Frequency response compensation method and device for ultra-wideband pulse signal, equipment and medium

Abstract

The application provides a frequency response compensation method, device and equipment of an ultra-wideband pulse signal and a medium, and relates to the technical field of signal processing. The method comprises the following steps: performing fast Fourier transform on the ultra-wideband pulse signal to obtain a complex spectrum corresponding to the ultra-wideband pulse signal; inputting the real part and the imaginary part of the complex spectrum into a physically constrained neural network to obtain a target compensation filter output by the physically constrained neural network; performing complex multiplication in the frequency domain on the complex spectrum and the target compensation filter to obtain a compensated spectrum; and performing inverse fast Fourier transform on the compensated spectrum to obtain a recovered pulse signal corresponding to the ultra-wideband pulse signal. The technical scheme of the application can realize high-precision compensation under the condition of limited training samples, so that the target compensation filter approximates to an ideal filter, and can be physically explained, thereby improving the robustness to noise, nonlinearity and time-varying channels and accelerating the convergence speed of the physically constrained neural network in the training stage.

Description

Technical Field

[0001] This invention relates to the field of signal processing technology, and in particular to a method, apparatus, device, and medium for frequency response compensation of ultra-wideband pulse signals. Background Technology

[0002] In ultra-wideband pulse communication systems, the signal frequency range is as wide as tens of megahertz (GHz). After passing through antennas, amplifiers, transmission lines and environmental channels, severe amplitude unevenness, group delay distortion and nonlinear distortion will occur. These distortions are often negligible in narrowband systems, but become extremely serious in ultra-wideband systems.

[0003] Traditional frequency domain compensation methods are usually based on linear channel models, which make it difficult to accurately recover broadband signals. At the same time, traditional methods rely on high-precision channel estimation, which is extremely sensitive to noise, nonlinearity and time-varying characteristics.

[0004] In recent years, neural network methods have been used to model complex channel characteristics, but purely data-driven networks have two problems: (1) they require a large amount of training data; and (2) they lack physical interpretability and stability, and are prone to overfitting. Therefore, how to achieve stable and interpretable frequency domain compensation while maintaining the adaptive capability of data-driven networks is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] This invention provides a method, apparatus, device, and medium for frequency response compensation of ultra-wideband pulse signals. It addresses the shortcomings of existing pulse signal compensation networks, which require large amounts of training data, lack physical interpretability and stability, are prone to overfitting, and cannot perform stable and interpretable frequency domain compensation. The invention trains the network using a physically constrained loss function determined by the channel frequency response, achieving high-precision compensation under limited training sample conditions. It employs a physically constrained neural network to determine the target compensation filter, making the target compensation filter approximate an ideal filter and enabling physical interpretation. This accurately fits the system response, improving robustness to noise, nonlinearity, and time-varying channels. Furthermore, by decomposing the complex spectrum into real and imaginary parts and inputting them into the physically constrained neural network, it accelerates the convergence speed of the physically constrained neural network during the training phase.

[0006] This invention provides a frequency response compensation method for ultra-wideband pulse signals, comprising the following steps.

[0007] Acquire an ultra-wideband pulse signal and perform a fast Fourier transform on the ultra-wideband pulse signal to obtain the complex spectrum corresponding to the ultra-wideband pulse signal; The complex spectrum is decomposed into real and imaginary parts and input into a physical constraint neural network to obtain the target compensation filter output by the physical constraint neural network; the physical constraint neural network is based on sample ultra-wideband pulse signals and trained by a physical constraint loss function; the physical constraint loss function is determined based on the channel frequency response corresponding to the sample ultra-wideband pulse signals. The complex spectrum and the target compensation filter are multiplied by a complex number in the frequency domain to obtain the compensated spectrum; Perform an inverse fast Fourier transform on the compensated spectrum to obtain the recovered pulse signal corresponding to the ultra-wideband pulse signal.

[0008] According to the frequency response compensation method for ultra-wideband pulse signals provided by the present invention, the step of decomposing the complex spectrum into real and imaginary parts and inputting them into a physical constraint neural network to obtain a target compensation filter output by the physical constraint neural network includes: The real part and the imaginary part are input into the input layer of the physical constraint neural network to obtain the feature map output by the input layer; The feature map is input into the hidden layer of the physical constraint neural network to obtain the real part and imaginary part features output by the hidden layer; The real and imaginary features are input into the output layer of the physical constraint neural network to obtain the target compensation filter output by the output layer.

