Optical vernier effect intensity demodulation method based on intelligent algorithm

Abstract

This invention discloses an optical vernier effect intensity demodulation method based on an intelligent algorithm. This method does not rely on expensive spectral analyzers during the demodulation stage; it only requires measuring the relative light intensity at the sampling wavelength. The neural network trained with calibration data then demodulates the target metric. The training data for the neural network is provided by the sensor calibration process. During calibration, the vernier effect spectrum varying with the target metric is first measured using a spectral analyzer. Then, the measured spectrum undergoes a Fast Fourier Transform, optimized spectral fitting, spectral reconstruction, and interpolation sequentially. Finally, wavelength sampling is performed on the reconstructed spectrum, and the relative light intensity ratios between different wavelengths are used as a dataset to train the neural network. The demodulation system consists of multiple DFB lasers of various wavelengths, a beam combiner, a 3dB coupler, a sensor, a signal photodetector, a reference photodetector, and a signal processing module. The wavelengths of the DFB lasers are consistent with the sampling wavelengths used in the calibration process. The driving circuit sequentially switches the DFB lasers of different wavelengths. The relative light intensity is measured by two photodetectors, and the relative light intensity ratios between different wavelengths are provided as input parameters to the intelligent algorithm in the signal processing module. The intelligent algorithm then calculates and outputs the target metric value. The demodulation stage of this invention does not require a spectral analyzer or spectral envelope extraction, and has the advantages of low cost, real-time demodulation, easy integration and convenient application.

Description

Technical Field

[0001] This invention relates to the field of fiber optic sensing, and in particular to a method for demodulating the intensity of the optical vernier effect using intelligent algorithms. Background Technology

[0002] Fiber optic sensors have experienced rapid development and widespread application due to their advantages such as small size, corrosion resistance, electromagnetic interference resistance, and radiation resistance. To improve sensitivity and resolution, the optical vernier effect has been extensively studied and applied in fiber optic sensors in recent years. The principle of the optical vernier effect is similar to that of a vernier caliper; by connecting two interferometers with similar free spectral ranges but different sensitivities to the measured quantity in series or parallel, the optical vernier effect can be generated. When a small change occurs in the measured quantity, the envelope of the superimposed spectrum of the two interferometers will shift significantly. After experimental calibration, the sensor can measure the measured parameter by detecting the change in the spectral envelope. Compared to a single interferometer, the optical vernier effect can improve the sensor's sensitivity by several times or even tens of times; generally, the greater the difference in sensitivity between the two sensors to the measured quantity, the higher the sensitivity.

[0003] Interferometers that generate the optical vernier effect mainly include the Michelson interferometer, Mach-Zehnder interferometer, Sagnac interferometer, and Fabry-Perot interferometer. Any suitable combination of two interferometers of the same or different types can produce the optical vernier effect. Although the structure and measured parameters of optical vernier effect sensors vary, existing demodulation methods are all based on wavelength demodulation. The basic principle is as follows: light emitted from a broadband light source passes through the sensor; the output spectrum is recorded using a spectrometer; the envelope of the spectrum is extracted through data processing on a computer; and demodulation is achieved through the functional relationship between the shift of the characteristic wavelength of the spectral envelope and the external measurand.

[0004] Existing optical vernier effect wavelength demodulation methods suffer from two main problems: first, the demodulation system relies on a spectrometer to measure the spectrum, resulting in high system costs; second, post-processing of the spectrum is required to extract the spectral envelope, making the demodulation process complex and difficult to integrate and implement in real time. These two drawbacks significantly limit the practical application of optical vernier effect sensors. Summary of the Invention

[0005] To overcome the problems of existing optical vernier effect demodulation methods, this invention proposes an optical vernier effect intensity demodulation method based on intelligent algorithms.

