Remote microwave photon broadband chirp signal down-conversion intelligent receiving device and method

Abstract

The invention discloses a remote microwave photon broadband chirp signal down-conversion intelligent receiving device and method. The device comprises a high-frequency receiving antenna, an electric amplifier, a Mach-Zehnder modulator, a slave laser, a master laser, an optical attenuator, an optical amplifier, a photoelectric detector, a low-pass filter, an analog-to-digital converter, an intelligent processing module, a first optical fiber and a second optical fiber. A high-frequency chirp signal received by the high-frequency receiving antenna is amplified by the electric amplifier and then is sent to the Mach-Zehnder modulator; laser generated by the master laser passes through the optical attenuator and then is sent to the slave laser through the first optical fiber to generate an optical carrier signal, and the optical carrier signal is modulated by the Mach-Zehnder modulator and then sent to the optical amplifier through the second optical fiber for optical power amplification; and then the signals are sequentially sent to a photoelectric detector, a low-pass filter and an analog-to-digital converter for down-conversion, filtering and analog-to-digital conversion, and are sent to an intelligent processing module for processing. The device has the characteristics of low complexity, high sensitivity, long distance and weak signal reception.

Description

Technical Field

[0001] The present invention relates to the field of microwave photonics technology, and particularly to a remote microwave photon broadband chirp signal down-conversion intelligent receiving device and method. Background Art

[0002] The broadband chirp signal direct down-conversion intelligent receiving device is applied to the fields of high-frequency broadband wireless communication systems and radar systems, and is used for the efficient down-conversion receiving processing of broadband chirp signals. Due to the characteristics of large bandwidth and high resolution of broadband chirp signals, their down-conversion processing needs to achieve effective reception under the influence of noise.

[0003] In the existing scheme, the remote receiving unit usually adopts a microwave source external modulation architecture to realize the down-conversion of broadband chirp signals. An external high-power microwave source is used to drive an electro-optic modulator to load the radio frequency chirp signal onto an optical carrier, and then it is transmitted to the central unit through an optical fiber for optoelectronic conversion and signal processing. For example, in 2022, the research group of Shilong Pan from Nanjing University of Aeronautics and Astronautics proposed a deep learning-based time-frequency domain signal recovery method for signal distortion problems in fiber-connected radar networks (Zhou Y, Zhang F, Pan S. “Deep-learning-based time–frequency domain signal recovery for fiber-connected radar networks” Optics Letters, 47(1):50-53, 2022). This scheme has defects such as high cost, high maintenance difficulty, and high complexity, and cannot meet the actual needs of dynamically switching signal reception in scenarios such as radar and communication integration.

[0004] Therefore, there is an urgent need to study a remote microwave photon broadband chirp signal down-conversion intelligent receiving device to achieve the reception of weak signals with remote broadband chirp signals of low complexity. Summary of the Invention

[0005] The purpose of the present invention is to provide a remote microwave photon broadband chirp signal down-conversion intelligent receiving device and method with low complexity and high sensitivity.

[0006] The technical solution for achieving the purpose of the present invention is as follows: A remote microwave photon broadband chirp signal down-conversion intelligent receiving device includes a remote receiving unit and a central unit. The central unit is connected to the remote receiving unit by a first optical fiber, and the remote receiving unit is connected to the central unit by a second optical fiber;

[0007] The remote receiving unit includes a high-frequency receiving antenna, an electrical amplifier, a Mach-Zehnder modulator, and a slave laser; the central unit includes a master laser, an optical attenuator, an optical amplifier, a photodetector, a low-pass filter, an analog-to-digital converter, and an intelligent processing module;

[0008] The output port of the high-frequency receiving antenna is connected to the input port of the electrical amplifier, and the output port of the electrical amplifier is connected to the modulation port of the Mach-Zehnder modulator; the output port of the master laser is connected to the input port of the optical attenuator, the output port of the optical attenuator is connected to the input port of the slave laser through the first optical fiber, and the output port of the slave laser is connected to the input port of the Mach-Zehnder modulator; the output port of the Mach-Zehnder modulator is connected to the input port of the optical amplifier through the second optical fiber, and the output port of the optical amplifier, the photodetector, the low-pass filter, the analog-to-digital converter, and the intelligent processing module are connected in sequence;

[0009] The high-frequency chirp signal is received by the high-frequency receiving antenna, power-amplified and filtered by the electrical amplifier, and then sent to the Mach-Zehnder modulator to be modulated onto the optical carrier signal; the laser generated by the master laser is power-adjusted and split by the optical attenuator, and then sent to the slave laser of the remote receiving unit through the first optical fiber to generate a single-cycle optical carrier signal, which is output to the Mach-Zehnder modulator. After electro-optic conversion by the Mach-Zehnder modulator, a modulated optical signal is generated and transmitted to the optical amplifier of the central unit through the second optical fiber for optical power amplification, and then sent to the photodetector. The photodetector performs beat frequency on the optical signal to realize down-conversion of the high-frequency chirp signal, and then is sent to the low-pass filter and the analog-to-digital converter in sequence to complete filtering and analog-to-digital conversion of the signal, and finally is sent to the intelligent processing module for signal processing.

