High-sensitivity magnetic signal receiving method and device based on spin exchange relaxation free (SERF) atomic spin effect

Abstract

This invention discloses a high-sensitivity magnetic signal receiving method and apparatus based on the SERF atomic spin effect, comprising the following steps: S1, establishing a SERF polarization system and acquiring an optical response signal; S2, converting the optical response signal into a linearly enhanced voltage signal; S3, performing interference suppression and frequency band extraction operations on the voltage signal to output a high signal-to-noise ratio (SNR) signal; S4, performing nonlinear amplitude modulation on the high SNR signal to output a stable signal stream; S5, converting the stable signal stream into a time-frequency feature map, inputting it into a convolutional neural network to extract frequency features, and generating a modulation signal; S6, mixing the modulation signal with two reference signals respectively to recover the binary data stream; S7, performing interference correction and bit error control on the binary data stream and outputting the communication result. This invention fully integrates magnetic sensing, signal processing, and intelligent demodulation technologies, achieving stable reception and high-precision reconstruction under extremely weak magnetic signals.

Description

Technical Field

[0001] This invention relates to the field of magnetic signal detection technology, and in particular to a high-sensitivity magnetic signal receiving method and apparatus based on the SERF atomic spin effect. Background Technology

[0002] With the increasing demand for electromagnetic communication in complex environments, traditional communication methods based on radio wave propagation have become severely limited in underwater, underground, and shielded areas. Magnetic induction communication, as a short-range communication method that does not require an ionizing propagation medium, has strong penetration, and is less affected by environmental factors, has received widespread attention in recent years.

[0003] Atomic magnetometers, especially those based on the SERF effect, are gradually becoming an important sensing technology for detecting weak magnetic signals due to their femtotes-level magnetic field detection sensitivity. SERF magnetometers achieve high-precision sensing of changes in extremely weak magnetic fields by using laser pumping to create atomic spin polarization and utilizing the spin precession phenomenon induced by an external magnetic field.

[0004] In existing technologies, directly using a SERF magnetometer for receiving communication signals will face the following significant problems:

[0005] First, the SERF magnetometer has physical characteristics such as narrow bandwidth and limited dynamic range, making it difficult to directly adapt to the complex changes in amplitude and frequency of communication signals. It is prone to signal saturation or distortion, which limits the actual communication performance.

[0006] Secondly, changes in the ambient magnetic field can severely affect the zero-field operating state of the SERF magnetometer, leading to signal drift or increased noise, making it difficult to maintain stable communication demodulation accuracy.

[0007] In addition, the raw optical signal output by the SERF magnetometer contains nonlinear response and background noise, and its waveform characteristics are fundamentally different from those of conventional carrier signals. Traditional demodulation methods are difficult to apply directly, and there is an urgent need to combine advanced signal processing algorithms, machine learning techniques and gain control mechanisms in system design.

[0008] Therefore, how to provide a high-sensitivity magnetic signal receiving method and device based on the SERF atomic spin effect is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0009] One objective of this invention is to propose a high-sensitivity magnetic signal receiving method and device based on the SERF atomic spin effect. This invention fully utilizes high-precision magnetic field sensing technology, nonlinear signal modulation and demodulation methods, convolutional neural network frequency feature extraction mechanisms, and adaptive filtering and gain control algorithms. It describes in detail the complete processing flow for achieving stable reception and accurate recovery of communication signals under extremely weak magnetic signal conditions, and has the advantages of high sensitivity, strong anti-interference ability, high signal recovery accuracy, and strong adaptability to complex environments.

[0010] The high-sensitivity magnetic signal receiving method based on the SERF atomic spin effect according to embodiments of the present invention includes the following steps:

[0011] S1. Establish a SERF polarization system and collect the optical response signal formed under the action of a low-frequency magnetic field signal;

[0012] S2. The analytic envelope of the optical response signal is constructed using Hilbert transform. The analytic envelope is then input into a least-squares support vector machine to fit the phase mapping, and the output voltage signal is linearly enhanced.

[0013] S3. Input the voltage signal into the dual-channel FIR filter to perform interference suppression and frequency band extraction operations respectively. Based on the error between the two outputs and the preset desired signal, update the filter coefficients through the NLMS algorithm and superimpose the two outputs to generate a high signal-to-noise ratio signal.

[0014] S4. Extract the root mean square amplitude of the high signal-to-noise ratio signal within the sliding window, compare it with the preset reference value to determine the amplitude modulation coefficient, and use the automatic gain control method to perform nonlinear amplitude modulation on the signal to output a stable signal stream.

[0015] S5. The stationary signal stream is converted into a time-frequency feature map by short-time Fourier transform, and then input into a convolutional neural network to extract frequency features. Based on the frequency features, the main frequency component of the stationary signal stream is separated to generate a modulation signal.

[0016] S6. Mix the modulation signal with the two reference signals respectively, perform low-pass filtering and definite integration on the mixing result, and recover the binary data stream based on the reference signal corresponding to the larger integral.

[0017] S7. Perform interference correction and error control on the binary data stream, and output the communication result.

[0018] Optionally, the SERF polarization system uses an atomic chamber filled with alkali metal atoms and buffer gas within a magnetically shielded structure. A heating device is used to maintain the temperature of the chamber within the operating range of a spin-free exchange relaxation state. Circularly polarized pump light and linearly polarized probe light are used to irradiate the chamber to form a stable spin polarization direction. A compensating magnetic field is applied through a triaxial coil array to lock the operating point at a zero magnetic field position. The photoelectric detection unit performs photoelectric conversion on the polarization change of the transmitted light and outputs an optical response signal.

