Reference interferometer optical fiber hydrophone detection system and method based on LSTM neural network

Abstract

The invention provides a reference interferometer fiber hydrophone detection system and method based on an LSTM neural network. The system comprises a data acquisition module and a data demodulation module. The data acquisition module comprises a light source modulation unit, a signal interferometer and a reference interferometer. A light source is modulated through the light source modulation unit, light signals are output to the signal interferometer and the reference interferometer, and then interference signals are obtained through the signal interferometer and the reference interferometer respectively. Two paths of optical signals have a phase difference, so that working points are mismatched, and different nonlinear distortions are introduced. A reference interference signal is demodulated through a data demodulation module, additional nonlinear distortion introduced by working point mismatch is compensated through an LSTM neural network so as to obtain light source phase noise matched with a signal interferometer, and finally a de-noised optical phase signal is obtained through differential operation. According to the system provided by the invention, additional nonlinear distortion introduced by mismatching of the working point of the reference interferometer can be compensated, so that phase noise suppression which is more accurate than that of a traditional method is realized, and the noise suppression capability of the optical fiber hydrophone system is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of optical technology, and in particular to a reference interferometer fiber optic hydrophone detection system and method based on LSTM neural network. Background Art

[0002] The detection of underwater ultra-low-frequency (ULF) acoustic signals has important applications in marine science research, military defense, and environmental monitoring. Due to the propagation characteristics of ULF acoustic signals and the complex underwater environment, traditional hydrophones often face challenges with insufficient sensitivity and electromagnetic interference when detecting these signals. Fiber-optic hydrophones utilize fiber-optic sensing technology to achieve highly sensitive detection of underwater sound waves. Their optical properties enable them to operate over a wide frequency band, covering low- to high-frequency acoustic signals. Compared to traditional electroacoustic equipment, fiber-optic hydrophones utilize passive sensors that are unaffected by electromagnetic interference, ensuring stable signal transmission and reliable detection.

[0003] The fiber optic hydrophone's sensor probe is a complete interferometer. One arm of the Michelson interferometer serves as the sensing arm, and the other as the reference arm. When detecting underwater acoustic pressure signals, the acoustic wave signal acts on the sensing arm, which is wrapped around a sensitivity-enhancing structure. This causes the length of the optical fiber in the sensing arm to change, while the length of the optical fiber in the reference arm, which is wrapped around a rigid material, remains unchanged. Consequently, the phase difference between the two arms changes. The change in the optical fiber interferometer phase difference is proportional to the change in acoustic pressure, and the interferometer phase difference signal can be directly correlated to changes in underwater acoustic pressure.

[0004] Light source phase noise (light source phase noise refers to the random fluctuations in the phase of the light wave output by a laser or other coherent light source over time) is the main source of background noise in the fiber optic hydrophone system. The reference interferometer fiber optic hydrophone offsets the system's light source phase noise based on the principle that the optical path difference between the two interferometers is equal. However, in order to introduce a reference interferometer, an interferometric fiber optic coupler needs to be added. Due to the phase difference between the two output ports of the interferometric fiber optic coupler, there is a phase difference between the phase noise of the reference interferometer and the detection interferometer. This phase difference is superimposed on the time-varying operating point of the laser and passes through the unbalanced interferometer, which also has a time-varying characteristic. It is difficult to directly compensate for this phase offset using hardware, which ultimately affects the noise cancellation effect. Summary of the Invention

[0005] In order to solve the above problems, an embodiment of the present application provides a reference interferometer fiber optic hydrophone detection system based on an LSTM neural network, which can solve the above problems.

[0006] To this end, the following technical solutions are adopted in the embodiments of the present application:

[0007] The embodiment of the present application provides a reference interferometer fiber optic hydrophone detection system based on LSTM neural network, which includes: a data acquisition module and a data demodulation module. The data acquisition module is used to collect the interference signal generated by the fiber optic hydrophone during the detection process; wherein, the data acquisition module includes a light source modulation unit, a signal interferometer and a reference interferometer; the light source modulation unit includes a first fiber optic coupler, the light source modulation unit is used to modulate the light source, and output the modulated light signal through the first output end and the second output end of the first fiber optic coupler; wherein, due to its own properties, the light signals output by the two output ends have a phase difference; the signal interferometer is placed in an underwater acoustic environment, connected to the first output end of the first fiber optic coupler, and receives the modulated light signal; the signal interferometer is based on the acoustic-optical coupling mechanism, and phase modulates the light signal through the underwater acoustic signal to obtain a first interference signal; the reference interferometer is placed in a sound-proof and vibration-proof environment, connected to the second output end of the first fiber optic coupler, and receives the modulated light signal for detecting the phase noise of the light source homologous to the signal interferometer, and obtaining a second interference signal; the data demodulation module Block, the data demodulation module is used to: demodulate the first interference signal and the second interference signal to obtain the phase changes of the signal interferometer and the reference interferometer respectively; input the phase change of the reference interferometer into a pre-trained LSTM neural network to capture the change characteristics of the phase offset over time, and obtain the predicted result of the phase change of the reference interferometer; wherein the pre-trained LSTM neural network is trained with the phase change of the reference interferometer as input and the phase change of the signal interferometer as output; the pre-trained LSTM neural network is used to compensate for the additional nonlinear distortion introduced by the working point mismatch between the signal interferometer and the reference interferometer; wherein the working point mismatch is caused by the phase difference between the optical signals output from the two output ends of the first optical fiber coupler; the predicted result of the phase change of the reference interferometer is differentially processed with the phase change of the signal interferometer to eliminate the light source phase noise of the signal interferometer and obtain the denoised optical phase signal.

