Adaptive doppler signal denoising filter

Abstract

This invention discloses an adaptive Doppler signal noise reduction filter, comprising: a sensor interface impedance matching and electrical isolation unit, an analog-to-digital conversion and Doppler signal preprocessing unit, a Doppler signal multi-scale wavelet decomposition hardware unit, a Doppler background noise statistical characteristic estimation unit, a hierarchical adaptive threshold hardware calculation and nonlinear correction unit, a wavelet decomposition frequency domain coefficient inverse transform hardware reconstruction unit, a noise reduction effect quantitative evaluation and parameter closed-loop optimization unit, and a power management module. This invention solves the reconstruction oscillation problem of the hard threshold function and the constant amplitude deviation defect of the soft threshold function by dynamically adjusting the hierarchical adaptive threshold based on the hyperbolic tangent square operator and using a fully hardware-based design. It filters out noise while completely preserving the amplitude and phase information of the Doppler signal, achieving high-fidelity noise reduction with low hardware resource consumption and meeting low-latency processing requirements in various scenarios.

Description

Technical Field

[0001] This invention relates to a signal noise reduction filter, specifically an adaptive Doppler signal noise reduction filter. Background Technology

[0002] Doppler signals are widely used in industrial and civilian fields such as underwater acoustic detection, industrial flow velocity measurement, and medical ultrasound detection. In practical applications, echo Doppler signals are often extremely weak and accompanied by complex background noise, which can easily lead to the effective signal being drowned out by the noise.

[0003] Traditional signal denoising hardware cannot adapt to the real-time changes in Doppler signals as the measured target's motion changes. Existing wavelet thresholding denoising has good time-frequency characteristics but still has significant drawbacks. For example, denoising filters based on hard thresholding functions, while preserving signal edge features, exhibit signal discontinuity at the threshold point, leading to pseudo-Gibbs oscillations during signal reconstruction that disrupt the phase continuity of the Doppler signal. Denoising filters based on soft thresholding functions, while maintaining good continuity, suffer from a constant deviation between the processed and original coefficients, resulting in energy attenuation and amplitude distortion in the reconstructed signal. Improved denoising filters cannot simultaneously address the dual problems of oscillation and amplitude deviation.

[0004] Furthermore, existing publicly available wavelet hierarchical adaptive thresholding denoising schemes, such as CN120929820A, calculate their thresholds based solely on noise intensity and signal length, failing to consider the inherent characteristic of Doppler signal noise exhibiting a 1 / f distribution with the number of decomposition layers, thus failing to achieve dynamic adaptation of the hierarchical thresholds. Simultaneously, the threshold functions of existing schemes cannot simultaneously address the oscillation problem of hard thresholds and the constant deviation defect of soft thresholds, and are mostly open-loop software algorithms, unable to dynamically optimize denoising parameters through hardware closed-loop feedback, making them unsuitable for the real-time high-fidelity denoising requirements of Doppler signals in complex scenarios such as underwater acoustic detection. Summary of the Invention

[0005] Purpose of the invention: The purpose of this invention is to provide an adaptive Doppler signal noise reduction filter that can accurately distinguish signals in the frequency domain, reduce resource consumption, improve real-time noise reduction effect, and enhance scene adaptability.

[0006] Technical Solution: This invention provides an adaptive Doppler signal noise reduction filter, which is integrated into an FPGA or ASIC hardware circuit. It includes: a sensor interface impedance matching and electrical isolation unit, an analog-to-digital conversion and Doppler signal preprocessing unit, a Doppler signal multi-scale wavelet decomposition hardware unit, a Doppler background noise statistical characteristic estimation unit, a hierarchical adaptive threshold hardware calculation and nonlinear correction unit, a wavelet decomposition frequency domain coefficient inverse transform hardware reconstruction unit, a noise reduction effect quantitative evaluation and parameter closed-loop optimization unit, and a power management module that supplies power to the entire filtering system.