[0019] Optionally, S2 specifically includes:

[0020] S21. Perform a Hilbert transform operation on the optical response signal to construct a complex analytic signal composed of a real part signal and a Hilbert imaginary part signal, and maintain a structure consistent with the optical response signal in the time dimension. The real part signal represents the optical response signal, and the Hilbert imaginary part signal represents the orthogonal component obtained after performing a 90-degree phase shift on the optical response signal.

[0021] S22. Calculate the analytic envelope obtained by summing the squares of the real and imaginary parts and taking the square root based on the complex analytic signal, and perform moving average processing.

[0022] S23. Amplitude sampling is performed on the analytical envelope to construct multiple signal segments of equal length. Each signal segment consists of a set number of amplitude sampling points to form a sample set. The least squares support vector machine algorithm is called. With the sample set as input, the mapping relationship between the analytical envelope and the phase shift is established by solving the least squares form of the mapping coefficients. The phase shift represents the instantaneous phase change caused by the external magnetic field in the atomic spin response trajectory corresponding to the analytical envelope. The atomic spin response trajectory refers to the dynamic response process formed by the change of the net spin polarization direction of alkali metal atoms with time under the action of an external magnetic field.

[0023] S24. Based on the established mapping relationship, input the analytical envelope into the least squares support vector machine algorithm in time order, and output the fitted phase sequence;

[0024] S25. Perform linear amplitude expansion processing on the fitted phase mapping sequence, and generate a linearly enhanced voltage signal based on the amplitude expansion result.

[0025] Optionally, S3 specifically includes:

[0026] S31. Input the voltage signal into the dual-channel FIR filter and perform finite impulse response filtering operations in the interference path and the frequency band path respectively to obtain the interference suppression signal and the frequency band extraction signal.

[0027] S32. Compare the interference suppression signal and the frequency band extraction signal with the preset desired signal point by point to generate the interference error sequence and the frequency band error sequence.

[0028] S33. Based on two error sequences, the weight increment at each time point in the corresponding path is calculated using the NLMS algorithm to generate an interference coefficient sequence and a frequency band coefficient sequence. The two coefficient sequences are then input into the corresponding paths in the dual-channel FIR filter. Combined with the preset amplitude boundary constraints, the weight parameters of the two paths are dynamically adjusted.

[0029] S34. Perform amplitude superposition operation on the interference suppression signal and the frequency band extraction signal after weight adjustment to output a high signal-to-noise ratio signal.

[0030] Optionally, S33 specifically includes:

[0031] S331. Set a sliding window of fixed length in the interference error sequence and the frequency band error sequence, and extract the sequence segments in the sliding window according to the time point to obtain the interference error segment and the frequency band error segment.

[0032] S332. Calculate the difference between the maximum and minimum amplitude of the two error segments respectively to form the interference fluctuation index and the frequency band fluctuation index.

[0033] S333. Compare the interference fluctuation index and frequency band fluctuation index with the preset fluctuation threshold. If the threshold is exceeded, the step size parameter of the corresponding path is lowered. If the threshold is lower, the step size parameter is raised. The step size parameter represents the adjustment ratio used to update the filter weight.

[0034] S334. Based on the adjusted step size parameters of each path, the NLMS algorithm is used to calculate the weight increment of the interference error sequence and the frequency band error sequence to form the interference coefficient sequence and the frequency band coefficient sequence.

[0035] S335. Compare the interference coefficient sequence and the frequency band coefficient sequence with the preset amplitude boundary of the corresponding path. If the amplitude exceeds the boundary, compress the amplitude linearly to the boundary limit in the original sign direction. Input the compression result into the corresponding path of the dual-channel FIR filter and add it to the original weight parameters at the time point to dynamically update the weight parameters.

[0036] Optionally, S4 specifically includes:

[0037] S41. Construct a sliding window with a fixed length for a high signal-to-noise ratio signal, set the window displacement step size, and make the window move continuously in the time series. Extract amplitude points in each window in time order to form a window amplitude set.

[0038] S42. Calculate the root mean square magnitude in the window magnitude set by summation of squares, and perform Min-Max normalization on the root mean square result based on the set maximum and minimum magnitudes to generate a root mean square magnitude sequence, while keeping the time index of the root mean square magnitude sequence consistent with the starting position of the sliding window.

[0039] S43. Compare the mean square amplitude sequence point by point with the preset reference value, and construct the amplitude adjustment coefficient sequence based on the comparison results;

[0040] S44. Perform point-by-point amplitude scaling on the high signal-to-noise ratio signal according to the amplitude modulation coefficient sequence. During the scaling process, perform linear interpolation on the amplitude modulation coefficient at each time point to generate an amplitude modulation signal sequence.

[0041] S45. Perform automatic gain control on the amplitude-modulated signal sequence, calculate the gain adjustment amount according to the amplitude reference trajectory, and calculate the hyperbolic tangent function value based on the gain adjustment amount at each time point. Multiply the function value with the corresponding amplitude-modulated signal amplitude to generate a gain adjustment sequence. The amplitude reference trajectory is constructed by the normalized mean square amplitude sequence and the amplitude modulation coefficient sequence.

[0042] S46. Multiply the gain adjustment sequence and the amplitude modulation signal sequence value by value at each time point, and perform amplitude smoothing and dynamic range compression to output a stable signal stream.

[0043] Optionally, S5 specifically includes:

[0044] S51. According to the set time frame length and time displacement step, the stationary signal stream is divided into continuous time frames. In each time frame, a short-time Fourier transform is performed to extract the corresponding frequency domain amplitude sequence. The frequency domain amplitude sequences of all time frames are arranged in chronological order to construct a time-frequency feature map.

[0045] S52. Input the time-frequency feature map into the pre-trained convolutional neural network, and extract the corresponding frequency feature sequence based on the amplitude distribution change in the frequency dimension of each time frame. The frequency feature sequence is composed of feature points output by the convolutional neural network in each time frame.