[0008] In one embodiment, during the detection process of the fiber optic hydrophone, the light source modulation unit is placed on the water surface.

[0009] In one embodiment, the light source modulation unit includes a signal generator, a laser, and a first fiber coupler; wherein the output end of the signal generator is connected to the modulation end of the laser; the output end of the laser is connected to the optical input port of the first fiber coupler; the signal generator modulates the output optical frequency of the laser by introducing a high-frequency carrier signal, and introduces an additional low-frequency sinusoidal signal to ensure the accuracy of the modulation depth estimation.

[0010] In one embodiment, during the signal modulation process performed by the signal generator, an additional low-frequency sinusoidal signal is introduced to ensure the accuracy of the modulation depth estimation.

[0011] In one embodiment, the signal interferometer includes a second fiber optic coupler, a first reference arm and a second sensing arm; the reference arm includes a first short optical fiber and a first Faraday rotator; the sensing arm includes a first long optical fiber and a second Faraday rotator; wherein the first light input port of the second fiber optic coupler is connected to the first light output port of the first fiber optic coupler; the first light output port of the second fiber optic coupler is connected to the first Faraday rotator via a first short optical fiber; the second light output port of the second fiber optic coupler is connected to the second Faraday rotator via the first long optical fiber; wherein the second fiber optic coupler is used to receive the light output by the first fiber optic coupler signal, and distributes the optical signal to the first Faraday rotator mirror and the second Faraday rotator mirror in a certain proportion; the first sensing arm is placed in an underwater acoustic environment, and the first reference arm is isolated from the underwater acoustic environment; the optical signals in the first reference arm and the first sensing arm return to their original paths after passing through the first Faraday rotator mirror and the second Faraday rotator mirror respectively, meet and interfere in the second optical fiber coupler, and generate an interference signal; wherein, in the process of detecting the underwater acoustic signal, the length of the first long optical fiber changes under the action of the underwater acoustic signal, the phase difference between the first reference arm and the first sensing arm changes, and the returned optical signals meet and interfere in the second optical fiber coupler to obtain a first interference signal.

[0012] In one embodiment, the reference interferometer includes a third fiber coupler, a second reference arm, and a second sensing arm; the second reference arm includes a second short optical fiber and a third Faraday rotator mirror; the second sensing arm includes a second long optical fiber and a fourth Faraday rotator mirror; wherein the first optical input port of the third fiber coupler is connected to the second optical output port of the first fiber coupler; the first optical output port of the third fiber coupler is connected to the third Faraday rotator mirror through the second short optical fiber; the second optical output port of the third fiber coupler is connected to the fourth Faraday rotator mirror through the second long optical fiber; wherein the third fiber coupler is used to receive the optical signal output by the first optical coupler and distribute the optical signal to the third Faraday rotator mirror and the fourth Faraday rotator mirror according to a certain ratio; in the process of detecting the phase noise of the light source homologous to the signal interferometer, the optical signals in the second reference arm and the second sensing arm respectively pass through the third Faraday rotator mirror and the fourth Faraday rotator mirror and then return to the original path, meet in the third fiber coupler and interfere to obtain a second interference signal.

[0013] In one embodiment, the data acquisition module further includes: a first detector for converting the first interference signal into an electrical signal; a second detector for converting the second interference signal into an electrical signal; and a data acquisition card for acquiring electrical signals corresponding to the first interference signal and the second interference signal.

[0014] In one embodiment, the second optical input port of the second optical fiber coupler is connected to the optical signal input port of the first detector; the second optical input port of the third optical fiber coupler is connected to the optical signal input port of the second detector; the electrical signal output port of the first detector is connected to the first electrical signal input port of the data acquisition card; the electrical signal output port of the second detector is connected to the second electrical signal input port of the data acquisition card; and the output end of the data acquisition card is connected to the input end of the data demodulation module.

[0015] In one embodiment, the process of demodulating the first interference signal and the second interference signal to obtain the phase change of the signal interferometer and the reference interferometer, respectively, specifically includes: performing baseband demodulation and double frequency demodulation on the first interference signal and the second interference signal, and then filtering out the high-frequency components in the first interference signal and the second interference signal by a low-pass filter; obtaining relevant parameters by an ellipse fitting algorithm; and calculating the phase change of the filtered signal interferometer and the reference interferometer by an inverse tangent operation method or a differential cross multiplication method.

[0016] An embodiment of the present application also provides a reference interferometer fiber optic hydrophone detection method based on an LSTM neural network, the method comprising: respectively collecting interference signals generated by a signal interferometer and a reference interferometer; demodulating the interference signals to obtain phase changes of the signal interferometer and the reference interferometer, respectively; inputting the phase change of the reference interferometer into the LSTM neural network to capture the time-varying characteristics of the phase offset and obtain a predicted result of the phase change of the reference interferometer; wherein the pre-trained LSTM neural network is trained with the phase change of the reference interferometer as input and the phase change of the signal interferometer as output; performing differential processing on the predicted result of the phase change of the reference interferometer and the phase change of the signal interferometer, and passing through a high-pass filter to obtain a denoised optical phase signal.

