Receiving end signal correction method and device based on terahertz wireless communication and medium

Abstract

The invention discloses a receiving end signal correction method and device based on terahertz wireless communication and a medium, belongs to the technical field of signal processing, and is used for solving the technical problems that the existing receiving end signal correction method of wireless communication neglects the existence of noise and interference and the accuracy of signal correction needs to be improved. The method comprises the following steps: acquiring a first signal received by a receiving end, and converting the first signal from a time domain to a frequency domain to obtain a first amplitude, a first phase and a first frequency of the first signal, the first signal being a signal of an original signal sent by a sending end reaching the receiving end through a wireless communication channel; correcting at least one of the first amplitude, the first phase and the first frequency based on a machine learning model to obtain a corrected second amplitude, a corrected second phase and a corrected second frequency; a corrected second signal is determined based on the corrected second amplitude, the second phase, and the second frequency. According to the method, the accuracy of signal correction is improved.

Description

Technical Field

[0001] The present invention relates to the field of signal processing technology, and in particular to a receiving end signal correction method, device and medium based on terahertz wireless communication. Background Art

[0002] A wireless communication system usually consists of a transmitter and a receiver. The transmitter is responsible for converting information into electromagnetic wave signals and sending them out through the antenna; the transmitted signal reaches the receiver through the channel, and the receiver is responsible for receiving the electromagnetic wave signals and restoring them to the original information.

[0003] Due to the complexity of the channel (such as multipath effects, interference, noise, etc.), the signal received by the receiving end has errors compared to the signal sent by the transmitting end. For example, when terahertz waves propagate in the air, they are easily affected by absorption and scattering, resulting in a shorter transmission distance and greater signal attenuation. The terahertz wave signal received by the receiving end has a larger error than the terahertz wave signal sent by the transmitting end.

[0004] Therefore, after receiving the signal, the receiving end needs to perform signal correction to ensure the accuracy and reliability of the signal. However, the existing signal correction method for the receiving end of wireless communication ignores the existence of noise and interference, and the error between the corrected signal and the transmitted signal is large, and the accuracy of signal correction needs to be improved. Summary of the invention

[0005] The present invention provides a receiving end signal correction method, device and medium based on terahertz wireless communication, so as to solve the technical problems that the existing receiving end signal correction method of wireless communication ignores the existence of noise and interference, the error between the corrected signal and the transmitted signal is large, and the accuracy of signal correction needs to be improved.

[0006] In a first aspect, the present invention provides a receiving end signal correction method based on terahertz wireless communication, comprising:

[0007] Acquire a first signal received by a receiving end, and convert the first signal from a time domain to a frequency domain to obtain a first amplitude, a first phase, and a first frequency of the first signal, wherein the first signal is a signal of an original signal sent by a transmitting end that reaches the receiving end through a wireless communication channel;

[0008] Based on the machine learning model, at least one of the first amplitude, the first phase and the first frequency is corrected respectively to obtain a corrected second amplitude, a second phase and a second frequency;

[0009] A corrected second signal is determined based on the corrected second amplitude, second phase and second frequency.

[0010] In a feasible implementation manner, after converting the first signal from the time domain to the frequency domain, the method further includes:

[0011] The attenuated frequency band in the first signal is enhanced by using a filter to obtain a first signal after frequency domain enhancement.

[0012] In a feasible implementation manner, performing enhancement processing on the attenuated frequency band in the first signal by using a filter includes:

[0013] Acquire the original signal sent by the transmitting end;

[0014] comparing the first signal with the original signal corresponding to the first signal to determine an attenuation frequency band;

[0015] Based on the attenuation frequency band, adjusting the filter weight corresponding to the attenuation frequency band in the learnable filter group;

[0016] The attenuated frequency band in the first signal is enhanced by using the learnable filter group.

[0017] In a feasible implementation, based on a machine learning model, at least one of the first amplitude, the first phase, and the first frequency is corrected respectively to obtain a corrected second amplitude, second phase, and second frequency, including:

[0018] Based on a generative adversarial network, the first amplitude is corrected to obtain a corrected first amplitude, the generator in the generative adversarial network is used to generate a corrected second amplitude, and the discriminator in the generative adversarial network is used to evaluate whether the second amplitude is consistent with the amplitude of the original signal sent by the transmitter.