[0046] S53. Based on the frequency feature sequence, locate the frequency point with the largest amplitude in each time frame, define it as the main frequency component, and arrange all the main frequency components in chronological order to form the main frequency trajectory.

[0047] S54. Compare the main frequency trajectory frame by frame, identify the direction of change of the main frequency position in adjacent time frames, mark the frequency jump points, and convert all frequency jump points into modulation signals arranged in time.

[0048] Optionally, S52 specifically includes:

[0049] S521. Perform Min-Max normalization on all amplitude points in the time-frequency feature map, and extract the maximum and minimum amplitudes within the corresponding time period at each frequency position.

[0050] S522. Using the difference between the maximum and minimum amplitude as the normalization interval, the amplitude at the corresponding frequency position in each time frame is linearly scaled according to this interval to generate an amplitude matrix.

[0051] S523. The amplitude matrix is ​​truncated at equal intervals according to the preset time period length to form a set of feature segments divided by time.

[0052] S524. Using the feature fragments as input samples of the convolutional neural network, extract the amplitude distribution features in the frequency dimension and form a frequency feature sequence.

[0053] Optionally, S6 specifically includes:

[0054] S61. Mix the modulation signal with the preset reference signal A and reference signal B respectively, perform low-pass filtering on the two mixing results, and complete the definite integration operation within the set integration time period to obtain the integral quantity A and the integral quantity B.

[0055] S62. Compare the integral quantities of A and B within the same integration time period, and generate a judgment result based on the reference signal corresponding to the larger integral quantity.

[0056] S63. Arrange the judgment results in chronological order to form a binary data stream.

[0057] A high-sensitivity magnetic signal receiving device based on the SERF atomic spin effect according to an embodiment of the present invention includes:

[0058] The signal acquisition module is used to establish the SERF polarization system and acquire the optical response signal formed under the action of a low-frequency magnetic field signal.

[0059] The photoelectric conversion module is used to construct the analytical envelope of the optical response signal using Hilbert transform, input the analytical envelope into a least-squares support vector machine to fit the phase mapping, and output a linearly enhanced voltage signal.

[0060] The filtering enhancement module is used to input the voltage signal into the dual-channel FIR filter to perform interference suppression and frequency band extraction operations respectively. Based on the error between the two outputs and the preset desired signal, the filter coefficients are updated by the NLMS algorithm, and the two outputs are superimposed to generate a high signal-to-noise ratio signal.

[0061] The dynamic amplitude modulation module is used to extract the root mean square amplitude of a high signal-to-noise ratio signal within a sliding window, compare it with a preset reference value to determine the amplitude modulation coefficient, and use an automatic gain control method to perform nonlinear amplitude modulation on the signal to output a smooth signal stream.

[0062] The frequency domain analysis module is used to convert the stationary signal stream into a time-frequency feature map through short-time Fourier transform, and input it into a convolutional neural network to extract frequency features. Based on the frequency features, the main frequency component of the stationary signal stream is separated to generate a modulation signal.

[0063] The mixing decision module is used to mix the modulated signal with two reference signals respectively, perform low-pass filtering and definite integration on the mixing result, and recover the binary data stream based on the reference signal corresponding to the larger integral.

[0064] The decoding and rectification module is used to perform interference correction and error control on the binary data stream and output the communication results.

[0065] The beneficial effects of this invention are:

[0066] First, by introducing the SERF atomic spin effect into the magnetic signal receiving system, this invention constructs a highly sensitive spin polarization detection mechanism, breaking through the bottlenecks of traditional coil receiving devices in terms of thermal noise, size limitations, and response bandwidth. This effectively improves the detection capability of extremely weak magnetic signals and lays the foundation for reliable communication in low-frequency, non-line-of-sight environments.

[0067] Secondly, this invention, based on the operating characteristics of the SERF magnetometer, designs a signal processing flow that includes Hilbert transform, support vector machine phase fitting, adaptive FIR filtering, dynamic amplitude modulation, and automatic gain control, enabling accurate extraction of the dominant frequency component and suppression of background interference. Simultaneously, by combining short-time Fourier analysis and convolutional neural network frequency feature extraction techniques, the ability to analyze and identify time-varying magnetic signals is improved, resolving the nonlinear output and interference susceptibility issues of the SERF magnetometer in communication applications.

[0068] Finally, this invention constructs a decoding mechanism based on reference signal mixing and integral decision at the receiving end, and integrates error control and interference correction strategies to realize a complete closed-loop receiving link from atomic spin response to binary communication data. The overall system has technical advantages of high sensitivity, high signal restoration accuracy, and strong anti-interference capability, and is particularly suitable for high-reliability magnetic induction communication applications in radio wave-constrained scenarios such as underwater, underground, and electromagnetically shielded environments. Attached Figure Description

[0069] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0070] Figure 1 This is a flowchart of the high-sensitivity magnetic signal receiving method based on the SERF atomic spin effect proposed in this invention;

[0071] Figure 2 This is a flowchart of the adaptive filtering enhancement process for the high-sensitivity magnetic signal receiving method based on the SERF atomic spin effect proposed in this invention.

[0072] Figure 3 This is a block diagram of the high-sensitivity magnetic signal receiving device based on the SERF atomic spin effect proposed in this invention. Detailed Implementation

[0073] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0074] refer to Figure 1-2 A high-sensitivity magnetic signal receiving method based on the SERF atomic spin effect includes the following steps:

[0075] S1. Establish a SERF polarization system and collect the optical response signal formed under the action of a low-frequency magnetic field signal;

[0076] S2. The analytic envelope of the optical response signal is constructed using Hilbert transform. The analytic envelope is then input into a least-squares support vector machine to fit the phase mapping, and the output voltage signal is linearly enhanced.

