Signal detection method and apparatus, and storage medium

Abstract

The present application discloses a signal detection method, an apparatus, and a storage medium, in which the signal detection method includes the steps of: obtaining at least one set of historical signal samples, including a historical channel estimate, a historical received signal, and a historical modulated signal; and, when the current channel estimate and the received signal to be measured are obtained, inputting the current channel estimate and the received signal to be measured into a pre-trained preset neural network to perform signal detection and obtain a signal estimate of the received signal to be measured, where the pre-trained preset neural network is obtained by training the preset neural network through at least one set of historical signal samples.

Description

Technical Field

[0001] This application is filed based on a Chinese patent application with application number 202210831551.7 and filing date July 15, 2022, and claims the priority of this Chinese patent application. The entire content of this Chinese patent application is incorporated herein by reference.

[0002] This application relates to the field of communication technologies, and particularly to a signal detection method, a signal detection device, a computer storage medium, and a computer program product.

Background Art

[0003] Multiple-Input Multiple-Output (MIMO) spatial division multiplexing technology and Orthogonal Frequency Division Multiplexing (OFDM) technology are highly competitive technologies in wireless communication systems. The MIMO spatial division multiplexing technology enables a system to easily obtain spatial diversity gain and channel capacity gain when the system bandwidth and transmission bandwidth do not change. The OFDM technology transmits data in parallel using a plurality of orthogonal sub-carriers, thereby greatly reducing the data rate of each path and adding a time guard interval, and thus has strong resistance to multipath interference and frequency selective fading.

[0004] Signal detection is a crucial technology in OFDM and MIMO systems. Related signal detection methods mainly encompass three aspects: linear detection, nonlinear detection, and optimal detection. While all can implement the signal detection process, their detection performance is low and they cannot satisfy the ever-increasing needs for signal detection. For example, optimal detection algorithms, such as maximum likelihood detection algorithms, are complex because they require a global search of all possible transmitted symbol regions for the received signal. Linear detection algorithms include zero-forcing algorithms and least mean squares error detection algorithms, which are not computationally complex but do not offer high detection accuracy. Nonlinear detection algorithms include spherical decoding algorithms and successive interference rejection algorithms, which improve detection accuracy by increasing detection complexity. [Overview of the project] [Problems that the invention aims to solve]

[0005] The embodiments of this application provide a signal detection method, a signal detection device, a computer storage medium, and a computer program product, thereby improving signal detection performance. [Means for solving the problem]

[0006] According to the first aspect, an embodiment of the present application includes the step of obtaining at least one set of historical signal samples, which include a historical channel estimate, a historical received signal, and a historical modulated signal. The process includes the step of, when the current channel estimate and the target received signal are acquired, inputting the current channel estimate and the target received signal into a pre-trained preset neural network to perform signal detection and obtain a signal estimate of the target received signal, The pre-trained preset neural network provides a signal detection method obtained by training the preset neural network through at least one set of historical signal samples.

[0007] According to a second aspect, the embodiment of the present application further provides a signal detection device that includes a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein when the processor executes the computer program, the device realizes the signal detection method described above.

[0008] According to a third aspect, the embodiment of the present application further provides a computer-readable storage medium that stores computer-executable instructions for performing the signal detection method described above.

[0009] According to a fourth aspect, an embodiment of the present application further provides a computer program product wherein the computer program or computer instruction is stored in a computer-readable storage medium, the processor of a computer device reads the computer program or computer instruction from the computer-readable storage medium, and the processor executes the computer program or computer instruction to cause the computer device to execute the signal detection method described above.

[0010] In this embodiment, by training a preset neural network based on historical signal samples including acquired historical channel estimates, historical received signals, and historical modulated signals to obtain a neural network model that satisfies the requirements, signal detection is performed using a pre-trained preset neural network model instead of a detection model in related technologies. By adopting a data model dual-drive method in which channel estimates and received signals are simultaneously input to the preset neural network for training or detection, the preset neural network can learn channel-related features more easily, improving signal detection performance, significantly reducing the amount of information required for training or detection, possessing good environmental adaptability and generalizability, meeting the ever-increasing needs for signal detection, and filling a technological gap in related methods. [Brief explanation of the drawing]

[0011] [Figure 1]This is a flowchart of a signal detection method according to one embodiment of the present invention. [Figure 2] This is a flowchart illustrating a signal detection method according to one embodiment of the present invention, in which at least one set of historical signal samples is acquired. [Figure 3] This is a flowchart showing the signal detection method according to one embodiment of the present invention, before signal detection is performed by inputting the current channel estimate and the received signal to be measured into a pre-trained preset neural network. [Figure 4] This is a flowchart illustrating a signal detection method according to one embodiment of the present invention, in which a pre-trained preset neural network is obtained by training a preset neural network. [Figure 5] This is a flowchart for obtaining a first real number array, which is a training sample, in a signal detection method according to one embodiment of the present invention. [Figure 6] This is a flowchart illustrating how to obtain a second real number array, which is the label of a training sample, in a signal detection method according to one embodiment of the present invention. [Figure 7] This is a flowchart showing the signal detection method according to one embodiment of the present invention, after obtaining the estimated signal value of the historically received signal. [Figure 8] This is a flowchart illustrating how to obtain an estimated signal value of a received signal to be measured in a signal detection method according to one embodiment of the present invention. [Figure 9] This is a flowchart for obtaining a third real number array in a signal detection method according to another embodiment of the present invention. [Figure 10] This is a flowchart showing the results after obtaining the real signal estimation result of the received signal to be measured in a signal detection method according to one embodiment of the present invention. [Figure 11] This is a flowchart illustrating how to obtain the complex signal estimation result of a received signal to be measured in a signal detection method according to one embodiment of the present invention. [Figure 12] This is an execution flowchart of a signal detection method according to one embodiment of the present invention. [Figure 13]It is a schematic diagram of the principle of a MIMO system based on a preset neural network according to an embodiment of the present application. [Figure 14] In the signal detection method according to another embodiment of the present application, it is a flowchart before performing signal detection by inputting a current channel estimation value and a received signal to be measured into a preset neural network trained in advance. [Figure 15] In the signal detection method according to another embodiment of the present application, it is a flowchart for obtaining a signal estimation value of a received signal to be measured. [Figure 16] It is a schematic diagram of the principle of an OFDM system based on a preset neural network according to an embodiment of the present application. [Figure 17] It is a schematic diagram of a signal detection device according to an embodiment of the present application.