[0019] In a feasible implementation, based on a machine learning model, at least one of the first amplitude, the first phase, and the first frequency is corrected respectively to obtain a corrected second amplitude, second phase, and second frequency, including:

[0020] Based on a recurrent neural network and / or a long short-term memory network, the first phase is corrected to obtain a corrected second phase.

[0021] In a feasible implementation manner, based on a recurrent neural network and / or a long short-term memory network, the first phase is corrected to obtain a corrected second phase, including:

[0022] Inputting the first phase into a recurrent neural network to obtain an intermediate phase corrected by the recurrent neural network;

[0023] The intermediate phase is input into a long short-term memory network to obtain a corrected second phase.

[0024] In a feasible implementation, based on a machine learning model, at least one of the first amplitude, the first phase, and the first frequency is corrected respectively to obtain a corrected second amplitude, second phase, and second frequency, including:

[0025] Performing dimensionality reduction processing on the first frequency by an encoder in the autoencoder to extract features of the frequency shift;

[0026] The feature of the frequency shift is input into a decoder in the autoencoder to obtain a corrected second frequency.

[0027] In a feasible implementation manner, after determining the corrected second signal, the method further includes:

[0028] Determine a quality parameter of the second signal, the quality parameter comprising at least one of a signal-to-noise ratio, a bit error rate, a phase offset, a frequency error, and a frequency response, the phase offset being an offset between a second phase and a phase of the original signal, the frequency error being an error between a center frequency in the second frequency and a frequency of the original signal, and the frequency response being used to characterize a relationship between the second signal and the original signal;

[0029] Based on the quality parameters, parameters of the machine learning model are adjusted.

[0030] In a second aspect, the present invention provides a receiving-end signal correction device based on terahertz wireless communication, the device comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, so that the at least one processor can execute the receiving-end signal correction method based on terahertz wireless communication described in any of the above-mentioned embodiments.

[0031] In a third aspect, the present invention provides a non-volatile computer storage medium, which is a non-volatile computer-readable storage medium, and the non-volatile computer-readable storage medium stores at least one program, each of which includes instructions, and when the instructions are executed by a terminal, the terminal executes the receiving end signal correction method based on terahertz wireless communication described in any of the above embodiments.

[0032] Compared with the prior art, the receiving end signal correction method, device and medium based on terahertz wireless communication provided by the present invention have the following beneficial technical effects:

[0033] (1) After acquiring the first signal received by the receiving end, the machine learning model of the present invention can correct the amplitude, phase and frequency of the first signal respectively, so as to more comprehensively capture the detailed information contained in the first signal, so that the corrected second amplitude, second phase and second frequency are the same or similar to the amplitude, phase and frequency of the original signal sent by the transmitting end, thereby reducing the deviation of the amplitude, phase and frequency of the first signal caused by noise and interference in the wireless communication system; the corrected second signal obtained based on the corrected second amplitude, second phase and second frequency reduces the error between the corrected second signal and the original signal sent by the transmitting end, thereby improving the accuracy of signal correction at the receiving end of wireless communication.

[0034] (2) Based on the advantages of different types of machine learning models, the present invention can adopt a suitable machine learning model for the first amplitude, first phase, and first frequency in the first signal, and perform multi-branch network error correction on the first signal to restore the original signal to the greatest extent and reduce signal errors.

[0035] (3) The present invention can also identify and enhance the attenuation band in the first signal through a learnable filter group, thereby reducing the interference of the attenuation in the wireless communication system to the first signal, so that the first signal after frequency domain enhancement can more completely reflect the characteristics of the original signal sent by the transmitter.

