Amplitude-frequency demodulation algorithm for extracting respiratory signal from pulse wave signal

Abstract

The invention provides an amplitude-frequency demodulation algorithm for extracting respiratory signals through amplitude-frequency demodulation and through remote photoplethysmography (rPPG) or contact photoplethysmography (PPG), which comprises the following specific steps: step 1, preprocessing pulse wave signals (BVP signals) acquired through the rPPG or the PPG, and performing frequency domain heart rate tracking; step 2, performing frequency modulation coherent demodulation according to the current heart rate; 3, recovering a BVP carrier signal according to a frequency modulation demodulation result; 4, amplitude modulation breathing signal demodulation is carried out according to the recovered carrier signals; and step 5, carrying out fusion on the demodulation signals. The step can overcome the dependence of a traditional method, namely a peak detection algorithm, on the sampling frequency and the signal-to-noise ratio of the BVP signal, and the signal-to-noise ratio of the extracted breathing signal is improved to a certain extent.

Description

Technical Field

[0001] The present invention belongs to the field of biomedical signal processing, and is specifically an amplitude-frequency demodulation algorithm for extracting respiratory signals through remote photoplethysmography (rPPG) or contact photoplethysmography (PPG) Technical Background

[0002] As one of the important parameters for monitoring the physiological state of the human body, respiratory signals are widely used in clinical diagnosis and health management. Traditional respiratory signal extraction methods usually rely on indicators related to heart rate variability (HRV), such as respiratory induced baseline intensity variation (RIIV), respiratory induced frequency variation (RIFV) and respiratory induced amplitude variation (RIAV). These indicators are based on the phenomenon of respiratory sinus arrhythmia (RSA), that is, during inhalation, lung stretch receptors inhibit vagal neuron activity, resulting in an increase in heart rate; during exhalation, the inhibitory effect weakens and the heart rate decreases, resulting in periodic fluctuations in heart rate.

[0003] Although these traditional methods perform well under ideal conditions, they have high requirements for signal sampling frequency, video frame rate and signal-to-noise ratio (SNR). In practical applications, especially in complex environments or under low SNR conditions, the accuracy of these methods is significantly reduced, and respiratory signals are difficult to extract reliably, limiting their widespread application. Therefore, there is an urgent need for a more robust respiratory signal extraction algorithm that can maintain high accuracy and stability in low SNR environments. Summary of the invention

[0004] The present invention proposes an amplitude-frequency demodulation algorithm based on BVP signal, which is used to extract respiratory signals through remote photoplethysmography (rPPG) or contact photoplethysmography (PPG). The algorithm effectively solves the limitations of traditional methods in low signal-to-noise ratio environments by preprocessing, frequency tracking, FM demodulation, AM demodulation and signal fusion of the pulse wave signal (BVP signal) collected by rPPG or PPG. Compared with the traditional extraction method based on heart rate variability (HRV), the algorithm of the present invention is more robust in complex environments and can ensure high-precision extraction of respiratory signals even when the sampling frequency and video frame rate are not ideal.

[0005] Specifically, the main innovations of the present invention are embodied in the following aspects:

[0006] 1. Optimization of the preprocessing process: By filtering, resampling and singular value filtering the BVP signal, low-frequency noise and artifacts are removed to enhance signal quality. Compared with the traditional method that relies on high-quality sampling conditions, the present invention introduces a dimensionality reduction algorithm in signal preprocessing, which is particularly suitable for remote rPPG signals, can effectively improve the signal-to-noise ratio, and remain stable under a variety of complex backgrounds.

[0007] 2. Frequency domain tracking and FM coherent demodulation: The present invention accurately captures the frequency changes of the respiratory signal by tracking the fundamental frequency and its harmonics in the BVP signal. In the frequency modulation and demodulation process, a digital coherent demodulation algorithm is used, which can effectively extract the FM component in the respiratory signal under low signal-to-noise ratio conditions. Especially when affected by motion artifacts or changes in ambient light, the frequency demodulation step of the present invention can maintain a high degree of anti-interference.

