5G mobile phone signal denoising method and system based on artificial intelligence

Abstract

The invention discloses a 5G mobile phone signal denoising method and system based on artificial intelligence, and relates to the technical field of communication. The system comprises a signal acquisition module; a preprocessing module; a post-processing module; the 5G mobile phone signal is denoised by creatively fusing the Fourier transform, the wavelet transform and the convolutional neural network technology, the denoising effect is qualitatively improved, and field tests show that the error rate is averagely reduced by more than 40%, and the data transmission rate is stably improved by about 30%; the system has super-strong adaptive capacity and can be dynamically adjusted along with a complex and changeable 5G signal environment, and a cloud updating mechanism guarantees long-term efficient operation; all the modules are organically integrated in the mobile phone, extra heavy hardware is not needed, the cost is controllable, large-scale popularization is easy, and the 5G mobile phone communication stability and the user experience are comprehensively improved.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a 5G mobile phone signal denoising method and system based on artificial intelligence. Background Art

[0002] 5G mobile phone signal transmission faces many difficulties of noise interference. In urban canyons with tall buildings, signals are repeatedly blocked and reflected by buildings, causing multipath fading noise, greatly disordered phase, and sharply deteriorating signal quality at the receiving end. In indoor scenes, electromagnetic radiation released by various electronic devices will also interfere with sound. For example, when the microwave oven in the home is working, the leaked microwave frequency band overlaps with the 5G frequency band, instantly generating high-intensity pulse noise, drowning out weak 5G signals; densely placed computers, printers and other equipment in the office continuously radiate messy electromagnetic signals, making 5G mobile phone signal transmission difficult. In addition, natural environmental factors also have a great impact. When it rains heavily, the scattering and absorption of electromagnetic waves by raindrops generates rain attenuation noise, resulting in a significant loss of signal energy; in active ionosphere areas, ionosphere flickering causes irregular fluctuations in signal strength and phase, just like reefs on the signal transmission channel, threatening stable data transmission. These complex and intertwined noises seriously distort 5G signals, causing aggravated signal distortion, a surge in data transmission bit error rate, frozen video call images, interrupted voice, high latency in online games, and slow progress bars for large file downloads, greatly affecting users' communication experience.

[0003] Traditional strategies for dealing with signal noise mostly rely on filter-based methods. Simple filters, such as low-pass filters, can only mechanically block signal components above the set frequency, and are unable to cope with the complex and changeable spectrum of 5G noise. 5G communication frequency band resources are dynamically allocated, and the signal spectrum width and center frequency of different application scenarios are flexible. Filters with fixed frequency parameters are difficult to accurately match, and often mistakenly filter out useful signal bands as noise, resulting in blurred signal details. Although adaptive filters try to dynamically adjust parameters, they rely on the statistical characteristics of signals and noise. In actual 5G scenarios, the suddenness and non-stationarity of noise make their assumptions invalid, and they cannot track the trajectory of noise changes in real time, and the denoising effect is unsatisfactory.

[0004] In view of the diverse usage scenarios of 5G mobile phones, the wide range of noise sources and the bizarre and changeable characteristics, it is urgent to develop a 5G mobile phone signal denoising method and system based on artificial intelligence according to the above-mentioned related technologies. Summary of the invention

[0005] The purpose of the present invention is to provide a 5G mobile phone signal denoising method and system based on artificial intelligence, which utilizes the powerful learning and pattern recognition capabilities of artificial intelligence technology to deeply analyze the 5G signal and noise characteristics, achieve accurate denoising, and ensure high-quality signal transmission.

[0006] To achieve the above object, the present invention provides the following technical solutions: A 5G mobile phone signal denoising system based on artificial intelligence, including the following modules: Signal acquisition module: used to collect 5G mobile phone signals, connected to the antenna and other signal receiving components in the 5G mobile phone, and able to obtain the original signal with noise; Preprocessing module: including Fourier transform module, wavelet transform module and convolutional neural network module, which respectively perform Fourier transform, wavelet transform and convolutional neural network model operations on the collected original signals; The convolutional neural network module in the preprocessing module is composed of a convolutional layer, a pooling layer and a fully connected layer, and is used to receive the preprocessed signal data and perform feature extraction and denoising; Post-processing module: Perform an inverse transform operation on the denoised signal features output by the pre-processing module, convert the signal back to the time domain, and obtain the final denoised 5G mobile phone signal.