[0036] (4) The present invention can monitor the signal quality of the corrected second signal through quality parameters such as the signal-to-noise ratio, bit error rate, phase offset and frequency response of the second signal to ensure the accuracy and robustness of signal recovery. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0038] Figure 1 A flow chart of a receiving end signal correction method based on terahertz wireless communication provided by the present invention;

[0039] Figure 2 A schematic diagram of the structure of a receiving-end signal correction device based on terahertz wireless communication provided by the present invention. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the specific embodiments of the present invention and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0041] A signal correction method in the related art obtains the initial level of the initial output signal at the output end of the signal source and the actual level of the actual output signal at the output end of the processing circuit; if they are the same, the actual output signal is output to the signal receiving end; if they are not the same, the actual output signal is corrected and the corrected actual output signal is output to the signal receiving end. However, this method is overly dependent on the accuracy of the initial level. If the obtained initial level itself has errors or drifts, the corrected signal may also be inaccurate. In addition, this method mainly focuses on level matching and ignores the existence of noise and interference. In practical applications, if the signal is affected by noise or interference, this correction method may output an erroneous signal as a correct signal.

[0042] Another signal correction method in the related art includes: receiving a test signal; performing coupling processing on the test signal to obtain a coupled signal; performing low-noise amplification processing on the coupled signal to obtain a low-noise amplified signal; transmitting the signal again to obtain a test signal; determining an amplitude difference and a phase difference based on the current amplified test signal and the last amplified test signal; and correcting a pre-received echo signal based on the amplitude difference and the phase difference to obtain a correct echo signal. Although the low-noise amplification of the coupled signal here can improve the signal strength, it may also amplify part of the noise, thereby affecting the signal quality, so that there may still be residual noise in the corrected signal, affecting the accuracy of the system's echo signal.

[0043] The above two signal correction methods both ignore the existence of noise and interference, the error between the corrected signal and the transmitted signal is large, and the accuracy of signal correction needs to be improved. In view of the above shortcomings in the prior art, the present invention provides a receiving end signal correction method, device and medium based on terahertz wireless communication to solve the above shortcomings and achieve more efficient wireless communication.

[0044] The technical solution proposed by the present invention is described in detail below with reference to the accompanying drawings.

[0045] Figure 1 The flowchart of the receiving end signal correction method based on terahertz wireless communication provided by the present invention. Figure 1 As shown, the method includes the following execution steps:

[0046] S101. Acquire a first signal received by a receiving end, and convert the first signal from a time domain to a frequency domain to obtain a first amplitude, a first phase, and a first frequency of the first signal.

[0047] The first signal received by the receiving end in the present invention is a transmission signal sent by the transmitting end in the wireless communication system, and is a signal that reaches the receiving end after being transmitted through the wireless communication channel.

[0048] For the convenience of analysis and processing, the first signal can be converted from the time domain to the frequency domain. The key features of the frequency domain signal include amplitude, phase and frequency. After the time-frequency conversion, the amplitude, phase and frequency of the first signal can be obtained, that is, the first amplitude, the first phase and the first frequency.

[0049] In some embodiments, the first signal may be converted from the time domain to the frequency domain by Fourier transform.

[0050] In the process of transmitting the original signal sent by the transmitter to the receiver through the wireless communication channel, some frequency bands may be distorted or attenuated due to absorption and scattering. As a possible implementation method, the first signal after time-frequency conversion can be enhanced in the frequency domain, focusing on restoring the frequency band information that is severely attenuated. After converting the first signal from the time domain to the frequency domain, the method also includes: enhancing the attenuated frequency band in the first signal through a filter to obtain the first signal after frequency domain enhancement.

[0051] As a possible implementation method, the attenuation frequency band in the first signal is enhanced through a filter, including: obtaining the original signal sent by the transmitter; comparing the first signal with the original signal corresponding to the first signal to determine the attenuation frequency band; based on the attenuation frequency band, adjusting the filtering weight corresponding to the attenuation frequency band in a learnable filter group; enhancing the attenuation frequency band in the first signal through the learnable filter group.

[0052] S102. Based on the machine learning model, at least one of the first amplitude, the first phase and the first frequency is corrected respectively to obtain a corrected second amplitude, a second phase and a second frequency.

[0053] Due to the influence of noise, attenuation and other transmission errors, the first signal is usually a time-domain signal with noise, so the first signal needs to be corrected.