[0009] According to the frequency response compensation method for ultra-wideband pulse signals provided by the present invention, the step of inputting the feature map into the hidden layer of the physically constrained neural network to obtain the real part features and imaginary part features output by the hidden layer includes: The feature map is input into the complex feature extraction layer in the hidden layer to obtain a sequence of multiple frequency points output by the complex feature extraction layer; Input all the frequency point sequences into the frequency domain global correlation layer in the hidden layer to obtain the real part features and the imaginary part features output by the frequency domain global correlation layer.

[0010] According to the frequency response compensation method for ultra-wideband pulse signals provided by the present invention, the step of inputting all the frequency point sequences into the frequency domain global correlation layer in the hidden layer to obtain the real part features and the imaginary part features output by the frequency domain global correlation layer includes: Input all the frequency point sequences into the transformer coding layer in the frequency domain global correlation layer to obtain multiple word sequences output by the transformer coding layer; Input all the word sequences into the position coding layer in the frequency domain global association layer to obtain the frequency order relationship output by the position coding layer; The frequency order relationship and all the word sequences are input into the multi-head self-attention mechanism layer in the frequency domain global association layer to obtain the real part features and the imaginary part features output by the multi-head self-attention mechanism layer.

[0011] According to the frequency response compensation method for ultra-wideband pulse signals provided by the present invention, the physical constraint loss function is determined based on the following method: The physical inverse transform constraints are determined based on the channel frequency response and the compensation filter during the training phase. The energy regularization term and frequency smoothness constraint are determined based on the compensation filter in the training phase. The physical constraint loss function is determined based on the physical inverse transformation constraint, the energy regularization term, and the frequency smoothness constraint.

[0012] According to the frequency response compensation method for ultra-wideband pulse signals provided by the present invention, the step of determining the physical inverse transform constraint based on the channel frequency response and the compensation filter during the training phase includes: ; in, This represents the physical inverse transformation constraint. This represents the total number of frequency points. Indicates the first The compensation filter for the training phase corresponding to each frequency point. Indicates the first Channel frequency response corresponding to each frequency point; The determination of the energy regularization term and frequency smoothness constraint based on the compensation filter during the training phase includes: ; ; in, This represents the energy regularization term. This indicates the frequency smoothness constraint. Indicates the first The compensation filter for the training phase corresponding to each frequency point.

[0013] According to the present invention, a frequency response compensation method for an ultra-wideband pulse signal is provided, wherein the frequency range of the ultra-wideband pulse signal is 10 kHz to 18 MHz.

[0014] The present invention also provides a frequency response compensation device for ultra-wideband pulse signals, comprising the following modules: An acquisition module is used to acquire an ultra-wideband pulse signal and perform a fast Fourier transform on the ultra-wideband pulse signal to obtain the complex spectrum corresponding to the ultra-wideband pulse signal. A network module is used to decompose the complex spectrum into real and imaginary parts and input them into a physical constraint neural network to obtain a target compensation filter output by the physical constraint neural network; the physical constraint neural network is based on sample ultra-wideband pulse signals and trained using a physical constraint loss function; the physical constraint loss function is determined based on the channel frequency response corresponding to the sample ultra-wideband pulse signals. The compensation module is used to perform a frequency domain complex multiplication of the complex spectrum and the target compensation filter to obtain the compensated spectrum; The recovery module is used to perform a fast inverse Fourier transform on the compensated spectrum to obtain the recovered pulse signal corresponding to the ultra-wideband pulse signal.

[0015] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the frequency response compensation method for ultra-wideband pulse signals as described above.

[0016] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the frequency response compensation method for ultra-wideband pulse signals as described above.

[0017] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the frequency response compensation method for ultra-wideband pulse signals as described above.

[0018] This invention provides a method, apparatus, device, and medium for frequency response compensation of ultra-wideband pulse signals. The method involves acquiring an ultra-wideband pulse signal and performing a Fast Fourier Transform (FFT) on it to obtain a complex spectrum. The complex spectrum is then decomposed into real and imaginary parts and input into a physically constrained neural network (PCN) to obtain a target compensation filter output by the PCN. The PCN is trained based on sample ultra-wideband pulse signals using a physical constraint loss function. This physical constraint loss function is determined based on the channel frequency response of the sample ultra-wideband pulse signal. The complex spectrum and the target compensation filter are multiplied in the frequency domain to obtain the compensated spectrum. Finally, an inverse Fast Fourier Transform (IFFT) is performed on the compensated spectrum to obtain the recovered pulse signal corresponding to the ultra-wideband pulse signal. This invention's technical solution trains the network using a physical constraint loss function determined by the channel frequency response, enabling high-precision compensation under limited training sample conditions. A physical constraint neural network with introduced physical constraints is used to determine the target compensation filter, making the target compensation filter approximate the ideal filter, thereby enabling physical interpretation, preventing overfitting, and improving robustness to noise, nonlinearity, and time-varying channels. Furthermore, the complex spectrum is decomposed into real and imaginary parts and input into the physical constraint neural network, which accelerates the convergence speed of the physical constraint neural network during the training phase. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating the frequency response compensation method for ultra-wideband pulse signals provided by the present invention.