[0010] Further, the master laser and the slave laser adopt distributed feedback lasers.

[0011] Further, the master laser of the central unit realizes the generation of the optical carrier signal by injecting the remote light into the slave laser of the remote receiving unit.

[0012] Further, the optical amplifier and the photodetector of the central unit are used to realize local optical domain amplification and optoelectronic conversion of the broadband chirp signal.

[0013] Further, the low-pass filter and the analog-to-digital converter of the central unit are used to realize low-pass filtering and digital conversion of the broadband chirp signal.

[0014] Further, the intelligent processing module of the central unit is used to perform adaptive processing on the low-frequency broadband chirp signal and accurately recover the down-converted broadband signal submerged by noise.

[0015] Further, the signal processing process of the intelligent processing module is specifically as follows:

[0016] Step 1: Perform Fourier transform on the low-frequency broadband chirp signal output by the analog-to-digital converter, and mark N clean samples and N contaminated samples according to the frequency domain suppression ratio;

[0017] Step 2: Perform data transformation on N clean samples and N contaminated samples to obtain time / frequency two-dimensional matrix data;

[0018] Step 3: Group the two-dimensional matrix data without crossover according to 8:1:1, and label them as the training set, validation set, and test set respectively;

[0019] Step 4: Construct a lightweight deep convolutional neural network;

[0020] Step 5: Use the training set to train the model parameters of the lightweight deep convolutional neural network; use the validation set to output the signal frequency domain suppression ratio, tune the lightweight deep convolutional neural network and prevent overfitting; use the test set to output the signal frequency domain suppression ratio to evaluate the adaptability of the lightweight deep convolutional neural network;

[0021] Step 6: Construct a lightweight deep convolutional neural network model, perform data preprocessing on the time domain of the weak signal generated by the analog-to-digital converter, and use the lightweight deep convolutional neural network model to optimize the two-dimensional matrix data;

[0022] Step 7: Complete the inverse transformation through data post-processing and output the optimized signal.

[0023] Furthermore, the construction of the lightweight deep convolutional neural network in Step 4 is as follows:

[0024] The constructed lightweight deep convolutional neural network consists of eight layers:

[0025] The first layer is a convolutional layer + ReLU layer, with a convolutional kernel size of 3×3, and the number of input channels and output channels is 1 and 64, respectively, to achieve two-dimensional data feature extraction;

[0026] The second layer is a convolutional layer + BN layer + ReLU layer to achieve two-dimensional data feature extraction, with a convolutional kernel size of 3×3, and the number of input channels and output channels is 64 and 64, respectively, to further refine the feature expression;

[0027] The third and fourth layers are both convolutional layer + BN layer + ReLU layer, with a convolutional kernel size of 3×3, and the number of input channels and output channels is 128 and 128, respectively, to achieve complex noise capture and processing;

[0028] The fifth, sixth, and seventh layers are all convolutional layer + BN layer + LReLU layer to process the weak parts that are easily submerged by noise, with a convolutional kernel size of 3×3, and the number of input channels and output channels is 256 and 256;

[0029] The eighth layer is a convolutional layer to output two-dimensional matrix data, with a convolutional kernel size of 3×3, and the number of input channels and output channels is 256 and 1.

[0030] A method for intelligent reception of down-conversion of a remote microwave photon broadband chirp signal, which is based on the above-mentioned intelligent reception device for down-conversion of a remote microwave photon broadband chirp signal, and the specific process is as follows:

[0031] The high-frequency chirp signal is received by the antenna, power-amplified and filtered by the electrical amplifier, and then sent to the Mach-Zehnder modulator to be modulated onto the optical carrier signal;

[0032] The laser generated by the master laser is power-adjusted and split by the optical attenuator, and then sent through the first optical fiber to the slave laser in the remote receiving unit to generate a single-cycle optical carrier signal, which is output to the Mach-Zehnder modulator. After electro-optic conversion by the Mach-Zehnder modulator, the modulated optical signal is transmitted through the second optical fiber to the optical amplifier in the central unit for optical power amplification, and then sent to the photodetector;

[0033] The photodetector performs beat frequency on the optical signal to achieve down-conversion of the high-frequency chirp signal, and then sequentially sends it to the low-pass filter and the analog-to-digital converter to complete filtering and analog-to-digital conversion of the signal;

[0034] Finally, it is sent to the intelligent processing module for signal processing.