Embodiments for Carrying out the Invention

[0012] To make the object, technical solution and advantages of the present application clearer, the present application will be described in more detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described in this specification are only for the purpose of interpreting the present application and not for limiting the present application.

[0013] Although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in an order different from that of the flowchart. Terms such as "first", "second", etc. in the specification, claims, and the above drawings are for distinguishing similar objects and not necessarily for explaining a specific order or sequence.

[0014] In recent years, artificial intelligence (AI) technology, particularly deep learning, has achieved great success in many fields, including computer vision, natural language processing, and speech recognition. Intelligent wireless communication enabled by AI technology is considered one of the mainstream directions for future 6G development. Its fundamental idea is to significantly improve the performance of wireless communication systems through the organic integration of wireless communication technology and AI technology. Physical layer AI design includes two main methods: end-to-end communication link design based on AI technology and communication module algorithm design based on AI technology. Currently, the idea behind communication module algorithm design based on AI technology is mainly data-driven, meaning that one or more communication modules are treated as unknown black boxes, and instead, deep learning networks are used. These deep learning networks rely mainly on training with large amounts of data, resulting in a heavy training burden, requiring significant time and effort, and their application scenarios are relatively single, lacking strong applicability and failing to meet current signal detection needs.

[0015] Based on this, the present application provides a signal detection method, a signal detection device, a computer storage medium, and a computer program product. The signal detection method according to an embodiment includes: obtaining at least one set of historical signal samples including a historical channel estimation value, a historical received signal, and a historical modulation signal; when a current channel estimation value and a received signal to be measured are obtained, inputting the current channel estimation value and the received signal to be measured into a preset neural network trained in advance to perform signal detection and obtain a signal estimation value of the received signal to be measured. The preset neural network trained in advance is obtained by training the preset neural network through at least one set of historical signal samples. In this embodiment, based on the obtained historical signal samples including the historical channel estimation value, the historical received signal, and the historical modulation signal, the preset neural network is trained to obtain a neural network model that meets the requirements. Instead of the detection model in the related art, the signal detection is performed using the preset neural network model trained in advance. By adopting the data model dual-drive method of simultaneously inputting the channel estimation value and the received signal into the preset neural network for training or detection, the preset neural network can more easily learn the channel-related features, not only improving the signal detection performance, but also significantly reducing the amount of information required for training or detection, having good environmental adaptability and generalization ability, being able to meet the increasing signal detection needs day by day, and filling the technical gap in the related method.

[0016] Hereinafter, the embodiments of the present application will be further described with reference to the drawings.

[0017] As shown in FIG. 1, FIG. 1 is a flowchart of a signal detection method according to an embodiment of the present application. The signal detection method includes: a step S110 of obtaining at least one set of historical signal samples including a historical channel estimation value, a historical received signal, and a historical modulation signal; Step S120, in which, when the current channel estimate and the target received signal are acquired, the current channel estimate and the target received signal are input to a pre-trained preset neural network to perform signal detection and obtain a signal estimate of the target received signal, wherein the pre-trained preset neural network is obtained by training the preset neural network through at least one set of historical signal samples, but is not limited to this.

[0018] In this step, by training a preset neural network based on historical signal samples including acquired historical channel estimates, historical received signals, and historical modulated signals to obtain a neural network model that satisfies the requirements, signal detection is performed using a pre-trained preset neural network model instead of a detection model in related technologies. By adopting a data model dual-drive scheme in which channel estimates and received signals are simultaneously input to the preset neural network for training or detection, the preset neural network can learn channel-related features more easily, improving signal detection performance, significantly reducing the amount of information required for training or detection, possessing good environmental adaptability and generalization, meeting the ever-increasing needs for signal detection, and filling a technological gap in related methods.

[0019] As can be seen from this, compared to data-driven deep learning networks of related technologies, the embodiment of the present application is based on data model dual drive, that is, it is based on conventional wireless communication systems and improves model performance by utilizing a deep learning network instead of a related detection module without changing the structure of the wireless communication system. Compared to data-driven deep learning networks which mainly rely on training with large amounts of data, the data model dual drive deep learning network in the embodiment of the present application relies only on the communication model or algorithm model and is based on existing models of the physical layer, thereby significantly reducing the amount of information required for training, upgrading, or detection, and possesses good environmental adaptability and generalization ability, as well as broad development prospects.

[0020] In one embodiment, the embodiment of the present application can be applied to OFDM technology or MIMO technology, but is not limited thereto. For example, it can be applied to an OFDM receiver based on MIMO technology, that is, the preset neural network in the embodiment of the present application can be integrated into the OFDM receiver, and in the OFDM receiver, the preset neural network can be trained, and after receiving relevant information, a signal can be detected based on the pre-trained preset neural network. The basic principle is consistent with the embodiment of the present application, and a detailed explanation is omitted. Those skilled in the art will understand that, depending on a specific application scenario, they may choose a method to which the signal detection method of the embodiment of the present application can be applied, but is not limited thereto. For example, with further development of network systems, it is not excluded that multiple application scenarios applicable to the signal detection method of the embodiment of the present application may be derived.

[0021] In one embodiment, the preset neural network is: Deep neural networks, Convolutional neural networks, Residual convolutional neural network, This includes, but is not limited to, at least one of the following: and residual convolutional neural networks with attention mechanisms.