[0017] The reference interferometer fiber optic hydrophone detection system based on the LSTM neural network provided by the present invention realizes more accurate phase noise suppression than traditional methods by dynamically learning and compensating for the additional nonlinear distortion introduced by the reference interferometer operating point mismatch in real time through the LSTM neural network. This significantly improves the noise suppression capability of the fiber optic hydrophone system and the detection capability of ultra-low frequency underwater acoustic signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The following is a brief introduction to the drawings required for describing the embodiments or prior art.

[0019] Figure 1 This is a schematic diagram of the structure of a reference interferometer fiber optic hydrophone detection system based on an LSTM neural network provided in an embodiment of the present application;

[0020] Figure 2 A schematic diagram of a data demodulation process provided in an embodiment of the present application;

[0021] Figure 3 This is another data demodulation process diagram provided in an embodiment of the present application;

[0022] Figure 4 A phase demodulation result comparison diagram provided in an embodiment of the present application;

[0023] Figure 5 A phase noise power spectrum diagram provided in an embodiment of the present application;

[0024] Figure 6 This is a flow chart of a reference interferometer fiber optic hydrophone detection method based on an LSTM neural network provided in an embodiment of the present application. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0026] In the description of this application, the terms "center", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings. They are only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, they should not be understood as limiting this application.

[0027] The terms "first" and "second" in this specification and claims are used to distinguish between different objects, rather than to describe a specific order of objects. For example, "a first operational amplifier" and "a second operational amplifier" are used to distinguish between different operational amplifiers, rather than to describe a specific order of operational amplifiers.

[0028] In the description of this application, it should be noted that, unless otherwise expressly specified or limited, the terms "installed," "connected," and "connected" should be understood in a broad sense, and may refer to, for example, a fixed connection, a detachable connection, a conflicting connection, or an integral connection. A person skilled in the art can understand the specific meanings of the above terms in this application based on the specific circumstances. In the embodiments of this application, "contact" or "coupling" may refer to direct contact between components or contact between components through an adhesive or thermally conductive adhesive.

[0029] In the description of this specification, specific features, structures, materials or characteristics may be combined in an appropriate manner in any one or more embodiments or examples.

[0030] Figure 1 This is a schematic diagram of the structure of a reference interferometer fiber optic hydrophone detection system based on an LSTM neural network provided in an embodiment of the present application. Figure 1 As shown, the reference interferometer fiber optic hydrophone detection system based on the LSTM neural network includes a data acquisition module 100 and a data demodulation module 200.

[0031] The data acquisition module 100 includes a light source modulation unit 103, a signal interferometer 101, and a reference interferometer 102. Specifically, the data acquisition module 100 is used to acquire interference signals generated by the fiber optic hydrophone during detection.

[0032] Furthermore, the light source modulation unit 103 is used to modulate the light source. Specifically, the light source modulation unit 103 includes a first optical fiber coupler 3 , and outputs a modulated optical signal through a first output terminal and a second output terminal of the optical fiber coupler 3 .

[0033] Specifically, due to the inherent properties of the fiber coupler 3, there is a phase difference between the optical signals output from its first and second output ends. For example, a 50:50 or 90:10 splitting ratio will introduce a phase difference of π / 2 between the two output ends. Geometric asymmetry (such as core offset and bending), material inhomogeneities, and path length differences in the coupler can also introduce additional phase differences. This phase difference can cause the interferometer system to deviate from its optimal operating point, resulting in a mismatch between the operating points of the signal interferometer and the reference interferometer. This shifts the system's response to small phase changes from the linear region (sensitive region) to the nonlinear region, thereby introducing nonlinear distortion.

[0034] Signal interferometer 101 is connected to the first output end of first fiber coupler 3 and receives the modulated optical signal. Signal interferometer 101 can be placed in an underwater acoustic environment to detect underwater acoustic signals using optical signals and obtain an interference signal. Based on an acousto-optic coupling mechanism, the signal interferometer phase modulates the optical signal with the underwater acoustic signal to obtain a first interference signal, which can reflect changes in sound pressure.

[0035] Reference interferometer 102 is connected to the second output end of first fiber coupler 3 and receives the modulated optical signal. Reference interferometer 102 can be placed in a sound-proof and vibration-proof environment to detect the phase noise of the light source cognate with signal interferometer 101 through the optical signal, and obtain a second interference signal, which can reflect the noise variation between reference interferometer 102 and signal interferometer 101.

[0036] The data demodulation module 200 utilizes a pre-trained LSTM neural network to compensate for the additional nonlinear distortion introduced by the operating point mismatch between the signal interferometer 101 and the reference interferometer 102. This operating point mismatch is caused by the phase difference between the optical signals output from the two output ends of the first fiber coupler. Specifically, the data demodulation module 200 can achieve the following:

[0037] The first interference signal and the second interference signal are demodulated to obtain phase changes of the signal interferometer and the reference interferometer respectively.