[0007] A 5G mobile phone signal denoising method based on artificial intelligence comprises the following steps: Step S1. Signal acquisition: First, the signal acquisition module in the 5G mobile phone is used to collect the noisy 5G mobile phone signal to obtain the noisy 5G mobile phone signal. ; Step S2. Preprocessing operation: preprocessing the noisy 5G mobile phone signal through the preprocessing module; Step S3. Post-processing: The post-processing module performs an inverse Fourier transform on the signal features after the pre-processing operation to obtain the final denoised 5G mobile phone signal.

[0008] Furthermore, the preprocessing operation in step S2 includes the following steps: Step S201. Fourier transform: the collected signal Perform Fourier transform and get , process the frequency domain signal according to the threshold method to obtain the frequency domain signal after preliminary processing ; Step S202. Wavelet transform: performing wavelet transform on the signal processed by Fourier transform; Step S203. Convolutional neural network denoising: Input the signal data processed by Fourier transform and wavelet transform into the trained convolutional neural network, and obtain the denoised signal features according to the calculation order of convolution layer, pooling layer and fully connected layer.

[0009] Furthermore, the formula for processing the frequency domain signal using the threshold method in step S201 is: , Where N is the length of the signal, k = 0, 1, ..., N-1. In the frequency domain, noise usually appears as a high-frequency component. The frequency domain signal is processed by setting an appropriate threshold T. ,for The frequency components of >T are retained. The frequency components ≤T are set to 0, and the frequency domain signal after preliminary processing is obtained. .

[0010] Furthermore, the process of wavelet transformation in step S202 is as follows: Step S2021. For signal , its continuous wavelet transform formula is: ψ , Where a is the scale factor, b is the translation factor, ψ is the wavelet function; Step S2022. In the discrete case, set the discrete signal , discrete wavelet transform is realized by filter bank, the signal is passed through low-pass filter and high pass filter Get the low frequency component cA and the high frequency component cD; Step S2023. During denoising, since the high-frequency component cD contains more noise, a soft threshold method is used to denoise the signal, and the signal denoised by wavelet transform is obtained. .

[0011] Furthermore, the process of the soft threshold denoising in step S2023 is as follows: The formula of soft threshold method is: , The wavelet coefficients of the high frequency component cD Perform denoising, where is the threshold value, and the denoised wavelet coefficient c is obtained , and then combined with the low-frequency component cA to obtain the signal after wavelet transform denoising .

[0012] Furthermore, the calculation process of the convolutional neural network model denoising in step S203 is as follows: Step S2031. Data preprocessing: The signal data after Fourier transform and wavelet transform is used as the input of the convolutional neural network, which is recorded as a matrix I, whose size is M×N, where M and N are the dimensions of the matrix; Step S2032. Convolution layer calculation: The size of the convolution kernel K in the convolution layer is k×k, where k is the side length of the convolution kernel and the step length is s. For the lth layer in the convolution layer, the calculation of the jth element of the i-th feature map is as follows:

[0013] in is the previous layer, i.e., the l-1 layer element, is the convolution kernel element of the lth layer, is the bias term of the lth layer, is the activation function, ReLU function ; Step S2033. Pooling layer calculation: Use maximum pooling, the pooling window size is p×p, and for the i-th feature map in the pooling layer, the calculation of the j-th element is: , Where m=0,1,…,p-1,n=0,1,…,p-1,x is the output of the convolutional layer; Step S2034. Fully connected layer calculation: After multiple convolutional layers and pooling layers, the obtained feature map is expanded into a one-dimensional vector , the output z of the fully connected layer is: z= , in is the weight, is a vector The elements in , b is the bias term.

[0014] Furthermore, the training process of the neural network model in step S203 is as follows: Step S20301. Data preparation: Collect a large amount of noisy and noise-free 5G mobile phone signal sample data, and divide the signal samples into training set, validation set and test set in a ratio of 7:2:1; Step S20302. Define the loss function: use the mean square error as the original loss function, add the L2 regularization term to the original loss function, and define the signal data after Fourier transform and wavelet transform preprocessing; Step S20303. Select optimization algorithm: Use stochastic gradient descent optimization algorithm. In each iteration, update the parameters according to the gradient of the loss function with respect to weight w and bias b. The weight update formula is w , the bias update formula is b ,in is the learning rate; Step S20304. Training process: For each sample in the training set, the signal data preprocessed by Fourier transform and wavelet transform is input into the convolutional neural network, the output is calculated and the loss is calculated according to the loss function, and then the weights and biases of the network are updated using the optimization algorithm, and this process is repeated until the loss on the validation set no longer decreases.