[0022] The examples of preset neural networks shown above may include those for offline training and online deployment, those for unified offline training followed by online deployment, or those for unified online training and deployment, and are not limited thereto. Those skilled in the art may also consider setting up appropriate preset neural networks in addition to the examples shown above, depending on factors such as the characteristics and needs of a specific application scenario, but a detailed explanation will not be omitted here.

[0023] In one embodiment, the historical channel estimate and historical received signal are distinguished from the current channel estimate and the measured received signal. That is, the historical channel estimate and historical received signal are used to train a preset neural network, while the current channel estimate and the measured received signal are signal parameters detected by the preset neural network. However, as the training process continues, the current channel estimate and the measured received signal may also become new historical channel estimates and historical received signals. In other words, the training of the preset neural network can be dynamic and has a constant transferability over time, allowing more signal samples to be used to train the preset neural network and achieve a better training effect. In subsequent embodiments, the training process of the preset neural network will be described in more detail.

[0024] In one embodiment, training a preset neural network through at least one set of historical signal samples to obtain a pre-trained preset neural network, and then inputting historical channel estimates and historical received signals as a set of detection data into the preset neural network to perform re-detection, and determining the actual training effect of the preset neural network based on the detection results, is advantageous for further determining whether the overall training process of the preset neural network is prone to errors.

[0025] In one embodiment, the historical channel estimate, historical received signal, current channel estimate, and the measured received signal may be obtained by sequentially transmitting a set of signals, or by individually transmitting each corresponding signal, and are not limited thereto.

[0026] In one embodiment, the specific number of historical signal samples can be set according to a particular application scenario; that is, the training samples for the preset neural network can be set according to a particular application scenario, and are not limited thereto.

[0027] As shown in Figure 2, in one embodiment of the present invention, if there are multiple sets of historical signal samples and each wireless fading channel corresponds to one of the multiple sets of historical signal samples, step S110 is performed. Step S111 involves acquiring a historically modulated signal transmitted by the transmitting side, Step S112 involves performing predetermined signal processing on the historical modulation signal in each wireless fading channel to obtain a received pilot signal, a local pilot signal, and a historical received signal at the receiving end. The procedure includes, but is not limited to, step S113, which involves multiplying the received pilot signal by the conjugate of the local pilot signal to obtain a historical channel estimate.

[0028] In this step, there are multiple sets of historical signal samples, and each radio fading channel corresponds to multiple sets of historical signal samples. By acquiring the historical modulated signal transmitted by the transmitter, predetermined signal processing is performed on the historical modulated signal in each radio fading channel, and the receiver obtains a received pilot signal, a local pilot signal, and a historical received signal. Furthermore, the received pilot signal is multiplied by the conjugate of the local pilot signal to obtain an estimated historical channel value. In other words, the amplitude of the received signal changes randomly due to the change in channel, causing signal fading. Therefore, each radio fading channel can acquire at least one set of completely different historical signal samples. In this way, multiple historical signal samples can be generated in total to train a preset neural network, and the training effect can be greatly guaranteed.

[0029] In one embodiment, the transmitter and receiver are relative concepts as a set, namely the transmitter corresponding to the initial source of the signal and the receiver corresponding to the backend result of the signal. This does not limit the specific representation of the transmitter and receiver, but merely serves to clearly explain the operating principle of the embodiment of this application. Specific transmitters and receivers can be set in particular application scenarios and are not limited thereto.

[0030] In one embodiment, there may be multiple methods for obtaining the history modulated signal transmitted by the transmitter, but these are limited here. For example, by obtaining at least one set of transmitted bitstream signals transmitted by the transmitter, it is possible to encode and modulate at least one more set of transmitted bitstream signals to obtain a history modulated signal. In other words, a history modulated signal can be obtained by preprocessing the initial transmitted bitstream signal, or the history modulated signal can be directly obtained from the transmitter by adjusting the signal to be processed into a history modulated signal by the transmitter beforehand.

[0031] In one embodiment, there may be multiple specific methods for obtaining a historical channel estimate by multiplying the received pilot signal by the conjugate of the local pilot signal, but this is not limited to such methods. For example, the historical channel estimate can be obtained by multiplying the received pilot signal by the conjugate of the local pilot signal using the least-squares LS channel estimation algorithm, but this is not limited to this method, or it can be implemented using other conventional communication algorithms common in this field, but this is not limited to this method. The least-squares method is a mathematical model applied to data processing fields such as error estimation, uncertainty, system identification and prediction, and forecasting, and is well known to those skilled in the art, so a detailed explanation is omitted here.

[0032] In one embodiment, there may be multiple methods for performing predetermined signal processing on a hierarchical modulated signal, but these methods are not limited here. For example, a fast Fourier transform can be performed on a hierarchical modulated signal to obtain a first output signal after adding a cyclic prefix, then this first output signal can be passed through different radio fading channels to obtain different second output signals, and the cyclic prefix can be removed and an inverse Fourier transform can be performed on the second output signals to restore the signal, ultimately obtaining a received pilot signal, a local pilot signal, and a hierarchical received signal. Alternatively, for example, a hierarchical modulated signal can be input to a predetermined network model, and demodulation and equalization processing can be performed on the hierarchical modulated signal through the predetermined network model to ultimately obtain a received pilot signal, a local pilot signal, and a hierarchical received signal.

[0033] As shown in Figure 3, in one embodiment of the present invention, if there are multiple sets of historical signal samples, before step S120, The method further includes, but is not limited to, step S130, which involves using the estimated historical channel and the historical received signal as training samples for each set of historical signal samples, using the historical modulated signal as the label for the training sample, and training the preset neural network to obtain a pre-trained preset neural network.

