A DAS-based indoor covert listening system

Abstract

The application discloses a kind of indoor concealed listening system based on DAS, belong to optical fiber acoustic wave sensing technical field, including sequentially connected DAS optical path, signal acquisition demodulation module integrated in FPGA end and noise reduction module integrated in PC end, DAS optical path is coupled with the reference light after attenuator with backscattering light of acoustic wave information, signal acquisition demodulation module is used to receive the electrical signal after AD module, IQ demodulation module, phase difference module and phase unwrapping module to obtain amplitude data and real phase data, noise reduction module is used to audio data noise reduction enhancement, realize sound play and phase diagram, spectrum diagram display;The scheme provided by the application effectively improves the definition of system sensing sound restoration, can acquire and process sound signal in real time, meet the application requirement of indoor concealed listening in actual scene, provide reliable support for important place safety monitoring, ruins rescue and the like.

Description

Technical Field

[0001] This invention relates to the field of distributed fiber optic acoustic wave sensing technology, and in particular to an indoor covert listening system based on DAS. Background Technology

[0002] When environmental changes occur along an optical fiber, such as sound or vibration causing minute strain, the phase of the light changes. By detecting these phase changes, vibration or sound information along the fiber can be obtained. Optical cables are critical infrastructure in global communication networks, carrying massive amounts of voice, data, and image information. In disaster relief efforts following earthquakes or other rubble, fiber optic listening devices can penetrate deep into confined spaces within the ruins to help rescuers detect the sounds of trapped individuals, such as shouts and knocking, providing crucial clues and improving rescue efficiency and the survival rate of those trapped. For important and highly secure indoor locations such as bank vaults, confidential archives, and museums, concealed fiber optic listening technology can monitor for abnormal sounds in real time, such as sounds of thieves cutting metal, picking locks, or drilling, enabling timely detection of security risks and triggering alarms, providing additional security for the location. However, this requires overcoming numerous technical challenges and integrating multiple advanced technologies to effectively detect, demodulate, reduce noise, and restore sound signals within the optical cable, while ensuring the system's security and reliability.

[0003] Currently, most research in the field of DAS (Digital Audio System) sound demodulation remains at the simulation stage. While these studies can theoretically simulate and analyze the sound demodulation process, existing research often fails to acquire and process sound signals in real-world scenarios in a timely manner, leading to signal delays and loss, and thus failing to meet the needs of real-time applications. Furthermore, in terms of sound reconstruction, the influence of noise, interference, and signal attenuation in the real environment results in significant differences between the reconstructed sound and the original sound, making it difficult to guarantee authenticity. Summary of the Invention

[0004] This invention aims to provide an indoor concealed listening system based on Direct Amplifier (DAS), addressing the shortcomings of existing DAS systems which suffer from poor sound demodulation, remain at the simulation stage, and cannot meet the demands of real-time applications. The system connects to existing indoor fiber optic lines. When sound is generated indoors, the sound wave signal is converted from an optical signal to an electrical signal in the DAS optical path. The signal acquisition and demodulation module integrated on the FPGA converts the electrical signal into a digital signal and processes it to obtain amplitude and true phase data. The amplitude and true phase data are then quantized and encoded into audio data in a noise reduction and restoration module integrated on the PC, where noise removal and voice enhancement are performed. Finally, high-quality real-time playback and display of the sound wave signal are achieved. This invention achieves highly efficient operation throughout the entire process from signal acquisition to processing. Utilizing existing indoor fiber optic cables or cables laid along walls, high-definition voice listening can be enabled within a 50-kilometer range along the fiber optic cable.

[0005] The objective of this invention is achieved through the following technical solution: an indoor concealed listening system based on DAS, used to listen for vibration or sound information of the optical fiber under test, comprising: a DAS optical path, a signal acquisition and demodulation module integrated on the FPGA end, and a noise reduction and restoration module integrated on the PC end connected in sequence.

