Brillouin distributed optical fiber sensing demodulation method based on frequency spectrum inverse Fourier transform
By combining spectral inverse Fourier transform and deep neural network, the problems of time-consuming measurement and high hardware cost in BOTDR systems are solved, and efficient and real-time Brillouin fiber optic sensing demodulation is achieved, which is suitable for complex scenarios such as industrial monitoring.
Patent Information
- Application Number
- CN202510855746.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-25
- Publication Date
- 2025-10-21
AI Technical Summary
The existing Brillouin optical time-domain reflectometry (BOTDR) system relies on frequency scanning, resulting in long measurement times and high hardware costs. Although the short-time Fourier transform (STFT) method improves efficiency, it requires secondary demodulation processing and has problems such as high consumption of computing resources.
By adopting a method based on spectral inverse Fourier transform and combining it with deep neural network, a nonlinear mapping between time domain signal and Brillouin frequency shift is directly established through single wide spectrum acquisition and signal preprocessing. This simplifies the hardware design, eliminates the electric domain bandpass filtering module, and uses neural network for efficient demodulation.
It significantly improves measurement efficiency and accuracy, reduces hardware complexity and cost, achieves real-time and highly reliable sensor demodulation, and is suitable for complex sensing scenarios.
Smart Images
Figure CN120820183A_ABST
Abstract
Description
Technical field:
[0001] The present invention belongs to the technical field of optical fiber sensing, and in particular relates to a Brillouin distributed optical fiber sensing demodulation method based on spectrum inverse Fourier transform. Background technology:
[0002] Optical fiber, with its low loss, corrosion resistance, and electromagnetic interference resistance, holds significant application value in communications and sensing. Distributed fiber-optic sensing technology, by analyzing optical effects within optical fibers, enables spatially distributed measurement of physical quantities such as temperature, strain, and vibration. Current mainstream technologies rely on three scattering effects: Rayleigh scattering-based systems are primarily used for acoustic signal monitoring due to their vibration sensitivity; Raman scattering-based systems, with their temperature sensitivity, have achieved large-scale application in temperature sensing; and Brillouin scattering-based systems, with their combined temperature and strain sensitivity, are becoming a research focus in the field of distributed sensing.
[0003] Incident light propagating in an optical fiber interacts with phonons, generating scattered light with a frequency different from that of the incident light. This scattered light includes Stokes light with a frequency shift down and anti-Stokes light with a frequency shift up. The power spectrum of the scattered light exhibits a nearly Lorentzian lineshape, and the offset of its center frequency from the center frequency of the incident light is called the Brillouin frequency shift (BFS). Because the Brillouin frequency shift varies with physical quantities in the optical fiber environment (such as temperature and strain), the distribution of these physical quantities can be inferred by measuring the Brillouin frequency shift. Brillouin optical time-domain analysis (BOTDA) and Brillouin optical time-domain reflectometry (BOTDR) are two commonly used methods for measuring Brillouin signals. Taking Brillouin optical time-domain reflectometry (BOTDR) as an example, this technique scans the power of scattered light at different frequency differences to obtain the Brillouin scattering spectrum. By extracting the Brillouin frequency shift at each optical fiber location and establishing a corresponding relationship between it and the corresponding physical quantity, precise measurement of the physical quantity can be achieved.
[0004] BOTDR technology offers the advantages of high spatial resolution and single-ended access. Its precise positioning capability is crucial for the rapid repair of damaged modules in scenarios such as industrial process monitoring and aircraft structural health monitoring. This requires the system to combine high-precision demodulation capabilities with fast response characteristics to accurately capture the amplitude, boundaries, and spatial distribution characteristics of physical quantity changes. An improved method based on the short-time Fourier transform (STFT) eliminates the need for frequency sweeping and enables rapid response to physical quantity detection.
[0005] However, traditional BOTDR systems require frequency scanning to acquire high-resolution Brillouin spectra, resulting in multiple repeated measurements and excessively high time-domain sampling rates. This not only significantly increases measurement time but also leads to inefficient data processing and escalating hardware costs. Existing post-processing methods (such as image denoising and neural network fitting) primarily process the raw data obtained from frequency scanning, making it difficult to fundamentally improve the efficiency bottleneck in the data acquisition stage.