[0034] In this step, using the historical channel estimate and the historical received signal as training samples makes it easier for the preset neural network to learn the characteristics of different radio fading channels when it is trained, thereby achieving a better training effect and realizing a complete training flow without using other training samples. This reduces the parameters and time required for training, and the stability of the preset neural network training can be improved by using the historical modulation signal associated with the historical channel estimate and the historical received signal as labels for the training samples.

[0035] The operating principles of each of the above embodiments will be explained below with specific examples.

[0036] Example 1 First, we need to determine a preset neural network. In a MIMO system with N receiving antennas, assuming the received signal is represented by Y,

number

number

number

number

number

number

number

[0037] Next, LS channel estimates

number

number

number

[0038] As shown in Figure 4, step S130 will be further described in one embodiment of the present application, and step 130 is, Step S131 involves combining the acquired historical channel estimate with the real and imaginary parts of the historical received signal to obtain a first real number array which is a training sample. Step S132 involves combining the real and imaginary parts of the acquired history-modulated signal to obtain a second real number array which is the labels of the training samples, The process includes, but is not limited to, step S133, which involves inputting a first real number array and a second real number array into a preset neural network for training and obtaining a signal estimate of a historically received signal.

[0039] In this step, the historical channel estimate and the real and imaginary parts of the historical received signal are combined to obtain a first real number array, which is the training sample, and the real and imaginary parts of the acquired historical modulated signal are combined to obtain a second real number array, which is the label of the training sample. In other words, the input forms of the historical channel estimate, historical received signal, and historical modulated signal are unified and set to a form that satisfies the input requirements of the preset neural network, thereby making it easier for the preset neural network to be trained on the real number array to obtain the signal estimate of the historical received signal. Furthermore, in certain application scenarios, the corresponding real and imaginary values ​​can represent the values ​​of the antenna port and receiving antenna, respectively, meaning that the form of the real number array can be better adapted to the computation of the neural network.

[0040] In one embodiment, the preset neural network may be of a type other than the real-valued neural network described above, and accordingly, the input to the preset neural network may vary depending on the type of preset neural network. In other words, steps S131 to S133 in the above embodiment do not limit the content input to the preset neural network, and it is also possible to select appropriate training input content according to a specific scenario, but this is not limited here.

[0041] As shown in Figure 5, step S131 will be further described in one embodiment of the present application, and step S131 is, Step S1311 involves obtaining the first real and first imaginary parts of the historical channel estimate and obtaining the second real and second imaginary parts of the historical received signal, The process includes, but is not limited to, step S1312, which involves arranging the first real part, the first imaginary part, the second real part, and the second imaginary part in a predetermined order to obtain a first real number array that serves as a training sample.

[0042] In this step, by obtaining the first real part and the first imaginary part of the historical channel estimate, and the second real part and the second imaginary part of the historical received signal, a set of real numbers corresponding to the historical channel estimate and the historical received signal can be generated. Furthermore, by arranging the corresponding first real part, first imaginary part, second real part, and second imaginary part in each set of real numbers in a predetermined order, a first array of real numbers containing the historical channel estimate and the historical received signal can be obtained.

[0043] In one embodiment, the pre-set arrangement order is not limited and can be arranged based on the real and imaginary parts corresponding to the historical channel estimate and the historical received signal, respectively. This includes, but is not limited to, arranging the real and imaginary parts of the historical channel estimate in adjacent positions, arranging the real and imaginary parts of the historical received signal in adjacent positions, or arranging them sequentially according to a specific application scenario.

[0044] As shown in Figure 6, in one embodiment of the present invention, when the hierarchical modulated signal includes a modulated signal stream that includes a plurality of hierarchical modulated signals, step S132 will be described further, and step S132 is, Step S1321 involves acquiring the real and imaginary parts of each hierarchical modulation signal, The method includes, but is not limited to, step S1322, which arranges the real and imaginary parts of each hierarchical modulation signal in a predetermined order to obtain a second array of real numbers which are the labels of the training samples.

[0045] In this step, if the hierarchical modulated signal includes a modulated signal stream containing multiple hierarchical modulated signals, then the hierarchical modulated signal corresponds to a single signal layer. By obtaining the real and imaginary parts of each hierarchical modulated signal within it, it is equivalent to obtaining the real and imaginary parts of the entire hierarchical modulated signal. Next, the real and imaginary parts of each hierarchical modulated signal are arranged in a predetermined order to obtain a second array of real numbers which are the labels of the training samples.

[0046] In one embodiment, the pre-set arrangement order is not limited and can be arranged based on the real and imaginary parts of each hierarchical modulation signal, that is, the real and imaginary parts of each hierarchical modulation signal can be arranged in adjacent positions, and the real and imaginary parts of adjacent hierarchical modulation signals can be arranged in adjacent positions, or they can be arranged sequentially according to a specific application scenario, but are not limited thereto.

[0047] As shown in Figure 7, in one embodiment of the present invention, after step S133, Step S134: Based on the signal estimate of the historical received signal and the second real number array, obtain the current value of the preset loss function corresponding to the preset neural network. The process further includes, but is not limited to, step S135, which determines a pre-trained preset neural network when the current value of a pre-set loss function matches a pre-set convergence condition, and steps S134 to S135 are used to determine a pre-trained preset neural network.

[0048] In this step, a loss function is calculated from the signal estimate obtained through training and a predetermined second real number array. Based on the calculated value, it is determined whether or not a predetermined convergence condition is met. If it is determined that the calculated value does not meet the predetermined convergence condition, the preset neural network is further trained and optimized to obtain a new signal estimate. By analogy, a new loss function value is obtained by calculating the new signal estimate, and the decision is made again, allowing the preset neural network to be continuously optimized. As a result, the trained signal estimate continuously approaches the transmitted signal and can flexibly support data bit sources of different lengths and different modulation schemes. Therefore, it can be reliably applied to correlation scenarios for calculating the loss function, and ultimately a preset neural network that meets the requirements can be trained.