[0021] Figure 2 This is a schematic diagram of the training phase of the physical constraint neural network provided by the present invention.

[0022] Figure 3 This is a schematic diagram of the output target compensation filter of the physical constraint neural network provided by the present invention.

[0023] Figure 4 This is a schematic diagram of the frequency response compensation device for ultra-wideband pulse signals provided by the present invention.

[0024] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0026] To address the aforementioned problems in the prior art, this invention provides a frequency response compensation method for ultra-wideband pulse signals. Figure 1 This is a flowchart illustrating the frequency response compensation method for ultra-wideband pulse signals provided by the present invention, as shown below. Figure 1 As shown, the method includes the following steps 110 to 140.

[0027] Step 110: Obtain the ultra-wideband pulse signal and perform a fast Fourier transform on the ultra-wideband pulse signal to obtain the complex spectrum corresponding to the ultra-wideband pulse signal.

[0028] Specifically, the ultra-wideband pulse signal transmitted through the channel can be acquired by a signal receiver. This ultra-wideband pulse signal is a time-domain pulse signal, and the signal receiver can be, for example, a high-performance oscilloscope or an analog-to-digital converter (ADC). The acquired ultra-wideband pulse signal can then be processed... Point-based Fast Fourier Transform (FFT) is used to obtain the complex spectrum corresponding to the ultra-wideband pulse signal. .in, The total number of frequency points, The value can be 2048 or 4096 to ensure sufficient frequency resolution.

[0029] In one embodiment, the frequency range of the ultra-wideband pulse signal is 10 kHz to 18 MHz.

[0030] Specifically, the frequency range of the ultra-wideband pulse signal is from 10 kHz to 18 GHz.

[0031] In the above embodiments, the frequency range of the ultra-wideband pulse signal can be set to cover the entire frequency band from 10kHz to 18GHz, adapting to frequency dependence and nonlinear distortion.

[0032] Step 120: Decompose the complex spectrum into real and imaginary parts and input them into a physical constraint neural network to obtain the target compensation filter output by the physical constraint neural network; the physical constraint neural network is based on the sample ultra-wideband pulse signal and trained by a physical constraint loss function; the physical constraint loss function is determined based on the channel frequency response corresponding to the sample ultra-wideband pulse signal.

[0033] Specifically, since neural networks typically process real-valued data, this invention designs a dual-channel input mechanism. This allows for the processing of complex-valued spectra. Decompose into real part and the virtual part Therefore, it can be based on the real part and the virtual part Construct shape as The input tensor, This represents the batch size, and 2 indicates that the number of channels is 2. This input tensor can be further fed into a Physics-Informed Neural Equalizer (PINE) to obtain the target compensation filter output by the Physics-Informed Neural Equalizer.

[0034] Optionally, to accelerate network convergence, the amplitude of the complex spectrum can be normalized before decomposing it into real and imaginary parts and inputting it into the physically constrained neural network, mapping the amplitude to... The phase of the complex spectrum remains unchanged or is normalized to the interval. .

[0035] The physical constraint neural network is trained based on sample ultra-wideband pulse signals and trained using a physical constraint loss function; the physical constraint loss function is determined based on the channel frequency response corresponding to the sample ultra-wideband pulse signals.

[0036] For example, Figure 2 This is a flowchart illustrating the training phase of the physically constrained neural network provided by this invention, as shown below. Figure 2As shown, during the training phase, sample ultra-wideband pulse signals can be generated using a combination of a pulse signal source, a power amplifier, and a photodetector. The parameters of these sample ultra-wideband pulse signals can be pre-set, including, for example, carrier frequency, pulse width, and period. Furthermore, a vector network analyzer or known pilot signals can be used to collect multiple sets of channel frequency responses under different environments. The sample data is then divided into training and validation sets. An Adaptive Moment Estimation (Adam) optimizer is used to iteratively train the PINE network. The training termination condition can be that the physical constraint loss function no longer decreases or the preset number of iterations (e.g., 500) is reached. This training method can employ offline training and online inference, and can be deployed on FPGA or GPU hardware for acceleration.