[0038] The phase change of the reference interferometer is input into a pre-trained LSTM neural network, which captures the temporal characteristics of the phase offset and produces a predicted phase change of the reference interferometer. Specifically, the predicted phase change of the reference interferometer represents the phase noise of the light source that matches the signal interferometer. In other words, the phase change of the reference interferometer can be predicted using a pre-trained LSTM neural network, thereby obtaining the phase noise of the light source that matches the signal interferometer.

[0039] A pre-trained LSTM neural network can be obtained by training the LSTM neural network model. Specifically, the phase change of the reference interferometer can be used as the input label of the LSTM neural network model, and the phase change of the signal interferometer can be used as the output label of the LSTM neural network model. Before model training, the training data can be pre-processed, such as through noise reduction, normalization, segmented sampling, and data set partitioning. During the training process, the model can be continuously optimized and adjusted using a pre-designed loss function until a trained LSTM neural network model is obtained.

[0040] Finally, the predicted result of the phase change of the reference interferometer can be differentially processed with the phase change of the signal interferometer to eliminate the light source phase noise of the signal interferometer and finally obtain the denoised optical phase signal.

[0041] It can be seen from this embodiment that by implementing LSTM real-time dynamic learning and prediction of time-varying phase offsets through the data demodulation module 200, a more matched light source phase noise can be obtained, thereby achieving more accurate phase noise suppression, which can significantly improve the noise suppression capability of the fiber optic hydrophone system and improve the detection capability of ultra-low frequency underwater acoustic signals.

[0042] In one embodiment, during the detection process of the fiber optic hydrophone, the light source modulation unit 103 is placed on the water surface.

[0043] It's easy to understand that placing the light source modulation unit 103 on the water surface completely eliminates the need to power the hydrophone's wet-end. By applying an additional low-frequency sinusoidal signal from the laser 2 via a signal generator 1 at the dry end (on the water surface), rather than applying an additional low-frequency sinusoidal signal to the wet-end interferometer via active components, the need for powering the wet-end is eliminated, improving device safety and further reducing manufacturing and operating costs.

[0044] In one embodiment, continue with reference to Figure 1 The light source modulation unit 103 further includes a signal generator 1 and a laser 2.

[0045] Specifically, the output end of signal generator 1 is connected to the modulation end of laser 2. The output end of laser 2 is connected to the optical input port of first fiber coupler 3. Signal generator 1 can generate an electrical signal to modulate the output characteristics of laser 2. Laser 2 then converts the electrical signal generated by signal generator 1 into an optical signal and distributes the output optical signal to first fiber coupler 3.

[0046] In addition, the signal generator 1 modulates the output optical frequency of the laser 2 by introducing a high-frequency carrier signal.

[0047] It is worth noting that the signal generator 1 is an electronic test instrument that can generate user-defined waveforms. It can generate waveforms of almost any shape, including complex modulation signals, pulse sequences, noise signals, etc.

[0048] Based on the previous embodiment, an additional low-frequency sinusoidal signal can optionally be introduced during the signal modulation process of the signal generator 1 to ensure the accuracy of the modulation depth estimation. The frequency of the additional low-frequency sinusoidal signal can be much lower than the frequency band of the signal to be measured, for example, the frequency of the additional low-frequency sinusoidal signal can be 5 Hz and the frequency of the optical signal can be 1000 Hz; the frequency of the additional low-frequency sinusoidal signal can be 3 Hz and the frequency of the optical signal can be 800 Hz, etc.

[0049] Based on the previous embodiment, optionally, after the data demodulation module 200 performs differential processing on the predicted result of the phase change of the reference interferometer and the phase change of the signal interferometer, the additional low-frequency sinusoidal signal introduced by the light source modulation unit 103 or the signal generator 1 can be filtered out by a high-pass filter to eliminate the interference of the additional low-frequency sinusoidal signal on the optical phase signal.

[0050] In one embodiment, preferably, the laser 2 is a single-frequency narrow-linewidth fiber laser. A single-frequency narrow-linewidth fiber laser is a high-performance laser light source whose core features are extremely high frequency purity (single frequency) of the output laser and extremely narrow spectral linewidth.

[0051] In one embodiment, preferably, the first optical fiber coupler 3 is a 1*2 coupler with a splitting ratio of 50:50.

[0052] It's worth noting that the phase difference between the two output ports of the 1x2 fiber coupler causes a phase difference between the phase noise of the reference interferometer and the detection interferometer. This phase difference is determined by the physical structure of the coupler (such as fiber spacing and coupling length) and the propagation characteristics of the light field. During the coupling process, the interaction of the evanescent waves of the light field causes a phase delay, resulting in a fixed phase difference.

[0053] In one embodiment, continue with reference to Figure 1 The signal interferometer 101 includes a second fiber coupler 4 and a first reference arm 20 and a second sensor arm 30. The first reference arm 20 includes a first short optical fiber 16 and a first Faraday rotator mirror 6. The second sensor arm 30 includes a first long optical fiber 5 and a second Faraday rotator mirror 7.