[0049] In one embodiment, the pre-set loss function is: Mean Squared Error (MSE) loss function, Root Mean Squared Error (RMSE) loss function, Mean Absolute Error (MAE) loss function, Euclidean distance loss function, cosine distance loss function, The function includes at least one of the following linear weight functions: MSE loss function, RMSE loss function, MAE loss function, Euclidean distance loss function, and cosine distance loss function.

[0050] Since the above loss function is well known to those skilled in the art, a detailed explanation thereof is omitted here. Alternatively, those skilled in the art may, but are not limited here, set up corresponding pre-configured loss functions for preset neural networks depending on specific application scenarios.

[0051] The operating principles of each of the above embodiments will be explained below with specific examples.

[0052] Example 2 First, we construct the training input for a preset neural network based on the MIMO system, and since this preset neural network is a real-valued neural network,

number

number

number

number

number

[0053] Next, labels for the training samples of the preset neural network are generated, and the model is trained. Referring to Figure 12, the transmitted bitstream sent by the transmitter is encoded and modulated, and then the transmitted modulated signal, including the multilayer MIMO signal, is generated.

number

number

number

number

number

number

number

[0054] Here,

number

number

number

number

number

[0055] As shown in Figure 8, step S120 will be further described in one embodiment of the present application, and step S120 is, Step S121 involves combining the acquired current channel estimate with the real and imaginary parts of the received signal to be measured to obtain a third real number array. The method includes, but is not limited to, step S122, which involves inputting a third real number array into a pre-trained preset neural network to perform signal detection and obtain a real number signal estimation result of the received signal to be measured.

[0056] In this step, the current channel estimate and the real and imaginary parts of the received signal to be measured are combined to obtain a third real number array as a detection input. This supplies the preset neural network with corresponding input parameters that satisfy the requirements. In other words, the input configuration of the current channel estimate and the received signal to be measured is unified and set to a configuration that satisfies the input requirements of the preset neural network, so that the preset neural network can detect the real number array and obtain a real signal estimation result of the received signal to be measured.

[0057] In one embodiment, the preset neural network may output other types of signals in addition to the real signal estimation results output above, and accordingly, the input to the preset neural network may vary depending on the output type of the preset neural network, but is not limited thereto.

[0058] As shown in Figure 9, step S121 will be further described in one embodiment of the present invention, and step S121 is, Step S1211 involves obtaining the third real and third imaginary parts of the current channel estimate and obtaining the fourth real and fourth imaginary parts of the received signal to be measured. The method includes, but is not limited to, step S1212, which involves arranging the third real part, the third imaginary part, the fourth real part, and the fourth imaginary part in a predetermined order to obtain the third real number array.

[0059] In this step, by obtaining the third real part and the third imaginary part of the current channel estimate, and the fourth real part and the fourth imaginary part of the received signal to be measured, a set of real numbers corresponding to the current channel estimate and the received signal to be measured can be generated. Furthermore, by arranging the corresponding third real part, third imaginary part, fourth real part, and fourth imaginary part in each set of real numbers in a predetermined order, a third real number array containing the current channel estimate and the received signal to be measured can be obtained.

[0060] In one embodiment, the pre-set arrangement order is not limited and can be arranged based on the real and imaginary parts corresponding to the current channel estimate and the received signal to be measured. This includes, but is not limited to, arranging the real and imaginary parts of the current channel estimate in adjacent positions, arranging the real and imaginary parts of the received signal to be measured in adjacent positions, or arranging them sequentially according to a specific application scenario.

[0061] As shown in Figure 10, in one embodiment of the present application, a step embodiment after step S122 will be further described, and after step S122, Step S123 involves performing a complex transformation process on the real signal estimation result of the received signal to be measured to obtain the complex signal estimation result of the received signal to be measured. The process further includes, but is not limited to, step S124, which involves demodulating and decoding the complex signal estimation result to obtain an output signal.

[0062] In this step, by obtaining the complex signal estimation result of the received signal to be measured from the real signal estimation result of the received signal to be measured, demodulation and decoding can be performed on the complex signal estimation result to obtain the final output signal.

[0063] As shown in Figure 11, in one embodiment of the present invention, when the real signal estimation result includes a real signal estimation array that includes the real and imaginary parts of a plurality of sequentially arranged hierarchical signals, step S123 will be described further, and step S123 is, Step S1231 involves complex-transforming the real and imaginary parts of each hierarchical signal in the real-number signal estimation array to obtain complex-transformed data corresponding to each hierarchical signal, The method includes, but is not limited to, step S1232, which involves arranging each complex transformation data in a predetermined order to obtain a complex signal estimation result for the received signal to be measured.

[0064] In this step, the real signal estimation result includes a real signal estimation array containing the real and imaginary parts of multiple hierarchical signals arranged in a predetermined order. Therefore, by complex transforming the real and imaginary parts of each hierarchical signal in the real signal estimation array, complex transformed data corresponding to each hierarchical signal can be obtained. Furthermore, by arranging these complex transformed data, the complex signal estimation result of the signal to be measured can be obtained accurately and reliably.

[0065] The operating principles of each of the above embodiments will be explained below with specific examples.

[0066] Example 3 As shown in Figure 12, online signal detection can be performed using a preset neural network, and after the preset neural network has been trained, it can be placed on the receiving side of the MIMO system to perform online signal detection. When performing online signal detection, the input parameters of the preset neural network can be obtained in the same way as in Example 1 and Example 2, the preset neural network performs online detection, and the output of the preset neural network, i.e., the estimated value of the transmitted signal, can be obtained, and the conversion from a real number array to equalized complex number data can be completed as follows.

number

[0067] Here, complex(·) represents an operation that constructs a single complex-valued data, and the complex-valued data obtained through the transformation is...

number

[0068] The operating principles of each of the above embodiments will be explained below with a complete process example.