[0015] Furthermore, the process of defining the loss function in step S20302 is as follows: Step S203021. Use the mean square error as the original loss function. For the signal data preprocessed by Fourier transform and wavelet transform, that is, the input x in the training sample and the corresponding noise-free target output y, the original loss function ,in is the predicted output of the convolutional neural network, and N is the number of samples; Step S203022. Add the L2 regularization term to the original loss function. Let w be the weight parameter in the network, and λ be the regularization parameter, i.e., the hyperparameter. Then, the loss function after adding L2 regularization is: , in, represents the sum of squares of all weights in the network.

[0016] Furthermore, the post-processing process in step S3 is as follows: For the discrete Fourier transform , its discrete Fourier inverse transform formula is: , The denoised signal characteristics of the convolutional neural network output in the frequency domain are , perform inverse discrete Fourier transform to obtain the time domain signal The specific calculation formula is: , Where N is the length of the signal, which is consistent with N in the Fourier transform of step S201, n=0,1,…,N-1. In actual calculation, the value of N is determined according to the signal length set in the processing process. For each k value in, k=0,1,…,N-1, calculate according to the formula The value of Multiply and sum, and finally divide by N to get The value at each n point is obtained, thereby completing the conversion from frequency domain to time domain and obtaining the final denoised time domain signal.

[0017] Compared with the prior art, the present invention has the following beneficial effects: Compared with traditional denoising methods, the denoising effect of the present invention has achieved a qualitative leap. After field testing, the bit error rate is reduced by more than 40% on average, and the data transmission rate is stably increased by about 30%. The system has super adaptive capabilities and can dynamically adjust to the complex and changeable 5G signal environment. The cloud update mechanism ensures long-term and efficient operation. All modules are organically integrated into the mobile phone, without the need for additional bulky hardware, and the cost is controllable. It is easy to promote on a large scale, and the communication stability and user experience of 5G mobile phones are improved in all aspects. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A schematic diagram of a module of a 5G mobile phone signal denoising system based on artificial intelligence provided by an embodiment of the present invention; Figure 2 A flowchart of the specific steps of a 5G mobile phone signal denoising method based on artificial intelligence provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0020] The execution subject of the method of this embodiment is a terminal, which can be a mobile phone, tablet computer, PDA, notebook or desktop computer, etc. Of course, it can also be other devices with similar functions, which is not limited in this embodiment.

[0021] See also Figures 1 to 2 The present invention provides a 5G mobile phone signal denoising method and system based on artificial intelligence, and the method is applied to 5G mobile phone signal denoising based on artificial intelligence, comprising: A 5G mobile phone signal denoising system based on artificial intelligence, including the following modules: Signal acquisition module 1: used to collect 5G mobile phone signals, connected to the antenna and other signal receiving components in the 5G mobile phone, and can obtain the original signal with noise; Preprocessing module 2: It includes Fourier transform module, wavelet transform module and convolutional neural network module, which respectively perform Fourier transform, wavelet transform and convolutional neural network model operation processing on the collected original signal; The convolutional neural network module in the preprocessing module 2 consists of a convolutional layer, a pooling layer, and a fully connected layer, which is used to receive the preprocessed signal data and perform feature extraction and denoising; Post-processing module 3: Perform an inverse transform operation on the denoised signal features output by the pre-processing module 2, convert the signal back to the time domain, and obtain the final denoised 5G mobile phone signal.

[0022] A 5G mobile phone signal denoising method based on artificial intelligence comprises the following steps: Step S1. Signal acquisition: First, the signal acquisition module 1 in the 5G mobile phone is used to collect the noisy 5G mobile phone signal to obtain the noisy 5G mobile phone signal. ; Step S2. Preprocessing operation: preprocessing the noisy 5G mobile phone signal through the preprocessing module 2; Step S3. Post-processing: The post-processing module 3 performs an inverse Fourier transform on the signal features after the pre-processing operation to obtain the final denoised 5G mobile phone signal.