[0006] While improved methods based on the short-time Fourier transform (STFT) reduce the number of measurements to a single one, improving measurement efficiency to a certain extent, the STFT calculation process is time-consuming, and the derived data is still raw, undemodulated data. Therefore, it still needs to be demodulated using a Brillouin spectrum fitting algorithm, further increasing measurement time. Summary of the invention:
[0007] The technical problem to be solved by the present invention is to provide a Brillouin distributed optical fiber sensing demodulation method based on spectral inverse Fourier transform, which overcomes the efficiency bottleneck and system complexity defects in the existing Brillouin optical time domain reflection technology: the traditional BOTDR system relies on the frequency scanning mechanism, which leads to a sharp increase in measurement time and high-frequency time domain sampling causes the hardware cost to rise, and the existing short-time Fourier transform (STFT)-based method achieves a single measurement but still requires secondary demodulation processing and consumes a lot of computing resources. By designing a single wide spectrum acquisition mode and a deep neural network demodulation architecture, the data redundancy problem caused by repeated scanning of multiple frequency points is solved, and the time delay of the traditional spectrum fitting algorithm is eliminated. At the same time, the electrical domain bandpass filtering module is removed through hardware system optimization, so that the measurement efficiency is improved while ensuring the measurement accuracy, and the requirements for real-time, low cost and high reliability in complex sensing scenarios are met.
[0008] The technical solution of the present invention is to provide a Brillouin distributed optical fiber sensing demodulation method based on spectrum inverse Fourier transform, comprising the following steps:
[0009] S1. Parameter Control and Dataset Construction: By modulating the external physical parameters applied to the optical fiber under test and synchronously adjusting the characteristic parameters of optical and other sensor devices, a dynamic correlation between the Brillouin frequency shift and the Brillouin scattering spectrum is established. A random phase sequence is generated, and the time-frequency domain scattering spectrum is converted into a Brillouin time-domain response signal with wide-spectrum characteristics through inverse Fourier transform. In the generated time-domain waveform, an optimization algorithm is used to select time-domain signals with amplitude fluctuations less than 50% as training samples, and the corresponding Brillouin frequency shift is used as a supervision label to construct a wide-spectrum time-domain signal-Brillouin frequency shift mapping relationship dataset.
[0010] S2. Neural Network Modeling and Training: Constructing a deep neural network architecture and using the mapping relationship dataset to perform network parameter optimization training; optimizing network weight parameters through a backpropagation algorithm and establishing a nonlinear mapping model from wide-spectrum time-domain signals to Brillouin frequency shifts until the model reaches a preset convergence standard;
[0011] S3. Signal acquisition and preprocessing: Using a hardware-simplified Brillouin optical time-domain reflectometry system, under the condition of a fixed RF signal source center frequency, the broadband time-domain response signal of the tested optical fiber at a specific scattering frequency point is acquired to complete signal preprocessing;
[0012] S4. Physical quantity demodulation and reconstruction: The preprocessed time domain signal is input into a well-trained neural network model, and the corresponding Brillouin frequency shift distribution is output. The temperature field or strain field distribution of the measured optical fiber is reconstructed through the Brillouin frequency shift-physical quantity conversion algorithm.
[0013] This method significantly reduces hardware complexity by eliminating the time loss associated with traditional frequency scanning mechanisms and adopting a single-shot wide-spectrum acquisition mode. Combined with the efficient demodulation capabilities of deep neural networks, this method improves data processing efficiency by two orders of magnitude while maintaining measurement accuracy.
[0014] Preferably, in step S1, the external physical parameters include but are not limited to temperature field distribution, strain field distribution and their spatial coordinates.
[0015] Preferably, in step S1, the characteristic parameters include Brillouin scattering spectrum linewidth, detection pulse width, photodetector bandwidth and signal-to-noise ratio threshold.
[0016] Preferably, in step S1, each of the training samples is single-channel wide-spectrum Brillouin time-domain waveform data.
[0017] Preferably, in step S1, the supervisory label is a Brillouin frequency shift feature of the corresponding time domain waveform.
[0018] Preferably, in step S1, the optimization algorithm is a genetic algorithm.