[0007] The input terminal of the sensor interface impedance matching and electrical isolation unit receives the externally input analog Doppler signal, and the output terminal of the sensor interface impedance matching and electrical isolation unit is connected to the input terminal of the analog-to-digital conversion and Doppler signal preprocessing unit to complete the impedance matching and electrical isolation of the Doppler signal.

[0008] The first output of the analog-to-digital conversion and Doppler signal preprocessing unit is connected to the input of the Doppler signal multi-scale wavelet decomposition hardware unit, and the second output of the analog-to-digital conversion and Doppler signal preprocessing unit is connected to the first input of the noise reduction effect quantization evaluation and parameter closed-loop optimization unit. This unit is used to convert the analog Doppler signal into a digital signal and complete anti-aliasing filtering and amplitude normalization processing to output a standardized digital Doppler signal.

[0009] The first output of the Doppler signal multi-scale wavelet decomposition hardware unit is connected to the input of the Doppler background noise statistical characteristic estimation unit, and the second output of the Doppler signal multi-scale wavelet decomposition hardware unit is connected to the first input of the hierarchical adaptive threshold hardware calculation and nonlinear correction unit. The Coiflet wavelet basis is used to perform 3 to 5 layers of wavelet decomposition on the input digital Doppler signal, which is used to output the wavelet decomposition frequency domain coefficients and decomposition layer index corresponding to each layer.

[0010] The output of the Doppler background noise statistical characteristic estimation unit is connected to the second input of the hierarchical adaptive threshold hardware calculation and nonlinear correction unit. It has a built-in hardware sorting circuit and a fixed-point parallel hardware operation unit, which is used to calculate the background noise statistical standard deviation of the Doppler signal based on the high-frequency detail components obtained by the first layer wavelet decomposition and through the median absolute deviation statistical method.

[0011] The feedback input of the hierarchical adaptive threshold hardware calculation and nonlinear correction unit is connected to the output of the noise reduction effect quantification evaluation and parameter closed-loop optimization unit. The output of the hierarchical adaptive threshold hardware calculation and nonlinear correction unit is connected to the input of the wavelet decomposition frequency domain coefficient inverse transform hardware reconstruction unit. It is used to generate the hierarchical adaptive threshold of the corresponding decomposition level according to the current decomposition level, the length of the current layer wavelet decomposition frequency domain coefficient, the statistical standard deviation of Doppler background noise, and the feedback optimization parameters of the noise reduction effect quantification evaluation and parameter closed-loop optimization unit, and to perform nonlinear compensation correction processing on the nonlinear wavelet coefficients of the wavelet decomposition frequency domain coefficients, that is, to correct the wavelet decomposition frequency domain coefficients based on the hierarchical adaptive threshold.

[0012] The formula for calculating the hierarchical adaptive threshold is:

[0013]

[0014] in, The statistical standard deviation of background noise. This is the attenuation adjustment coefficient, with an initial value of 0.1 and an adjustment range of 0.02 to 0.2. This represents the current decomposition level. For the first The actual length of the frequency domain coefficients in layer wavelet decomposition. For the first The layer has a fixed natural logarithm value. This is the attenuation coefficient for the stratified threshold.

[0015] The calculation formula for the nonlinear compensation and correction processing of the nonlinear wavelet coefficients is as follows:

[0016]

[0017] in, These are the frequency domain coefficients of the original wavelet decomposition. The magnitude of the frequency domain coefficients of the original wavelet decomposition is given. For the first Layer-specific adaptive threshold, For nonlinear compensation adjustment coefficients, For symbol extraction function, The function is a hyperbolic tangent function, and the nonlinear compensation adjustment coefficient is... Adjusted in real time based on the statistical standard deviation of background noise, the calculation formula is as follows: , The preset adjustable parameter has a value range of 0.05~0.5, with a preferred value of 0.2; the hyperbolic tangent square value is pre-stored in a 16-bit wide pre-calculated parameter read-only memory mapping table, and is read in a single clock cycle through address indexing;

[0018] The wavelet decomposition frequency domain coefficient inverse transform hardware reconstruction unit adopts an inverse wavelet transform pipeline cascaded FIR digital filter array corresponding to the Doppler signal multi-scale wavelet decomposition hardware unit. Its first output is connected to the second input of the noise reduction effect quantification evaluation and parameter closed-loop optimization unit. The second output outputs the final noise-reduced Doppler signal, which is used to perform wavelet inverse transform on the wavelet decomposition frequency domain coefficients after wavelet coefficient nonlinear compensation correction, and output the noise-reduced Doppler signal.