[0054] The present invention processes the first amplitude, first phase, and first frequency of the first signal respectively: only one of the first amplitude, the first phase, and the first frequency can be corrected through a machine learning model, or two of the first amplitude, the first phase, and the first frequency can be corrected through a machine learning model, or all three of the first amplitude, the first phase, and the first frequency can be corrected through a machine learning model respectively.

[0055] The machine learning model used to correct the first amplitude, the machine learning model used to correct the first phase, and the machine learning model used to correct the first frequency can be the same.

[0056] Different types of machine learning models have different advantages. In order to improve the accuracy of correction, different machine learning models can be used to correct the first amplitude, the first phase and the first frequency respectively according to the characteristics of the first amplitude, the first phase and the first frequency.

[0057] The amplitude represents the strength of the signal, which is usually affected by factors such as noise, attenuation, and multipath effects. In the frequency domain, the amplitude information mainly reflects the energy distribution of the signal. The goal of the machine learning model used to correct the first amplitude is to restore or enhance the correct strength of the signal.

[0058] As a possible implementation, the present invention can use Generative Adversarial Networks (GAN) to correct the first amplitude of the first signal to obtain the corrected second amplitude, because GAN has a good application in amplitude recovery, especially in signal enhancement and compensation, which can reduce the amplitude distortion caused by attenuation or noise. The generator of the generative adversarial network is responsible for generating the corrected second amplitude, while the discriminator evaluates whether the generated second amplitude is consistent with the original signal, and continuously improves the accuracy of amplitude recovery through adversarial training.

[0059] By introducing a GAN-based denoising mechanism and generating simulated noise scenarios for adversarial network training, the network can be more generalized, more efficiently identify and suppress complex noise, and improve bit error rate and transmission stability.

[0060] Phase reflects the position of the signal on the time axis and is usually affected by time delay, phase offset or multipath propagation. Phase optimization is crucial because even if the amplitude is correct, the wrong phase will still cause signal distortion or interference. As a possible implementation method, the first phase is corrected based on a recurrent neural network (RNN) and / or a long short-term memory network (LSTM) to obtain a corrected second phase.

[0061] In phase correction, RNN or LSTM can learn the temporal dependencies of the signal, especially in the problem of delay and phase drift. They can capture the long-term dependencies in the signal and thus optimize the phase effectively.

[0062] In some embodiments, one of RNN and LSTM can be used to correct the first phase of the first signal to obtain a corrected second signal; or both RNN and LSTM machine learning models can be used at the same time to combine the advantages of both for phase correction.

[0063] As a possible implementation method, the first phase is corrected based on a recurrent neural network and / or a long short-term memory network to obtain a corrected second phase, including: inputting the first phase into the recurrent neural network to obtain an intermediate phase corrected by the recurrent neural network; inputting the intermediate phase into the long short-term memory network to obtain the corrected second phase.

[0064] The RNN and LSTM networks in phase correction optimize the delay and phase drift of the signal by combining time series modeling and memory capabilities. Specifically, the RNN and LSTM models accept the original phase information of the input signal (such as time series data or frequency domain features) as input and capture the time correlation and long-term dependency of the signal through their recursive structure. The output of the model is the corrected phase value.

[0065] In this process, RNN is used to quickly capture short-term temporal dependencies, while LSTM solves the problem of traditional RNN being easy to forget on long sequences through its memory unit and gating mechanism. The combination of the two ensures that the network can accurately model short-term characteristics and effectively deal with long-term phase drift problems.

[0066] The optimization effect is reflected in the fact that the generated correction phase can effectively reduce the error caused by time delay and reduce the impact of phase mismatch on signal transmission quality, thereby improving the overall fidelity and stability of the signal. By introducing this time series modeling technology, the original phase information of the signal can be restored more accurately compared to the correction method based only on amplitude or static analysis.

[0067] Frequency offset is usually caused by factors such as multipath effect, frequency drift, carrier frequency deviation, etc. Frequency correction mainly focuses on recovering the correct frequency position of the signal to ensure correct demodulation and synchronization of the signal. As a possible implementation method, the first frequency can be corrected by an autoencoder to obtain a corrected second frequency.

