backoff mechanism. The performance detection methods employed may include, but are not limited to: Method 1: When a severe throughput degradation is detected, such as a throughput loss of ≤30%, the system automatically switches to a non-AI-based signal processing algorithm.

[0160] Method 2: It can calculate the rate of change in throughput over a period of time. Cyclic Redundancy Check (CRC) error rate and the rate of change of disturbance These parameters can be obtained from the function. f By combining the parentheses, the function value V can be expressed as: V = ( When V exceeds a certain threshold, a model update can be triggered.

[0161] In this embodiment, a two-level verification mechanism can be adopted for the model adaptation or tuning scheme based on the above-mentioned diagnosis-driven approach: First, the real symbols obtained during the enhanced detection phase can be compared with the AI ​​model prediction results to confirm the model aging effect. Then, after confirming the mismatch between the model and the environment, constrained fine-tuning can be implemented using the newly acquired dataset while keeping the model architecture unchanged. The incremental update strategy adopted in the above data acquisition phase fully utilizes the inherent symbiotic characteristics of the transmitter and receiver in digital communication, that is, the transmitted symbols and received signals form a naturally aligned input-output pair, reducing the need for ideal channel state information.

[0162] In some optional implementations of the embodiments of this disclosure, during the data acquisition phase, a CRC-based online sample collection and generation method can be implemented, such as... Figure 16a As shown, the sample generation process 700 according to embodiments of this disclosure may include, but is not limited to, the following steps: Step 701, send bits; wherein, the bits sent may include original bits and redundant bits, for example, the bits sent... The values ​​are 00110 10111, where 00110 is the original bit and 10111 is the redundant bit.

[0163] Step 702, decode the bits.

[0164] Step 703, Error Detection Stage: When the value C0 corresponding to the CRC status is 1, it indicates a data packet decoding error. At this time, the received signal is recorded. Channel estimation Redundant version of channel coding (Redundancy Version, RV), Demodulation Reference Signal Noise and interference .

[0165] Step 704, retransmission phase, for the first Each retransmission continues to record the received signal during each retransmission. Channel estimation Redundant version of channel coding Demodulation reference signal Noise and interference And the CRC status, until the value corresponding to the CRC status is 0, indicating that the transmitted bits... Correct decoding was achieved. For example, during the first retransmission, the decoded bits were 00010 10101. At this time, the CRC value C1 = 1, indicating that the data packet decoding was still incorrect. The value of this retransmission was recorded. , , , and Then a second retransmission is performed. The decoded bits are 00110 10111. At this time, the value C2 corresponding to the CRC status is 0, indicating that the data packet was decoded correctly. This retransmission is recorded. , , , , and the decoded raw bits .

[0166] Step 705, Sample generation stage, based on the original bits obtained by correct decoding and the corresponding initial pass and repass phases respectively Generate the corresponding encoded recovered bits. , , …, , as the label of the sample, where K is the total number of retransmissions. Then, the received signal recorded in steps 703 and 704 above will be... Channel estimation Demodulation reference signal and noise and interference As input to the sample, the input and output of the sample are matched to generate samples 0, 1, 2, ..., K.

[0167] For example, such as Figure 16b As shown, the original bits obtained after correct decoding 00110 corresponds to the initial transmission, the first retransmission, and the second retransmission, respectively. , and Then based on the original bit and each It can generate the corresponding encoded recovered bits. , , Using these as labels for the samples, the initial transmission, the first retransmission, and the second retransmission are recorded respectively. , and }、{ , and },and{ , and } are used as inputs to the samples, and then the outputs of the samples, namely sample 0, sample 1 and sample 2, are obtained.

[0168] In this embodiment, correctly retransmitted bits can be used as labels to recover previously erroneously transmitted bits, thereby expanding the training dataset.

[0169] In the above embodiments, by learning the spatiotemporal and frequency domain channel characteristics, sample efficiency can be improved. Effective model adjustment can be completed with only a very small amount of training data (e.g., ≤1% of the initial training set). Furthermore, seamless switching between AI-based and non-AI-based modes during runtime can be achieved based on a pre-configured switching protocol. For example, a switching threshold (such as the aforementioned throughput loss threshold) can be pre-configured, and switching occurs when the threshold is reached. Moreover, no dedicated hardware support is required, effectively controlling costs. Meanwhile, the validation of the updated model is completed based on real-time performance monitoring when the enhanced AI receiver is restarted, realizing a closed-loop optimization and adaptation system for dynamic wireless environments.

[0170] In some optional implementations of the embodiments of this disclosure, the priority of the generated samples can be determined based on the number of retransmissions and the bit error rate. Optionally, sample quality indicators... It can be defined as: s= .