[0019] The output of the noise reduction effect quantification evaluation and parameter closed-loop optimization unit is connected to the feedback input of the hierarchical adaptive threshold hardware calculation and nonlinear correction unit. It is used to calculate the signal-to-noise ratio and root mean square error based on the original digital Doppler signal output by the analog-to-digital conversion and Doppler signal preprocessing unit and the noise-reduced signal output by the wavelet decomposition frequency domain coefficient inverse transform hardware reconstruction unit. With the optimization objectives of maximizing the Doppler signal-to-noise ratio and minimizing the amplitude error of the reconstructed signal, the attenuation adjustment coefficient adjustment value is generated and fed back to the hierarchical adaptive threshold hardware calculation and nonlinear correction unit in a closed loop to realize the dynamic optimization of the threshold.

[0020] Preferably, the sensor interface impedance matching and electrical isolation unit is an impedance matching interface circuit with built-in electrostatic protection unit and electrical isolation unit, and built-in TVS electrostatic protection unit and magnetic coupling electrical isolation unit, used to match the output characteristics of the Doppler sensor with the input requirements of the analog-to-digital conversion and Doppler signal preprocessing unit, so as to ensure the complete transmission of the Doppler signal.

[0021] Preferably, the analog-to-digital conversion and Doppler signal preprocessing unit incorporates a 16-bit or higher analog-to-digital conversion chip, an 8th-order anti-aliasing filter circuit, and an amplitude normalization circuit. It achieves analog-to-digital conversion of the analog Doppler signal through a preset sampling rate, removes out-of-band interference using anti-aliasing filtering, normalizes the signal amplitude, and outputs a normalized digital Doppler signal with a bit width matching that of subsequent processing. The preset sampling rate is set based on the highest effective frequency of the Doppler signal in the tested scene and the number of multi-scale wavelet decomposition levels, with a value range of 1kHz to 20MHz. It must satisfy the Nyquist sampling theorem and match the frequency band division requirements of wavelet decomposition.

[0022] Preferably, the Doppler signal multi-scale wavelet decomposition hardware unit adopts a pipelined cascaded FIR digital filter array, selects the coif5 wavelet basis to perform 4-level wavelet decomposition on the digital Doppler signal, and the pipelined cascaded FIR digital filter array adopts a 16-order tap design with a coefficient quantization accuracy of 16-bit fixed-point numbers.

[0023] Preferably, the Doppler background noise statistical characteristic estimation unit incorporates a hardware sorting circuit and a fixed-point parallel hardware arithmetic unit. Based on the high-frequency detail components obtained from the first-level wavelet decomposition, it calculates the background noise statistical standard deviation of the signal using the median absolute deviation statistical method. The calculation formula is as follows:

[0024]

[0025] in, The background noise statistical standard deviation of the Doppler signal. The first wavelet decomposition yields the... High-frequency detail components The median extraction function sorts the input discrete numerical sequence in ascending order using a hardware sorting circuit and extracts the value at the middle position of the sequence. The hardware sorting circuit is a fully pipelined parallel digital logic circuit, consisting of cascaded pairwise numerical comparators and multiplexers, and completes sorting and median output in a single clock cycle.

[0026] Preferably, the hierarchical adaptive threshold hardware calculation and nonlinear correction unit includes a decomposition level index register, a hierarchical threshold attenuation coefficient lookup table unit, a statistical item pre-calculation unit, a threshold calculation hardware circuit, and a nonlinear correction processing unit;

[0027] The decomposition level index register is used to store the current decomposition level index and call the corresponding logarithmic value in the pre-calculated parameter read-only storage mapping table, and combine it with the attenuation adjustment coefficient to complete the threshold calculation.