[0037] In one embodiment, the physical constraint loss function is determined based on the following: The physical inverse transform constraints are determined based on the channel frequency response and the compensation filter during the training phase. The energy regularization term and frequency smoothness constraint are determined based on the compensation filter in the training phase. The physical constraint loss function is determined based on the physical inverse transformation constraint, the energy regularization term, and the frequency smoothness constraint.

[0038] Specifically, the physical inverse transform constraints can be determined based on the channel frequency response and the compensation filter used during the training phase. It can also be based on the compensation filter during the training phase. Determine the energy regularization term and frequency smoothness constraints .

[0039] Furthermore, the physical constraint loss function can be determined based on the physical inverse transformation constraint, the energy regularization term, and the frequency smoothness constraint. This physical constraint loss function... It can be expressed by the following formula: in, This represents the energy weight coefficient corresponding to the energy regularization term. This represents the frequency smoothing line weight coefficient corresponding to the frequency smoothing constraint.

[0040] In the above embodiments, unlike the traditional mean squared error (MSE) loss that only relies on "input-output" matching, optimizing the physical constraint neural network based on this physical constraint loss function can make the network output controlled by physical constraints and will not be affected by non-causal or divergent filtering responses.

[0041] In one embodiment, determining the physical inverse transform constraints based on the channel frequency response and the compensation filter during the training phase includes: ; in, This represents the physical inverse transformation constraint. This represents the total number of frequency points. Indicates the first The compensation filter for the training phase corresponding to each frequency point. Indicates the first Channel frequency response corresponding to each frequency point; The determination of the energy regularization term and frequency smoothness constraint based on the compensation filter during the training phase includes: ; ; in, This represents the energy regularization term. This indicates the frequency smoothness constraint. Indicates the first The compensation filter for the training phase corresponding to each frequency point.

[0042] In the above embodiments, the effect of the energy regularization term is that if there are deep fading frequencies in the channel, direct inversion would lead to infinite gain, thus amplifying noise. The frequency smoothness constraint limits the total energy of the compensation filter, preventing unphysically extreme high gain and protecting the transmit / receive amplifiers from entering the saturation region. The effect of the frequency smoothness constraint is that the frequency response of real physical devices (such as filters and amplifiers) is continuous and smooth, while noise is abrupt. By minimizing the first-order difference between adjacent frequency points, the frequency smoothness constraint forces the network to ignore random noise disturbances and only learn the smooth physical response of the system.

[0043] In one embodiment, the step of decomposing the complex spectrum into real and imaginary parts and inputting them into a physical constraint neural network to obtain the target compensation filter output by the physical constraint neural network includes: The real part and the imaginary part are input into the input layer of the physical constraint neural network to obtain the feature map output by the input layer; The feature map is input into the hidden layer of the physical constraint neural network to obtain the real part and imaginary part features output by the hidden layer; The real and imaginary features are input into the output layer of the physical constraint neural network to obtain the target compensation filter output by the output layer.

[0044] Specifically, Figure 3This is a flowchart illustrating the physical constraint neural network output target compensation filter provided by the present invention, as shown below. Figure 3 As shown, a physically constrained neural network includes an input layer, a hidden layer, and an output layer. The real and imaginary parts of a complex spectrum can be input into the input layer of the physically constrained neural network to obtain a feature map output by the input layer. The input layer includes a real part channel and an imaginary part channel; the real part of the complex spectrum is input to the real part channel, and the imaginary part of the complex spectrum is input to the imaginary part channel.

[0045] Furthermore, the feature map can be input into the hidden layer of a physically constrained neural network to obtain the real part features output by the hidden layer. and the characteristics of the imaginary part .

[0046] Furthermore, the real and imaginary features are input into the output layer of the physical constraint neural network. The output layer synthesizes the real and imaginary features and outputs a target compensation filter. .

[0047] In the above embodiments, the input layer is set to dual channels (real and imaginary parts), which accelerates the convergence speed of the physical constraint neural network during the training phase, and the real and imaginary features output by the hidden layer are synthesized into a target compensation filter, thereby improving the compensation accuracy.

[0048] In one embodiment, the step of inputting the feature map into the hidden layer of the physically constrained neural network to obtain the real and imaginary features output by the hidden layer includes: The feature map is input into the complex feature extraction layer in the hidden layer to obtain a sequence of multiple frequency points output by the complex feature extraction layer; Input all the frequency point sequences into the frequency domain global correlation layer in the hidden layer to obtain the real part features and the imaginary part features output by the frequency domain global correlation layer.