[0008] 3. Improvement of the AM demodulation process: In terms of amplitude demodulation, the present invention extracts the amplitude change of the respiratory signal by performing coherent amplitude demodulation on the recovered carrier signal. This technology can effectively separate the heart rate change from the amplitude change of the respiratory signal, improve the extraction accuracy of the respiratory signal, and solve the problem of the traditional method's dependence on a high signal-to-noise ratio environment in amplitude extraction.

[0009] 4. Signal fusion and optimization: In order to further improve the quality of the demodulated respiratory signal, the present invention combines the results of FM demodulation and AM demodulation, calculates their correlation coefficients and performs linear fusion according to the signal-to-noise ratio. This step maximizes the improvement of the signal-to-noise ratio, enhances the robustness of the demodulated signal, and enables the extracted respiratory signal to maintain a high accuracy in all kinds of environments.

[0010] In general, the amplitude-frequency demodulation algorithm of the present invention greatly improves the extraction accuracy and adaptability of respiratory signals through multi-level signal processing and fusion, and is suitable for the extraction and application of remote and contact BVP signals in low signal-to-noise ratio environments and complex backgrounds. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 This is a diagram of the algorithm framework provided by the present invention.

[0012] Figure 2 This is a processing flow chart of the amplitude-frequency demodulation algorithm of the respiratory signal provided by the present invention.

[0013] Figure 3 The present invention provides a signal flow chart of the amplitude-frequency demodulation algorithm of the respiratory signal.

[0014] Figure 4 This is a schematic diagram showing the final effect of extracting the energy spectrum of the respiratory signal using this method and the comparison with the effect of the traditional RIFV method. DETAILED DESCRIPTION

[0015] The present invention proposes a respiratory signal extraction method combining FM and AM demodulation by analyzing the modulation relationship of the pulse wave signal. Figure 1 As shown. The pulse wave signal is assumed to be:

[0016]

[0017] Among them, m(t) is the breathing signal, k f is the FM sensitivity, A a +k a m(t) represents the amplitude modulation part of the respiratory signal, and k0m(t) is the additive respiratory component. Based on this hypothetical model, the present invention extracts the frequency modulation (RIFV) and amplitude modulation (RIAV) components in the signal through FM and AM demodulation respectively. The flow chart is as follows: Figure 2 shown.

[0018] 1. Signal preprocessing

[0019] The signal preprocessing step in the present invention performs multi-level processing on the collected BVP signal to eliminate the influence of environmental interference and noise on the signal. First, the BVP signal is processed by a bandpass filter to filter out unnecessary low-frequency and high-frequency noise and retain the frequency band related to the respiratory signal. Next, the signal is resampled to adapt to the sampling frequency of different devices to ensure the stability of the signal for subsequent processing. For rPPG signals, signal dimensionality reduction technology and signal enhancement algorithms are additionally used to further optimize the signal-to-noise ratio of the signal. Finally, the singular value filtering method is used to process the PPG signal to remove artifacts and other interferences to ensure the quality of the input signal. The signal time domain waveform is as follows: Figure 3 .a.

[0020] The filtered signal is Fourier transformed to obtain the spectrum, and the heart rate estimation is obtained by frequency domain tracking in the time-spectrum spectrum.

[0021] 2. FM coherent demodulation

[0022] The pre-processed BVP signal is subjected to frequency modulation coherent demodulation to extract the frequency components of the respiratory signal. This process uses a digital coherent demodulation algorithm to focus on the fundamental frequency and the first harmonic frequency, demodulates the frequency modulation signal caused by respiratory sinus arrhythmia (RSA), and obtains the frequency modulated respiratory signal waveform within the time window.