[0035] Furthermore, the signal processing process of the intelligent processing module is specifically as follows:

[0036] Step 1: Perform Fourier transform on the low-frequency broadband chirp signal output by the analog-to-digital converter, and mark N clean samples and N contaminated samples according to the frequency domain suppression ratio;

[0037] Step 2: Perform data transformation on the N clean samples and N contaminated samples to obtain two-dimensional matrix data in the time / frequency two-dimensional dimension;

[0038] Step 3: Group the two-dimensional matrix data without intersection according to 8:1:1, and mark them as the training set, the validation set, and the test set respectively;

[0039] Step 4: Construct a lightweight deep convolutional neural network, which consists of eight layers in total:

[0040] The first layer is a convolutional layer + ReLU layer, the convolutional kernel size is 3×3, the number of input channels and output channels is 1 and 64, respectively, to realize two-dimensional data feature extraction;

[0041] The second layer is a convolutional layer + BN layer + ReLU layer to realize two-dimensional data feature extraction, the convolutional kernel size is 3×3, the number of input channels and output channels is 64 and 64, respectively, to further refine the feature expression;

[0042] The third layer and the fourth layer are both convolutional layer + BN layer + ReLU layer, the convolutional kernel size is 3×3, the number of input channels and output channels is 128 and 128, respectively, to realize complex noise capture and processing;

[0043] The fifth, sixth, and seventh layers are all convolutional layer + BN layer + LReLU layer to process the weak parts that are easily submerged by noise. The convolution kernel size is 3×3, and the number of input channels and output channels is 256 and 256;

[0044] The eighth layer is a convolutional layer that outputs two-dimensional matrix data. The convolution kernel size is 3×3, and the number of input channels and output channels is 256 and 1;

[0045] Step 5: Use the training set to train the model parameters of the lightweight deep convolutional neural network; use the validation set to output the signal frequency domain suppression ratio, optimize the lightweight deep convolutional neural network, and prevent overfitting; use the test set to output the signal frequency domain suppression ratio to evaluate the adaptability of the lightweight deep convolutional neural network;

[0046] Step 6: Build a lightweight deep convolutional neural network model to preprocess the data in the time domain of the weak signal generated by the analog-to-digital converter, and use the lightweight deep convolutional neural network model to optimize the two-dimensional matrix data;

[0047] Step 7: Complete the inverse transformation through post-data processing and output the optimized signal.

[0048] Compared with the prior art, the significant advantages of the present invention are: (1) The present device realizes the generation of single-cycle state through the remote optical injection method, without the need for a high-precision high-frequency local oscillator, reducing the demand for precision electrical equipment; (3) In the remote receiving unit, the high-frequency chirped signal is modulated into an optical signal and then transmitted to the central unit through an optical fiber. The down-conversion is completed through the optical amplifier, the photodetector, and the low-pass filter. The architecture of the lightweight distributed receiving unit and the centralized central unit reduces the maintenance difficulty of the device; (3) The weak signal receiving ability of the device is improved through the intelligent processing module of the central unit, reducing the use of high-precision equipment by the device in the way of software compensating for hardware, reducing the device complexity, and improving the device sensitivity. Description of the Drawings

[0049] Figure 1 It is a schematic structural diagram of a remote microwave photon broadband chirped signal down-conversion intelligent receiving device of the present invention.

[0050] Figure 2 It is a frequency domain diagram of a high-frequency linear chirped signal after electrical amplification provided in an embodiment of the present invention.

[0051] Figure 3 It is a spectrogram of a high-frequency signal modulated by an optical carrier provided in an embodiment of the present invention.

[0052] Figure 4 It is a frequency domain diagram of a low-noise high-frequency signal output by a photodetector provided in an embodiment of the present invention.

[0053] Figure 5 This is the frequency-domain diagram of the down-converted signal with high noise provided in the embodiment of the present invention.

[0054] Figure 6 This is the time-frequency diagram of the down-converted signal with high noise provided in the embodiment of the present invention.

[0055] Figure 7 This is the time-frequency diagram of the restored down-converted signal provided in the embodiment of the present invention.

[0056] Figure 8 This is the frequency-domain diagram of the restored down-converted signal provided in the embodiment of the present invention.

[0057] In the figure: 1. High-frequency receiving antenna; 2. Electrical amplifier; 3. Mach-Zehnder modulator; 4. Slave laser; 5. Master laser; 6. Optical attenuator; 7. First optical fiber; 8. Second optical fiber; 9. Optical amplifier; 10. Photoelectric detector; 11. Low-pass filter; 12. Analog-to-digital converter; 13. Intelligent processing module. Detailed implementation manners

[0058] The following further elaborates on the present invention in detail in conjunction with the accompanying drawings and specific embodiments.