[0023] The pre-processing operation in step S2 includes the following steps: Step S201. Fourier transform: the collected signal Perform Fourier transform and get , process the frequency domain signal according to the threshold method to obtain the frequency domain signal after preliminary processing ; Step S202. Wavelet transform: performing wavelet transform on the signal processed by Fourier transform; Step S203. Convolutional neural network denoising: Input the signal data processed by Fourier transform and wavelet transform into the trained convolutional neural network, and obtain the denoised signal features according to the calculation order of convolution layer, pooling layer and fully connected layer.

[0024] The formula for processing the frequency domain signal by the threshold method in step S201 is: , Where N is the length of the signal, k = 0, 1, ..., N-1. In the frequency domain, noise usually appears as a high-frequency component. The frequency domain signal is processed by setting an appropriate threshold T. ,for The frequency components of >T are retained. The frequency components ≤T are set to 0, and the frequency domain signal after preliminary processing is obtained. .

[0025] The process of wavelet transformation in step S202 is as follows: Step S2021. For signal , its continuous wavelet transform formula is: ψ , Where a is the scale factor, b is the translation factor, ψ is the wavelet function; Step S2022. In the discrete case, set the discrete signal , discrete wavelet transform is realized by filter bank, the signal is passed through low-pass filter and high pass filter Get the low frequency component cA and the high frequency component cD; Step S2023. During denoising, since the high-frequency component cD contains more noise, a soft threshold method is used to denoise the signal, and the signal denoised by wavelet transform is obtained. .

[0026] The process of soft threshold denoising in step S2023 is as follows: The formula of soft threshold method is: The wavelet coefficients of the high frequency component cD Perform denoising, where is the threshold value, and the denoised wavelet coefficient c is obtained , and then combined with the low-frequency component cA to obtain the signal after wavelet transform denoising .

[0027] The calculation process of the convolutional neural network model denoising in step S203 is as follows: Step S2031. Data preprocessing: The signal data after Fourier transform and wavelet transform is used as the input of the convolutional neural network, which is recorded as a matrix I, whose size is M×N, where M and N are the dimensions of the matrix; Step S2032. Convolution layer calculation: The size of the convolution kernel K in the convolution layer is k×k, where k is the side length of the convolution kernel and the step length is s. For the lth layer in the convolution layer, the calculation of the jth element of the i-th feature map is as follows:

[0028] in is the previous layer, i.e., the l-1 layer element, is the convolution kernel element of the lth layer, is the bias term of the lth layer, is the activation function, ReLU function ; Step S2033. Pooling layer calculation: Use maximum pooling, the pooling window size is p×p, and for the i-th feature map in the pooling layer, the calculation of the j-th element is: , Where m=0,1,…,p-1,n=0,1,…,p-1,x is the output of the convolutional layer; Step S2034. Fully connected layer calculation: After multiple convolutional layers and pooling layers, the obtained feature map is expanded into a one-dimensional vector , the output z of the fully connected layer is: z= , in is the weight, is a vector The elements in , b is the bias term.

[0029] The training process of the neural network model in step S203 is as follows: Step S20301. Data preparation: Collect a large amount of noisy and noise-free 5G mobile phone signal sample data, and divide the signal samples into training set, validation set and test set in a ratio of 7:2:1; Step S20302. Define the loss function: use the mean square error as the original loss function, add the L2 regularization term to the original loss function, and define the signal data after Fourier transform and wavelet transform preprocessing; Step S20303. Select optimization algorithm: Use stochastic gradient descent optimization algorithm. In each iteration, update the parameters according to the gradient of the loss function with respect to weight w and bias b. The weight update formula is: , the bias update formula is b ,in is the learning rate; Step S20304. Training process: For each sample in the training set, the signal data preprocessed by Fourier transform and wavelet transform is input into the convolutional neural network, the output is calculated and the loss is calculated according to the loss function, and then the weights and biases of the network are updated using the optimization algorithm, and this process is repeated until the loss on the validation set no longer decreases.

[0030] The process of defining the loss function in step S20302 is as follows: Step S203021. Using mean square error as the original loss function, for the signal data preprocessed by Fourier transform and wavelet transform, that is, the input x in the training sample and the corresponding noise-free target output y, the original loss function is: , in is the predicted output of the convolutional neural network, and N is the number of samples; Step S203022. Add the L2 regularization term to the original loss function. Let w be the weight parameter in the network, and λ be the regularization parameter, i.e., the hyperparameter. Then, the loss function after adding L2 regularization is: , in, represents the sum of squares of all weights in the network.