[0006] The DAS optical path includes a narrow-linewidth laser, a coupler, an acousto-optic modulator (AOM), an erbium-doped amplifier (EDFA), a circulator, a 3dB coupler, and a balanced detector. The narrow-linewidth laser generated by the narrow-linewidth laser is split into two parts by the coupler, one part serving as the probe light and the other as the local reference light. The probe light is modulated into a pulse by the acousto-optic modulator (AOM), then amplified by the erbium-doped amplifier (EDFA) to generate a probe light pulse, which is then injected into the fiber under test through the circulator. Back Rayleigh scattering light is formed in the fiber under test. The back Rayleigh scattering light returns to the circulator and coherently interacts with the local reference light after passing through the attenuator in the 3dB coupler. Finally, it is converted into an electrical signal output by the balanced detector.

[0007] The signal acquisition and demodulation module includes an AD module, an IQ demodulation module, a phase differential module, and a phase unwinding module. The AD module acquires electrical signals and converts them into digital signals. The IQ demodulation module performs IQ demodulation on the digital signals to obtain phase data and amplitude data. The phase data is processed by the phase differential module to obtain differential phase data. The differential phase data is then unwound by the phase unwinding module to obtain the true phase data. The amplitude data and the true phase data are directly transmitted to the noise reduction and restoration module.

[0008] The noise reduction and restoration module includes a speech noise reduction module and an audiovisual module. The speech noise reduction module quantizes and encodes amplitude data and true phase data to obtain audio data. The audio data is then denoised using Wiener filtering based on spectral subtraction to obtain low-noise frequency data. The low-noise frequency data is then processed by the OMLSA algorithm, which integrates Markov Decision Process (MDP) and IMCRA, to obtain enhanced audio data. The audiovisual module converts the enhanced audio data into sound for playback, extracts enhanced phase data from the enhanced audio data and displays it in the form of a phase diagram, and extracts enhanced spectral data from the enhanced audio data and displays it in the form of a spectrum diagram.

[0009] Preferably, the narrow linewidth laser has an operating wavelength of 1549.056 nm, a maximum optical power of 40 mW, and a linewidth of 2.98 kHz.

[0010] Preferably, the radio frequency signal of the acousto-optic modulator (AOM) has a frequency of 200MHz, a power of 38µw, a duty cycle of 4%, an extinction ratio of 60dB, and an optical pulse width of 4µs.

[0011] Preferably, the output power of the erbium-doped amplifier (EDFA) is set to 49 dBm, the output optical power is 1.8 mW, and the noise figure is 4.5 dB.

[0012] Preferably, the balanced detector has a common-mode rejection ratio of 32dB, a photoelectric response efficiency of 0.95A / W, a response wavelength range of 800 to 1700nm, and a response bandwidth of 200MHz.

[0013] Preferably, the AD module uses the AD9643 chip, with an input voltage range of -2V to 2V and a sampling rate of 250Msps.

[0014] Preferably, the digital signal output by the AD module is:

[0015]

[0016] in, Phase curve; A s The amplitude data is the amplitude of the signal facing away from the Rayleigh signal; ω s ω is the angular frequency of the probe light; n is the index of the discrete time series;

[0017] The collected digital signal E s (n) Orthogonal signals of the same frequency generated at the FPGA end Multiplying them together, we get the orthogonal I and Q signals:

[0018]

[0019] Phase data of the IQ signal is obtained through the arctangent function of the cordic signal. The square root of the sum of square roots gives the amplitude data A. s :

[0020]

[0021] Phase data Differential phase data is obtained through the phase difference module, and then the phase unwinding module is used to unwind and restore the vibration position and the true phase data of the monitored content.

[0022] Preferably, the low-noise frequency data is processed by the OMLSA algorithm, which integrates Markov Decision Process (MDP) and IMCRA, to obtain enhanced audio data. The specific steps are as follows: The OMLSA algorithm first initializes the operation, then divides the low-noise frequency data into frames. After framing, the first frame of data is processed by the outline_process in the OMLSA algorithm to obtain state variables. Zero data is created using outline_process, and the state variables are updated to the zero data. Next, based on the audio characteristics of the low-noise frequency data, a state space covering noise estimation error and signal-to-noise ratio is defined, and an action space for adjusting the parameters in IMCRA is defined. A reward function is set to evaluate the effect of parameter adjustment, and the state transition probability is determined. Then, the mapping strategy from the state variables to the action space is initialized. During the operation, the strategy is continuously evaluated based on the reward function calculation results, using the state variables as a benchmark. If the effect is not good, the mapping strategy is improved, and the parameters in IMCRA are adjusted. This process is iterated and optimized until the optimal parameters are obtained. The enhanced audio data is then calculated based on the optimal parameters.