[0028] The layered threshold decay coefficient lookup table unit is a pre-calculated parameter read-only storage mapping table, used to pre-store the fixed natural logarithmic values ​​corresponding to different decomposition layers.

[0029] The statistical item pre-calculation unit is used to output the corresponding statistical item based on the current layer number index;

[0030] The threshold calculation hardware circuit includes a multiplier and a divider, used to receive the background noise statistical standard deviation, statistical terms and damping attenuation coefficient, and output the hierarchical adaptive threshold corresponding to the decomposition level.

[0031] The nonlinear correction processing unit corrects the wavelet decomposition frequency domain coefficients based on a hierarchical adaptive threshold. The correction formula for the frequency domain coefficients of layer wavelet decomposition is as follows:

[0032]

[0033] in, These are the frequency domain coefficients of the original wavelet decomposition. The original wavelet decomposition frequency domain coefficients amplitude, For the first Layer-specific adaptive threshold, The nonlinear compensation adjustment coefficient is adjusted in real time based on the statistical standard deviation of the background noise. The calculation formula is as follows: , This is a preset adjustable parameter, with a value range of 0.05 to 0.5. For symbol extraction function, The hyperbolic tangent function is a continuously differentiable nonlinear smoothing operator with a range of . Its square operator The range of values ​​is It is used to achieve nonlinear smoothing compensation of the frequency domain coefficients of wavelet decomposition.

[0034] Preferably, the filter parameters of the inverse wavelet transform pipelined FIR digital filter array are fully matched with those of the Doppler signal multi-scale wavelet decomposition hardware unit, and a fully pipelined structure is adopted to complete the first-level reconstruction operation in a single clock cycle.

[0035] Preferably, the power management module is a linear regulated power supply circuit with an output ripple of less than 10mV, used to provide a stable and low-noise operating power supply for all modules.

[0036] Preferably, the first input terminal of the noise reduction effect quantification evaluation and parameter closed-loop optimization unit receives the original digital Doppler signal output by the analog-to-digital conversion and Doppler signal preprocessing unit, and the second input terminal receives the noise-reduced Doppler signal output by the wavelet decomposition frequency domain coefficient inverse transform hardware reconstruction unit. Based on the two input signals, the signal-to-noise ratio (SNR) and root mean square error (RMSE) are calculated to generate a dynamic adjustment value for the hierarchical threshold attenuation coefficient, which is then output to the hierarchical adaptive threshold hardware calculation and nonlinear correction unit. The calculation formulas for the SNR and RMS are as follows:

[0037]

[0038]

[0039] in, The original digital Doppler signal output by the analog-to-digital conversion and Doppler signal preprocessing unit; The noise-reduced Doppler signal is the output of the wavelet decomposition frequency domain coefficient inverse transform hardware reconstruction unit; N is the number of sampling points of a single frame signal, and the number of sampling points of a single frame signal is not less than 1024 points.

[0040] Preferably, the noise reduction effect quantification evaluation and parameter closed-loop optimization unit aims to maximize the Doppler signal-to-noise ratio (SNR) and minimize the reconstructed signal amplitude error. The dynamic adjustment value of the layered threshold attenuation coefficient is used to adjust the attenuation adjustment coefficient in the layered adaptive threshold hardware calculation and nonlinear correction unit. The preset step size is 0.02, the preset target threshold for SNR is 20dB, the upper limit of the root mean square error (RMSE) is 0.05, and the lower limit of the RMSE is 0.02. The optimization logic is as follows: when the SNR is lower than the preset target threshold and the RMSE is higher than the preset upper limit, the attenuation adjustment coefficient is lowered by a preset step size to reduce the threshold of the deep decomposition level, so as to retain more effective low-frequency signals; when the SNR is higher than the preset target threshold and the RMSE is lower than the preset lower limit, the attenuation adjustment coefficient is raised by a preset step size to increase the threshold of the high-frequency decomposition level, so as to enhance the noise suppression capability and achieve closed-loop dynamic optimization of the threshold.