[0049] Specifically, such as Figure 3 As shown, the hidden layer includes a complex feature extraction layer (Complex Convolution) and a frequency-domain global correlation layer (Frequency-domain Transformer). The feature map can be input into the complex feature extraction layer in the hidden layer; assuming the feature map is... ,in, Represents the real part of the feature map. The complex convolution kernel of the complex feature extraction layer represents the imaginary part of the feature map. The operation performed by the complex feature extraction layer is: The complex feature extraction layer can specifically include three stacked one-dimensional complex convolution layers. This complex feature extraction layer can extract small distortion features within a local frequency band (such as periodic ripple caused by impedance mismatch). Finally, the complex feature extraction layer can output multiple frequency point sequences. It is easy to understand that each frequency point sequence still includes real and imaginary parts.

[0050] Furthermore, all frequency point sequences By inputting the frequency domain global correlation layer into the hidden layer, we can obtain the real and imaginary features output by the frequency domain global correlation layer.

[0051] Traditional real-valued convolutions are difficult to effectively process the phase information of radio frequency signals. In the above embodiments, complex convolutions are performed through complex feature extraction layers, which can effectively and reliably process the phase information of signals.

[0052] In one embodiment, the step of inputting all the frequency point sequences into the frequency domain global correlation layer in the hidden layer to obtain the real part features and the imaginary part features output by the frequency domain global correlation layer includes: Input all the frequency point sequences into the transformer coding layer in the frequency domain global correlation layer to obtain multiple word sequences output by the transformer coding layer; Input all the word sequences into the position coding layer in the frequency domain global association layer to obtain the frequency order relationship output by the position coding layer; The frequency order relationship and all the word sequences are input into the multi-head self-attention mechanism layer in the frequency domain global association layer to obtain the real part features and the imaginary part features output by the multi-head self-attention mechanism layer.

[0053] Specifically, the frequency domain global correlation layer can also include a transformer coding layer, a positional encoding layer, and a multi-head self-attention layer. After inputting all frequency point sequences into the transformer coding layer in the frequency domain global correlation layer, the transformer coding layer processes all frequency point sequences... Treating it as a "single sequence" in natural language processing, we can then obtain multiple word sequences output by the transformer coding layer.

[0054] Furthermore, after inputting all word sequences into the position coding layer in the frequency domain global association layer, the position coding layer can clearly define the order relationship from low frequency to high frequency in a single sequence, and thus obtain the frequency order relationship output by the position coding layer.

[0055] Furthermore, after inputting the frequency order relationship and all word sequences into the multi-head self-attention mechanism layer in the frequency domain global association layer, the multi-head self-attention mechanism layer can calculate the correlation weight matrix corresponding to different frequency point sequences to achieve feature fusion across the entire frequency band, thereby obtaining the real and imaginary features output by the multi-head self-attention mechanism layer.

[0056] Because ultra-wideband signals span an extremely wide range (from kHz to GHz), fundamental distortion in the low-frequency band often exhibits nonlinear coupling with harmonic distortion in the high-frequency band. Traditional convolutional layers have limited receptive fields, making it difficult to capture this long-distance dependence. In the above embodiment, the combination of Transformer, positional coding, and multi-head self-attention mechanism improves the receptive field and achieves frequency point fusion across the entire frequency band, laying the foundation for subsequent signal recovery.

[0057] Step 130: Multiply the complex spectrum and the target compensation filter by a complex number in the frequency domain to obtain the compensated spectrum.

[0058] Specifically, the complex spectrum can be... and target compensation filter Performing complex multiplication in the frequency domain yields the compensated spectrum. .

[0059] Step 140: Perform a fast inverse Fourier transform on the compensated spectrum to obtain the recovered pulse signal corresponding to the ultra-wideband pulse signal.

[0060] Specifically, the compensated spectrum can be... An inverse fast fourier transform (IFFT) is performed to obtain the recovered pulse signal corresponding to the ultra-wideband pulse signal. It should be noted that the technical solution of this invention can be applied simultaneously to ultra-wideband communication systems, radar waveform shaping, millimeter-wave front-end linearization, optoelectronic link compensation, and other scenarios.

[0061] This invention provides a frequency response compensation method for ultra-wideband pulse signals. The method involves acquiring an ultra-wideband pulse signal and performing a Fast Fourier Transform (FFT) on it to obtain a complex spectrum. The complex spectrum is then decomposed into real and imaginary parts and input into a physically constrained neural network (PCN) to obtain a target compensation filter output by the PCN. The PCN is trained based on sample ultra-wideband pulse signals using a physical constraint loss function determined by the channel frequency response of the sample ultra-wideband pulse signal. The complex spectrum and the target compensation filter are multiplied in the frequency domain to obtain the compensated spectrum. Finally, an inverse Fast Fourier Transform (IFFT) is performed on the compensated spectrum to obtain the recovered pulse signal corresponding to the ultra-wideband pulse signal. This invention's technical solution trains the network using a physical constraint loss function determined by the channel frequency response, enabling high-precision compensation under limited training sample conditions. A physical constraint neural network with introduced physical constraints is used to determine the target compensation filter, making the target compensation filter approximate the ideal filter, thereby enabling physical interpretation, preventing overfitting, and improving robustness to noise, nonlinearity, and time-varying channels. Furthermore, the complex spectrum is decomposed into real and imaginary parts and input into the physical constraint neural network, which accelerates the convergence speed of the physical constraint neural network during the training phase.