[0023] Preferably, the audiovisual module includes a QMediaPlayer toolbox, a PhaseWidget data container, and a frequencyWidget data container. The QMediaPlayer toolbox is used to play and listen to the enhanced audio data. The enhanced audio data is subjected to Fourier transform to extract enhanced phase data. The enhanced audio data is subjected to Fast Fourier Transform to extract enhanced spectrum data. The enhanced phase data is transmitted to the PhaseWidget data container to realize phase diagram display, and the enhanced spectrum data is transmitted to the frequencyWidget data container to realize spectrum diagram display.

[0024] The present invention has the following beneficial effects: The solution provided by the present invention effectively improves the clarity of the sound reproduction of the system, can acquire and process sound signals in real time, meets the application needs of indoor covert listening in actual scenarios, and provides reliable support for safety monitoring of important places, rubble rescue and other tasks.

[0025] The solution provided by this invention converts acoustic signals into electrical signals via a DAS optical path. An AD module acquires and converts the electrical signals into digital signals. These digital signals undergo IQ demodulation, differential phase analysis, and dewinding to obtain amplitude and true phase data. In the noise reduction and restoration module, the audio data is processed by Wiener filtering based on spectral subtraction, and an algorithm that fuses MDP with a traditional IMCRA-based OMLSA to obtain enhanced audio data. In the OMLSA algorithm based on the fusion of MDP and IMCRA, MDP dynamically optimizes the IMCRA parameters, breaking the traditional fixed parameter mode. This allows for real-time parameter adjustment based on different noise environments and audio characteristics, resulting in accurate noise removal and speech enhancement of the enhanced audio data. It effectively filters out noise above 2kHz and enhances the energy of the human voice portion of the audio data. The audiovisual module plays the enhanced audio data and displays its phase and spectral data, providing users with an intuitive and real-time listening experience. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the indoor covert listening system based on DAS according to the present invention;

[0027] Figure 2 The time and frequency domain diagrams show the first Wiener filter based on spectral subtraction and the second OMLSA denoising based on IMCRA.

[0028] Figure 3 Time-domain and frequency-domain graphs of existing algorithms for spectral subtraction and wavelet packet thresholding for denoising;

[0029] Figure 4 The time and frequency domain diagrams show the noise reduction process after Wiener filtering based on spectral subtraction and OMLSA algorithm based on the fusion of MDP and IMCRA.

[0030] Figure 5 The time-frequency diagrams for noise reduction using Wiener filtering based on spectral subtraction, OMLSA based on IMCRA, Wiener filtering based on spectral subtraction, and OMLSA algorithms based on the fusion of MDP and IMCRA are shown.

[0031] Figure 6 Diagram of the output electrical signal for the balanced detector. Detailed Implementation

[0032] In the following description, the present invention will be further illustrated with reference to specific embodiments in order to provide a thorough understanding of this application, but the present invention is not limited to the embodiments. In other instances, detailed descriptions of apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary details.

[0033] The indoor covert listening system described in this invention refers to a system capable of acquiring indoor sound information through specific methods. This includes sounds generated by people talking, objects falling, and slight movements. When indoor monitoring is required, this system can accurately capture this sound information, and using advanced audio processing technology, it can accurately determine the location of the sound source and reconstruct clear audio content, thereby obtaining information about the indoor target.

[0034] Figure 1 This is a schematic diagram of a DAS-based indoor covert listening system. This system is used to listen for vibrations or sound information from the optical fiber under test. It includes a DAS optical path, a signal acquisition and demodulation module integrated on the FPGA, and a noise reduction and restoration module integrated on the PC, connected in sequence. The system can be activated for listening along an optical fiber extending up to 50 kilometers. The optical fiber can utilize existing indoor communication network fiber optic cables or be laid along walls.