[0006] The technical solution adopted in this invention is: an optical vernier effect intensity demodulation method based on intelligent algorithms, the main steps of which are as follows: Figure 1As shown, the process includes: 1. Sensor spectral response measurement; 2. Extraction of interferometer characteristic parameters; 3. Spectral reconstruction and interpolation; 4. Wavelength sampling of the reconstructed spectrum; 5. Training the neural network; 6. Measuring relative light intensity; 7. Outputting the measurand. These seven steps can be divided into two stages: 1. Sensor calibration stage, including steps 1-4; 2. Sensor demodulation stage, including steps 5-7. The sensor calibration stage is responsible for collecting and generating the dataset required for training the neural network, and establishing the relationship between the relative light intensity ratio between sampling wavelengths and the measurand. The sensor demodulation stage measures the relative light intensity at the sampling wavelength and inputs the relative light intensity ratio between different wavelengths into the neural network, which then calculates and demodulates the measurand. Specifically, in step 1, a spectral analyzer records the vernier effect spectrum output by the sensor under different measurand effects. In step 2, a fast Fourier transform and optimal spectral fitting are performed on all the spectra recorded in step 1 to obtain the functional relationship between the interferometer optical path difference and the measurand. In step 3, the functional relationship between the optical path difference and the measurand obtained in step 2 is used to reconstruct the vernier effect and perform spectral interpolation. Step 4: Sample the light intensity of the spectrum generated in Step 3 at the sampling wavelength to obtain the functional relationship between the relative light intensity ratio between any two wavelengths and the optical path difference of the interferometer. Step 5: Train the neural network using the data generated in Step 4. After training the neural network using the sensor's calibration data, the demodulation stage only requires measuring the relative light intensity at the sampling wavelength. Inputting the relative light intensity ratio between different wavelengths into the neural network allows for the demodulation of the current optical path difference of the interferometer. The demodulated optical path difference can be used to reconstruct the vernier effect spectrum and evaluate the sensor's wavelength sensitivity. If it is not necessary to evaluate the sensor's sensitivity, the current value of the measurement can be directly obtained using the functional relationship between the optical path difference and the measurement.

[0007] Compared with existing technical solutions, the beneficial effects of this invention are: the demodulation system no longer relies on expensive spectral analyzers, but only needs to use a photodetector to measure the relative light intensity at the sampling wavelength, which can significantly reduce the cost of the demodulation system. In addition, the demodulation system does not need to extract the spectral envelope, but can directly calculate the measurement result by a trained neural network algorithm, and demodulation can be realized on embedded devices such as microcontrollers, improving the integration level and real-time demodulation capability of the demodulation system. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating the main implementation steps of the present invention.

[0009] Figure 2 This is a typical result obtained by fast Fourier transform of the vernier effect spectrum in an embodiment of the present invention.

[0010] Figure 3 This is a comparison between the measured spectrum and the reconstructed spectrum in an embodiment of the present invention.

[0011] Figure 4 This represents the optimized relationship between the optical path difference of the sensing cavity and the measurement target obtained in this embodiment of the invention.

[0012] Figure 5 This is a curve showing the relative light intensity ratio among the three sampling wavelengths in an embodiment of the present invention as a function of optical path difference.

[0013] Figure 6 This is a schematic diagram of a typical demodulation system in an embodiment of the present invention.

[0014] Figure 7 This is a comparison between the results obtained by demodulation using the present invention and the results obtained by the traditional wavelength demodulation method in the embodiments of the present invention.

[0015] Figure reference numerals: 1. Measurement of the sensor's spectral response; 2. Extraction of interferometer characteristic parameters; 3. Spectral reconstruction and interpolation; 4. Wavelength sampling of the reconstructed spectrum; 5. Training the neural network; 6. Measurement of relative light intensity; 7. Output of the measured quantity; 8. DFB-LD; 9. Beam combiner; 10. 3dB coupler; 11. Sensor; 12. Signal photodetector; 13. Reference photodetector; 14. Signal processing module. Detailed Implementation

[0016] The principles, implementation methods, and beneficial effects of the present invention will be described in detail below with reference to the embodiments.

[0017] The optical vernier effect spectrum is formed by superimposing the output spectra of two interferometers. Let the optical path difference between the resonant cavities of these two interferometers be Δ. L 1 and Δ L 2. The relative intensity spectrum of the sensor output after removing the DC component can be expressed as: (1) in λ The wavelength in a vacuum. A 1 and A 2 represents the intensity modulation coefficients of the two cavities, respectively. It can be seen that the wavelength... λ The relative light intensity of the output spectrum and four parameters ( A 1, Δ L 1, A 2, Δ L 2) Related. The optical path difference is proportional to the cavity length and the effective refractive index of the cavity medium. When the optical path difference changes due to the measurement, the intensity distribution of the output spectrum will change. Traditional wavelength demodulation methods achieve sensing by establishing the relationship between the wavelength shift of the spectral envelope and the measurement. However, essentially, the change in the spectral envelope is caused by the change in the optical path difference of the interferometer.