[0031] The post-processing process in step S3 is as follows: For the discrete Fourier transform , its discrete Fourier inverse transform formula is: , The denoised signal characteristics of the convolutional neural network output in the frequency domain are , perform inverse discrete Fourier transform to obtain the time domain signal The specific calculation formula is: , Where N is the length of the signal, which is consistent with N in the Fourier transform of step S201, n=0,1,…,N-1. In actual calculation, the value of N is determined according to the signal length set in the processing process. For each value of k in , calculated according to the formula The value of Multiply and sum, and finally divide by N to get The value at each n point is obtained, thereby completing the conversion from frequency domain to time domain and obtaining the final denoised time domain signal.

[0032] The office area in a large science and technology park was selected as the test site. The 5G base station coverage in this area is good, but there are many complex interference sources. There are a large number of electronic devices running continuously in the park, such as densely distributed computers, servers, various office appliances, etc. They continuously radiate electromagnetic noise, which interferes with 5G mobile phone signals; at the same time, most of the buildings in the park are modern buildings combining glass curtain walls and steel structures. The signals are frequently reflected and refracted during transmission, causing multipath fading noise, which seriously affects the quality of 5G mobile phone signals: Implementation steps: S1. Signal acquisition: The signal acquisition module 1 of the present invention is turned on in multiple 5G mobile phones participating in the test. The module is closely connected to the mobile phone antenna and other signal receiving components to collect the original 5G mobile phone signal with noise in real time at a sampling frequency of up to 10GHz. This high-frequency sampling ensures that subtle changes in the signal can be accurately captured, and key information will not be missed even in complex environments where the signal fluctuates rapidly. The collected signals are quickly stored in the built-in cache area of ​​the mobile phone in the form of digital sequences, and the collection timestamp is accurately marked to provide an accurate time reference for subsequent synchronous processing and analysis; S2. Preprocessing operations: S201. Fourier transform: , Where N is the length of the signal, k = 0, 1, ..., N-1. In the frequency domain, noise usually appears as a high-frequency component. The frequency domain signal is processed by setting an appropriate threshold T. ,for The frequency components of >T are retained. The frequency components ≤T are set to 0, and the frequency domain signal after preliminary processing is obtained. , retrieve the digital signal from the buffer area, send it to the Fourier transform module, and use the fast Fourier transform (FFT) algorithm to convert the time domain signal into a frequency domain signal. For the obtained frequency domain signal, set the threshold T based on the characteristic that noise is usually concentrated in the high-frequency component. In this test, after preliminary spectrum analysis of the environmental noise in the park, the threshold T is determined to be the amplitude corresponding to a specific frequency value. The frequency domain signal is processed according to the threshold method, that is, the frequency components with amplitudes greater than T in the frequency domain signal are retained, because these are most likely the effective components of the 5G signal; and the frequency components with amplitudes less than or equal to T are directly set to 0, so that some high-frequency noise is initially filtered out, and the frequency domain signal after preliminary processing is obtained; S202. Wavelet transform: Step S2021. For signal , its continuous wavelet transform formula is: ψ , Where a is the scale factor, b is the translation factor, ψ is the wavelet function; Step S2022. In the discrete case, set the discrete signal , discrete wavelet transform is realized by filter bank, the signal is passed through low-pass filter and high pass filter Get the low frequency component cA and the high frequency component cD; Step S2023. During denoising, since the high-frequency component cD contains more noise, a soft threshold method is used to denoise the signal, and the signal denoised by wavelet transform is obtained. ; The process of soft threshold denoising in step S2023 is as follows: The formula of soft threshold method is: , The wavelet coefficients of the high frequency component cD Perform denoising, where is the threshold value, and the denoised wavelet coefficient c is obtained , and then combined with the low-frequency component cA to obtain the signal after wavelet transform denoising , input the signal processed by Fourier transform into the wavelet transform module. The db4 wavelet in the Daubechies wavelet series is selected as the wavelet function, and based on the complexity of the signal, it is determined to perform a 4-layer discrete wavelet transform. Through the filter group, the signal passes through the low-pass filter and the high-pass filter respectively to obtain the low-frequency component cA and the high-frequency component cD. Since the high-frequency component cD contains a lot of noise, the soft threshold method is used for denoising. After in-depth analysis and multiple tests of the park noise samples, the soft threshold is determined to be an optimized value. According to the soft threshold method, the wavelet coefficients of the high-frequency component cD are finely denoised to obtain the denoised wavelet coefficients, which are then recombined with the low-frequency component to obtain the signal after wavelet transform denoising. While retaining the main outline of the signal, the signal effectively removes some detail noise; S203. Convolutional neural network denoising: The signal data after Fourier transform and wavelet transform is used as the input of the convolutional neural network. The data is organized into a matrix I of size M×N, where M=128 and N=128, to adapt to the input requirements of the convolutional neural network. The convolutional neural network model consists of carefully designed convolutional layers, pooling layers, and fully connected layers: Step S2031. Data preprocessing: The signal data after Fourier transform and wavelet transform is used as the input of the convolutional neural network, which is recorded as a matrix I, whose size is M×N, where M and N are the dimensions of the matrix; Step S2032. Convolution layer calculation: The size of the convolution kernel K in the convolution layer is k×k, where k is the side length of the convolution kernel and the step length is s. The convolution layer is configured with convolution kernels of various sizes. The first layer uses a 3×3 convolution kernel with a step length of s=1. For the lth layer in the convolution layer, the calculation of the jth element of the ith feature map strictly follows the formula:

[0033] in is the previous layer, i.e., the l-1 layer element, is the convolution kernel element of the lth layer, is the bias term of the lth layer, is the activation function, ReLU function ; The activation function f uses the ReLU function (f=max). Through the sliding scanning of these convolution kernels, the local features of the signal and noise are accurately extracted, such as identifying the tiny spectrum fluctuations caused by electromagnetic interference of electronic equipment; Step S2033. Pooling layer calculation: Use maximum pooling, the pooling window size is p×p, and for the i-th feature map in the pooling layer, the calculation of the j-th element is: , Where m=0,1,…,p-1,n=0,1,…,p-1,x is the output of the convolutional layer, and the pooling window size is 2×2. Through the pooling operation, redundant data is greatly reduced, the calculation process is accelerated, and the key features are highlighted, allowing the model to focus on more discriminative feature information; Step S2034. Fully connected layer calculation: After multiple convolutional layers and pooling layers, the obtained feature map is expanded into a one-dimensional vector , the output z of the fully connected layer is: z= , in is the weight, is a vector The elements in , b is the bias term, and the fully connected layer is like an "intelligent center" that integrates global features and outputs the denoised signal features based on the signal and noise feature mapping relationship learned in the early stage, accurately restoring the original appearance of the signal; S3. Post-processing: Discrete Fourier Transform , its discrete Fourier inverse transform formula is: , The denoised signal characteristics of the convolutional neural network output in the frequency domain are , perform inverse discrete Fourier transform to obtain the time domain signal The specific calculation formula is: , Where N is the length of the signal, which is consistent with N in the Fourier transform of step S201, n=0,1,…,N-1. In actual calculation, the value of N is determined according to the signal length set in the processing process. For each value of k in , calculated according to the formula The value of Multiply and sum, and finally divide by N to get The value at each n point is obtained, thereby completing the conversion from frequency domain to time domain and obtaining the final denoised time domain signal. The denoised signal characteristics output by the convolutional neural network are in the frequency domain, and the inverse discrete Fourier transform is used to convert it back to the time domain, and the calculation is performed according to the inverse discrete Fourier transform formula. In the actual calculation, since the previous Fourier transform sets the signal length N=1024, at this time, according to the formula, for each k value in (k=0,1,…,10241), the value is accurately calculated, then multiplied and summed, and finally divided by N to obtain the value at each n point (n=0,1,…,10241), thereby perfectly completing the conversion from frequency domain to time domain and obtaining the final denoised time domain signal.

[0034] Implementation effect: After a week of continuous testing in the office area of ​​the science and technology park, the 5G mobile phone with the denoising system of the present invention turned on showed significant differences from the comparison mobile phone without the denoising system turned on. In the present invention, the data is preprocessed by Fourier transform and wavelet transform, and then the convolutional neural network model is used for feature extraction. Combined with L2 regularization, the model calculation efficiency and adaptability are improved, and the denoising over-distorted signal can be restored. In terms of video calls, the mobile phone with the denoising system turned on has reduced the jamming phenomenon by more than 85%, the picture clarity is significantly improved, and the voice call is smooth and uninterrupted; in the data transmission scenario, the download rate of large files is increased by an average of 35%, the web page loading speed is accelerated by 40%, and the online game delay is reduced by 60%, which greatly improves the user's communication experience. At the same time, by regularly transmitting the processed signal samples back to the cloud server, the cloud performs in-depth analysis and evaluation based on the original pure signal at the base station end, and finds that the system of the present invention can effectively adapt to the complex and changeable noise environment in the park. When the noise changes due to dynamic factors such as the opening of new equipment and the flow of personnel, it still maintains stable and efficient denoising performance, which fully verifies the excellent adaptive denoising ability of the present invention and its practical value in actual complex scenarios.