[0019] Preferably, in step S2, the deep neural network includes a signal input layer, a feature extraction layer and a regression output layer, wherein: the number of input layer nodes is consistent with the number of time domain signal sampling points, and the output layer nodes correspond to Brillouin frequency shift prediction values; the feature extraction layer adopts a composite network structure, including a fully connected layer, a convolutional layer and an attention mechanism module; the training process adopts a k-fold cross-validation strategy, and the early stopping method is used to prevent overfitting.
[0020] Preferably, in step S3, the simplified hardware system omits the electric domain bandpass filtering module.
[0021] Preferably, in step S3, the signal preprocessing is data normalization.
[0022] Compared with the prior art, the present invention has the following advantages:
[0023] The present invention obtains a single, wide-band, one-dimensional Brillouin time-domain curve by fixing the frequency of the RF source, thus avoiding the time consumption associated with traditional frequency scanning and improving measurement efficiency and real-time performance. At the same time, it simplifies the hardware design, eliminates the need for an electrical bandpass filter, and reduces the complexity and cost of the system. Furthermore, by using an artificial neural network for high-precision demodulation, the Brillouin frequency shift can be accurately extracted from a single, wide-spectrum time-domain curve, significantly improving demodulation speed and accuracy. By reducing computing resource consumption and processing time, the present invention significantly improves data processing efficiency, and has higher system stability and reliability, making it suitable for more complex sensing application scenarios. Description of the drawings:
[0024] Figure 1 This is a flow chart of the signal demodulation method of the present invention.
[0025] Figure 2 This is a diagram showing the corresponding relationship between the Brillouin scattering spectrum and the Brillouin frequency shift of the present invention.
[0026] Figure 3 This is a schematic diagram of random phase generation according to the present invention.
[0027] Figure 4 This is a diagram of the wide spectrum time domain signal generation method of the present invention.
[0028] Figure 5 This is the feature extraction network model diagram of the present invention.
[0029] Figure 6 This is the temperature field demodulation result diagram of the present invention. Specific implementation method:
[0030] The present invention will be further described below with reference to the accompanying drawings:
[0031] A Brillouin distributed fiber optic sensing demodulation method based on spectrum inverse Fourier transform, such as Figure 1 As shown, the present invention performs Brillouin signal demodulation through the following steps:
[0032] 1) Set the sampling range of the Brillouin optical time domain sensor to 0MHz to 450MHz, with a scanning frequency interval of 2MHz, and a total of 226 frequency scanning points; set the time domain sampling rate to 1GHz, the local optical fiber length to be measured to 1000 meters, and a total of 10001 sampling points, where the interval between points is 0.1 meters; the Brillouin frequency shift position is set at 1MHz intervals, from 0MHz to 450MHz, with a total of 451 groups; the line width of the Brillouin scattering spectrum is set at 2MHz intervals, from 20MHz to 40MHz, with a total of 11 groups; the detection pulse width is set at 10ns intervals, from 10ns to 50ns, with a total of 5 groups; the detector bandwidth is fixed at 1.25GHz; the amplitude signal-to-noise ratio is set at 5dBm intervals, from 10dBm to 30dBm, with a total of 5 groups. A total of 124025 groups of characteristic parameters. Match the Brillouin frequency shift and Brillouin scattering spectrum of each group of data. Figure 2 The corresponding relationship between a set of Brillouin scattering spectra and Brillouin frequency shift is shown, where the x-axis coordinate position of the dotted line is the Brillouin frequency shift of 260MHz. Figure 3 As shown in the figure, a random phase sequence based on the Gaussian distribution model is generated, and for each set of Brillouin scattering spectra, the inverse Fourier transform is used to convert it into Figure 4 The Brillouin time-domain response signals with wide-spectrum characteristics are shown in Figure 2. A genetic algorithm was used to select 10 time-domain signal groups with amplitude fluctuations less than 50%. The generated time-domain waveforms were used as training samples, and the corresponding Brillouin frequency shifts were used as supervisory labels to construct a dataset of "wide-spectrum time-domain signal-Brillouin frequency shift" mapping relationships.