[0018] In this invention, during the sensor calibration stage: First, the optical path difference is calculated by performing a fast Fourier transform and spectral fitting on the measured spectrum, thereby calibrating the relationship between the optical path difference and the measured quantity; then, spectral interpolation is performed and the relative light intensity ratio between any two wavelengths is calculated by sampling at several specific wavelengths; finally, the relationship curve between the relative light intensity ratio between the sampled wavelengths and the optical path difference is used as a dataset to complete the training of the neural network.

[0019] In the demodulation stage of this invention: by measuring the relative light intensity at the sampling wavelength and using the relative light intensity ratio between different wavelengths as input parameters of the neural network, the measured quantity can be directly output after neural network processing. If necessary, the vernier effect spectrum can be easily reconstructed based on the relationship between the optical path difference and the measured quantity to meet the needs of sensor sensitivity evaluation.

[0020] The following is a further explanation with reference to an embodiment. This embodiment applies the invention to a temperature sensor formed by two fiber optic Fabry-Perot (FP) cavities connected in parallel. During the sensor calibration phase, one FP cavity (hereinafter referred to as the sensing cavity) is placed in a temperature-controlled chamber to sense temperature changes. The other FP cavity (hereinafter referred to as the reference cavity) is placed in a constant-temperature chamber to maintain a constant temperature. During the sensor calibration phase, light emitted from a broadband light source is sent to both sensors. The temperature is controlled by the temperature-controlled chamber, and the spectrum is recorded once every 2°C within the range of 26-44°C using a spectrometer, resulting in a total of 10 spectra.

[0021] First, a Fast Fourier Transform (FFT) is performed on the measured spectrum. Typical results are as follows: Figure 2 As shown. Based on the amplitude and frequency of the FFT curve, we can obtain ( A 1, Δ L 1, A 2, Δ L 2) The estimated value. From Figure 2 The results shown are as follows A 1≈0.65, A 2≈0.48, that is, corresponding to the amplitudes of the two peaks respectively. The optical path difference between the two cavities can be expressed by Δ L = λ ' 2 / F ,in λ ' is the average wavelength, F for Figure 2 The spatial frequencies of the two peaks shown. Figure 2 The results shown yield estimated values ​​Δ for the two optical path differences. L 1≈611.3μm, Δ L 2≈665.5μm.

[0022] Then, using the estimated values ​​as initial values, the precise values ​​of this set of parameters are obtained through spectral fitting. Spectral fitting minimizes the difference between the measured spectrum and the values ​​obtained using the parameter combination (…). A 1, Δ L 1, A 2, Δ L 2) The objective function is achieved by the mean square error between the reconstructed spectrum and the spectrum obtained by Equation 1. The optimization objective function is as follows: (2) in N The number of wavelength sampling points for the spectrometer is 5001 in this embodiment. This example uses a particle swarm optimization algorithm to solve this optimization problem, with 50 particles and a maximum of 5000 iterations. The optimized parameters are ( A 1 = 0.716, Δ L 1 = 608104.5nm A 2 = 0.584, Δ L 2=670245.5nm). The comparison between the spectrum reconstructed using these parameters and the measured spectrum is shown in the figure below. Figure 3 As shown, the measured spectrum can be well reproduced by reconstructing the spectrum, and the envelope characteristic wavelengths of the two are almost identical.

[0023] By performing the above processing on all 10 measured spectra and then linearly fitting the results, the functional relationship between the optical path difference of the sensing cavity and the measured quantity (temperature) can be obtained, such as... Figure 4 As shown in the figure. Within the sensor's measurement range, the optical path difference is interpolated at equal intervals according to this functional relationship to obtain the reconstructed spectra at different temperatures. In this embodiment, interpolation is performed in the range of 26-44℃, resulting in 501 reconstructed spectra at different temperatures. These spectra are sampled at three wavelengths: 1535.1nm, 1542.9nm, and 1554.9nm, to obtain the relative light intensity of each reconstructed spectrum at these three wavelengths. The relationship between the relative light intensity ratio among these three sampling wavelengths and the optical path difference of the sensing cavity is shown in the figure. Figure 5 As shown. This embodiment uses three sampling wavelengths, which can obtain three sets of data. The more sampling wavelengths, the larger the data volume and the higher the demodulation accuracy, but the complexity and cost of the demodulation system will also increase accordingly. To ensure sufficient accuracy, the number of sampling wavelengths should not be less than three. The selection of sampling wavelengths should be determined based on the envelope period of the vernier effect spectrum, and should not exceed one period of the envelope and should be distributed with equal wavelength intervals as much as possible. Using the relative intensity ratio between different wavelengths as a parameter can improve the anti-interference capability of the demodulation algorithm. In this example, the intelligent algorithm selected is a feedforward neural network, with... Figure 5The data shown is used as a dataset to train a feedforward neural network using the backpropagation algorithm. Experiments show that the optimal number of nodes in the hidden layer of the neural network is 6. After 1000 iterations, the mean square error of the objective function no longer decreases, indicating that the neural network training is complete.