[0035] Those skilled in the art will appreciate that the modules and method steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0036] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices, equipment and modules can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0037] In the several embodiments provided in the present application, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or units can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or equipment, which can be electrical, mechanical or other forms.

[0038] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0039] In addition, each functional module in each embodiment of the present invention may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0040] If the functions are implemented in the form of software function modules and sold or used as independent products, they 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 the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only storage server, random access storage server, disk or optical disk, etc., various media that can store program instructions.

[0041] In addition, it should be noted that the combination of the various technical features in this case is not limited to the combination described in the claims of this case or the combination described in the specific embodiments. All technical features described in this case can be freely combined or combined in any way unless there is a contradiction between them.

[0042] It should be noted that the above examples are only specific embodiments of the present invention, and the present invention is obviously not limited to the above examples, and there are many similar variations. All variations directly derived or associated from the contents disclosed by the technicians in this field should fall within the protection scope of the present invention.

[0043] The above are only preferred embodiments of the present invention and are not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A 5G mobile phone signal denoising system based on artificial intelligence, characterized in that: Includes the following modules: Signal acquisition module (1): used to collect 5G mobile phone signals and connected to the antenna signal receiving component in the 5G mobile phone to obtain the original signal with noise; Preprocessing module (2): including Fourier transform module, wavelet transform module and convolutional neural network module, which respectively perform Fourier transform, wavelet transform and convolutional neural network model operation processing on the collected original signal; The convolutional neural network module in the preprocessing module (2) is composed of a convolutional layer, a pooling layer and a fully connected layer, and is used to receive the preprocessed signal data and perform feature extraction and denoising processing; Post-processing module (3): performs an inverse transformation operation on the denoised signal features output by the pre-processing module (2), converts the signal back to the time domain, and obtains the final denoised 5G mobile phone signal.

2. A 5G mobile phone signal denoising method based on artificial intelligence, characterized in that: The following steps are involved: Step S1. Signal acquisition: First, the signal acquisition module (1) in the 5G mobile phone is used to acquire the noisy 5G mobile phone signal to obtain the noisy 5G mobile phone signal. ; Step S2. Preprocessing operation: preprocessing the noisy 5G mobile phone signal through the preprocessing module (2); Step S3. Post-processing: The post-processing module (3) performs an inverse Fourier transform on the signal features after the pre-processing operation to obtain the final denoised 5G mobile phone signal.

3. The 5G mobile phone signal denoising method based on artificial intelligence according to claim 2 is characterized in that: The pre-processing operation in step S2 includes the following steps: Step S201. Fourier transform: the collected signal Perform Fourier transform and get , process the frequency domain signal according to the threshold method to obtain the frequency domain signal after preliminary processing ; Step S202. Wavelet transform: performing wavelet transform on the signal processed by Fourier transform; Step S203. Convolutional neural network denoising: Input the signal data processed by Fourier transform and wavelet transform into the trained convolutional neural network, and obtain the denoised signal features according to the calculation order of convolution layer, pooling layer and fully connected layer.

4. The 5G mobile phone signal denoising method based on artificial intelligence according to claim 3 is characterized in that: The formula for processing the frequency domain signal using the threshold method in step S201 is: , Where N is the length of the signal, k = 0, 1, ..., N-1. In the frequency domain, noise usually appears as a high-frequency component. The frequency domain signal is processed by setting the threshold T. ,for The frequency components of >T are retained. The frequency components ≤T are set to 0, and the frequency domain signal after preliminary processing is obtained. .

5. The 5G mobile phone signal denoising method based on artificial intelligence according to claim 3 is characterized in that: The process of wavelet transformation in step S202 is as follows: Step S2021. For signal , its continuous wavelet transform formula is: ψ , Where a is the scale factor, b is the translation factor, ψ is the wavelet function; Step S2022. In the discrete case, set the discrete signal , discrete wavelet transform is realized by filter bank, the signal is passed through low-pass filter and high pass filter Get the low frequency component cA and the high frequency component cD; Step S2023. When denoising, the high-frequency component cD contains more noise, and the soft threshold method is used to denoise, and the signal after wavelet transform denoising is obtained. .