[0035] The DAS optical path includes a narrow-linewidth laser, a coupler, an acousto-optic modulator (AOM), an erbium-doped amplifier (EDFA), a circulator, an attenuator, a 3dB coupler, and a balanced detector. The fiber under test is connected to two ports of the circulator; one port of the circulator is connected to the erbium-doped amplifier (EDFA); the EDFA is connected to the acousto-optic modulator (AOM); the AOM is connected to the coupler and the narrow-linewidth laser; and the 3dB coupler is connected to the three ports of the circulator, the attenuator, and the balanced detector.

[0036] The narrow-linewidth laser generated by the narrow-linewidth laser is split into two parts by the coupler, with 90% used as probe light and 10% as local reference light. The probe light is modulated into a pulse by the acousto-optic modulator (AOM), then amplified by the erbium-doped amplifier (EDFA) to generate a probe light pulse, which is injected into the fiber under test through a circulator. Backscattered Rayleigh light is formed in the fiber under test, returning to the circulator and coherently coupled with the reference light (after attenuation) in a 3dB coupler. Finally, the signal is converted into an electrical signal output by a balanced detector. Figure 6 As shown.

[0037] High-coherence interferometry using narrow-linewidth lasers achieves nanometer-level measurement accuracy, far exceeding that of traditional light sources. Its operating wavelength is 1549.056 nm, with a maximum optical power of 40 mW and a linewidth of 2.98 kHz. The DAS optical path utilizes a high-coherence narrow-linewidth laser, reducing laser frequency drift and phase noise, ensuring stable coherence of the probe light during transmission through the optical fiber. The narrow-linewidth laser's optimized operating wavelength and linewidth parameters, combined with a specific acousto-optic modulator (AOM), erbium-doped amplifier (EDFA), and balanced detector output electrical signal, such as... Figure 6As shown. This optical path measurement and listening range reaches 50km, reducing signal attenuation and noise interference, and has a high signal-to-noise ratio.

[0038] The acousto-optic modulator (AOM) has a radio frequency signal frequency of 200MHz, a power of 38uw, and a duty cycle of 4%; a high extinction ratio of 60dB, reducing crosstalk of continuous light; and an optical pulse width of 4us.

[0039] Erbium-doped amplifiers (EDFAs) alter their operating current to change their optical power amplification. With an output power of 49 dBm and an output optical power of 1.8 mW, the noise figure is as low as 4.5 dB. If the pulsed optical power is low, considering factors such as fiber loss, the returned scattered light will be overwhelmed by noise. Increasing the maximum detection distance can be achieved by increasing the pulse width or increasing the input power. However, the pulse width limits the detection spatial resolution. Therefore, erbium-doped amplifiers (EDFAs) are used to provide higher detection optical power.

[0040] The balanced detector has a common-mode rejection ratio of 32dB, a photoelectric response efficiency of 0.95A / W, a response wavelength range of 800 to 1700nm, and a response bandwidth of up to 200MHz, meeting the detection requirements of 200MHz beat frequency signals. The balanced detector outputs a dual-path signal after coupling the local reference light and the probe light; the common-mode noise and DC bias are subtracted to obtain the electrical signal.

[0041] The FPGA uses a COE file to store orthogonal, cosine signals of the same frequency, triggering the AOM (Analog Array Component ). The rated frequency shift of the AOM center frequency is 200MHz. The signal acquisition and demodulation module includes an AD module, an IQ demodulation module, a phase differential module, and a phase unwinding module. The AD module acquires electrical signals and converts them into digital signals. The IQ demodulation module performs IQ demodulation on the digital signals to obtain phase data and amplitude data. The phase data is processed by the phase differential module to obtain differential phase data, which is then unwound by the phase unwinding module to obtain the true phase data. The amplitude data and the true phase data are directly transmitted to the noise reduction and restoration module.

[0042] The AD module uses the AD9643 chip, with an input voltage range of -2V to 2V, a sampling rate of 250Msps, a sampling data bit width of 16 bits, and a bandwidth of 250*16 / s. The AD module converts the acquired electrical signal into a digital signal.