[0033] 2) The feature extraction network adopts a composite structure, such as Figure 5 The input layer receives a 450-point time domain signal generated by inverse Fourier transform of the 226-dimensional Brillouin scattering spectrum. After standardization (sample-by-sample mean variance normalization), it is sequentially processed through a one-dimensional convolutional layer (kernel size 5, number of channels 64, step size 2, Swish activation), a bidirectional long short-term memory network (BiLSTM, hidden unit 128) to extract time series features, and introduces a channel attention module (compression ratio 16) to enhance the weight distribution of key waveform areas; after feature fusion, it is compressed to 256 dimensions by global average pooling. The vector is then nonlinearly mapped through two fully connected layers (128 and 64 neurons, ReLU activation, and dropout rate of 0.3), and finally a linear output layer generates a 1D Brillouin frequency shift value. During the training process, a 5-fold cross-validation strategy is used, and the total sample set (137,775 groups of parameters × 10 time domain signals / group = 1,377,750 samples) is divided into a training set, a validation set, and a test set (ratio 7:2:1) according to the parameter group. The optimizer uses Adam (initial learning rate 0.001, weight decay 1×10 -5 ), the loss function uses Huber loss combined with L2 regularization (coefficient 1×10 -4), with an early stopping mechanism set to terminate training if the validation set loss did not decrease for 15 consecutive epochs. The batch size was fixed at 128, the training epoch limit was 500, and the gradient clipping threshold was set to 2.0 to ensure training stability. Network parameter optimization training was performed using a dataset mapping a "broad-spectrum time-domain signal to Brillouin frequency shift" relationship. The network weight parameters were optimized using a backpropagation algorithm, and a nonlinear mapping model from a single broad-spectrum Brillouin time-domain signal to a Brillouin frequency shift was established until the model reached the preset convergence criteria.
[0034] 3) Using a Brillouin optical time-domain reflectometer without an electrical filter, the center frequency of the RF signal source was fixed at 10.4 GHz, and a Brillouin frequency spectrum with an average of 4096 pulses was acquired over a 1000-meter length of bend-insensitive (G.657) optical fiber. The incident pulse width was set to 20 ns, corresponding to a theoretical spatial resolution of 2 meters. The photodetector directly measured a broad spectrum time-domain curve with 10,001 sampling points at a time-domain sampling rate of 1 GSample / s, including a fiber segment heated to 70°C.
[0035] 4) Through the artificial neural network trained in step 2), the distribution result of the temperature field can be obtained from the Brillouin frequency spectrum data obtained in step 3), and the demodulation result is as follows: Figure 6 As shown. After using the Brillouin signal demodulation method based on the spectrum inverse Fourier transform network of the present invention, the recognition accuracy of the temperature event segment is extremely high. When the pulse width is 20ns, the measurement uncertainty of the temperature is 0.3°C, which shows that the present invention can demodulate the distribution information of the temperature event segment with extremely high accuracy. Compared with the Brillouin optical time-domain reflection technology based on STFT under the same parameter settings, the present invention does not need to extract the Brillouin scattering spectrum from the wide-spectrum time-domain signal, and the physical quantity demodulation time can be almost ignored. In addition, compared with the swept-frequency Brillouin scattering spectrum obtained with a frequency sampling interval of 1MHz and a frequency sampling range of 200MHz, the present invention can reduce the measurement time to 1 / 200 of it.
[0036] The present invention aims to overcome the technical limitations of the efficiency superposition of the "measurement-processing" dual links in traditional Brillouin signal demodulation systems and construct an end-to-end intelligent demodulation architecture: to address the repeated measurement problem caused by the frequency scanning mechanism in existing technologies, a demodulation paradigm based on wide-spectrum single-shot acquisition is proposed. The nonlinear mapping between the time domain signal and the Brillouin frequency shift is directly established through the inverse Fourier transform neural network, eliminating the time redundancy caused by multi-frequency point scanning in traditional methods; to address the computational bottlenecks of time-frequency conversion and secondary demodulation, a composite neural network structure with time series feature extraction capabilities is designed to achieve real-time analysis of time domain signals; by deeply coupling the feature screening capabilities of the neural network with the hardware architecture, the function of the traditional electrical domain bandpass filtering module is replaced, simplifying the system complexity while improving signal quality, and ultimately forming a new demodulation system that integrates single-shot measurement, intelligent demodulation and lightweight hardware to meet the dual needs of real-time responsiveness and high reliability in the field of industrial monitoring.