[0023] The demodulation is mainly performed in the frequency domain through CZT (Chirp Z-Transform). First, the rPPG signal x in the time domain is transformed using CZT. t Convert to frequency domain so that it can be processed within the frequency range. CZT is an efficient frequency domain analysis method that can focus on a specific frequency region for spectrum estimation, so it is more suitable for extracting characteristic signals in the frequency bands related to heart rate and breathing. The transformed signal is expressed as:

[0024] X(f)=CZT(x t )

[0025] Through this frequency domain representation, the frequency components of the signal can be concentrated at the center frequency f hrIn order to better extract the frequency change information. In the frequency domain, the signal is separated into orthogonal IQ components, namely cosine and sine components. The IQ signal represents the phase and amplitude changes of the signal and is used to demodulate the frequency modulation and amplitude modulation components. The orthogonal components are expressed as:

[0026] X I (f) = X(ff hr )+conj(X(f hr -f))

[0027] X Q (f) = X(ff hr )-conj(X(f hr -f))

[0028] These components can extract the phase information of the signal from the frequency domain. By performing an inverse CZT transform on the IQ signal in the frequency domain, it is converted back to the time domain. The two components represent the cosine and sine signals in the time domain, respectively, so as to calculate the instantaneous phase. The instantaneous phase is calculated by the ratio between the sine and cosine components:

[0029]

[0030] Derivative the instantaneous phase to obtain the frequency modulated breathing component m in the signal FM (t), the spectrum is as Figure 3 .c.

[0031] 3. Recover the carrier signal

[0032] According to the results of FM demodulation and the heart rate tracking signal, the original carrier signal is restored to reduce the impact of the breathing signal on the heart rate, provide a more stable carrier signal, and ensure the accuracy of subsequent demodulation. In order to perform coherent demodulation, the carrier signal needs to be based on the estimated heart rate frequency f hr Reconstruct. The reconstructed carrier signal is:

[0033] x carrier (t) = cos(2πf hr t+θ(t))

[0034] Where θ(t) is the instantaneous phase calculated during FM demodulation. This carrier signal is multiplied with the original rPPG signal to extract the amplitude modulation component in the signal and recover the carrier spectrum as Figure 3 .b.

[0035] 4. AM demodulation

[0036] Based on the recovered carrier signal, AM demodulation is performed to extract the amplitude component of the respiratory signal. Through AM demodulation technology, the amplitude change information of the signal is obtained. The filtered rPPG signal is multiplied with the reconstructed carrier signal to obtain the demodulated signal, which represents the amplitude modulation part of the respiratory signal, and then a bandpass filter (BPF) is applied. The signal after bandpass filtering is mainly concentrated in the respiratory frequency band, which is expressed as:

[0037] m AM (t) = BPF(x carrier (t)·x t )

[0038] The obtained amplitude modulated breathing signal m AM (t) spectrum as Figure 3 .d.

[0039] 5. Signal Fusion

[0040] By calculating the correlation coefficient of the AM and FM demodulated signals and performing linear fusion according to the signal-to-noise ratio, the fused signal has higher robustness and signal-to-noise ratio, and finally generates a high-quality respiratory signal, which is suitable for respiratory monitoring in different complex environments.

[0041] Through the above FM and AM demodulation steps, the frequency modulation and amplitude modulation components of the breathing signal are extracted. The final breathing signal is obtained by combining these components to obtain complete breathing information. The final breathing signal is expressed as:

[0042] m(t)=m AM (t)+k·m FM (t)

[0043] Where k is mainly composed of m AM (t) and m FM The variance and correlation coefficient of (t) are calculated. To simplify the calculation,

[0044]

[0045] The signal is low-pass filtered to remove high-frequency noise and finally a smoothed respiratory signal is obtained. The fusion signal spectrum is as follows: Figure 3 .e, the reference breathing signal spectrum is as follows Figure 3 .f, the breathing signal can be accurately extracted.

[0046] 6. Respiration rate calculation and signal output

[0047] The respiratory signal is aligned with the synchronized BVP signal (pulse wave signal) to ensure the timing consistency of the signal, and then the time-frequency analysis is performed to calculate the period of the respiratory signal and obtain the respiratory frequency. Figure 4,Comparing the effects of the traditional method and this method, it can be seen that the signal-to-noise ratio is greatly improved.