[0062] The frequency response compensation device for ultra-wideband pulse signals provided by the present invention is described below. The frequency response compensation device for ultra-wideband pulse signals described below can be referred to in correspondence with the frequency response compensation method for ultra-wideband pulse signals described above.

[0063] Figure 4 This is a schematic diagram of the structure of the frequency response compensation device for ultra-wideband pulse signals provided by the present invention, as shown below. Figure 4 As shown, the frequency response compensation device 400 for ultra-wideband pulse signals includes the following modules: The acquisition module 410 is used to acquire an ultra-wideband pulse signal and perform a fast Fourier transform on the ultra-wideband pulse signal to obtain the complex spectrum corresponding to the ultra-wideband pulse signal. Network module 420 is used to decompose the complex spectrum into real and imaginary parts and input them into a physical constraint neural network to obtain the target compensation filter output by the physical constraint neural network; the physical constraint neural network is based on sample ultra-wideband pulse signals and trained by a physical constraint loss function; the physical constraint loss function is determined based on the channel frequency response corresponding to the sample ultra-wideband pulse signals. The compensation module 430 is used to perform a frequency domain complex multiplication of the complex spectrum and the target compensation filter to obtain the compensated spectrum; The recovery module 440 is used to perform a fast inverse Fourier transform on the compensated spectrum to obtain the recovered pulse signal corresponding to the ultra-wideband pulse signal.

[0064] In one embodiment, network module 420 is specifically used for: The real part and the imaginary part are input into the input layer of the physical constraint neural network to obtain the feature map output by the input layer; The feature map is input into the hidden layer of the physical constraint neural network to obtain the real part and imaginary part features output by the hidden layer; The real and imaginary features are input into the output layer of the physical constraint neural network to obtain the target compensation filter output by the output layer.

[0065] In one embodiment, the network module 420 is further configured to: The feature map is input into the complex feature extraction layer in the hidden layer to obtain a sequence of multiple frequency points output by the complex feature extraction layer; Input all the frequency point sequences into the frequency domain global correlation layer in the hidden layer to obtain the real part features and the imaginary part features output by the frequency domain global correlation layer.

[0066] In one embodiment, the network module 420 is further configured to: Input all the frequency point sequences into the transformer coding layer in the frequency domain global correlation layer to obtain multiple word sequences output by the transformer coding layer; Input all the word sequences into the position coding layer in the frequency domain global association layer to obtain the frequency order relationship output by the position coding layer; The frequency order relationship and all the word sequences are input into the multi-head self-attention mechanism layer in the frequency domain global association layer to obtain the real part features and the imaginary part features output by the multi-head self-attention mechanism layer.

[0067] In one embodiment, the frequency response compensation device 400 for the ultra-wideband pulse signal further includes a loss function determination module, which determines the physical constraint loss function by means of: The physical inverse transform constraints are determined based on the channel frequency response and the compensation filter during the training phase. The energy regularization term and frequency smoothness constraint are determined based on the compensation filter in the training phase. The physical constraint loss function is determined based on the physical inverse transformation constraint, the energy regularization term, and the frequency smoothness constraint.

[0068] In one embodiment, the loss function determination module is specifically used for: ; in, This represents the physical inverse transformation constraint. This represents the total number of frequency points. Indicates the first The compensation filter for the training phase corresponding to each frequency point. Indicates the first Channel frequency response corresponding to each frequency point; The determination of the energy regularization term and frequency smoothness constraint based on the compensation filter during the training phase includes: ; ; in, This represents the energy regularization term. This indicates the frequency smoothness constraint. Indicates the first The compensation filter for the training phase corresponding to each frequency point.

[0069] In one embodiment, the frequency range of the ultra-wideband pulse signal is 10 kHz to 18 MHz.