[0059] Combined with Figure 1 , an intelligent receiving device for remote microwave photon broadband chirped signal down-conversion according to the present invention includes a remote receiving unit, a central unit, a first optical fiber 7, and a second optical fiber 8. The remote receiving unit includes a high-frequency receiving antenna 1, an electrical amplifier 2, a Mach-Zehnder modulator 3, and a slave laser 4; the central unit includes a master laser 5, an optical attenuator 6, an optical amplifier 9, a photoelectric detector 10, a low-pass filter 11, an analog-to-digital converter 12, and an intelligent processing module 13;

[0060] The central unit is connected to the remote receiving unit by the first optical fiber 7, and the remote receiving unit is connected to the central unit by the second optical fiber 8;

[0061] The output port of the high-frequency receiving antenna 1 is connected to the input port of the electrical amplifier 2, and the output port of the electrical amplifier 2 is connected to the modulation port of the Mach-Zehnder modulator 3; the output port of the master laser 5 is connected to the input port of the optical attenuator 6, the output port of the optical attenuator 6 is connected to the input port of the slave laser 4 through the first optical fiber 7, and the output port of the slave laser 4 is connected to the input port of the Mach-Zehnder modulator 3; the output port of the Mach-Zehnder modulator 3 is connected to the input port of the optical amplifier 9 through the second optical fiber 8, and the output port of the optical amplifier 9, the photoelectric detector 10, the low-pass filter 11, the analog-to-digital converter 12, and the intelligent processing module 13 are connected in sequence;

[0062] The high-frequency chirp signal is received by antenna 1, power-amplified and filtered by an electrical amplifier 2, and then sent to a Mach-Zehnder modulator 3 to be modulated onto an optical carrier signal. The laser generated by the master laser 5 is power-adjusted and split by an optical attenuator 6, and then sent through a first optical fiber 7 to a slave laser 4 in a remote receiving unit to generate an optical carrier signal in a single-cycle state, which is output to the Mach-Zehnder modulator 3. After electro-optic conversion by the Mach-Zehnder modulator 3, a modulated optical signal is generated, which is transmitted through a second optical fiber 8 to an optical amplifier 9 in the central unit for optical power amplification, and then sent to a photodetector 10. The photodetector 10 performs beat frequency on the optical signal to realize down-conversion of the high-frequency chirp signal, and then is successively sent to a low-pass filter 11 and an analog-to-digital converter 12 to complete filtering and analog-to-digital conversion of the signal, and finally sent to an intelligent processing module 13 for processing.

[0063] As a specific example, the master laser 5 and the slave laser 4 adopt distributed feedback lasers.

[0064] As a specific example, the master laser 5 in the central unit realizes the generation of the optical carrier signal by injecting remote light into the slave laser 4 in the remote receiving unit.

[0065] As a specific example, the optical amplifier 9 and the photodetector 10 in the central unit are used to realize local optical domain amplification and optoelectronic conversion of broadband chirp signals.

[0066] As a specific example, the low-pass filter 11 and the analog-to-digital converter 12 in the central unit are used to realize low-pass filtering and digital conversion of broadband chirp signals.

[0067] As a specific example, the intelligent processing module 13 in the central unit is used to perform adaptive processing on the low-frequency broadband chirp signal and accurately restore the down-converted broadband signal submerged by noise.

[0068] As a specific example, the signal processing process of the intelligent processing module 13 is specifically as follows:

[0069] Step 1: Perform Fourier transform on the low-frequency broadband chirp signal output by the analog-to-digital converter 12, and mark N clean samples and N contaminated samples according to the frequency domain suppression ratio;

[0070] Step 2: Perform data transformation on the N clean samples and N contaminated samples to obtain time / frequency two-dimensional matrix data;

[0071] Step 3: Group the two-dimensional matrix data without crossover according to 8:1:1, and mark them as a training set, a validation set, and a test set respectively;

[0072] Step 4: Construct a lightweight deep convolutional neural network;

[0073] Step 5: Use the training set to train the model parameters of the lightweight deep convolutional neural network; use the validation set to output the signal frequency domain suppression ratio, optimize the lightweight deep convolutional neural network and prevent overfitting; use the test set to output the signal frequency domain suppression ratio to evaluate the adaptability of the lightweight deep convolutional neural network;

[0074] Step 6: Construct a lightweight deep convolutional neural network model to perform data preprocessing on the time domain of the weak signal generated by the analog-to-digital converter 12, and use the lightweight deep convolutional neural network model to optimize the two-dimensional matrix data;

[0075] Step 7: Complete the inverse transformation through data post-processing and output the optimized signal.

[0076] As a specific example, the construction of the lightweight deep convolutional neural network described in step 4 is as follows:

[0077] The constructed lightweight deep convolutional neural network consists of eight layers:

[0078] The first layer is a convolutional layer + ReLU layer, the convolutional kernel size is 3×3, and the number of input channels and output channels is 1 and 64, respectively, to achieve two-dimensional data feature extraction;

[0079] The second layer is a convolutional layer + BN layer + ReLU layer to achieve two-dimensional data feature extraction, the convolutional kernel size is 3×3, and the number of input channels and output channels is 64 and 64, respectively, to further refine the feature expression;

[0080] The third and fourth layers are both convolutional layers + BN layers + ReLU layers, the convolutional kernel size is 3×3, and the number of input channels and output channels is 128 and 128, respectively, to achieve complex noise capture and processing;

[0081] The fifth, sixth, and seventh layers are all convolutional layers + BN layers + LReLU layers to process the weak parts that are easily submerged by noise, the convolutional kernel size is 3×3, and the number of input channels and output channels is 256 and 256;

[0082] The eighth layer is a convolutional layer to output two-dimensional matrix data, the convolutional kernel size is 3×3, and the number of input channels and output channels is 256 and 1.