[0069] Example 4 As shown in Figure 13, the OFDM system based on MIMO technology is provided with a MIMO signal detector and an LS channel estimation module based on a neural network (i.e., the preset neural network in the embodiment of this application), in which case L = 2 data layers, P = 2 antenna ports, T = 2 physical transmitting antennas, and R = 2 physical receiving antennas. The receiving frequency domain signal Y can be expressed by equation (2).

number

[0070] Here, X is the transmitted signal modulated by these two data layers, with dimension L×1, i.e., 2×1; H is the radio channel matrix between the transmitter and receiver, with dimension T×R, i.e., 2×2; and N is the additive white Gaussian noise, with dimension R×1, i.e., 2×1. Frequency domain received signal

number

number

number

[0071] The neural network-based MIMO signal detection method proposed in this example focuses on how to detect the transmitted signal X from the receiving frequency domain signal Y, and the specific steps are as follows.

[0072] Step 1: Determine the structure of a neural network-based MIMO signal detector. First,

number

number

number

number

number

number

[0073] Step 2: Build the training input for the MIMO signal detector based on the convolutional neural network. Using the same method as in Example 1 above,

number

number

number

number

[0074] Step 3: After generating labels for the MIMO signal detector training samples, perform model training. The transmitter transmits a modulated signal.

number

number

number

number

[0075] Furthermore, the modulated signal

number

number

[0076] Here,

number

number

number

number

number

number

[0077] Here, n is the number of samples in the batch size used for training each time. When training the model, the CNN is trained using the Adam optimizer based on the MSE loss function. Network training is determined to be complete when the loss function has converged to a certain extent, and at this time, the model parameters are saved and used for subsequent online signal detection.

[0078] Step 4: Perform online MIMO signal detection. Once CNN training is complete, it can be placed on the receiving side of the OFDM system to perform online signal detection. When detecting online signals, refer to the processing steps in Step 1 and Step 2, similarly obtain the input parameters for the CNN, perform online detection, and obtain the output of the CNN, i.e., the estimated value of the transmitted signal. This estimated value of the transmitted signal contains a corresponding real number array. Next, the conversion from the real number array to equalized complex number data is completed as follows.

number

[0079] Here, complex(·) represents an operation that constructs a single complex-valued data, and the complex-valued data obtained through the transformation is...

number

[0080] As can be seen from this, the signal detection method of the embodiment of the present invention has the output being data after equalization and layer demapping, the modulated signal on the transmitting side is used as the training label, and the MSE is used as the loss function. This allows the estimated value to continuously approach the transmitted signal, and it can flexibly support data bitstreams of different lengths and different modulation schemes, thus providing excellent versatility.

[0081] As shown in Figure 14, in one embodiment of the present invention, if there are multiple sets of historical signal samples and the preset neural network includes an enhanced channel estimation subnetwork and a channel equalization subnetwork, before step S120, Step S140 involves inputting the historical channel estimates into the enhanced channel estimation subnetwork for each set of historical signal samples to perform channel enhancement estimation training and obtain the first estimation result. The method further includes, but is not limited to, step S150, which involves inputting the first estimation result and the historical received signal into a channel equalization subnetwork to perform channel equalization training and obtain a pre-trained preset neural network.

[0082] In this step, the preset neural network includes two subnetworks: a reinforced channel estimation subnetwork and a channel equalization subnetwork. Each subnetwork has its own distinct capabilities. Therefore, for each set of historical signal samples, the historical channel estimate is input into the reinforced channel estimation subnetwork to perform channel reinforcement estimation training and obtain a first estimation result. Then, the first prediction result and the historical received signal are input into the channel equalization subnetwork to perform channel equalization training and obtain a pre-trained preset neural network. This allows for the achievement of a similarly effective and reliable training effect compared to a preset neural network with integrated functionality.

[0083] In one embodiment, the enhanced channel estimation subnetwork and the channel equalization subnetwork may be trained together, or trained separately, or a person skilled in the art may select a corresponding training method according to a specific application scenario to train the enhanced channel estimation subnetwork and the channel equalization subnetwork, etc., but is not limited thereto.

[0084] As shown in Figure 15, in one embodiment of the present invention, step S120 is Step S125 involves inputting the current channel estimate into a pre-trained enhanced channel estimation subnetwork to perform channel enhancement estimation and obtain a second estimation result. The process includes, but is not limited to, step S126, which involves inputting the second estimation result and the received signal to be measured into a pre-trained channel equalization subnetwork to perform channel equalization and obtain a signal estimate of the received signal to be measured.

[0085] In this step, if signal detection is required, the current channel estimate can be input into a pre-trained enhanced channel estimation subnetwork to perform channel enhancement estimation, thereby directly obtaining a second estimation result. The second estimation result and the received signal to be measured can then be input into a pre-trained channel equalization subnetwork to perform channel equalization, thereby obtaining a signal estimate of the received signal to be measured. Testing and calculating both the channel estimation result and the final signal estimate is advantageous for improving the signal detection accuracy of the preset neural network.

[0086] The operating principles of steps S140-S150 and S125-S126 are the same as those of the related steps in the previously described embodiment. The only difference is that the preset neural network in the previously described embodiment is an integrated structure. In this application, the preset neural network is divided into two subnetworks, and the functions of one preset neural network can be fully realized through the cooperation of the two subnetworks. Since the detailed principles and processes of the preset neural network have been described in detail in the previously described embodiment, the step examples of this embodiment can refer to the detailed principles and processes of the preset neural network in the previously described embodiment. To avoid redundancy, they will not be described in detail here.

[0087] The operating principles of each of the above embodiments will be explained below with reference to other complete process examples.