[0070] The frequency response compensation device for ultra-wideband pulse signals provided by this invention acquires an ultra-wideband pulse signal and performs a Fast Fourier Transform (FFT) on it to obtain a complex spectrum corresponding to the ultra-wideband pulse signal. The complex spectrum is decomposed into real and imaginary parts and input into a physically constrained neural network (FSN) to obtain a target compensation filter output by the FSN. The FSN is trained based on sample ultra-wideband pulse signals using a physically constrained loss function. The physically constrained loss function is determined based on the channel frequency response corresponding to the sample ultra-wideband pulse signal. The complex spectrum and the target compensation filter are multiplied in the frequency domain to obtain the compensated spectrum. An inverse Fast Fourier Transform (FFT) is then performed on the compensated spectrum to obtain the recovered pulse signal corresponding to the ultra-wideband pulse signal. This invention's technical solution trains the network using a physically constrained loss function determined by the channel frequency response, enabling high-precision compensation under limited training sample conditions. A physical constraint neural network with introduced physical constraints is used to determine the target compensation filter, making the target compensation filter approximate the ideal filter, thereby enabling physical interpretation, preventing overfitting, and improving robustness to noise, nonlinearity, and time-varying channels. Furthermore, the complex spectrum is decomposed into real and imaginary parts and input into the physical constraint neural network, which accelerates the convergence speed of the physical constraint neural network during the training phase.

[0071] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention, such as... Figure 5As shown, the electronic device may include: a processor 510, a communications interface 520, a memory 530, and a communication bus 540, wherein the processor 510, the communications interface 520, and the memory 530 communicate with each other via the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute a frequency response compensation method for ultra-wideband pulse signals, the method including: Acquire an ultra-wideband pulse signal and perform a fast Fourier transform on the ultra-wideband pulse signal to obtain the complex spectrum corresponding to the ultra-wideband pulse signal; The complex spectrum is decomposed into real and imaginary parts and input into a physical constraint neural network to obtain the target compensation filter output by the physical constraint neural network; the physical constraint neural network is based on sample ultra-wideband pulse signals and trained by a physical constraint loss function; the physical constraint loss function is determined based on the channel frequency response corresponding to the sample ultra-wideband pulse signals. The complex spectrum and the target compensation filter are multiplied by a complex number in the frequency domain to obtain the compensated spectrum; Perform an inverse fast Fourier transform on the compensated spectrum to obtain the recovered pulse signal corresponding to the ultra-wideband pulse signal.

[0072] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0073] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program that can be stored on a non-transitory computer-readable storage medium, wherein when the computer program is executed by a processor, the computer is capable of executing the frequency response compensation method for ultra-wideband pulse signals provided by the above methods, the method comprising: Acquire an ultra-wideband pulse signal and perform a fast Fourier transform on the ultra-wideband pulse signal to obtain the complex spectrum corresponding to the ultra-wideband pulse signal; The complex spectrum is decomposed into real and imaginary parts and input into a physical constraint neural network to obtain the target compensation filter output by the physical constraint neural network; the physical constraint neural network is based on sample ultra-wideband pulse signals and trained by a physical constraint loss function; the physical constraint loss function is determined based on the channel frequency response corresponding to the sample ultra-wideband pulse signals. The complex spectrum and the target compensation filter are multiplied by a complex number in the frequency domain to obtain the compensated spectrum; Perform an inverse fast Fourier transform on the compensated spectrum to obtain the recovered pulse signal corresponding to the ultra-wideband pulse signal.

[0074] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a frequency response compensation method for ultra-wideband pulse signals provided by the methods described above, the method comprising: Acquire an ultra-wideband pulse signal and perform a fast Fourier transform on the ultra-wideband pulse signal to obtain the complex spectrum corresponding to the ultra-wideband pulse signal; The complex spectrum is decomposed into real and imaginary parts and input into a physical constraint neural network to obtain the target compensation filter output by the physical constraint neural network; the physical constraint neural network is based on sample ultra-wideband pulse signals and trained by a physical constraint loss function; the physical constraint loss function is determined based on the channel frequency response corresponding to the sample ultra-wideband pulse signals. The complex spectrum and the target compensation filter are multiplied by a complex number in the frequency domain to obtain the compensated spectrum; Perform an inverse fast Fourier transform on the compensated spectrum to obtain the recovered pulse signal corresponding to the ultra-wideband pulse signal.

[0075] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0076] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0077] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for frequency response compensation of ultra-wideband pulse signals, characterized in that, include: Acquire an ultra-wideband pulse signal and perform a fast Fourier transform on the ultra-wideband pulse signal to obtain the complex spectrum corresponding to the ultra-wideband pulse signal; The complex spectrum is decomposed into real and imaginary parts and input into a physical constraint neural network to obtain the target compensation filter output by the physical constraint neural network; the physical constraint neural network is based on sample ultra-wideband pulse signals and trained by a physical constraint loss function; the physical constraint loss function is determined based on the channel frequency response corresponding to the sample ultra-wideband pulse signals. The complex spectrum and the target compensation filter are multiplied by a complex number in the frequency domain to obtain the compensated spectrum; Perform an inverse fast Fourier transform on the compensated spectrum to obtain the recovered pulse signal corresponding to the ultra-wideband pulse signal.