[0083] The present invention also provides a remote microwave photon broadband chirp signal down-conversion intelligent receiving method, which is based on the remote microwave photon broadband chirp signal down-conversion intelligent receiving device, and the specific process is as follows:

[0084] The high-frequency chirp signal is received by antenna 1, amplified in power and filtered by the electrical amplifier 2, and then sent to the Mach-Zehnder modulator 3 to be modulated onto the optical carrier signal;

[0085] The laser generated by the master laser 5 is subjected to power adjustment and beam splitting by the optical attenuator 6, and then enters the slave laser 4 of the remote receiving unit through the first optical fiber 7 to generate an optical carrier signal in a single-cycle state, which is output to the Mach-Zehnder modulator 3. After electro-optic conversion by the Mach-Zehnder modulator 3, a modulated optical signal is generated and transmitted to the optical amplifier 9 of the central unit through the second optical fiber 8 for optical power amplification, and then delivered to the photodetector 10.

[0086] The photodetector 10 performs beat frequency on the optical signal to realize down-conversion of the high-frequency chirp signal, and then sequentially delivers it to the low-pass filter 11 and the analog-to-digital converter 12 to complete filtering and analog-to-digital conversion of the signal.

[0087] Finally, it is delivered to the intelligent processing module 13 for signal processing.

[0088] As a specific example, the signal processing process of the intelligent processing module 13 is as follows:

[0089] Step 1: Perform Fourier transform on the low-frequency broadband chirp signal output by the analog-to-digital converter 12, and mark N clean samples and N contaminated samples according to the frequency domain suppression ratio.

[0090] Step 2: Perform data transformation on N clean samples and N contaminated samples to obtain time / frequency two-dimensional matrix data.

[0091] Step 3: Group the two-dimensional matrix data without crossover according to 8:1:1, and mark them as the training set, validation set, and test set respectively.

[0092] Step 4: Construct a lightweight deep convolutional neural network, which consists of eight layers in total:

[0093] The first layer is a convolutional layer + ReLU layer, with a convolution kernel size of 3×3, and the number of input channels and output channels is 1 and 64, respectively, to realize two-dimensional data feature extraction.

[0094] The second layer is a convolutional layer + BN layer + ReLU layer to realize two-dimensional data feature extraction, with a convolution kernel size of 3×3, and the number of input channels and output channels is 64 and 64, respectively, to further refine the feature expression.

[0095] The third layer and the fourth layer are both convolutional layers + BN layers + ReLU layers, with a convolution kernel size of 3×3, and the number of input channels and output channels is 128 and 128, respectively, to realize complex noise capture and processing.

[0096] The fifth layer, the sixth layer, and the seventh layer are all convolutional layers + BN layers + LReLU layers to process the weak parts that are easily submerged by noise, with a convolution kernel size of 3×3, and the number of input channels and output channels is 256 and 256, respectively.

[0097] The eighth layer is the output two-dimensional matrix data of the convolutional layer. The size of the convolutional kernel is 3×3, and the number of input channels and output channels is 256 and 1 respectively.

[0098] Step 5: Use the training set to train the model parameters of the lightweight deep convolutional neural network; use the validation set to output the signal frequency domain suppression ratio, optimize the lightweight deep convolutional neural network and prevent overfitting; use the test set to output the signal frequency domain suppression ratio to evaluate the adaptability of the lightweight deep convolutional neural network.

[0099] Step 6: Build a lightweight deep convolutional neural network model to perform data preprocessing on the weak signal time domain generated by the analog-to-digital converter 12, and use the lightweight deep convolutional neural network model to optimize the two-dimensional matrix data.

[0100] Step 7: Complete the inverse transformation through data post-processing and output the optimized signal.

[0101] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0102] Embodiment

[0103] This embodiment provides a process in which a linear chirp signal with a center frequency of 29 GHz and a bandwidth of 2 GHz is received and signal-amplified by the high-frequency receiving antennas 1 and the 20 dB electrical amplifier 2. Figure 2 It is the frequency domain diagram of the high-frequency linear chirp signal after electrical amplification.

[0104] The master laser 5 generates a laser with a frequency of 193.13 THz and a power of 17 dBm. After passing through the adjustable optical attenuator 6 with a connection loss of 0.4 dB, it is transmitted through the first optical fiber 7 with a length of 2.6 km to the slave laser 4 with a frequency of 193.1 THz and a power of 9.3 dBm. Since the laser injection-locked laser system can enter the single-cycle state working model, the slave laser 4 is connected to the Mach-Zehnder modulator 3. The linear chirp signal with a center frequency of 29 GHz and a bandwidth of 2 GHz is effectively modulated by the Mach-Zehnder modulator 3. Refer to Figure 3 It is the spectrogram of the high-frequency signal modulated by the optical carrier provided by this embodiment.