[0048] Through the above steps, the present invention can achieve high-precision extraction of rPPG signals from videos, and efficiently obtain respiratory signals through FM / AM demodulation technology. The method has significant advantages in signal strength enhancement, motion artifact suppression and real-time performance, and is suitable for physiological state monitoring in a variety of complex scenarios. Compared with traditional methods, the method of the present invention has a certain degree of improvement in signal-to-noise ratio, anti-interference ability and real-time performance, and has greater application prospects compared with existing methods.

[0049] The above is only the preferred combination used in the present invention. It should be pointed out that for those skilled in the art, several improvements and adjustments can be made without departing from the principles of the present invention. These improvements and adjustments should also be regarded as the scope of protection of the present invention.

Claims

1. An amplitude-frequency demodulation algorithm for extracting respiratory signals by remote photoplethysmography (rPPG) or contact photoplethysmography (PPG), the main features of which include: S1. Preprocess the pulse wave signal (BVP signal) collected by rPPG or PPG, including filtering, resampling, singular value filtering, etc., and perform frequency tracking to obtain the fundamental frequency and harmonic frequency in the current time window. S2. Perform coherent frequency modulation demodulation on the preprocessed BVP signal at the fundamental frequency and harmonics to obtain a frequency modulated breathing signal. S3. Restore the original carrier signal without amplitude modulation according to the frequency modulated signal and the tracked heart rate signal, and perform amplitude demodulation on the pulse wave signal at the heart rate and harmonics to obtain the breathing signal. S4. According to the respiratory signals obtained by amplitude demodulation and frequency demodulation, the correlation coefficient is calculated and linear fusion is performed according to the signal-to-noise ratio to obtain the final amplitude-frequency demodulated respiratory signal.

2. The method for extracting respiratory signals according to claim 1, preprocessing the BVP signal in S1, is mainly characterized in that 1. According to the hardware characteristics of the signal acquisition device, a variety of signal processing links are combined, including a post-bandpass filter, a signal dimension reduction algorithm for rPPG signals, a filtering algorithm, a signal enhancement algorithm, etc., a singular value filter for PPG devices, a median filter, etc. The signal-to-noise ratio of the BVP signal is maximized under the condition of introducing minimum nonlinear distortion 2. The time-frequency spectrum of the BVP signal in the frequency range of interest is obtained by using CZT, and the heart rate and its harmonic spectrum on the time-frequency spectrum are tracked to obtain the estimated heart rate of the current time window.

3. According to the method for extracting the respiratory signal according to claim 1, for the frequency modulated respiratory signal demodulation method in S2, the main feature is that a digital coherent demodulation algorithm is used on the spectrum to demodulate the frequency modulated respiratory signal caused by respiratory sinus arrhythmia at the fundamental frequency and the first harmonic to obtain the frequency modulated respiratory signal waveform within the time window.

4. The method for extracting a respiratory signal according to claim 1, wherein the method for demodulating an amplitude modulated respiratory signal in S3 is characterized in that The estimated heart rate value and the extracted frequency modulation component of the respiratory signal are used to restore the original signal without amplitude modulation, and the pulse wave signal is coherently amplitude demodulated at the heart rate and harmonics.

5. According to the method for extracting respiratory signals based on rPPG according to claim 1, for the S4 respiratory signal fusion method, the main feature is that the correlation coefficient and signal-to-noise ratio of multiple amplitude-frequency modulation respiratory signals in the time window are calculated, and linear fusion is performed according to the correlation and signal-to-noise ratio of different signals, the signal-to-noise ratio of the amplitude-frequency demodulated respiratory signal is maximized, and the signals of each time window are overlapped and added to obtain the final demodulated respiratory signal.

6. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method of extracting the amplitude-frequency demodulation algorithm of the respiratory signal from the pulse wave signal as described in any one of claims 1-5 is implemented.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 5.

Images