[0171] in, K is the total number of retransmissions. Indicates the first Second retransmission Indicates the first Number of error bits during retransmission This represents the total number of bits transmitted in each transmission. This sample quality metric comprehensively considers the total number of retransmissions. The larger the retransmission number *i*, the more complex the channel conditions experienced by the retransmitted data, and the more the current model cannot match the environment. Therefore, such samples need more weight; for example, it can be proportional to the bit error rate. Furthermore, samples with high bit error rates should also be given higher weights, indicating that the current model's predictive ability is insufficient.

[0172] In some optional implementations of the embodiments of this disclosure, during the model optimization stage, optionally, the above-mentioned sample quality indicators can be used. The values ​​are sorted from largest to smallest, prioritizing those based on... The model is updated based on samples with larger values.

[0173] The embodiments of this disclosure provide an AI-based intelligent receiver that addresses the problems of insufficient bit error rate performance and complex signal processing in receivers in wireless networks. It designs an AI-based receiver algorithm that can achieve signal equalization and demodulation of data symbols using only the received signal, pilot signal, and their channel and noise estimations, eliminating the need for channel interpolation, equalization, and demodulation in traditional algorithms. Specifically, this implementation arranges subframe signals into tensor form, utilizing multi-dimensional signal blocks containing information from the time, frequency, and spatial domains. Feature fusion is then performed, and the network implicitly learns the relationships between the received signals, thus obtaining the equalized and demodulated channel. Furthermore, by considering a network mode with dual supervision in the time and frequency domains and designing a multi-level loss function, the information in the time and frequency domains can be enhanced, further improving the neural network's ability to learn the relationships between multiple signal domains. This enhances signal decoding performance and reduces the bit error rate.

[0174] Figure 17 This is a block diagram of a wireless communication device according to an embodiment of the present disclosure.

[0175] Reference Figure 17 The wireless communication device 1300 may include a transceiver 1310 and a processor 1320. Specifically, the processor 1320 may be coupled to the transceiver 1310 and configured to perform the above-mentioned... Figure 4 , Figure 5 or Figure 6 The methods mentioned in the description. For details on the operations involved in the above methods, please refer to [link / reference]. Figure 4 , Figure 5 or Figure 6 The descriptions will not be repeated here. According to embodiments of this disclosure, the wireless communication device 1300 may be a base station-side receiver, a terminal-side receiver, a bypass device-side receiver, or a receiver of a relay node in a communication link, or may be a wireless communication device including the aforementioned receivers.

[0176] This disclosure also provides an electronic device including a processor, and optionally, a transceiver and / or memory coupled to the processor, the processor being configured to perform the steps of the method provided in any optional embodiment of this disclosure.

[0177] Figure 18 The diagram shows a structural schematic of an electronic device to which an embodiment of the present invention applies, such as... Figure 18 As shown, Figure 18The illustrated electronic device 1400 includes a processor 1401 and a memory 1403. The processor 1401 and the memory 1403 are connected, for example, via a bus 1402. Optionally, the electronic device 1400 may further include a transceiver 1404, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 1404 is not limited to one type, and the structure of the electronic device 1400 does not constitute a limitation on the embodiments of this disclosure. Optionally, the electronic device may be a receiver, such as a base station-side receiver or a terminal-side receiver.

[0178] Processor 1401 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It may implement or execute the various exemplary logic blocks, modules, and circuits described in connection with this disclosure. Processor 1401 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0179] Bus 1402 may include a pathway for transmitting information between the aforementioned components. Bus 1402 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 1402 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 18 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0180] The memory 1403 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media, other magnetic storage devices, or any other medium capable of carrying or storing computer programs and capable of being read by a computer, without limitation herein.

[0181] The memory 1403 is used to store computer programs that execute embodiments of the present disclosure, and is controlled by the processor 1401 to execute them. The processor 1401 is used to execute the computer programs stored in the memory 1403 to implement the steps shown in the foregoing method embodiments.

[0182] This disclosure provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can implement the steps and corresponding content of the aforementioned method embodiments.

[0183] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, can implement the steps and corresponding content of the aforementioned method embodiments.

[0184] The terms “first,” “second,” “third,” “fourth,” “1,” “2,” etc. (if present) in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in a sequence other than that shown in the figures or text.

[0185] It should be understood that although arrows indicate various operation steps in the flowcharts of the embodiments of this disclosure, the order in which these steps are implemented is not limited to the order indicated by the arrows. Unless explicitly stated herein, in some implementation scenarios of the embodiments of this disclosure, the implementation steps in each flowchart can be executed in other orders as required. Furthermore, some or all of the steps in each flowchart may include multiple sub-steps or multiple stages based on the actual implementation scenario. Some or all of these sub-steps or stages can be executed at the same time, and each sub-step or stage can also be executed at different times. In scenarios where execution times differ, the execution order of these sub-steps or stages can be flexibly configured as required, and the embodiments of this disclosure do not limit this.