[0024] A schematic diagram of the demodulation system is shown below. Figure 6 As shown, 8 represents three distributed feedback lasers (DFB-LDs) with wavelengths of 1535.1 nm, 1542.9 nm, and 1554.9 nm, respectively. The light from the three wavelengths is combined by a beam combiner 9 and then split into two paths by a 3dB coupler 10. One path is sent to a vernier-effect fiber optic sensor 11, and the other is sent to a reference photodetector 13. The light returning from the sensor 11 is again sent to a signal photodetector 12 via the 3dB coupler 10. During operation, the three DFB-LDs are switched on and off sequentially under circuit control. The relative light intensity at the three wavelengths can be calculated from the signals measured by the signal photodetector 12 and the reference photodetector 13. Finally, after processing by the neural network algorithm in the signal processing module 14, the measured value (temperature) can be directly obtained. The demodulation results of this invention are compared with those obtained by traditional wavelength demodulation methods. Figure 7 As shown. From Figure 7 (a) It can be seen that the sensor sensitivity obtained by this invention is 5.82 nm / ℃, while the sensitivity obtained by the traditional wavelength demodulation method is 5.86 nm / ℃, and the error between the two is negligible. From Figure 7 (b) It can be seen that the relative measurement errors of the two are almost identical. This embodiment fully demonstrates the beneficial effects of the present invention, namely: no expensive spectrometer is required during demodulation, which can significantly reduce the cost of the demodulation system; no spectral envelope extraction is required during demodulation, enabling real-time demodulation. The beneficial effects of the present invention are of great significance for promoting the practical application of optical vernier effect sensors.

[0025] This embodiment is only a partial implementation of the present invention, not all of it. The present invention proposes an intensity demodulation method for an optical vernier effect sensor based on an intelligent algorithm, independent of the sensor structure and the measurement being performed; the intensity demodulation system is not limited to three sampling wavelengths. All other implementations based on the embodiments of the present invention, without inventive effort, are within the protection scope of the present invention.

Claims

1. A method for demodulating the intensity of the optical vernier effect based on intelligent algorithms, characterized by: The system comprises two stages: calibration and demodulation. The calibration stage includes steps 1 (measuring the sensor's spectral response), 2 (extracting interferometer characteristic parameters), 3 (spectral reconstruction and interpolation), 4 (spectral wavelength sampling), and 5 (training the neural network). The demodulation stage includes steps 6 (measuring relative light intensity) and 7 (calculating and outputting the measured value via the neural network). The demodulation system consists of a DFB laser, a beam combiner, a 3dB coupler, a signal photodetector, a reference photodetector, and a signal processing module. It achieves this by measuring the relative light intensity at the sampled wavelength and using the relative intensity ratio between different wavelengths as a parameter input to the trained neural network, i.e., the intelligent algorithm. Demodulation of the target measurement: Step 1: Record the vernier effect spectrum output by the sensor under different target measurement conditions using a spectrometer; Step 2: Perform fast Fourier transform and optimal spectral fitting on all the spectra recorded in Step 1 to obtain the functional relationship between the interferometer optical path difference and the target measurement; Step 3: Reconstruct the vernier effect and perform spectral interpolation using the functional relationship between the optical path difference and the target measurement obtained in Step 2; Step 4: Sample the light intensity of the spectrum generated in Step 3 at the sampling wavelength to obtain the functional relationship between the relative light intensity ratio between any two wavelengths and the interferometer optical path difference; Step 5: Train the neural network using the data generated in Step 4.

2. The optical vernier effect intensity demodulation method based on intelligent algorithms according to claim 1, characterized in that: The intelligent algorithm is a feedforward neural network algorithm, with the relative light intensity ratio between sampling wavelengths as the input parameter and the optical path difference of the measured or sensing cavity as the output parameter.

3. The optical vernier effect intensity demodulation method based on intelligent algorithms according to claim 1, characterized in that: The wavelength of the DFB laser in the demodulation system is consistent with the sampling wavelength during the calibration process. The number of wavelengths is not less than 3. The wavelength distribution range is determined according to the period of the vernier effect spectral envelope, and does not exceed one period of the spectral envelope and is distributed with equal wavelength intervals as much as possible.

Images