[0043]

[0044] in, Phase curve; A s The amplitude data is the amplitude of the signal facing away from the Rayleigh signal; ω s The angular frequency of the probe light; n is the index of the discrete time series;

[0045] The collected digital signal E s (n) Orthogonal signals of the same frequency generated at the FPGA end Multiplying them together, we get the orthogonal I and Q signals:

[0046]

[0047] Phase data of the IQ signal is obtained through the arctangent function of the cordic signal. The square root of the sum of square roots gives the amplitude data A. s：

[0048]

[0049] Phase data is obtained through a differential phase module. The phase data obtained by IQ demodulation is a wound phase, with values ​​all within the interval (-π, π). A phase transition point exists, occurring at π with a transition amplitude of 2π. In the phase unwinding module, the differential phase data undergoes spatial unwinding followed by temporal unwinding to restore the vibration position and the true phase data of the monitored content. This true phase data is stored in DDR4, and an interrupt signal is sent to the x86 CPU. Upon detecting the interrupt, the x86 CPU immediately reads the result from DDR4 to the PC-side QT via the PCIe high-speed interface.

[0050] The noise reduction and restoration module includes a speech noise reduction module and an audiovisual module. Before performing noise reduction and speech enhancement, the speech noise reduction module needs to quantize and encode the amplitude data and true phase data to obtain audio data. Wiener filtering based on spectral subtraction first reads the audio data. In spectral subtraction, the audio data undergoes spectrum calculation to obtain its spectral data, which is then smoothed. Over-subtraction denoising in spectral subtraction suppresses noise in the smoothed spectral data, and the noise residual is processed to obtain initially denoised audio data. After spectral subtraction denoising, noise estimation is performed based on the initially denoised audio data to provide noise data support for subsequent processing. In the Wiener filtering stage, the initially denoised audio data undergoes spectrum calculation to obtain new spectral data. Then, the Wiener filter is applied to the calculated new spectral data for filtering, resulting in low-noise frequency data after Wiener filtering optimization.