[0077] S3. Input the voltage signal into the dual-channel FIR filter to perform interference suppression and frequency band extraction operations respectively. Based on the error between the two outputs and the preset desired signal, update the filter coefficients through the NLMS algorithm and superimpose the two outputs to generate a high signal-to-noise ratio signal.

[0078] S4. Extract the root mean square amplitude of the high signal-to-noise ratio signal within the sliding window, compare it with the preset reference value to determine the amplitude modulation coefficient, and use the automatic gain control method to perform nonlinear amplitude modulation on the signal to output a stable signal stream.

[0079] S5. The stationary signal stream is converted into a time-frequency feature map by short-time Fourier transform, and then input into a convolutional neural network to extract frequency features. Based on the frequency features, the main frequency component of the stationary signal stream is separated to generate a modulation signal.

[0080] S6. Mix the modulation signal with the two reference signals respectively, perform low-pass filtering and definite integration on the mixing result, and recover the binary data stream based on the reference signal corresponding to the larger integral.

[0081] S7. Perform interference correction and error control on the binary data stream, and output the communication result.

[0082] In this embodiment, the SERF polarization system uses an atomic gas chamber filled with alkali metal atoms and buffer gas within a magnetically shielded structure. A heating device is used to maintain the temperature of the gas chamber within the operating range of a spin-free exchange relaxation state. Circularly polarized pump light and linearly polarized probe light are used to irradiate the gas chamber to form a stable spin polarization direction. A compensating magnetic field is applied through a triaxial coil array to lock the operating point at a zero magnetic field position. The photoelectric detection unit performs photoelectric conversion on the polarization change of the transmitted light and outputs an optical response signal.

[0083] In this embodiment, S2 specifically includes:

[0084] S21. Perform a Hilbert transform operation on the optical response signal to construct a complex analytic signal composed of a real part signal and a Hilbert imaginary part signal, and maintain a structure consistent with the optical response signal in the time dimension. The real part signal represents the optical response signal, and the Hilbert imaginary part signal represents the orthogonal component obtained after performing a 90-degree phase shift on the optical response signal.

[0085] S22. Calculate the analytic envelope obtained by summing the squares of the real and imaginary parts and taking the square root based on the complex analytic signal, and perform moving average processing.

[0086] S23. Amplitude sampling is performed on the analytical envelope to construct multiple signal segments of equal length. Each signal segment consists of a set number of amplitude sampling points to form a sample set. The least squares support vector machine algorithm is called. With the sample set as input, the mapping relationship between the analytical envelope and the phase shift is established by solving the least squares form of the mapping coefficients. The phase shift represents the instantaneous phase change caused by the external magnetic field in the atomic spin response trajectory corresponding to the analytical envelope. The atomic spin response trajectory refers to the dynamic response process formed by the change of the net spin polarization direction of alkali metal atoms with time under the action of an external magnetic field.

[0087] S24. Based on the established mapping relationship, input the analytical envelope into the least squares support vector machine algorithm in time order, and output the fitted phase sequence;

[0088] S25. Perform linear amplitude expansion processing on the fitted phase mapping sequence, and generate a linearly enhanced voltage signal based on the amplitude expansion result.

[0089] In this embodiment, S3 specifically includes:

[0090] S31. Input the voltage signal into the dual-channel FIR filter and perform finite impulse response filtering operations in the interference path and the frequency band path respectively to obtain the interference suppression signal and the frequency band extraction signal.

[0091] S32. Compare the interference suppression signal and the frequency band extraction signal with the preset desired signal point by point to generate the interference error sequence and the frequency band error sequence.

[0092] S33. Based on two error sequences, the weight increment at each time point in the corresponding path is calculated using the NLMS algorithm to generate an interference coefficient sequence and a frequency band coefficient sequence. The two coefficient sequences are then input into the corresponding paths in the dual-channel FIR filter. Combined with the preset amplitude boundary constraints, the weight parameters of the two paths are dynamically adjusted.

[0093] S34. Perform amplitude superposition operation on the interference suppression signal and the frequency band extraction signal after weight adjustment to output a high signal-to-noise ratio signal.

[0094] In this embodiment, S33 specifically includes:

[0095] S331. Set a sliding window of fixed length in the interference error sequence and the frequency band error sequence, and extract the sequence segments in the sliding window according to the time point to obtain the interference error segment and the frequency band error segment.

[0096] S332. Calculate the difference between the maximum and minimum amplitude of the two error segments respectively to form the interference fluctuation index and the frequency band fluctuation index.

[0097] S333. Compare the interference fluctuation index and frequency band fluctuation index with the preset fluctuation threshold. If the threshold is exceeded, the step size parameter of the corresponding path is lowered. If the threshold is lower, the step size parameter is raised. The step size parameter represents the adjustment ratio used to update the filter weight.

[0098] S334. Based on the adjusted step size parameters of each path, the NLMS algorithm is used to calculate the weight increment of the interference error sequence and the frequency band error sequence to form the interference coefficient sequence and the frequency band coefficient sequence.

[0099] S335. Compare the interference coefficient sequence and the frequency band coefficient sequence with the preset amplitude boundary of the corresponding path. If the amplitude exceeds the boundary, compress the amplitude linearly to the boundary limit in the original sign direction. Input the compression result into the corresponding path of the dual-channel FIR filter and add it to the original weight parameters at the time point to dynamically update the weight parameters.

[0100] In this embodiment, the process of calculating the weight increment using the NLMS algorithm specifically includes:

[0101] Obtain the error value at the current time point, where the error value is the difference between the current output of the filter and the preset desired signal;

[0102] Obtain the processed signal segment at the current time point. The processed signal segment is the result of truncating the error sequence corresponding to the current path within the weight length range, and is used as the reference input for the weight vector.