[0054] Specifically, the second optical input port of the second fiber coupler 4 is connected to the first optical output port of the first fiber coupler 3. The first optical output port of the second fiber coupler 4 is connected to the first Faraday rotator 6 via a first short optical fiber 16. The second optical output port of the second fiber coupler 4 is connected to the second Faraday rotator 7 via a first long optical fiber 5.

[0055] Furthermore, the operating principle of the signal interferometer 101 is as follows: the first sensor arm 20 is placed in an underwater acoustic environment, while the first reference arm 30 is isolated from the environment. The optical signals in the first reference arm 20 and the first sensor arm 30, respectively, pass through the first Faraday rotator mirror 6 and the second Faraday rotator mirror 7, then return along their original paths. They meet and interfere within the second fiber coupler 4, generating an interference signal. Furthermore, during the detection of the underwater acoustic signal, the length of the first long optical fiber 5 changes under the influence of the underwater acoustic signal, altering the phase difference between the first reference arm 20 and the first sensor arm 30. The returning optical signals then meet and interfere within the second fiber coupler 4, generating a first interference signal.

[0056] In one embodiment, the second fiber coupler 4 is preferably a 2*2 coupler with a splitting ratio of 50:50. The first Faraday rotator 6 is preferably a 45° Faraday rotator. The second Faraday rotator 7 is preferably a 45° Faraday rotator.

[0057] In one embodiment, the reference interferometer 102 includes a third fiber coupler 8 , a second long optical fiber 9 , a second short optical fiber 17 , a third Faraday rotator mirror 10 , and a fourth Faraday rotator mirror 11 .

[0058] The reference interferometer 102 includes a third fiber coupler 8, a second reference arm 40, and a second sensing arm 50. The second reference arm 40 includes a second short fiber 17 and a third Faraday rotator mirror 10. The second sensing arm 50 includes a second long fiber 9 and a fourth Faraday rotator mirror 11.

[0059] Specifically, the first optical input port of the third fiber coupler 8 is connected to the second optical output port of the first fiber coupler 3. The first optical output port of the third fiber coupler 8 is connected to the third Faraday rotator 10 via a second short optical fiber 17. The second optical output port of the third fiber coupler 8 is connected to the fourth Faraday rotator 11 via a second long optical fiber 9.

[0060] Furthermore, the operating principle of the reference interferometer 102 is as follows: the third fiber coupler 8 is used to receive the optical signal output by the first optical coupler 3 and distribute the optical signal in a certain ratio to the third Faraday rotator mirror 10 and the fourth Faraday rotator mirror 11. During the detection of phase noise from the light source co-originating with the signal interferometer 101, the optical signals in the second reference arm 40 and the second sensor arm 50 respectively pass through the third Faraday rotator mirror 10 and the fourth Faraday rotator mirror 11 and then return along their original paths. They meet and interfere in the third fiber coupler 8 to obtain a second interference signal.

[0061] In one embodiment, the third fiber coupler 8 is preferably a 2*2 coupler with a splitting ratio of 50:50. The third Faraday rotator mirror 10 is preferably a 45° Faraday rotator mirror. The fourth Faraday rotator mirror 11 is preferably a 45° Faraday rotator mirror.

[0062] In one embodiment, continue with reference to Figure 1 The data acquisition module 100 also includes a first detector 12 , a second detector 13 , and a data acquisition card 14 .

[0063] Specifically, the interference signal input from the signal interferometer 101 can be converted into an electrical signal by the first detector 12. The interference signal input from the reference interferometer 102 can be converted into an electrical signal by the second detector 13. The data acquisition card 14 receives the electrical signals input from the first detector 12 and the second detector 13, respectively, thereby obtaining the interference signals corresponding to the signal interferometer 101 and the reference interferometer 102.

[0064] In one embodiment, continue with reference to Figure 1 The first optical input port of the second optical fiber coupler 4 is connected to the optical signal input port of the first detector 12. The second optical input port of the third optical fiber coupler 8 is connected to the optical signal input port of the second detector 13. The electrical signal output port of the first detector 12 is connected to the first electrical signal input port of the data acquisition card 14. The electrical signal output port of the second detector 12 is connected to the second electrical signal input port of the data acquisition card 14. The output end of the data acquisition card 14 is connected to the input end of the data demodulation module 200.

[0065] Furthermore, the interference signal generated by the signal interferometer 101 can be output from another input port of the second fiber coupler 4, converted into an electrical signal by the first detector 12, then collected by the data acquisition card 14, and demodulated by the data demodulation module 200. The interference signal generated by the reference interferometer 102 is output from an input port of the third fiber coupler 8, undergoes photoelectric conversion by the second detector 13, and finally enters the data demodulation module 200 for demodulation.

[0066] In one embodiment, the first detector 12 and the second detector 13 preferably have the same detection bandwidth, and the detection bandwidths of the first detector 12 and the second detector 13 satisfy the Nyquist sampling theorem. The Nyquist sampling theorem states that to reconstruct the original continuous signal without distortion, the sampling frequency must be at least twice the highest frequency of the signal.

[0067] In one embodiment, the present solution further provides a method for demodulating the first interference signal and the second interference signal. This embodiment may include the following steps:

[0068] Step 1: perform baseband demodulation and double frequency demodulation on the first interference signal and the second interference signal, and then filter out high-frequency components in the first interference signal and the second interference signal through a low-pass filter.