[0068] The autoencoder in frequency correction learns and corrects the frequency offset in the signal through its compression and reconstruction mechanism. Specifically, the autoencoder takes the first frequency of the first signal as input, performs dimensionality reduction processing on the first frequency through the encoder, and extracts the characteristics of the frequency offset. Then, the decoder reconstructs the signal based on the extracted characteristics and outputs the corrected second frequency or the frequency offset correction value.

[0069] The encoder is responsible for capturing the patterns and regularities of frequency offsets and helping the network identify complex frequency offset characteristics, while the decoder learns how to restore the signal and corrects frequency drift, carrier frequency deviation and other problems. The entire autoencoder network is trained end-to-end to make the frequency position of the reconstructed signal as close as possible to the correct frequency of the original signal.

[0070] The optimization effect is reflected in the fact that the autoencoder can accurately correct the frequency error, restore the correct frequency position of the signal, and ensure the accuracy of demodulation and synchronization. Compared with traditional frequency correction methods (such as phase-locked loops or frequency filtering), the autoencoder can more flexibly adapt to nonlinear frequency drift in complex environments, providing higher robustness and accuracy for the frequency recovery of the first signal.

[0071] S103: Determine a corrected second signal based on the corrected second amplitude, second phase, and second frequency.

[0072] The present invention can fuse the corrected second amplitude, second phase and second frequency through the output of the machine learning model to obtain a corrected frequency domain signal, and then convert the corrected frequency domain signal back to the time domain through an inverse Fourier transform to obtain a corrected second signal.

[0073] By correcting the first amplitude, the first phase and the first frequency in the first signal respectively, the details of the signal can be captured more comprehensively and the accuracy of signal reconstruction can be improved.

[0074] As a possible implementation, the present invention can measure the accuracy of signal recovery and the performance of the machine learning model used for signal correction by the quality parameter of the second signal. After determining the corrected second signal, the method further includes: determining the quality parameter of the second signal, the quality parameter including at least one of signal-to-noise ratio, bit error rate, phase offset, frequency error and frequency response. Based on the quality parameter, adjust the parameters of the machine learning model.

[0075] The signal-to-noise ratio is an indicator of signal quality. The higher the signal-to-noise ratio, the better the signal quality. The signal-to-noise ratio (SNR) of the second signal can be calculated by comparing the signal power with the noise power: Among them, P signal is the signal power in the second signal, P noise is the noise power in the second signal. The higher the SNR, the better the signal quality of the second signal.

[0076] The Bit Error Rate (BER) is a measure of the probability of bit errors during transmission and is an important indicator of the performance of a communication system. The BER is calculated by comparing the difference between the received bit stream and the originally sent bit stream to determine the number of errors.

[0077] The present invention can determine the number of bit errors in the second signal by the difference between the bit stream corresponding to the second signal and the bit stream corresponding to the original signal, and determine the bit error rate of the second signal by the proportion of the number of bit errors in the total number of codes.

[0078] The phase offset is an offset between the second phase and the phase of the original signal sent by the transmitting end.

[0079] The frequency error is an error between a center frequency in the second frequency and a frequency of an original signal sent by the sending end.

[0080] Frequency response is a function that describes the response characteristics of a system to input signals of different frequencies. It is usually expressed as a function of the amplitude ratio and phase difference of the system output and input as a function of frequency. It reflects the system's ability to process signals in the frequency domain and can be used to analyze the distortion, gain and attenuation of signals when they pass through the system.

[0081] The amplitude response is the modulus of the frequency response, which indicates the gain or attenuation of the system for different frequency components. If the amplitude response of a frequency component is less than 1 (or negative gain, usually expressed in decibels (dB) on a logarithmic scale), it means that the frequency signal component is attenuated after propagation or passing through the system.

[0082] Phase response is the angular portion of the frequency response, which describes the time delay or phase shift of the signal at different frequency components after the signal passes through the system. This delay can cause signal distortion, especially in multi-band signals.

[0083] The frequency response of the second signal is used to characterize the relationship between the second signal and the original signal sent by the transmitter, that is, the frequency response of the wireless communication system composed of the transmitter, the receiver and the wireless communication channel.