[0103] The sum of squared amplitudes of the processed signal segments is calculated and used as a normalization factor to prevent instability in weight updates caused by fluctuations in the energy of the input signal.

[0104] Multiply the error value by the step size parameter, then by the processed signal segment, and divide by the normalization factor to obtain the weight increment at the current time point.

[0105] In this embodiment, S4 specifically includes:

[0106] S41. Construct a sliding window with a fixed length for a high signal-to-noise ratio signal, set the window displacement step size, and make the window move continuously in the time series. Extract amplitude points in each window in time order to form a window amplitude set.

[0107] S42. Calculate the root mean square magnitude in the window magnitude set by summation of squares, and perform Min-Max normalization on the root mean square result based on the set maximum and minimum magnitudes to generate a root mean square magnitude sequence, while keeping the time index of the root mean square magnitude sequence consistent with the starting position of the sliding window.

[0108] S43. Compare the mean square amplitude sequence point by point with the preset reference value, and construct the amplitude adjustment coefficient sequence based on the comparison results;

[0109] S44. Perform point-by-point amplitude scaling on the high signal-to-noise ratio signal according to the amplitude modulation coefficient sequence. During the scaling process, perform linear interpolation on the amplitude modulation coefficient at each time point to generate an amplitude modulation signal sequence.

[0110] S45. Perform automatic gain control on the amplitude-modulated signal sequence, calculate the gain adjustment amount according to the amplitude reference trajectory, and calculate the hyperbolic tangent function value based on the gain adjustment amount at each time point. Multiply the function value with the corresponding amplitude-modulated signal amplitude to generate a gain adjustment sequence. The amplitude reference trajectory is constructed by the normalized mean square amplitude sequence and the amplitude modulation coefficient sequence.

[0111] S46. Multiply the gain adjustment sequence and the amplitude modulation signal sequence value by value at each time point, and perform amplitude smoothing and dynamic range compression to output a stable signal stream. Specifically, this includes:

[0112] The gain adjustment sequence is multiplied by the amplitude modulation signal sequence at time points to generate a preliminary shaped signal stream;

[0113] The amplitude change smoothing process is performed on the initial shaped signal stream. By detecting the abrupt slope of the continuous time period in the amplitude change rate curve, a linear transition segment is inserted in the time period when the slope exceeds the preset threshold to correct abnormal changes.

[0114] The amplitude of the initial shaped signal stream is monitored in real time. If the amplitude exceeds the set maximum allowable range, it is compressed to the preset upper limit range according to the hyperbolic tangent function. If the amplitude is lower than the preset lower limit, it is replaced with the set minimum control amplitude. Finally, a stable signal stream with continuous time and stable amplitude range is output.

[0115] In this embodiment, S5 specifically includes:

[0116] S51. According to the set time frame length and time displacement step, the stationary signal stream is divided into continuous time frames. In each time frame, a short-time Fourier transform is performed to extract the corresponding frequency domain amplitude sequence. The frequency domain amplitude sequences of all time frames are arranged in chronological order to construct a time-frequency feature map.

[0117] S52. Input the time-frequency feature map into the pre-trained convolutional neural network, and extract the corresponding frequency feature sequence based on the amplitude distribution change in the frequency dimension of each time frame. The frequency feature sequence is composed of feature points output by the convolutional neural network in each time frame.

[0118] S53. Based on the frequency feature sequence, locate the frequency point with the largest amplitude in each time frame, define it as the main frequency component, and arrange all the main frequency components in chronological order to form the main frequency trajectory.

[0119] S54. Compare the main frequency trajectory frame by frame, identify the direction of change of the main frequency position in adjacent time frames, mark the frequency jump points, and convert all frequency jump points into modulation signals arranged in time.

[0120] In this embodiment, S52 specifically includes:

[0121] S521. Perform Min-Max normalization on all amplitude points in the time-frequency feature map, and extract the maximum and minimum amplitudes within the corresponding time period at each frequency position.

[0122] S522. Using the difference between the maximum and minimum amplitude as the normalization interval, the amplitude at the corresponding frequency position in each time frame is linearly scaled according to this interval to generate an amplitude matrix.

[0123] S523. The amplitude matrix is ​​truncated at equal intervals according to the preset time period length to form a set of feature segments divided by time.

[0124] S524. Using the feature fragments as input samples of the convolutional neural network, extract the amplitude distribution features in the frequency dimension and form a frequency feature sequence.

[0125] In this embodiment, the convolutional neural network consists of several two-dimensional convolutional layers, activation function layers, and pooling layers. The two-dimensional convolutional layers are used to extract amplitude distribution features in the frequency and time dimensions. The activation function layers perform nonlinear transformations on the convolutional outputs. The pooling layers perform region dimensionality reduction and feature preservation on the activation results.

[0126] In this embodiment, S6 specifically includes:

[0127] S61. Mix the modulation signal with the preset reference signal A and reference signal B respectively, perform low-pass filtering on the two mixing results, and complete the definite integration operation within the set integration time period to obtain the integral quantity A and the integral quantity B.

[0128] S62. Compare the integral quantities of A and B within the same integration time period, and generate a judgment result based on the reference signal corresponding to the larger integral quantity.

[0129] S63. Arrange the judgment results in chronological order to form a binary data stream.

[0130] refer to Figure 3 A high-sensitivity magnetic signal receiving device based on the SERF atomic spin effect includes:

[0131] The signal acquisition module is used to establish the SERF polarization system and acquire the optical response signal formed under the action of a low-frequency magnetic field signal.