[0069] Exemplary, reference Figure 2 , the first interference signal I generated by the signal interferometer s Expressed as:

[0070]

[0071] The second interference signal I generated by the reference interferometer r It can be expressed as:

[0072]

[0073] Where A s 、A r is the DC light intensity, B s 、B r is the interference light intensity, C s 、C r is the phase modulation depth, ω0 is the carrier angular frequency, is the optical phase signal, is the light source phase noise, To introduce an additional low-frequency sinusoidal signal, the phase difference between the two output ports of the first fiber coupler 3 acts on the unbalanced interferometer to generate an additional working point offset.

[0074] to I s , I r Perform Bessel function expansion, I s , I r Expressed as:

[0075]

[0076] The first interference signal is demodulated at the base frequency and demodulated at the double frequency by the second interference signal, and then the interference signal is filtered by a low-pass filter. s , the second interference signal I r Multiply them with the baseband signal Gcos(ω0t) and the double frequency signal Hcos(2ω0t) respectively, and then pass them through a low-pass filter with a cutoff frequency of 1 / 2 times the modulation frequency to obtain the output S S1 、S S2 、S R1 、S R2 , expressed as:

[0077]

[0078] Step 2: Obtain relevant parameters by ellipse fitting algorithm.

[0079] Exemplary, reference Figure 2 , by ellipse fitting algorithm (EFA), S S1 、S S2 and S R1 、S R2 An elliptic curve can be formed, which is expressed as:

[0080]

[0081] Where, The relevant parameters J2(C s ) / J1(C s )、J2(C r ) / J1(C r ), Expressed as:

[0082]

[0083] Step 3: Calculate the phase change corresponding to the filtered interference signal by using an inverse tangent operation method or a differential cross multiplication method.

[0084] Exemplary, reference Figure 2 , by taking the filtered output S S1 、S S2 and S R1 、S R2 Divide them two by two and perform inverse tangent operation, then calculate the relevant parameters J2(C s ) / J1(C s )、J2(C r ) / J1(C r ) is substituted into the equation, and the phase change of the two interferometers is obtained, which is expressed as:

[0085]

[0086] refer to Figure 3 , or the filtered output S S1 、S S2 and S R1 、S R2 Cross-multiplication, subtraction and integration of two differentials, and then the relevant parameters Substituting, we can also get the phase change of the two interferometers, which is expressed as:

[0087]

[0088] In one embodiment, in order to more clearly illustrate the specific process of predicting the phase change of the reference interferometer through the pre-trained LSTM neural network, a specific example is used for illustration. In this example:

[0089] Specifically, the reference interferometer phase change (which can be obtained using the To represent) input into the LSTM neural network model, the implementation of LSTM (Long Short-Term Memory) includes three control gates: forget gate F t , update gate U t , output gate O t,The three control gates jointly manage the information flow and retain relevant historical data.

[0090] The forget gate is responsible for deciding which information in the memory cell state should be discarded. To do this, it takes the current input X t and the hidden layer state l t-1 , use the sigmoid function to process it and generate a value between 0 and 1. When the value is 0, it means that the information is completely discarded, and when the value is 1, it means that the information is completely retained. t It can be expressed as:

[0091] F t =σ(W f ·[x t ,l t-1 ]+p f )

[0092] Where W f is the weight matrix of the forget gate, p f is the forget gate bias. The update gate uses the same method as the forget gate to identify new information that needs to be incorporated into the memory cell state.

[0093] Calculate the candidate memory cell value through the hyperbolic tangent function tanh Its range is -1 to 1.

[0094]

[0095] Set the old memory cell value c s-1 With the forget gate F t Multiply to discard old information that is no longer useful. With update gate U t Multiply to achieve different degrees of update of the current memory cell value, then the current memory cell value c s Expressed as:

[0096]

[0097] The value of the output gate O t The same as the forget gate operation principle, by calculating the value of the output gate to determine which part of the value to output. The current memory cell value is mapped to the range of -1 to 1 through the tanh activation function, and then the two are multiplied to obtain the current network state value l t :

[0098] l t =O t tanh(c s )

[0099] The network output at the current moment is y t Expressed as:

[0100] y t =tanh(l t )

[0101] The characteristics of phase shift changing with time are obtained through the pre-trained LSTM neural network and an accurate prediction is made. The output result of the pre-trained LSTM neural network is differentially processed with the demodulation result of the signal interferometer (that is, the phase change of the reference interferometer predicted by the pre-trained LSTM neural network is subtracted from the phase change of the signal interferometer). Optionally, the additional low-frequency sinusoidal signal introduced by the light source modulation unit 103 or the signal generator 1 can be filtered out through a high-pass filter to obtain the denoised optical phase signal.

[0102] This example also provides the comparison results of LSTM neural network and other algorithms applied to this solution. Figure 4 The time domain signals of different algorithms before high-pass filtering, i.e., the low-frequency reference signals, are compared. The blue line represents the output of the signal interferometer, the red line represents the output of the reference interferometer without LSTM neural network compensation, and the yellow line represents the output after LSTM neural network compensation. The experimental results show that after the LSTM neural network compensates for the additional nonlinear distortion introduced by the operating point mismatch, the output of the reference interferometer approaches the output of the signal interferometer, with the yellow and blue lines coinciding, demonstrating the significant effectiveness of the LSTM neural network compensation.