6. The 5G mobile phone signal denoising method based on artificial intelligence according to claim 5 is characterized in that: The process of the soft threshold method denoising in step S2023 is as follows: The formula of soft threshold method is: The wavelet coefficients of the high frequency component cD Perform denoising, where is the threshold value, and the denoised wavelet coefficient c is obtained , and then combined with the low-frequency component cA to obtain the signal after wavelet transform denoising .

7. The 5G mobile phone signal denoising method based on artificial intelligence according to claim 3 is characterized in that: The calculation process of the convolutional neural network model denoising in step S203 is as follows: Step S2031. Data preprocessing: The signal data after Fourier transform and wavelet transform is used as the input of the convolutional neural network, which is recorded as a matrix I, whose size is M×N, where M and N are the dimensions of the matrix; Step S2032. Convolution layer calculation: The size of the convolution kernel K in the convolution layer is k×k, where k is the side length of the convolution kernel and the step length is s. For the lth layer in the convolution layer, the calculation of the jth element of the i-th feature map is as follows: ， in is the previous layer, i.e., the l-1 layer element, is the convolution kernel element of the lth layer, is the bias term of the lth layer, is the activation function, ReLU function ; Step S2033. Pooling layer calculation: Use maximum pooling, the pooling window size is p×p, and for the i-th feature map in the pooling layer, the calculation of the j-th element is: , Where m=0,1,…,p-1,n=0,1,…,p-1,x is the output of the convolutional layer; Step S2034. Fully connected layer calculation: After multiple convolutional layers and pooling layers, the obtained feature map is expanded into a one-dimensional vector , the output z of the fully connected layer is: z= , in is the weight, is a vector The elements in , b is the bias term.

8. The 5G mobile phone signal denoising method based on artificial intelligence according to claim 3 is characterized in that: The training process of the neural network model in step S203 is as follows: Step S20301. Data preparation: Collect a large amount of noisy and noise-free 5G mobile phone signal sample data, and divide the signal samples into training set, validation set and test set in a ratio of 7:2:1; Step S20302. Define the loss function: use the mean square error as the original loss function, add the L2 regularization term to the original loss function, and define the signal data after Fourier transform and wavelet transform preprocessing; Step S20303. Select optimization algorithm: Use stochastic gradient descent optimization algorithm. In each iteration, update the parameters according to the gradient of the loss function with respect to weight w and bias b. The weight update formula is w , the bias update formula is b ,in is the learning rate; Step S20304. Training process: For each sample in the training set, the signal data preprocessed by Fourier transform and wavelet transform is input into the convolutional neural network, the output is calculated and the loss is calculated according to the loss function, and then the weights and biases of the network are updated using the optimization algorithm, and this process is repeated until the loss on the validation set no longer decreases.

9. The 5G mobile phone signal denoising method based on artificial intelligence according to claim 8 is characterized in that: The process of defining the loss function in step S20302 is as follows: Step S203021. Use the mean square error as the original loss function. For the signal data preprocessed by Fourier transform and wavelet transform, that is, the input x in the training sample and the corresponding noise-free target output y, the original loss function ,in is the predicted output of the convolutional neural network, and N is the number of samples; Step S203022. Add the L2 regularization term to the original loss function. Let w be the weight parameter in the network, and λ be the regularization parameter, i.e., the hyperparameter. Then, the loss function after adding L2 regularization is: , in, represents the sum of squares of all weights in the network.

10. The 5G mobile phone signal denoising method based on artificial intelligence according to claim 2 is characterized in that: The post-processing process in step S3 is as follows: For the discrete Fourier transform , its discrete Fourier inverse transform formula is: , The denoised signal characteristics of the convolutional neural network output in the frequency domain are , perform inverse discrete Fourier transform to obtain the time domain signal The specific calculation formula is: , Where N is the length of the signal, which is consistent with N in the Fourier transform of step S201, n=0, 1, ..., N-1, and the value of N is determined according to the signal length set during the processing. For each k value in, k=0,1,…,N-1, calculate according to the formula The value of Multiply and sum, and finally divide by N to get The value at each n point is obtained, thereby completing the conversion from frequency domain to time domain and obtaining the final denoised time domain signal.

Images