[0132] The photoelectric conversion module is used to construct the analytical envelope of the optical response signal using Hilbert transform, input the analytical envelope into a least-squares support vector machine to fit the phase mapping, and output a linearly enhanced voltage signal.

[0133] The filtering enhancement module is used to input the voltage signal into the dual-channel FIR filter to perform interference suppression and frequency band extraction operations respectively. Based on the error between the two outputs and the preset desired signal, the filter coefficients are updated by the NLMS algorithm, and the two outputs are superimposed to generate a high signal-to-noise ratio signal.

[0134] The dynamic amplitude modulation module is used to extract the root mean square amplitude of a high signal-to-noise ratio signal within a sliding window, compare it with a preset reference value to determine the amplitude modulation coefficient, and use an automatic gain control method to perform nonlinear amplitude modulation on the signal to output a smooth signal stream.

[0135] The frequency domain analysis module is used to convert the stationary signal stream into a time-frequency feature map through short-time Fourier transform, and input it into a convolutional neural network to extract frequency features. Based on the frequency features, the main frequency component of the stationary signal stream is separated to generate a modulation signal.

[0136] The mixing decision module is used to mix the modulated signal with two reference signals respectively, perform low-pass filtering and definite integration on the mixing result, and recover the binary data stream based on the reference signal corresponding to the larger integral.

[0137] The decoding and rectification module is used to perform interference correction and error control on the binary data stream and output the communication results.

[0138] Example 1:

[0139] To verify the feasibility of this invention in practice, it was applied to a closed underground environment with strong magnetic interference and radio wave shielding characteristics for communication reception testing. This environment is characterized by: high background magnetic noise, complex spatial structure, ineffective traditional wireless communication methods, significant attenuation of the magnetic signal during propagation, background magnetic field fluctuations ranging from 100 nT to 700 nT within the test area, and communication signal frequencies ranging from 50 Hz to 1.5 kHz, exhibiting strong non-stationary characteristics and abrupt modulation features.

[0140] In this scenario, the high-sensitivity magnetic signal receiving device based on the SERF atomic spin effect described in this invention is deployed. By constructing a SERF working system including a laser pumping path, heating control, magnetic shielding structure, photoelectric detection channel, and triaxial compensation coil, stable sensing and response to extremely weak modulated magnetic signals are achieved. The system uses the optical response signal as input and utilizes Hilbert transform and least squares support vector machine to construct a phase mapping relationship to obtain a linearly enhanced voltage signal. The voltage signal is processed by a dual-channel FIR filter and the Normalized Least Mean Square (NLMS) algorithm to perform interference suppression and frequency band extraction, respectively. Then, a stable signal stream is generated through amplitude normalization, amplitude modulation coefficient calculation, and automatic gain control. The receiving system further uses short-time Fourier transform to extract time-frequency feature maps and uses a convolutional neural network to extract frequency change trajectories, identify the dominant frequency component, and recover the magnetic modulation signal. Finally, the modulated signal and reference frequency are mixed, low-pass filtered, and integrated for decision, outputting binary communication data, and a communication result is generated by combining an error control mechanism.

[0141] To compare performance, a traditional coil receiver was selected as a reference. Under the same test conditions, signal-to-noise ratio, bit error rate, signal recovery rate, frequency identification accuracy, system response time, and maximum communication distance were compared and evaluated. The table below summarizes the test data:

[0142] Table 1. Performance Comparison Data Between SERF-Based Magnetic Reception Method and Traditional Coil Reception System

[0143] Test number Background noise (nT) Transmission distance (m) Average signal-to-noise ratio (dB) Bit error rate (%) Signal recovery rate (%) System response time (s) Stable communication status Frequency recognition accuracy (%) 1 100 15 21.4 0.5 99.2 0.23 yes 98.6 2 300 20 19.6 0.7 98.3 0.26 yes 97.9 3 450 25 17.8 1.3 96.7 0.31 yes 96.4 4 600 30 14.1 3.6 92.5 0.34 yes 94.2 5 700 35 12.7 4.9 89.1 0.38 no 90.5 6 300 20 (Coil Method) 3.2 12.4 68.4 0.81 no 84.3 7 450 25 (Coil Method) 2.7 17.2 60.3 0.89 no 79.8

[0144] As shown in Table 1, the receiving system of this invention maintains a high signal-to-noise ratio and a low bit error rate consistently within a background noise range of 100 nT to 600 nT. The average signal-to-noise ratio can reach over 17.8 dB, the bit error rate is controlled within 1.5%, the signal recovery rate generally exceeds 96%, and the frequency identification accuracy is consistently above 94%. Even under high noise interference conditions of 600 nT, the communication link can still be maintained stably.

[0145] In contrast, traditional coil receiving methods, under the same conditions, have a signal-to-noise ratio of less than 3.5 dB, a bit error rate exceeding 12%, and experience communication interruptions and a significant decrease in signal recovery rate after further interference. Regarding response time, the system of this invention generally has a response time of less than 0.35 seconds, far superior to the coil method's response delay of over 0.8 seconds, demonstrating strong real-time signal processing capabilities.

[0146] Furthermore, the present invention achieved a stable receiving distance of 37.2 meters in the longest communication distance test, while the communication performance of the coil system dropped sharply after exceeding 25 meters, indicating that the present invention not only has advantages in near-field sensitivity, but also has significant technological breakthroughs in long-distance transmission capability.

[0147] In summary, this invention, through its innovative SERF magnetic receiver architecture and multi-level signal processing strategy, systematically solves the core problems of traditional magnetic communication technology in terms of sensitivity, stability, bandwidth adaptability, and adaptability to complex environments, and possesses extremely high practical value and promising engineering application prospects.