[0051] The low-noise frequency (LNV) data input is based on the OMLSA algorithm, which integrates Markov Decision Process (MDP) and IMCRA. During the initialization phase, a series of parameters are set according to the input LNV data. Regarding the Fast Fourier Transform (FFT) parameters, the reference sampling frequency is set to 16000 Hz, the reference window length is 256 sampling points, and the reference overlap window length is 0.75 times the reference window length, i.e., 192 sampling points. In the noise spectrum estimation parameters, the frequency smoothing window size is 1, which determines the window width during frequency smoothing; the cyclic averaging parameter is set to 0.9, which controls the dependence on historical data, with a value closer to 1 indicating a higher dependence on historical data; the local search window size is 8, used to search for relevant data within a local area to assist noise estimation; the observation window size is 15, used to observe data characteristics within a certain time range; in the parameters related to signal absence prior probability estimation, the threshold parameter for signal absence is set to 1.67, the minimum noise estimation bias factor is 1.66, another threshold parameter for determining signal absence is 4.6, and the minimum noise threshold parameter for determining estimation is 3. The time-varying smoothing parameter for the conditional probability of speech presence is 0.85, used to adjust the smoothness in the speech presence probability calculation process and to compensate for the bias parameter. The signal missing prior probability smoothing parameter is 0.7, used to smooth the estimated value of the signal missing prior probability. The local search window size is 1, and the global search window size is 15, used to search for relevant information in the local and global ranges to estimate the signal missing prior probability, respectively. The frequency range correlation high-frequency cutoff frequency is set to 10 kHz, and the low-frequency cutoff frequency is set to 50 Hz to limit the frequency range. The minimum prior probability threshold is set to 0.005. The prior signal-to-noise ratio (SNR) estimation smoothing parameter is 0.95 to smooth the estimated value of the prior SNR, and the minimum SNR threshold is set to -18 dB. Flags are also set: a broadband flag of 1 indicates whether a broadband correlation processing strategy is used; a pitch flag of 1 determines whether tonal features in low-noise frequencies are considered; and a non-stationarity flag set to "high" indicates a high degree of non-stationarity in low-noise frequencies. When the sampling rate of the input low-noise frequency data differs from the reference sampling frequency, some of the above parameters are updated proportionally. After initializing the required parameters in IMCRA, the Markov Decision Process (MDP) applies the smoothing operation cyclic average parameter (a1), the time-varying smoothing parameter of the conditional probability of speech presence (a2), the smoothing parameter of the prior signal-to-noise ratio estimation (a3), and the smoothing parameter of the prior probability of signal missing (a4) that have been initialized and set in IMCRA.Simultaneously, based on the MDP framework, the noise estimation error and signal-to-noise ratio of the enhanced audio data are incorporated into the state space. Then, adjustments to the IMCRA parameters a1, a2, a3, and a4 are performed as the action space. A reward function is defined to calculate the change in noise estimation error and signal-to-noise ratio of the enhanced audio, used to evaluate the effectiveness of parameter adjustments in speech enhancement. When the noise estimation error is large, a1 is prioritized for adjustment, followed by a2, a3, and a4, providing an initial decision direction for subsequent parameter adjustments. After setting the state space, action space, reward function, and initial decision, the low-noise frequency data is divided into frames according to a set number of frames. First, a short-time Fourier transform is performed on the current frame's audio data to obtain frequency domain data. Then, a discrete Fourier transform is performed on the current frame's speech data to obtain power spectrum data. Noise estimation and speech enhancement are performed based on the frequency domain data and power spectrum data. Before speech enhancement, a Hamming window is created, and statistical data of the window function are calculated based on the transformed frequency domain data. Based on statistical information, in the `outline_process` mode, firstly, all-zero output data of the same length as the low-noise frequency data and all-zero data for storing intermediate results are created. Next, the number of frames is calculated based on the low-noise frequency data and the parameters determined during initialization. For each frame, if the data length is insufficient, padding is performed to obtain padded frame data. Subsequently, the padded frame data is multiplied by the created and normalized window function, allowing the frequency domain features to be accurately extracted by subsequent Fast Fourier Transform (FFT) to obtain new frequency domain data. After obtaining the new frequency domain data, the presence of a signal is determined based on a pre-set threshold condition. If the condition is met, the flag is set to the corresponding state. During the noise reduction process, the current frame audio data undergoes a Discrete Fourier Transform to obtain power spectrum data, and intermediate quantities related to the prior signal-to-noise ratio are calculated. The results are updated considering the minimum signal-to-noise ratio threshold. Then, the current frame audio data undergoes a Discrete Fourier Transform to obtain power spectrum data, which is then frequency smoothed and combined with the power spectrum data from the current frame and the previous frame updated to the minimum noise estimate. Next, the probability of speech presence is calculated based on the power spectrum data of the minimum sound estimation value. During this process, the long-term and short-term results of noise estimation are updated according to different situations, while time-varying smoothing is performed simultaneously. After calculating the local and global prior probabilities based on the current frame, they are converted to decibel representation. The local and global probabilities, the final probability of speech presence, are then updated based on the decibel values. The prior signal-to-noise ratio is recalculated to obtain the total gain. After obtaining the total gain, the frequency domain data after gain adjustment is converted back to time domain data. At this point, the time domain signal is multiplied again by the normalized window function to update the current frame data for speech enhancement. After obtaining the current frame data for speech enhancement, the policy evaluation method is called, and the noise estimation error or signal-to-noise ratio change between the current and previous frames is used as a reward to update the MDP state space.Based on the evaluation results, the mapping strategy improvement method is invoked. If the reward is negative, the strategy is adjusted; for example, if increasing a1 has a poor effect, it is changed to decreasing it. If the reward is positive, the strategy is maintained or optimized (the key parameters a1, a2, a3, and a4 are adjusted sequentially in small steps of 0.01). If the current strategy suggests increasing a1, a1 is increased by 0.01. If the reward is positive, the state variable is updated, the strategy is maintained, and a1 is increased by 0.01 again, and the state variable is updated again according to the new result. Finally, based on the adjusted parameters and the current frame data, combined with the noise estimation error and signal-to-noise ratio effect, the MDP state space is updated. The current frame data is continuously iterated to gradually optimize the IMCRA parameters. After the current frame data of the low noise frequency data has been processed by the OMLSA algorithm with optimized IMCRA parameters, the number of processed frames is updated. After applying the OMLSA algorithm with optimized IMCRA parameters to all frame data, the enhanced audio data is obtained.