2. The frequency response compensation method for ultra-wideband pulse signals according to claim 1, characterized in that, The step of decomposing the complex spectrum into real and imaginary parts and inputting them into a physical constraint neural network to obtain the target compensation filter output by the physical constraint neural network includes: The real part and the imaginary part are input into the input layer of the physical constraint neural network to obtain the feature map output by the input layer; The feature map is input into the hidden layer of the physical constraint neural network to obtain the real part and imaginary part features output by the hidden layer; The real and imaginary features are input into the output layer of the physical constraint neural network to obtain the target compensation filter output by the output layer.

3. The frequency response compensation method for ultra-wideband pulse signals according to claim 2, characterized in that, The step of inputting the feature map into the hidden layer of the physical constraint neural network to obtain the real and imaginary features output by the hidden layer includes: The feature map is input into the complex feature extraction layer in the hidden layer to obtain a sequence of multiple frequency points output by the complex feature extraction layer; Input all the frequency point sequences into the frequency domain global correlation layer in the hidden layer to obtain the real part features and the imaginary part features output by the frequency domain global correlation layer.

4. The frequency response compensation method for ultra-wideband pulse signals according to claim 3, characterized in that, The step of inputting all the frequency point sequences into the frequency domain global correlation layer in the hidden layer to obtain the real part features and the imaginary part features output by the frequency domain global correlation layer includes: Input all the frequency point sequences into the transformer coding layer in the frequency domain global correlation layer to obtain multiple word sequences output by the transformer coding layer; Input all the word sequences into the position coding layer in the frequency domain global association layer to obtain the frequency order relationship output by the position coding layer; The frequency order relationship and all the word sequences are input into the multi-head self-attention mechanism layer in the frequency domain global association layer to obtain the real part features and the imaginary part features output by the multi-head self-attention mechanism layer.

5. The frequency response compensation method for ultra-wideband pulse signals according to claim 1, characterized in that, The physical constraint loss function is determined based on the following method: The physical inverse transform constraints are determined based on the channel frequency response and the compensation filter during the training phase. The energy regularization term and frequency smoothness constraint are determined based on the compensation filter in the training phase. The physical constraint loss function is determined based on the physical inverse transformation constraint, the energy regularization term, and the frequency smoothness constraint.

6. The frequency response compensation method for ultra-wideband pulse signals according to claim 5, characterized in that, The determination of physical inverse transform constraints based on the channel frequency response and the compensation filter during the training phase includes: ； in, This represents the physical inverse transformation constraint. This represents the total number of frequency points. Indicates the first The compensation filter for the training phase corresponding to each frequency point. Indicates the first Channel frequency response corresponding to each frequency point; The determination of the energy regularization term and frequency smoothness constraint based on the compensation filter during the training phase includes: ； ； in, This represents the energy regularization term. This indicates the frequency smoothness constraint. Indicates the first The compensation filter for the training phase corresponding to each frequency point.

7. The frequency response compensation method for ultra-wideband pulse signals according to any one of claims 1 to 6, characterized in that, The frequency range of the ultra-wideband pulse signal is 10 kHz to 18 MHz.

8. A frequency response compensation device for ultra-wideband pulse signals, characterized in that, include: An acquisition module is used to acquire an ultra-wideband pulse signal and perform a fast Fourier transform on the ultra-wideband pulse signal to obtain the complex spectrum corresponding to the ultra-wideband pulse signal. A network module is used to decompose the complex spectrum into real and imaginary parts and input them into a physical constraint neural network to obtain a target compensation filter output by the physical constraint neural network; the physical constraint neural network is based on sample ultra-wideband pulse signals and trained using a physical constraint loss function; the physical constraint loss function is determined based on the channel frequency response corresponding to the sample ultra-wideband pulse signals. The compensation module is used to perform a frequency domain complex multiplication of the complex spectrum and the target compensation filter to obtain the compensated spectrum; The recovery module is used to perform a fast inverse Fourier transform on the compensated spectrum to obtain the recovered pulse signal corresponding to the ultra-wideband pulse signal.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the frequency response compensation method for ultra-wideband pulse signals as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the frequency response compensation method for ultra-wideband pulse signals as described in any one of claims 1 to 7.

Images