[0105] The modulated optical signal is connected to the optical amplifier 9 with a gain of 20 dB through the second optical fiber 8 with a length of 2.6 km, and is photoelectrically converted by the PIN photodetector 10 to output an analog electrical signal. Refer to Figure 4In this embodiment, the low-noise high-frequency signal is subjected to beat frequency on the optical signal by the photodetector 10 to realize the down-conversion of the high-frequency chirp signal, and then is successively sent to the low-pass filter 11 and the analog-to-digital converter 12 to complete the filtering and analog-to-digital conversion of the signal. Due to the push-pull effect, the linear chirp signal with a single-cycle operating frequency point of 27.55 GHz, a center frequency of 29 GHz, and a bandwidth of 2 GHz is converted into a linear chirp signal with a frequency of 1.45 GHz and a bandwidth of 2 GHz, and finally sent to the intelligent processing module 13 for processing.

[0106] The signal processing process of the intelligent processing module 13 is specifically as follows:

[0107] Step 1: Perform Fourier transform on the low-frequency wideband chirp signal output by the analog-to-digital converter 12, and mark 500 clean samples and 500 contaminated samples according to the frequency domain suppression ratio. Figure 5 The frequency domain diagram of the high-noise down-converted signal provided for this embodiment;

[0108] Step 2: After performing data transformation on 500 clean samples and 500 contaminated samples, obtain time / frequency two-dimensional matrix data in the two dimensions. Figure 6 The time-frequency diagram of the high-noise down-converted signal provided for this embodiment;

[0109] Step 3: Group the two-dimensional matrix data without intersection according to 8:1:1, and mark them as the training set, the validation set, and the test set respectively.

[0110] Step 4: Construct a lightweight deep convolutional neural network, and the structure is successively: the first layer is a convolutional layer + ReLU layer, the convolution kernel size is 3×3, the number of input channels and output channels is 1 and 64, to realize the extraction of two-dimensional data features; the second layer is a convolutional layer + BN layer + ReLU layer to realize the extraction of two-dimensional data features, the convolution kernel size is 3×3, the number of input channels and output channels is 64 and 64, to further refine the feature expression; the third layer and the fourth layer are both convolutional layer + BN layer + ReLU layer, the convolution kernel size is 3×3, the number of input channels and output channels is 128 and 128, to realize the capture and processing of complex noise; the fifth layer, the sixth layer, and the seventh layer are all convolutional layer + BN layer + LReLU layer to process the weak parts that are easily submerged by noise, the convolution kernel size is 3×3, the number of input channels and output channels is 256 and 256; the eighth layer is a convolutional layer to output two-dimensional matrix data, the convolution kernel size is 3×3, the number of input channels and output channels is 256 and 1;

[0111] Step 5: Use the training set to train the model parameters of the lightweight deep convolutional neural network. Adopt the adaptive moment estimation optimization method with a learning rate of 0.001 and iterate 500 times to update the parameters of each layer of the convolutional neural network. Use the validation set to output the signal frequency domain suppression ratio to optimize the lightweight deep convolutional neural network and prevent overfitting. Use the test set to output the signal frequency domain suppression ratio to evaluate the adaptability of the lightweight deep convolutional neural network.

[0112] Step 6: Construct a lightweight deep convolutional neural network model to preprocess the time domain data of the weak signal generated by the analog-to-digital converter 12. Use the lightweight deep convolutional neural network model to optimize the two-dimensional matrix data, and finally complete the inverse transformation through data post-processing to output the optimized signal. Figure 7 This is the time-frequency diagram of the down-converted signal after recovery provided by this embodiment. Compare Figure 6 and Figure 7 , and the useful signal is recovered within the same power range for the high-noise down-converted time-frequency signal.

[0113] Step 7: Complete the inverse transformation through data post-processing, and the output frequency domain suppression ratio of the optimized signal is 16.5 dB. Figure 8 This is the frequency domain diagram of the down-converted signal after recovery provided by this embodiment.

[0114] In this embodiment, the remote optical injection master laser 5, the optical amplifier 9, the photodetector 10, and the low-pass filter 11 are used to realize the direct down-conversion of the broadband signal. The architecture of the lightweight distributed receiving unit and the centralized central unit reduces the maintenance difficulty of the device. The intelligent processing module of the lightweight deep convolutional neural network adaptively processes the broadband signal, refines the feature expression through the stepped channel number, and processes the weak signal submerged by noise, reducing the use of high-precision equipment by the device in the way of software compensating for hardware, and accurately recovering the down-converted broadband signal.

[0115] Obviously, the above embodiments are only examples clearly described and not limitations on the implementation manners. For those of ordinary skill in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation manners here. And the obvious changes or modifications derived therefrom are still within the protection scope of the present invention.