[0084] Terahertz waves (0.1 to 10 THz) have abundant spectrum resources and have the potential for large bandwidth and extremely high data transmission rate. They are considered to be the key technology for future 6G or even 7G communications. However, terahertz waves are susceptible to absorption and scattering when propagating in the air, resulting in a short transmission distance, large signal attenuation, and a large error between the signal received by the receiving end and the signal sent by the transmitting end. Therefore, how to improve the transmission efficiency of terahertz wave communication and overcome the challenges in its propagation has become an urgent problem to be solved. Taking terahertz wave communication as an example, the receiving end signal correction method based on terahertz wireless communication provided by the present invention is explained.

[0085] The core function of this method is to ensure that the data can be restored to the original signal as accurately as possible at the receiving end, reducing interference from factors such as noise, attenuation, and scattering.

[0086] In this method, the received terahertz wave signal undergoes multiple processing steps, including signal feature extraction, frequency domain enhancement, time-frequency joint analysis, and multi-branch network error correction, in order to restore the original signal sent by the transmitter to the greatest extent and reduce the signal error. The specific workflow is as follows.

[0087] First, a terahertz wave signal from a terahertz wave transmission link is received, that is, a first signal, which is usually affected by noise, attenuation and other transmission errors. The first signal is usually a time domain signal with noise, expressed as y(t), representing the received time domain signal.

[0088] The main characteristics of the terahertz signal include amplitude, phase and frequency. Amplitude represents the strength or power of the terahertz signal, which is usually affected by factors such as attenuation, noise and multipath effects. Phase represents the phase information of the terahertz signal in the frequency domain, which may be offset due to multipath propagation effects. Frequency is affected by the frequency characteristics of the channel, and the attenuation degree of different frequency bands is different.

[0089] After receiving the terahertz signal, Fourier transform is first used to perform time-frequency domain conversion to convert the time domain signal into a frequency domain signal for subsequent processing.

[0090] During the transmission of terahertz wave signals, some frequency bands may be distorted or attenuated due to absorption and scattering. Therefore, the present invention can convert the first signal from the time domain to the frequency domain, and then enhance the first signal in the frequency domain, focusing on restoring the frequency band information that is severely attenuated.

[0091] Frequency domain enhancement can use a learnable filter group to dynamically adjust the frequency band filtering weights so that the model retains more information for severely attenuated frequency bands and improves the final signal restoration effect.

[0092] The method for identifying attenuated frequency bands includes: converting the time domain signal y(t) of the first signal into a frequency domain signal Y(f) through Fourier transform, and then comparing the signal strength and noise level of each frequency band of the received first signal with the original signal sent by the transmitter to identify those attenuated frequency bands caused by factors such as the propagation medium (such as the atmosphere).

[0093] By using a learnable filter bank to enhance the received terahertz wave signal in the frequency domain and restore the signal strength in the attenuated frequency band, the terahertz signal can more completely reflect the characteristics of the original signal. Compared with general noise reduction schemes, this method can better adapt to the frequency band characteristics of terahertz wave signals and improve the quality of reconstructed signals.

[0094] In frequency domain signals, amplitude, phase and frequency are three very critical parameters that have a direct impact on the quality and accuracy of the signal. After frequency domain enhancement, different types of networks have different advantages in optimizing these three parameters. The following is a specific analysis of amplitude, phase and frequency to select the most suitable network architecture.

[0095] Amplitude represents the strength of the signal, which is usually affected by factors such as noise, attenuation, and multipath effects. In the frequency domain, amplitude information mainly reflects the energy distribution of the signal. The goal of optimizing the amplitude is to restore or enhance the correct strength of the signal. This branch uses a generative adversarial network (GAN) to correct the signal amplitude because GAN has a good application in amplitude recovery, especially in signal enhancement and compensation, which can reduce amplitude distortion caused by attenuation or noise. The generator of the generative adversarial network is responsible for generating the corrected amplitude, while the discriminator evaluates whether the generated amplitude is consistent with the real signal, and continuously improves the accuracy of amplitude recovery through adversarial training.

[0096] By introducing a denoising mechanism based on a generative adversarial network (GAN), a simulated noise scenario adversarial network training is generated, making it more generalizable, able to more efficiently identify and suppress complex noise, and improve bit error rate and transmission stability.