[0052] The noise reduction effect without MDP algorithm integration is as follows Figure 2 The time-domain and frequency-domain diagrams are shown below. Figure 3 For audio data processed by existing speech enhancement algorithms based on convolutional recurrent networks, through Figure 2 The data shows that the audio has a large amount of residual background noise, a low signal-to-noise ratio, and dispersed energy. Figure 3 The overall waveform of the time domain plot is more regular, but there are still some minor fluctuations and residual background noise. Figure 3 The frequency domain diagram shows that there is still some energy distribution in the non-speech frequency region, but Figure 3 Existing algorithms are complex and difficult to implement in real time. This system, by integrating the MDP algorithm, enhances audio data noise reduction and enhancement in both the time and frequency domains. Figure 4 As shown, compared to Figure 3 and Figure 2 The time-domain waveform is smoother, and the energy in the non-speech frequency region of the frequency domain is further reduced. The time-frequency diagrams of noise reduction algorithms based on Wiener filtering (spectral subtraction), OMLSA (IMCRA), Wiener filtering (spectral subtraction), and OMLSA (MDP and IMCRA fusion) are shown below. Figure 5 As shown, Figure 5 As shown in the right figure, after fusion, noise above 2kHz has been basically removed compared to the unfused left side; the energy of human voice audio segments below 2kHz is significantly enhanced, and the energy of some noise is reduced; the overall background noise removal and speech enhancement effects are stronger than existing algorithms, and the algorithm complexity is lower than existing algorithms.

[0053] The audiovisual module includes the QMediaPlayer toolbox. The QMediaPlayer toolbox is used to play and listen to the enhanced audio data, perform Fourier transform on the enhanced audio data to extract enhanced phase data, perform Fast Fourier transform on the enhanced audio data to extract enhanced spectrum data, transmit the enhanced phase data to the PhaseWidget data container to realize phase diagram display, and transmit the enhanced spectrum data to the frequencyWidget data container to realize spectrum diagram display.

Claims

1. A DAS-based indoor concealed listening system for listening to vibration or sound information of the optical fiber under test, characterized in that, include: The DAS optical path, the signal acquisition and demodulation module integrated on the FPGA side, and the noise reduction and restoration module integrated on the PC side are connected in sequence. The DAS optical path includes a narrow-linewidth laser, a coupler, an acousto-optic modulator (AOM), an erbium-doped amplifier (EDFA), a circulator, a 3dB coupler, and a balanced detector. The narrow-linewidth laser generated by the narrow-linewidth laser is split into two parts by the coupler, one part serving as the probe light and the other as the local reference light. The probe light is modulated into a pulse by the acousto-optic modulator (AOM), then amplified by the erbium-doped amplifier (EDFA) to generate a probe light pulse, which is then injected into the fiber under test through the circulator, forming backscattered Rayleigh light in the fiber. The backscattered Rayleigh light returns to the circulator and coherently interacts with the local reference light (after passing through an attenuator) in the 3dB coupler. Finally, it is converted into an electrical signal output by the balanced detector. The signal acquisition and demodulation module includes an AD module, an IQ demodulation module, a phase differential module, and a phase unwinding module. The AD module acquires electrical signals and converts them into digital signals. The IQ demodulation module performs IQ demodulation on the digital signals to obtain phase data and amplitude data. The phase data is processed by the phase differential module to obtain differential phase data. The differential phase data is then unwound by the phase unwinding module to obtain the true phase data. The amplitude data and the true phase data are directly transmitted to the noise reduction and restoration module. The noise reduction and restoration module includes a speech noise reduction module and an audiovisual module. The speech noise reduction module quantizes and encodes amplitude data and true phase data to obtain audio data. The audio data is then denoised using Wiener filtering based on spectral subtraction to obtain low noise frequency data. The low noise frequency data is then processed by the OMLSA algorithm based on the fusion of Markov decision process (MDP) and IMCRA to obtain enhanced audio data. The audiovisual module converts the enhanced audio data into sound for playback, extracts enhanced phase data from the enhanced audio data and displays it in the form of a phase diagram, and extracts enhanced spectrum data from the enhanced audio data and displays it in the form of a spectrum diagram. The low-noise frequency (LNV) data is processed by the OMLSA algorithm, which integrates Markov Decision Process (MDP) and IMCRA, to obtain enhanced audio data. The specific steps are as follows: First, the OMLSA algorithm initializes the operation. Then, the LNV data is framed. After framing, the first frame is processed using the `outline_process` function in the OMLSA algorithm to obtain state variables. Zero data is created using `outline_process`, and the state variables are updated to the zero data position. Next, based on the audio characteristics of the LNV data, a state space encompassing noise estimation error and signal-to-noise ratio is defined. An action space for adjusting parameters in the IMCRA is defined, and a reward function is set to evaluate the effect of parameter adjustment, determining the state transition probability. Then, a mapping strategy from state variables to the action space is initialized. During operation, the strategy is continuously evaluated based on the state variables and the reward function calculation results. If the effect is unsatisfactory, the mapping strategy is improved, and the parameters in the IMCRA are adjusted. This iterative optimization continues until the optimal parameters are obtained, and the enhanced audio data is calculated based on the optimal parameters.