Claims

1. A remote microwave photon broadband chirped signal down-conversion intelligent receiving device, characterized in that, It includes a remote receiving unit and a central unit. The central unit is connected to the remote receiving unit by a first optical fiber (7), and the remote receiving unit is connected to the central unit by a second optical fiber (8). The remote receiving unit includes a high-frequency receiving antenna (1), an electrical amplifier (2), a Mach-Zehnder modulator (3), and a slave laser (4); the central unit includes a master laser (5), an optical attenuator (6), an optical amplifier (9), a photodetector (10), a low-pass filter (11), an analog-to-digital converter (12), and an intelligent processing module (13). The output port of the high-frequency receiving antenna (1) is connected to the input port of the electrical amplifier (2), and the output port of the electrical amplifier (2) is connected to the modulation port of the Mach-Zehnder modulator (3); the output port of the master laser (5) is connected to the input port of the optical attenuator (6), the output port of the optical attenuator (6) is connected to the input port of the slave laser (4) in the remote receiving unit through the first optical fiber (7), and the output port of the slave laser (4) is connected to the input port of the Mach-Zehnder modulator (3); the output port of the Mach-Zehnder modulator (3) is connected to the input port of the optical amplifier (9) through the second optical fiber (8), and the output port of the optical amplifier (9), the photodetector (10), the low-pass filter (11), the analog-to-digital converter (12), and the intelligent processing module (13) are connected in sequence. The high-frequency chirp signal is received by the high-frequency receiving antenna (1), power-amplified and filtered by the electrical amplifier (2), and then sent to the Mach-Zehnder modulator (3) to be modulated onto the optical carrier signal; the laser generated by the master laser (5) is power-adjusted and split by the optical attenuator (6), and then sent through the first optical fiber (7) to the slave laser (4) in the remote receiving unit to generate a single-cycle optical carrier signal, which is output to the Mach-Zehnder modulator (3). After electro-optic conversion by the Mach-Zehnder modulator (3), a modulated optical signal is generated and transmitted through the second optical fiber (8) to the optical amplifier (9) in the central unit for optical power amplification, and then sent to the photodetector (10). The photodetector (10) performs beat frequency on the optical signal to realize the down-conversion of the high-frequency chirp signal, and then is sent to the low-pass filter (11) and the analog-to-digital converter (12) in sequence to complete filtering and analog-to-digital conversion of the signal, and finally sent to the intelligent processing module (13) for signal processing.

2. The remote microwave photon broadband chirped signal down-conversion intelligent receiving device according to claim 1, wherein The master laser (5) and the slave laser (4) adopt distributed feedback lasers.

3. The remote microwave photon broadband chirped signal down-conversion intelligent receiving device according to claim 1, wherein The master laser (5) in the central unit realizes the generation of the optical carrier signal by injecting remote light into the slave laser (4) in the remote receiving unit.

4. The remote microwave photon broadband chirped signal down-conversion intelligent receiving device according to claim 1, wherein The optical amplifier (9) and the photodetector (10) in the central unit are used to realize local optical domain amplification and optoelectronic conversion of the broadband chirp signal.

5. The remote microwave photon broadband chirped signal down-conversion intelligent receiving device according to claim 1, characterized in that, The low-pass filter (11) and the analog-to-digital converter (12) in the central unit are used to realize low-pass filtering and digital conversion of the broadband chirp signal.

6. The remote microwave photon broadband chirped signal down-conversion intelligent receiving device according to claim 1, characterized in that, The intelligent processing module (13) in the central unit is used to perform adaptive processing on the low-frequency broadband chirp signal and accurately recover the down-converted broadband signal submerged by noise.

7. The remote microwave photon broadband chirped signal down-conversion intelligent receiving device according to claim 1, characterized in that, The signal processing process of the intelligent processing module (13) is specifically as follows: Step 1: Perform Fourier transform on the low-frequency wideband chirp signal output by the analog-to-digital converter (12), and mark N clean samples and N contaminated samples according to the frequency domain suppression ratio; Step 2: Perform data transformation on the N clean samples and N contaminated samples to obtain time / frequency two-dimensional matrix data; Step 3: Group the two-dimensional matrix data without crossover according to 8:1:1, and mark them as the training set, validation set, and test set respectively; Step 4: Construct a lightweight deep convolutional neural network; Step 5: Use the training set to train the model parameters of the lightweight deep convolutional neural network; use the validation set to output the signal frequency domain suppression ratio to optimize the lightweight deep convolutional neural network and prevent overfitting; use the test set to output the signal frequency domain suppression ratio to evaluate the adaptability of the lightweight deep convolutional neural network; Step 6: Construct a lightweight deep convolutional neural network model, perform data preprocessing on the time domain of the weak signal generated by the analog-to-digital converter (12), and use the lightweight deep convolutional neural network model to optimize the two-dimensional matrix data; Step 7: Complete the inverse transformation through data post-processing and output the optimized signal.