[0148] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A high-sensitivity magnetic signal receiving method based on the SERF atomic spin effect, characterized in that, Includes the following steps: S1. Establish a SERF polarization system and collect the optical response signal formed under the action of a low-frequency magnetic field signal; S2. The analytic envelope of the optical response signal is constructed using Hilbert transform. The analytic envelope is then input into a least-squares support vector machine to fit the phase mapping, and the output voltage signal is linearly enhanced. S3. Input the voltage signal into the dual-channel FIR filter to perform interference suppression and frequency band extraction operations respectively. Based on the error between the two outputs and the preset desired signal, update the filter coefficients through the NLMS algorithm and superimpose the two outputs to generate a high signal-to-noise ratio signal. S4. Extract the root mean square amplitude of the high signal-to-noise ratio signal within the sliding window, compare it with the preset reference value to determine the amplitude modulation coefficient, and use the automatic gain control method to perform nonlinear amplitude modulation on the signal to output a stable signal stream. S5. The stationary signal stream is converted into a time-frequency feature map by short-time Fourier transform, and then input into a convolutional neural network to extract frequency features. Based on the frequency features, the main frequency component of the stationary signal stream is separated to generate a modulation signal. S5 specifically includes: S51. According to the set time frame length and time displacement step, the stationary signal stream is divided into continuous time frames. In each time frame, a short-time Fourier transform is performed to extract the corresponding frequency domain amplitude sequence. The frequency domain amplitude sequences of all time frames are arranged in chronological order to construct a time-frequency feature map. S52. Input the time-frequency feature map into the pre-trained convolutional neural network, and extract the corresponding frequency feature sequence based on the amplitude distribution change in the frequency dimension of each time frame. The frequency feature sequence is composed of feature points output by the convolutional neural network in each time frame. S53. Based on the frequency feature sequence, locate the frequency point with the largest amplitude in each time frame, define it as the main frequency component, and arrange all the main frequency components in chronological order to form the main frequency trajectory. S54. Compare the main frequency trajectory frame by frame, identify the direction of change of the main frequency position in adjacent time frames, mark the frequency jump points, and convert all frequency jump points into modulation signals arranged in time. S6. Mix the modulation signal with the two reference signals respectively, perform low-pass filtering and definite integration on the mixing result, and recover the binary data stream based on the reference signal corresponding to the larger integral. S7. Perform interference correction and error control on the binary data stream, and output the communication result.

2. The high-sensitivity magnetic signal receiving method based on the SERF atomic spin effect according to claim 1, characterized in that, The SERF polarization system uses an atomic chamber filled with alkali metal atoms and buffer gas within a magnetically shielded structure. A heating device maintains the chamber temperature within the operating range of a spin-free exchange relaxation state. Circularly polarized pump light and linearly polarized probe light are used to irradiate the chamber to form a stable spin polarization direction. A compensating magnetic field is applied through a triaxial coil array to lock the operating point at a zero magnetic field position. The photoelectric detection unit performs photoelectric conversion on the polarization change of the transmitted light and outputs an optical response signal.

3. The high-sensitivity magnetic signal receiving method based on the SERF atomic spin effect according to claim 1, characterized in that, S2 specifically includes: S21. Perform a Hilbert transform operation on the optical response signal to construct a complex analytic signal composed of a real part signal and a Hilbert imaginary part signal, and maintain a structure consistent with the optical response signal in the time dimension. The real part signal represents the optical response signal, and the Hilbert imaginary part signal represents the orthogonal component obtained after performing a 90-degree phase shift on the optical response signal. S22. Calculate the analytic envelope obtained by summing the squares of the real and imaginary parts and taking the square root based on the complex analytic signal, and perform moving average processing. S23. Amplitude sampling is performed on the analytical envelope to construct multiple signal segments of equal length. Each signal segment consists of a set number of amplitude sampling points to form a sample set. The least squares support vector machine algorithm is called. With the sample set as input, the mapping relationship between the analytical envelope and the phase shift is established by solving the least squares form of the mapping coefficients. The phase shift represents the instantaneous phase change caused by the external magnetic field in the atomic spin response trajectory corresponding to the analytical envelope. The atomic spin response trajectory refers to the dynamic response process formed by the change of the net spin polarization direction of alkali metal atoms with time under the action of an external magnetic field. S24. Based on the established mapping relationship, input the analytical envelope into the least squares support vector machine algorithm in time order, and output the fitted phase sequence; S25. Perform linear amplitude expansion processing on the fitted phase mapping sequence, and generate a linearly enhanced voltage signal based on the amplitude expansion result.

4. The high-sensitivity magnetic signal receiving method based on the SERF atomic spin effect according to claim 1, characterized in that, S3 specifically includes: S31. Input the voltage signal into the dual-channel FIR filter and perform finite impulse response filtering operations in the interference path and the frequency band path respectively to obtain the interference suppression signal and the frequency band extraction signal. S32. Compare the interference suppression signal and the frequency band extraction signal with the preset desired signal point by point to generate the interference error sequence and the frequency band error sequence. S33. Based on two error sequences, the weight increment at each time point in the corresponding path is calculated using the NLMS algorithm to generate an interference coefficient sequence and a frequency band coefficient sequence. The two coefficient sequences are then input into the corresponding paths in the dual-channel FIR filter. Combined with the preset amplitude boundary constraints, the weight parameters of the two paths are dynamically adjusted. S34. Perform amplitude superposition operation on the interference suppression signal and the frequency band extraction signal after weight adjustment to output a high signal-to-noise ratio signal.