[0088] Example 5 As shown in Figure 16, the OFDM system based on MIMO technology is provided with a MIMO signal detector and an LS channel estimation module based on a neural network (i.e., a preset neural network in the embodiment of this application), in which case L = 4 data layers, P = 4 antenna ports, T = 4 physical transmitting antennas, and R = 4 physical receiving antennas. The receiving frequency domain signal Y can be expressed by equation (4).

number

[0089] Here, X is the transmitted signal modulated by these four data layers, with dimensions L×1, i.e., 4×1; H is the radio channel matrix between the transmitter and receiver, with dimensions T×R, i.e., 4×4; and N is the additive white Gaussian noise, with dimensions R×1, i.e., 4×1. Frequency domain received signal

number

number

number

[0090] The neural network-based MIMO signal detection method proposed in this example focuses on how to detect the transmitted signal X from the receiving frequency domain signal Y, and the specific steps are as follows.

[0091] Step 1: Determine the structure of a neural network-based MIMO signal detector. First,

number

number

number

number

number

number

[0092] Step 2: Build the training input for the reinforced channel estimation subnetwork, using the same method as in Example 1 above.

number

number

number

number

number

number

[0093] Then, we construct the training input for the channel equalization subnetwork,

number

number

number

[0094] Here,

number

number

[0095] Step 3: After generating labels for the MIMO signal detector training samples, perform model training. The transmitter transmits a modulated signal.

number

number

number

number

[0096] Furthermore, the modulated signal

number

number

[0097] Here,

number

number

number

number

number

number

[0098] Here, n is the number of samples in the batch size used for training each time. When training the model, the residual neural network is trained using an Adam optimizer based on the Euclidean distance loss function. Network training is determined to be complete when the loss function has converged to a certain extent, and at this time, the model parameters are saved and used for subsequent online signal detection.

[0099] Step 4: After performing online MIMO signal detection and completing residual neural network training, the system can be placed on the receiving side of the OFDM system to perform online signal detection. When detecting online signals, refer to the processing steps in Steps 1 and 2, similarly obtain the input parameters of the residual neural network, perform online detection, and obtain the output of the residual neural network, i.e., the estimated value of the transmitted signal. This estimated value of the transmitted signal contains a corresponding real number array, and the conversion from the real number array to equalized complex number data is completed as follows.

number

[0100] Here, complex(·) represents an operation that constructs a single complex-valued data, and the complex-valued data obtained through the transformation is...

number

[0101] As can be seen from this, the signal detection method of the embodiment of the present invention has an output that is equalized and layered demapping data, uses the modulated signal on the transmitting side as the training label, and uses the Euclidean distance as the loss function. As a result, the estimated value continuously approaches the transmitted signal, and it can flexibly support data bitstreams of different lengths and different modulation schemes, thus possessing excellent versatility.

[0102] Furthermore, as shown in Figure 17, one embodiment of the present invention further discloses a signal detection device 100 that implements the signal detection method in any of the above embodiments, comprising at least one processor 110 and at least one memory 120 for storing at least one program, wherein at least one program is executed by the at least one processor 110.

[0103] Furthermore, one embodiment of the present invention further discloses a computer-readable storage medium that stores computer-executable instructions for performing the signal detection method in any of the above embodiments.

[0104] Furthermore, one embodiment of the present application further discloses a computer program product which includes a computer program or computer instruction stored in a computer-readable storage medium, wherein the processor of the computer device reads the computer program or computer instruction from the computer-readable storage medium, and the processor executes the computer program or computer instruction, thereby enabling the computer device to perform the signal detection method of the previous embodiment.

[0105] Those skilled in the art will understand that all or some steps, systems, in the methods disclosed above may be implemented as software, firmware, hardware, or appropriate combinations thereof. Some or all physical components may be implemented as software executed by a processor, e.g., a central processing unit, a digital signal processor, or a microprocessor; or as hardware; or as an integrated circuit, e.g., an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-temporary media) and communication media (or temporary media). As is well known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technique for storing information (e.g., computer-readable instructions, data structures, program modules, or other data). Computer storage media may include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital multifunction disk (DVD) or other optical disk memory, magnetic cartridges, magnetic tapes, magnetic disk memory or other magnetic storage devices, or any other media that can be used to store desired information and are accessible by a computer. Furthermore, it is well known to those skilled in the art that communication media typically include computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carriers or other transmission mechanisms, and may include any information transmission medium.

Claims

1. A signal detection method, The steps include obtaining at least one set of historical signal samples, which include a historical channel estimate, a historical received signal, and a historical modulated signal; The process includes the step of, when the current channel estimate and the target received signal are acquired, inputting the current channel estimate and the target received signal into a pre-trained preset neural network to perform signal detection and obtain a signal estimate of the target received signal, The pre-trained preset neural network is obtained by training the preset neural network through at least one set of historical signal samples. There are multiple sets of the aforementioned historical signal samples, and each wireless fading channel corresponds to one of the multiple sets of the aforementioned historical signal samples. The step of obtaining at least one set of historical signal samples is: The steps include: acquiring the historically modulated signal transmitted by the transmitting side, In each of the aforementioned wireless fading channels, a predetermined signal processing is performed on the historical modulation signal to obtain a received pilot signal, a local pilot signal, and a historical reception signal at the receiving end. The steps include: multiplying the received pilot signal by the conjugate of the local pilot signal to obtain a historical channel estimate; including Signal detection method.

2. A step of obtaining a pre-trained preset neural network by training the preset neural network with at least one set of historical signal samples, The process includes the step of obtaining a pre-trained preset neural network by training the preset neural network with the estimated history channel value and the received history signal as training samples for each set of history signal samples, the history modulated signal as the label for the training sample, and so on. The signal detection method according to claim 1.

3. If the preset neural network includes an enhanced channel estimation subnetwork and a channel equalization subnetwork, the step of obtaining the pre-trained preset neural network is to train the preset neural network with at least one set of historical signal samples, For each set of historical signal samples, the historical channel estimates are input to the enhanced channel estimation subnetwork to perform channel enhancement estimation training and obtain a first estimation result. The process includes inputting the first estimation result and the history received signal into the channel equalization subnetwork to perform channel equalization training and obtain the pre-trained preset neural network. The signal detection method according to claim 1.