[0186] The above text and accompanying drawings are provided as examples only to help the reader understand this disclosure. They are not intended and should not be construed as limiting the scope of this disclosure in any way. Although certain embodiments and examples have been provided, it will be apparent to those skilled in the art, based on the content disclosed herein, that changes can be made to the illustrated embodiments and examples, and other similar implementations based on the technical concept of this disclosure can be adopted without departing from the scope of this disclosure, and these modifications and modifications are also within the protection scope of the embodiments of this disclosure.

Claims

1. A method performed by a receiver in a wireless communication network, comprising: The first data is obtained by fusing at least one of the acquired pilot signal, channel estimation, and noise estimation with the received signal. The first data is divided into multiple data blocks to be processed in at least one dimension of the time domain, frequency domain, and spatial domain. The plurality of data blocks to be processed are demodulated to obtain the original transmitted signal corresponding to the received signal.

2. The method according to claim 1, wherein, The step of dividing the first data into multiple data blocks to be processed in at least one dimension of the time domain, frequency domain, and spatial domain includes: The first data is augmented to obtain the second data; The second data is processed into the plurality of data blocks to be processed in at least one dimension of the time domain, frequency domain, and spatial domain.

3. The method according to claim 2, wherein, The methods for performing data augmentation on the first data include at least one of the following: Phase rotation; Row positions swapped; Column positions swapped; Mirror flip.

4. The method according to claim 1, further comprising: The information features of at least two dimensions in the time domain, frequency domain and spatial domain of each data block to be processed are fused to obtain multiple data blocks after feature fusion. The step of demodulating the plurality of data blocks to obtain the original transmitted signal corresponding to the received signal includes: The original transmitted signal is obtained by demodulating the multiple data blocks that have undergone feature fusion.

5. The method according to claim 1, wherein, The step of dividing the first data into multiple data blocks to be processed in at least one dimension of the time domain, frequency domain, and spatial domain includes: Based on the information features of the first data in the time domain, frequency domain, or spatial domain, the first data is divided into multiple intermediate data blocks. The information features of at least two dimensions in the time domain, frequency domain, and spatial domain of each intermediate data block are fused to obtain the plurality of data blocks to be processed.

6. The method according to claim 1, further comprising: At least one of the pilot signal, the channel estimation, and the noise estimation, as well as the received signal, are preprocessed.

7. The method according to claim 1, wherein, The demodulation process of the plurality of data blocks to be processed to obtain the original transmitted signal corresponding to the received signal includes: The multiple data blocks to be processed are input into an artificial intelligence (AI) network for demodulation processing, wherein the AI ​​network is used to output the original transmitted signal.

8. The method according to claim 7, wherein, The step of inputting the plurality of data blocks to be processed into an artificial intelligence (AI) network for demodulation processing includes: The multiple data blocks to be processed are processed through the embedding layer and coding layer of the AI ​​network to obtain a frequency domain output signal; Convert the frequency domain output signal into a first time domain output signal; The first time-domain output signal is converted into a second time-domain output signal through the denoising layer of the AI ​​network; The first time-domain output signal and the second time-domain output signal are weighted to obtain the original transmitted signal.

9. The method according to claim 8, wherein, The loss function for training the AI ​​network is obtained by weighting the frequency domain loss function corresponding to the frequency domain output signal and the time domain loss function corresponding to the second time domain output signal.

10. The method according to claim 8, wherein, The process of processing the multiple data blocks to be processed through the embedding layer and coding layer of the AI ​​network to obtain a frequency domain output signal includes: The multiple data blocks to be processed are jointly processed by the embedding layer to obtain multiple data blocks to be encoded. The multiple data blocks to be encoded are encoded by the coding layer to obtain the frequency domain output signal.

11. The method according to claim 10, wherein, The step of encoding the plurality of data blocks to be encoded through the coding layer to obtain the frequency domain output signal includes: The plurality of data blocks to be encoded are encoded by at least one multilayer perceptron in the coding layer corresponding to at least one dimension in the time domain, frequency domain, and spatial domain, to obtain the frequency domain output signal.

12. The method according to claim 1, further comprising: The pilot signal, the channel estimation, and the noise estimation are acquired based on multiple receiving antennas, along with the received signal.

13. A wireless communication device, comprising: A transceiver is configured to send and receive signals; and A processor, coupled to the transceiver, is configured to perform the method as described in any one of claims 1 to 12.

14. A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method of any one of claims 1 to 12.

Images