[0103] Further, Figure 5 The spectra of the residual system noise after high-pass filtering and noise removal are also compared. The blue line represents the self-noise of the signal interferometer. The red line represents the residual system noise of the reference interferometer before LSTM neural network compensation, while the yellow line represents the residual system noise after LSTM neural network compensation. The experimental results show that the residual noise energy after LSTM neural network compensation is significantly lower than the other two cases, with the yellow line showing the lowest residual noise.

[0104] As can be seen from this embodiment, since the phase offset introduced by the interferometric fiber coupler varies with the time-varying operating point of the laser, traditional hardware methods are difficult to directly compensate for this change. However, the time series modeling capability of LSTM can automatically learn and compensate for this time-varying characteristic, overcoming the limitations of hardware methods.

[0105] Figure 6 A flow chart of a reference interferometer fiber optic hydrophone detection method based on an LSTM neural network provided by the present invention is shown. The reference interferometer fiber optic hydrophone detection method based on an LSTM neural network provided by the present invention comprises the following steps:

[0106] Step S101 : collecting interference signals generated by the signal interferometer and the reference interferometer respectively.

[0107] Step S102 : Demodulate the interference signal to obtain phase changes of the signal interferometer and the reference interferometer, respectively.

[0108] In step S103, the phase change of the reference interferometer is input into a pre-trained LSTM neural network to capture the temporal characteristics of the phase offset and obtain a predicted result for the phase change of the reference interferometer. Specifically, the pre-trained LSTM neural network is trained using the phase change of the reference interferometer as input and the phase change of the signal interferometer as output.

[0109] Step S104 : performing differential processing on the predicted result of the phase variation of the reference interferometer and the phase variation of the signal interferometer to obtain a denoised optical phase signal.

[0110] The types, quantities, shapes, installation methods, and structures of the components of the LSTM neural network-based reference interferometer fiber-optic hydrophone detection system provided in the embodiments of this application are not limited to those described above. Any technical solution implemented under the principles of this application is within the scope of protection of this solution. Any technical solution, whether combined in a suitable manner, with one or more embodiments or illustrations in this specification is within the scope of protection of this solution.

[0111] Finally, it should be noted that the above embodiments are merely illustrative of the technical solutions of the present application. Those skilled in the art should understand that, although the present application has been described in detail with reference to the aforementioned embodiments, the technical solutions described in the aforementioned embodiments may be modified or some of the technical features thereof may be replaced with equivalents. However, such modifications or replacements do not deviate from the spirit and scope of the technical solutions in the various embodiments of the present application.

Claims

1. A reference interferometer fiber optic hydrophone detection system based on LSTM neural network, characterized in that: include: A data acquisition module is used to collect interference signals generated by the optical fiber hydrophone during the detection process; wherein the data acquisition module includes: A light source modulation unit, comprising a first optical fiber coupler, configured to modulate a light source and output a modulated optical signal through a first output end and a second output end of the first optical fiber coupler; wherein, due to the nature of the first optical fiber coupler, the optical signals outputted from the two output ends have a phase difference; a signal interferometer placed in an underwater acoustic environment, connected to the first output end of the first optical fiber coupler, and receiving the modulated optical signal; the signal interferometer, based on an acousto-optic coupling mechanism, phase-modulates the optical signal using the underwater acoustic signal to obtain a first interference signal; a reference interferometer, placed in a sound-proof and vibration-proof environment, connected to the second output end of the first optical fiber coupler, receiving the modulated optical signal, for detecting the phase noise of the light source cognate with the signal interferometer, and obtaining a second interference signal; A data demodulation module, wherein the data demodulation module is used to: Demodulating the first interference signal and the second interference signal to obtain phase changes of the signal interferometer and the reference interferometer respectively; Inputting the phase change of the reference interferometer into a pre-trained LSTM neural network to capture the time-varying characteristics of the phase offset and obtain a predicted result of the phase change of the reference interferometer; wherein the pre-trained LSTM neural network is trained using the phase change of the reference interferometer as input and the phase change of the signal interferometer as output; the pre-trained LSTM neural network is used to compensate for additional nonlinear distortion introduced by an operating point mismatch between the signal interferometer and the reference interferometer; wherein the operating point mismatch is caused by a phase difference between the optical signals output from the two output ends of the first optical fiber coupler; The predicted result of the phase change of the reference interferometer is differentially processed with the phase change of the signal interferometer to eliminate the light source phase noise of the signal interferometer and obtain the de-noised optical phase signal.

2. The reference interferometer fiber optic hydrophone detection system based on LSTM neural network according to claim 1 is characterized in that: During the detection process of the optical fiber hydrophone, the light source modulation unit is placed on the water surface.

3. The reference interferometer fiber optic hydrophone detection system based on LSTM neural network according to claim 1 is characterized in that: The light source modulation unit also includes a signal generator and a laser; wherein the output end of the signal generator is connected to the modulation end of the laser; the output end of the laser is connected to the optical input port of the first optical fiber coupler; the signal generator modulates the output optical frequency of the laser by introducing a high-frequency carrier signal.