[0097] Phase reflects the position of the signal on the time axis and is usually affected by delay, phase offset or multipath propagation. Phase optimization is crucial because even if the amplitude is correct, the wrong phase can still cause signal distortion or interference. This branch can use recurrent neural networks (RNN) and long short-term memory networks (LSTM). In phase correction, RNN or LSTM can learn the time dependency of the signal, especially in the problem of delay and phase drift. They can capture the long-term dependencies in the signal, thereby effectively optimizing the phase.

[0098] Frequency offset is usually caused by factors such as multipath effects, frequency drift, and carrier frequency deviation. Frequency correction focuses on recovering the correct frequency position of the signal to ensure correct demodulation and synchronization of the signal. This branch can use an autoencoder, which can learn the pattern of frequency offset by compressing and reconstructing the input signal. It corrects the frequency offset through the reconstruction process and recovers the correct frequency of the signal.

[0099] Through the output of the machine learning model, the amplitude, phase, and frequency information are corrected, and then the optimized values ​​are fused to obtain a corrected frequency domain signal. Then, the corrected frequency domain signal is converted back to the time domain through inverse Fourier transform to obtain the final corrected second signal

[0100]

[0101] Compared with the general scheme, which only extracts a single feature for denoising, this scheme can capture the details of the signal more comprehensively and improve the accuracy of signal reconstruction.

[0102] During the signal correction process, we continuously extract and monitor various signal parameters to ensure the accuracy of signal recovery. These parameters include signal-to-noise ratio, bit error rate, phase offset and frequency response.

[0103] This scheme uses frequency domain enhancement and multi-branch optimization to more comprehensively capture signal details, improve the accuracy of signal reconstruction, and the transmission efficiency and stability of terahertz wave communication.

[0104] The signal parameters in the whole process include signal-to-noise ratio, bit error rate, signal amplitude, phase offset, frequency response, etc. These parameters can be used to monitor the optimized signal quality in real time to ensure the accuracy and robustness of signal recovery.

[0105] Corresponding to the above-mentioned embodiments, the present invention further provides a receiving-end signal correction device based on terahertz wireless communication. Figure 2 A structural schematic diagram of a receiving-end signal correction device based on terahertz wireless communication provided in an embodiment of the present invention, the device may include: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, so that the at least one processor can execute the receiving-end signal correction method based on terahertz wireless communication as described in the above embodiment.

[0106] In a possible implementation of the present invention, the at least one processor described above is capable of executing the following steps: acquiring a first signal received by a receiving end, and converting the first signal from the time domain to the frequency domain to obtain a first amplitude, a first phase, and a first frequency of the first signal, where the first signal is a signal of the original signal sent by a transmitting end that reaches the receiving end via a wireless communication channel; based on a machine learning model, respectively correcting at least one of the first amplitude, the first phase, and the first frequency to obtain a corrected second amplitude, second phase, and second frequency; and determining a corrected second signal based on the corrected second amplitude, second phase, and second frequency.

[0107] In a specific implementation, the present invention can also be provided as a non-volatile computer storage medium, wherein the storage medium is a non-volatile computer-readable storage medium, and the non-volatile computer-readable storage medium stores at least one program, each program includes instructions, and when the instructions are executed by the terminal, the terminal executes the receiving end signal correction method based on terahertz wireless communication as described in the above embodiment.

[0108] In a possible implementation of the present invention, the aforementioned terminal executes, obtains a first signal received by a receiving end, and converts the first signal from the time domain to the frequency domain to obtain a first amplitude, a first phase and a first frequency of the first signal, where the first signal is a signal of the original signal sent by the transmitting end that reaches the receiving end through a wireless communication channel; based on a machine learning model, at least one of the first amplitude, the first phase and the first frequency is corrected respectively to obtain a corrected second amplitude, a second phase and a second frequency; based on the corrected second amplitude, the second phase and the second frequency, a corrected second signal is determined.