2. The indoor covert listening system based on DAS according to claim 1, characterized in that, The narrow linewidth laser operates at a wavelength of 1549.056 nm, has a maximum optical power of 40 mW, and a linewidth of 2.98 kHz.

3. The indoor covert listening system based on DAS according to claim 1, characterized in that, The radio frequency signal of the acousto-optic modulator (AOM) has a frequency of 200MHz, a power of 38µw, a duty cycle of 4%, an extinction ratio of 60dB, and an optical pulse width of 4µs.

4. The indoor covert listening system based on DAS according to claim 1, characterized in that, The output power of the erbium-doped amplifier (EDFA) is set to 49 dBm, the output optical power is 1.8 mW, and the noise figure is 4.5 dB.

5. The indoor covert listening system based on DAS according to claim 1, characterized in that, The balanced detector has a common-mode rejection ratio of 32 dB, a photoelectric response efficiency of 0.95 A / W, a response wavelength range of 800 to 1700 nm, and a response bandwidth of 200 MHz.

6. The indoor covert listening system based on DAS according to claim 1, characterized in that, The AD module uses the AD9643 chip, with an input voltage range of -2V to 2V and a sampling rate of 250Msps.

7. The indoor covert listening system based on DAS according to claim 1, characterized in that, The digital signal output by the AD module is: ； in, The phase curve; The amplitude data is for the direction away from the Rayleigh signal; ω is the angular frequency of the probe light; n is the index of the discrete time series; The acquired digital signal E s (n) Orthogonal signals of the same frequency generated at the FPGA terminal , Multiplying them together, we get the orthogonal I and Q signals: ； Phase data of the IQ signal is obtained through the arctangent function of the cordic signal. The square root of the sum of squares gives the amplitude data A. s： ； ； Phase data Differential phase data is obtained through the phase difference module, and then the phase unwinding module is used to unwind and restore the vibration position and the true phase data of the monitored content.

8. The indoor covert listening system based on DAS according to claim 1, characterized in that, The audiovisual module includes a QMediaPlayer toolbox, a PhaseWidget data container, and a frequencyWidget data container. The QMediaPlayer toolbox is used to play and listen to the enhanced audio data. The enhanced audio data is subjected to Fourier transform to extract enhanced phase data, and the enhanced audio data is subjected to Fast Fourier transform to extract enhanced spectrum data. The enhanced phase data is transmitted to the PhaseWidget data container to realize the phase diagram display, and the enhanced spectrum data is transmitted to the frequencyWidget data container to realize the spectrum diagram display.

Images