4. The reference interferometer fiber optic hydrophone detection system based on LSTM neural network according to claim 3 is characterized in that: During the signal modulation process of the signal generator, an additional low-frequency sinusoidal signal is introduced to ensure the accuracy of the modulation depth estimation.

5. The reference interferometer fiber optic hydrophone detection system based on LSTM neural network according to claim 3 is characterized in that: The signal interferometer includes a second fiber coupler, a first reference arm, and a second sensing arm; the reference arm includes a first short optical fiber and a first Faraday rotator mirror; the sensing arm includes a first long optical fiber and a second Faraday rotator mirror; wherein the first light input port of the second fiber coupler is connected to the first light output port of the first fiber coupler; the first light output port of the second fiber coupler is connected to the first Faraday rotator mirror via a first short optical fiber; and the second light output port of the second fiber coupler is connected to the second Faraday rotator mirror via the first long optical fiber. The second optical fiber coupler is used to receive the optical signal output by the first optical fiber coupler and distribute the optical signal to the first Faraday rotator mirror and the second Faraday rotator mirror in a certain proportion; the first sensor arm is placed in an underwater acoustic environment, and the first reference arm is isolated from the underwater acoustic environment; the optical signals in the first reference arm and the first sensor arm return to their original paths after passing through the first Faraday rotator mirror and the second Faraday rotator mirror respectively, meet and interfere in the second optical fiber coupler, and generate an interference signal; in the process of detecting the underwater acoustic signal, the length of the first long optical fiber changes under the action of the underwater acoustic signal, the phase difference between the first reference arm and the first sensor arm changes, and the returned optical signals meet and interfere in the second optical fiber coupler to obtain a first interference signal.

6. The reference interferometer fiber optic hydrophone detection system based on LSTM neural network according to claim 3 is characterized in that: The reference interferometer includes a third fiber coupler, a second reference arm, and a second sensing arm; the second reference arm includes a second short optical fiber and a third Faraday rotator mirror; the second sensing arm includes a second long optical fiber and a fourth Faraday rotator mirror; wherein the first light input port of the third fiber coupler is connected to the second light output port of the first fiber coupler; the first light output port of the third fiber coupler is connected to the third Faraday rotator mirror via the second short optical fiber; and the second light output port of the third fiber coupler is connected to the fourth Faraday rotator mirror via the second long optical fiber. The third optical fiber coupler is used to receive the optical signal output by the first optical fiber coupler and distribute the optical signal to the third Faraday rotator mirror and the fourth Faraday rotator mirror in a certain proportion. In the process of detecting the phase noise of the light source homologous to the signal interferometer, the optical signals in the second reference arm and the second sensor arm respectively pass through the third Faraday rotator mirror and the fourth Faraday rotator mirror and then return to the original path, meet in the third optical fiber coupler and interfere to obtain a second interference signal.

7. The reference interferometer fiber optic hydrophone detection system based on LSTM neural network according to claim 1, characterized in that: The data acquisition module also includes: a first detector, configured to convert the first interference signal into an electrical signal; a second detector, configured to convert the second interference signal into an electrical signal; The data acquisition card is used to acquire electrical signals corresponding to the first interference signal and the second interference signal.

8. The reference interferometer fiber optic hydrophone detection system based on LSTM neural network according to claims 4-6, characterized in that: The second optical input port of the second optical fiber coupler is connected to the optical signal input port of the first detector; the second optical input port of the third optical fiber coupler is connected to the optical signal input port of the second detector; The electrical signal output port of the first detector is connected to the first electrical signal input port of the data acquisition card; The electrical signal output port of the second detector is connected to the second electrical signal input port of the data acquisition card; the output end of the data acquisition card is connected to the input end of the data demodulation module.

9. The reference interferometer fiber optic hydrophone detection system based on LSTM neural network according to claim 1, characterized in that: The process of demodulating the first interference signal and the second interference signal to obtain phase changes of the signal interferometer and the reference interferometer, respectively, specifically includes: Performing baseband demodulation and double frequency demodulation on the first interference signal and the second interference signal, and then filtering out high-frequency components in the first interference signal and the second interference signal through a low-pass filter; The relevant parameters are obtained by the ellipse fitting algorithm; The phase changes of the filtered signal interferometer and the reference interferometer are calculated by using an inverse tangent operation method or a differential cross multiplication method.

10. A reference interferometer fiber optic hydrophone detection method based on LSTM neural network, characterized in that: The method includes: respectively collecting interference signals generated by the signal interferometer and the reference interferometer; Demodulating the interference signal to obtain phase changes of the signal interferometer and the reference interferometer respectively; Inputting the phase change of the reference interferometer into a pre-trained LSTM neural network to capture the time-varying characteristics of the phase offset and obtain a prediction result of the phase change of the reference interferometer; the pre-trained LSTM neural network is trained with the phase change of the reference interferometer as input and the phase change of the signal interferometer as output; The predicted result of the phase variation of the reference interferometer is differentially processed with the phase variation of the signal interferometer to obtain a de-noised optical phase signal.

Images