8. The remote microwave photon broadband chirped signal down-conversion intelligent receiving device according to claim 7, characterized in that The construction of the lightweight deep convolutional neural network in Step 4 is specifically as follows: Construct a lightweight deep convolutional neural network, which consists of eight layers in total: The first layer is a convolutional layer + ReLU layer, the convolutional kernel size is 3×3, the number of input channels and output channels is 1 and 64, respectively, to achieve two-dimensional data feature extraction; The second layer is a convolutional layer + BN layer + ReLU layer to achieve two-dimensional data feature extraction, the convolutional kernel size is 3×3, the number of input channels and output channels is 64 and 64, respectively, to further refine the feature expression; The third layer and the fourth layer are both convolutional layers + BN layers + ReLU layers, the convolutional kernel size is 3×3, the number of input channels and output channels is 128 and 128, respectively, to achieve complex noise capture and processing; The fifth layer, the sixth layer, and the seventh layer are all convolutional layers + BN layers + LReLU layers to process the weak parts that are easily submerged by noise, the convolutional kernel size is 3×3, the number of input channels and output channels is 256 and 256, respectively; The eighth layer is a convolutional layer to output two-dimensional matrix data, the convolutional kernel size is 3×3, the number of input channels and output channels is 256 and 1, respectively.

9. A down-conversion intelligent receiving method for a remote microwave photon broadband chirp signal, characterized in that, This method is based on the remote microwave photon broadband chirp signal down-conversion intelligent receiving device described in any one of claims 1 to 8, and the specific process is as follows: The high-frequency chirp signal is received by the antenna (1), and after being power-amplified and filtered by the electrical amplifier (2), it is sent to the Mach-Zehnder modulator (3) to be modulated onto the optical carrier signal; The laser generated by the master laser (5) is power-adjusted and split by the optical attenuator (6), and then sent through the first optical fiber (7) to the slave laser (4) of the remote receiving unit to generate a single-cycle state optical carrier signal, which is output to the Mach-Zehnder modulator (3). After electro-optic conversion by the Mach-Zehnder modulator (3), a modulated optical signal is generated and transmitted through the second optical fiber (8) to the optical amplifier (9) of the central unit for optical power amplification, and then delivered to the photodetector (10); The photodetector (10) performs beat frequency on the optical signal to realize high-frequency chirp signal down-conversion, and then successively sends it to the low-pass filter (11) and the analog-to-digital converter (12) to complete filtering and analog-to-digital conversion of the signal; Finally, it is sent to the intelligent processing module (13) for signal processing.

10. The down-conversion intelligent receiving method for remote microwave photon broadband chirp signals according to claim 9, characterized in that, The signal processing process of the intelligent processing module (13) is specifically as follows: Step 1: Perform Fourier transform on the low-frequency wideband chirp signal output by the analog-to-digital converter (12), and mark N clean samples and N contaminated samples according to the frequency domain suppression ratio; Step 2: Perform data transformation on N clean samples and N contaminated samples to obtain time / frequency two-dimensional two-dimensional matrix data; Step 3: Group the two-dimensional matrix data without intersection according to 8:1:1, and mark them as the training set, the validation set, and the test set respectively; Step 4: Construct a lightweight deep convolutional neural network, which has a total of eight layers: The first layer is a convolutional layer + ReLU layer, the convolution kernel size is 3×3, and the number of input channels and output channels is 1 and 64, respectively, to realize two-dimensional data feature extraction; The second layer is a convolutional layer + BN layer + ReLU layer to realize two-dimensional data feature extraction, the convolution kernel size is 3×3, and the number of input channels and output channels is 64 and 64, respectively, to further refine the feature expression; The third layer and the fourth layer are both convolutional layer + BN layer + ReLU layer, the convolution kernel size is 3×3, and the number of input channels and output channels is 128 and 128, respectively, to realize complex noise capture and processing; The fifth layer, the sixth layer, and the seventh layer are all convolutional layer + BN layer + LReLU layer to process the weak parts that are easily submerged by noise, the convolution kernel size is 3×3, and the number of input channels and output channels is 256 and 256; The eighth layer is a convolutional layer to output two-dimensional matrix data, the convolution kernel size is 3×3, and the number of input channels and output channels is 256 and 1; Step 5: Use the training set to train the model parameters of the lightweight deep convolutional neural network; use the validation set to output the signal frequency domain suppression ratio to optimize the lightweight deep convolutional neural network and prevent overfitting; use the test set to output the signal frequency domain suppression ratio to evaluate the adaptability of the lightweight deep convolutional neural network; Step 6: Construct a lightweight deep convolutional neural network model, perform data preprocessing on the time domain of the weak signal generated by the analog-to-digital converter (12), and use the lightweight deep convolutional neural network model to optimize the two-dimensional matrix data; Step 7: Complete the inverse transformation through data post-processing and output the optimized signal.

Images