[0037] The above description is only for the preferred embodiment of the present invention, which should not be understood as limiting the claims. Any equivalent process changes made using the present invention description are included in the patent protection scope of the present invention.
Claims
1. A Brillouin distributed optical fiber sensing demodulation method based on spectrum inverse Fourier transform, characterized by: The following steps are included: S1. Parameter Control and Dataset Construction: By modulating the external physical parameters applied to the optical fiber under test, the characteristic parameters of the sensor device are adjusted synchronously to establish a dynamic correlation between the Brillouin frequency shift and the Brillouin scattering spectrum; A random phase sequence is generated, and the time-frequency domain scattering spectrum is converted into a Brillouin time-domain response signal with wide-spectrum characteristics through inverse Fourier transform. From the generated time-domain waveform, an optimization algorithm is used to select time-domain signals with amplitude fluctuations less than 50% as training samples. The corresponding Brillouin frequency shift is used as the supervision label to construct a dataset of the wide-spectrum time-domain signal-Brillouin frequency shift mapping relationship. S2. Neural network modeling and training: building a deep neural network architecture and using the mapping relationship dataset to perform network parameter optimization training; The network weight parameters are optimized through the back-propagation algorithm to establish a nonlinear mapping model from wide-spectrum time-domain signals to Brillouin frequency shifts until the model reaches the preset convergence standard. S3. Signal acquisition and preprocessing: Using a hardware-simplified Brillouin optical time-domain reflectometry system, under the condition of a fixed RF signal source center frequency, the broadband time-domain response signal of the tested optical fiber at a specific scattering frequency point is acquired to complete signal preprocessing; S4. Physical quantity demodulation and reconstruction: The preprocessed time domain signal is input into a well-trained neural network model, and the corresponding Brillouin frequency shift distribution is output. The temperature field or strain field distribution of the measured optical fiber is reconstructed through the Brillouin frequency shift-physical quantity conversion algorithm.
2. The Brillouin distributed optical fiber sensing demodulation method based on spectrum inverse Fourier transform according to claim 1, characterized in that: In step S1, the external physical parameters include but are not limited to temperature field distribution, strain field distribution and their spatial coordinates.
3. The Brillouin distributed optical fiber sensing demodulation method based on spectrum inverse Fourier transform according to claim 1, characterized in that: In step S1, the characteristic parameters include Brillouin scattering spectrum linewidth, detection pulse width, photodetector bandwidth and signal-to-noise ratio threshold.
4. The Brillouin distributed optical fiber sensing demodulation method based on spectrum inverse Fourier transform according to claim 1, characterized in that: In step S1, each of the training samples is single-channel wide-spectrum Brillouin time-domain waveform data.
5. The Brillouin distributed optical fiber sensing demodulation method based on spectrum inverse Fourier transform according to claim 1, characterized in that: In step S1, the supervisory label is the Brillouin frequency shift feature of the corresponding time domain waveform.
6. The Brillouin distributed optical fiber sensing demodulation method based on spectrum inverse Fourier transform according to claim 1, characterized in that: In step S1, the optimization algorithm is a genetic algorithm.
7. The Brillouin distributed optical fiber sensing demodulation method based on spectrum inverse Fourier transform according to claim 1, characterized in that: In step S2, the deep neural network includes a signal input layer, a feature extraction layer and a regression output layer, wherein: the number of input layer nodes is consistent with the number of time domain signal sampling points, and the output layer nodes correspond to the Brillouin frequency shift prediction values; the feature extraction layer adopts a composite network structure, including a fully connected layer, a convolutional layer and an attention mechanism module; the training process adopts a k-fold cross-validation strategy, and the early stopping method is used to prevent overfitting.
8. The Brillouin distributed optical fiber sensing demodulation method based on spectrum inverse Fourier transform according to claim 1, characterized in that: In step S3, the simplified hardware system omits the electric domain bandpass filter module.
9. The Brillouin distributed optical fiber sensing demodulation method based on spectrum inverse Fourier transform according to claim 1, characterized in that: In step S3, the signal preprocessing is data normalization.