[0109] The various embodiments of the present invention are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0110] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0111] The above description is only an embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A receiving end signal correction method based on terahertz wireless communication, characterized in that: The method comprises: Acquire a first signal received by a receiving end, and convert the first signal from a time domain to a frequency domain to obtain a first amplitude, a first phase, and a first frequency of the first signal, wherein the first signal is a signal of an original signal sent by a transmitting end that reaches the receiving end through a wireless communication channel; Based on the machine learning model, at least one of the first amplitude, the first phase and the first frequency is corrected respectively to obtain a corrected second amplitude, a second phase and a second frequency; A corrected second signal is determined based on the corrected second amplitude, second phase and second frequency.

2. The receiving end signal correction method based on terahertz wireless communication according to claim 1 is characterized in that: After converting the first signal from the time domain to the frequency domain, the method further comprises: The attenuated frequency band in the first signal is enhanced by using a filter to obtain a first signal after frequency domain enhancement.

3. The receiving end signal correction method based on terahertz wireless communication according to claim 2 is characterized in that: The attenuated frequency band in the first signal is enhanced by a filter, including: Acquire the original signal sent by the transmitting end; comparing the first signal with the original signal corresponding to the first signal to determine an attenuation frequency band; Based on the attenuation frequency band, adjusting the filter weight corresponding to the attenuation frequency band in the learnable filter group; The attenuated frequency band in the first signal is enhanced by using the learnable filter group.

4. The receiving end signal correction method based on terahertz wireless communication according to claim 1 is characterized in that: Based on the machine learning model, at least one of the first amplitude, the first phase and the first frequency is corrected respectively to obtain a corrected second amplitude, a second phase and a second frequency, including: Based on a generative adversarial network, the first amplitude is corrected to obtain a corrected first amplitude, the generator in the generative adversarial network is used to generate a corrected second amplitude, and the discriminator in the generative adversarial network is used to evaluate whether the second amplitude is consistent with the amplitude of the original signal sent by the transmitter.

5. The receiving end signal correction method based on terahertz wireless communication according to claim 1, characterized in that: Based on the machine learning model, at least one of the first amplitude, the first phase and the first frequency is corrected respectively to obtain a corrected second amplitude, a second phase and a second frequency, including: Based on a recurrent neural network and / or a long short-term memory network, the first phase is corrected to obtain a corrected second phase.

6. The receiving end signal correction method based on terahertz wireless communication according to claim 5, characterized in that: Based on a recurrent neural network and / or a long short-term memory network, correcting the first phase to obtain a corrected second phase includes: Inputting the first phase into a recurrent neural network to obtain an intermediate phase corrected by the recurrent neural network; The intermediate phase is input into a long short-term memory network to obtain a corrected second phase.

7. The receiving end signal correction method based on terahertz wireless communication according to claim 1 is characterized in that: Based on the machine learning model, at least one of the first amplitude, the first phase and the first frequency is corrected respectively to obtain a corrected second amplitude, a second phase and a second frequency, including: Performing dimensionality reduction processing on the first frequency by an encoder in the autoencoder to extract features of the frequency shift; The feature of the frequency shift is input into a decoder in the autoencoder to obtain a corrected second frequency.

8. The receiving end signal correction method based on terahertz wireless communication according to claim 1, characterized in that: After determining the corrected second signal, the method further comprises: Determine a quality parameter of the second signal, the quality parameter comprising at least one of a signal-to-noise ratio, a bit error rate, a phase offset, a frequency error, and a frequency response, the phase offset being an offset between a second phase and a phase of the original signal, the frequency error being an error between a center frequency in the second frequency and a frequency of the original signal, and the frequency response being used to characterize a relationship between the second signal and the original signal; Based on the quality parameters, parameters of the machine learning model are adjusted.

9. A receiving-end signal correction device based on terahertz wireless communication, characterized in that: The device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, so that the at least one processor can execute the receiving end signal correction method based on terahertz wireless communication according to any one of claims 1-8.

10. A non-volatile computer storage medium, characterized in that: The storage medium is a non-volatile computer-readable storage medium, and the non-volatile computer-readable storage medium stores at least one program, each of which includes instructions, and when the instructions are executed by the terminal, the terminal executes the receiving end signal correction method based on terahertz wireless communication according to any one of claims 1-8.

Images