5. The high-sensitivity magnetic signal receiving method based on the SERF atomic spin effect according to claim 4, characterized in that, Specifically, S33 includes: S331. Set a sliding window of fixed length in the interference error sequence and the frequency band error sequence, and extract the sequence segments in the sliding window according to the time point to obtain the interference error segment and the frequency band error segment. S332. Calculate the difference between the maximum and minimum amplitude of the two error segments respectively to form the interference fluctuation index and the frequency band fluctuation index. S333. Compare the interference fluctuation index and frequency band fluctuation index with the preset fluctuation threshold. If the threshold is exceeded, the step size parameter of the corresponding path is lowered. If the threshold is lower, the step size parameter is raised. The step size parameter represents the adjustment ratio used to update the filter weight. S334. Based on the adjusted step size parameters of each path, the NLMS algorithm is used to calculate the weight increment of the interference error sequence and the frequency band error sequence to form the interference coefficient sequence and the frequency band coefficient sequence. S335. Compare the interference coefficient sequence and the frequency band coefficient sequence with the preset amplitude boundary of the corresponding path. If the amplitude exceeds the boundary, compress the amplitude linearly to the boundary limit in the original sign direction. Input the compression result into the corresponding path of the dual-channel FIR filter and add it to the original weight parameters at the time point to dynamically update the weight parameters.

6. The high-sensitivity magnetic signal receiving method based on the SERF atomic spin effect according to claim 1, characterized in that, S4 specifically includes: S41. Construct a sliding window with a fixed length for a high signal-to-noise ratio signal, set the window displacement step size, and make the window move continuously in the time series. Extract amplitude points in each window in time order to form a window amplitude set. S42. Calculate the root mean square magnitude in the window magnitude set by summation of squares, and perform Min-Max normalization on the root mean square result based on the set maximum and minimum magnitudes to generate a root mean square magnitude sequence, while keeping the time index of the root mean square magnitude sequence consistent with the starting position of the sliding window. S43. Compare the mean square amplitude sequence point by point with the preset reference value, and construct the amplitude adjustment coefficient sequence based on the comparison results; S44. Perform point-by-point amplitude scaling on the high signal-to-noise ratio signal according to the amplitude modulation coefficient sequence. During the scaling process, perform linear interpolation on the amplitude modulation coefficient at each time point to generate an amplitude modulation signal sequence. S45. Perform automatic gain control on the amplitude-modulated signal sequence, calculate the gain adjustment amount according to the amplitude reference trajectory, and calculate the hyperbolic tangent function value based on the gain adjustment amount at each time point. Multiply the function value with the corresponding amplitude-modulated signal amplitude to generate a gain adjustment sequence. The amplitude reference trajectory is constructed by the normalized mean square amplitude sequence and the amplitude modulation coefficient sequence. S46. Multiply the gain adjustment sequence and the amplitude modulation signal sequence value by value at each time point, and perform amplitude smoothing and dynamic range compression to output a stable signal stream.

7. The high-sensitivity magnetic signal receiving method based on the SERF atomic spin effect according to claim 1, characterized in that, Specifically, S52 includes: S521. Perform Min-Max normalization on all amplitude points in the time-frequency feature map, and extract the maximum and minimum amplitudes within the corresponding time period at each frequency position. S522. Using the difference between the maximum and minimum amplitude as the normalization interval, the amplitude at the corresponding frequency position in each time frame is linearly scaled according to this interval to generate an amplitude matrix. S523. The amplitude matrix is ​​truncated at equal intervals according to the preset time period length to form a set of feature segments divided by time. S524. Using the feature fragments as input samples of the convolutional neural network, extract the amplitude distribution features in the frequency dimension and form a frequency feature sequence.

8. The high-sensitivity magnetic signal receiving method based on the SERF atomic spin effect according to claim 1, characterized in that, S6 specifically includes: S61. Mix the modulation signal with the preset reference signal A and reference signal B respectively, perform low-pass filtering on the two mixing results, and complete the definite integration operation within the set integration time period to obtain the integral quantity A and the integral quantity B. S62. Compare the integral quantities of A and B within the same integration time period, and generate a judgment result based on the reference signal corresponding to the larger integral quantity. S63. Arrange the judgment results in chronological order to form a binary data stream.

9. A high-sensitivity magnetic signal receiving device based on the SERF atomic spin effect, comprising the high-sensitivity magnetic signal receiving method based on the SERF atomic spin effect as described in any one of claims 1 to 8, characterized in that, include: The signal acquisition module is used to establish the SERF polarization system and acquire the optical response signal formed under the action of a low-frequency magnetic field signal. The photoelectric conversion module is used to construct the analytical envelope of the optical response signal using Hilbert transform, input the analytical envelope into a least-squares support vector machine to fit the phase mapping, and output a linearly enhanced voltage signal. The filtering enhancement module is used to input the voltage signal into the dual-channel FIR filter to perform interference suppression and frequency band extraction operations respectively. Based on the error between the two outputs and the preset desired signal, the filter coefficients are updated by the NLMS algorithm, and the two outputs are superimposed to generate a high signal-to-noise ratio signal. The dynamic amplitude modulation module is used to extract the root mean square amplitude of a high signal-to-noise ratio signal within a sliding window, compare it with a preset reference value to determine the amplitude modulation coefficient, and use an automatic gain control method to perform nonlinear amplitude modulation on the signal to output a smooth signal stream. The frequency domain analysis module is used to convert the stationary signal stream into a time-frequency feature map through short-time Fourier transform, and input it into a convolutional neural network to extract frequency features. Based on the frequency features, the main frequency component of the stationary signal stream is separated to generate a modulation signal. The mixing decision module is used to mix the modulated signal with two reference signals respectively, perform low-pass filtering and definite integration on the mixing result, and recover the binary data stream based on the reference signal corresponding to the larger integral. The decoding and rectification module is used to perform interference correction and error control on the binary data stream and output the communication results.

Images