4. The step of inputting the current channel estimate and the received signal to be measured into a pre-trained preset neural network to perform signal detection and obtain a signal estimate of the received signal to be measured is as follows: The steps include: inputting the current channel estimate into the pre-trained enhanced channel estimation subnetwork to perform channel enhancement estimation and obtain a second estimation result; The second estimation result and the received signal to be measured are input to the pre-trained channel equalization subnetwork to perform channel equalization and obtain a signal estimate of the received signal to be measured. including The signal detection method according to claim 3.

5. The step of training the preset neural network using the estimated history channel value and the received history signal as training samples, the modulated history signal as the label for the training samples, is as follows: The steps include: obtaining a first real number array which is a training sample by combining the acquired historical channel estimate with the real and imaginary parts of the historical received signal; The steps include: combining the real and imaginary parts of the acquired history modulation signal to obtain a second real number array which is the label of the training sample; The steps include inputting the first real number array and the second real number array into the preset neural network for training to obtain the signal estimate of the history received signal, including The signal detection method according to claim 2.

6. The step of obtaining a first real number array which is a training sample by combining the acquired historical channel estimate with the real and imaginary parts of the historical received signal is as follows: The steps include obtaining the first real part and the first imaginary part of the estimated history channel value, and obtaining the second real part and the second imaginary part of the received history signal, The steps include: arranging the first real part, the first imaginary part, the second real part, and the second imaginary part in a predetermined order to obtain a first real number array which is a training sample; including The signal detection method according to claim 5.

7. The aforementioned hierarchical modulation signal includes a modulation signal layer comprising multiple hierarchical modulation signals, The step of combining the real and imaginary parts of the acquired history-modulated signal to obtain a second real number array which is the label of the training sample is: The steps include obtaining the real and imaginary parts of each of the aforementioned hierarchical modulation signals, The steps include: arranging the real and imaginary parts of each hierarchical modulation signal in a predetermined order to obtain a second array of real numbers which are the labels of the training samples; including The signal detection method according to claim 5.

8. After obtaining the signal estimate of the historical received signal, The steps include obtaining the current value of a preset loss function corresponding to the preset neural network based on the signal estimate of the historical received signal and the second real number array, The steps include determining the pre-trained preset neural network when the current value of the pre-set loss function matches the pre-set convergence condition, Based on this, the pre-trained preset neural network is determined. The signal detection method according to claim 5.

9. The aforementioned preset loss function is Mean Squared Error (MSE) Loss Function Root Mean Square Error RMSE Loss Function Mean absolute error (MAE) loss function, Euclidean distance loss function, cosine distance loss function, and includes at least one of the linear weight functions of the MSE loss function, RMSE loss function, MAE loss function, Euclidean distance loss function, and cosine distance loss function. The signal detection method according to claim 8.

10. The step of inputting the current channel estimate and the received signal to be measured into a pre-trained preset neural network to perform signal detection and obtain a signal estimate of the received signal to be measured is as follows: The steps include: obtaining a third real number array by combining the acquired current channel estimate with the real and imaginary parts of the measured received signal; The third real number array is input to a pre-trained preset neural network to perform signal detection and obtain the real number signal estimation result of the received signal to be measured. including The signal detection method according to claim 1.

11. The step of obtaining a third real number array by combining the acquired current channel estimate with the real and imaginary parts of the measured received signal is as follows: The steps include obtaining the third real part and the third imaginary part of the current channel estimate, and obtaining the fourth real part and the fourth imaginary part of the measured received signal, The steps include arranging the third real part, the third imaginary part, the fourth real part, and the fourth imaginary part in a predetermined order to obtain a third real number array, including The signal detection method according to claim 10.

12. After obtaining the real signal estimation result of the measured received signal, The steps include: performing a complex transformation process on the real signal estimation result of the received signal to be measured to obtain the complex signal estimation result of the received signal to be measured; The steps include demodulating and decoding the complex signal estimation result to obtain an output signal, Includes The signal detection method according to claim 10.

13. The real signal estimation result includes a real signal estimation array that includes the real and imaginary parts of a plurality of hierarchical signals arranged in order. The step of performing a complex transformation process on the real signal estimation result of the received signal to be measured to obtain the complex signal estimation result of the received signal to be measured is: The steps include: complex-transforming the real and imaginary parts of each of the hierarchical signals in the real-number signal estimation array to obtain complex-transformed data corresponding to each of the hierarchical signals; The steps include: arranging each of the aforementioned complex transformation data in a predetermined order to obtain the complex signal estimation result of the received signal to be measured; including The signal detection method according to claim 12.

14. The step of obtaining a historical channel estimate by multiplying the received pilot signal by the conjugate of the local pilot signal is, The process includes the step of obtaining a historical channel estimate by multiplying the received pilot signal by the conjugate of the local pilot signal using a least-squares LS channel estimation algorithm. The signal detection method according to claim 1.

15. The step of acquiring the historically modulated signal transmitted by the transmitting side is: The steps include obtaining at least one set of transmitted bitstream signals sent by the transmitting side, The steps include encoding and modulating at least one set of the transmitted bitstream signals to obtain a history-modulated signal, including The signal detection method according to claim 1.

16. The aforementioned preset neural network is Deep neural networks, Convolutional neural networks, Residual convolutional neural network, and includes at least one of the following: residual convolutional neural networks with attention mechanisms. The signal detection method according to claim 1.

17. At least one processor, At least one memory for storing at least one program, Includes, When at least one of the programs is executed by at least one of the processors, the signal detection method according to any one of claims 1 to 16 is realized. Signal detection device.

18. The processor stores executable programs, The program executable by the processor, when executed by the processor, implements the signal detection method described in any one of claims 1 to 16. Computer-readable storage medium.

Images