[0041] Beneficial Effects: Compared with existing technologies, this invention has the following significant advantages: 1. By introducing a layered threshold attenuation coefficient that varies with the number of decomposition layers, the threshold smoothly attenuates as the number of decomposition layers increases. This ensures sufficient suppression of high-frequency layer noise while avoiding excessive filtering of low-frequency effective frequency shift signals, solving the inherent defect of the existing fixed universal threshold's "one-size-fits-all" approach and perfectly adapting to the 1 / f distribution characteristics of Doppler signal noise; 2. The threshold function is continuously differentiable throughout, completely eliminating the reconstruction oscillation problem of hard threshold functions. Simultaneously, through the asymptotic convergence characteristics of the hyperbolic tangent square operator, unbiased estimation of the large-amplitude effective signal is achieved, solving the constant amplitude deviation defect of soft threshold functions. While filtering out noise, it retains 100% of the phase and amplitude information of the Doppler signal. 2. Employing a multi-scale wavelet decomposition hardware architecture, this invention accurately separates high-frequency noise from low-frequency effective frequency shift information, fully preserving the core characteristics of the Doppler signal. 3. This invention optimizes the attenuation adjustment coefficient through a closed-loop system using both signal-to-noise ratio (SNR) and root mean square error (RMSE). The threshold strategy can be dynamically adjusted based on the actual noise reduction effect, overcoming the limitation of existing open-loop fixed-parameter schemes that cannot adapt to different noise environments. In a 15dB low SNR input environment, the output SNR can reach up to 25.8dB, far exceeding the noise reduction performance of existing open-loop schemes. 4. Combining the advantages of time-frequency analysis and nonlinear correction, this invention achieves full-process hardware implementation through lookup tables and pipeline design, adapting to the Doppler signal processing needs of various scenarios such as underwater detection, industrial speed measurement, and medical testing. Attached Figure Description

[0042] Figure 1 This is a diagram of the overall architecture of the present invention;

[0043] Figure 2 This is a schematic diagram of the internal logic structure of the hierarchical adaptive threshold hardware calculation and nonlinear correction unit of the present invention;

[0044] Figure 3 This is a comparison diagram of the output waveforms of the noisy Doppler signal processed by the present invention. Detailed Implementation

[0045] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0046] like Figure 1 As shown, this invention provides an adaptive Doppler signal noise reduction filter. To achieve the above objectives, the specific hardware implementation and working process of this invention are as follows:

[0047] The adaptive Doppler signal noise reduction filter disclosed in this embodiment is configured with a sensor interface impedance matching and electrical isolation unit, an analog-to-digital conversion and Doppler signal preprocessing unit, and a power management module. It can be used as a front-end signal preprocessing unit for devices such as underwater acoustic Doppler current meters and industrial laser Doppler velocimeters. Each module of the filter is provided with a stable and low-noise operating power by a linear regulated power supply module, and the output ripple of the power supply module is less than 10mV.

[0048] First, the analog Doppler signal acquired by the front-end Doppler sensor is received through the sensor interface impedance matching and electrical isolation unit. After matching the sensor output characteristics with the input requirements of the subsequent analog-to-digital conversion and Doppler signal preprocessing unit, impedance matching and electrical isolation are performed on the analog Doppler signal to ensure the complete transmission of the Doppler signal and avoid signal reflection, attenuation, and electromagnetic interference. The sensor interface impedance matching and electrical isolation unit is an impedance matching interface circuit with a built-in TVS electrostatic protection unit and a magnetic coupling electrical isolation unit.

[0049] Secondly, the Doppler signal after interface conversion is sent to the analog-to-digital conversion and Doppler signal preprocessing unit. In this embodiment, for the echo signal of the underwater acoustic Doppler current meter, a preset sampling rate is set based on the highest effective frequency of the Doppler signal in the measured scene and the number of multi-scale wavelet decomposition levels. This rate satisfies the Nyquist sampling theorem and matches the frequency band division requirements of wavelet decomposition, with a range of 1kHz to 20MHz. In this embodiment, the sampling rate is set to 2000Hz. The analog-to-digital conversion and Doppler signal preprocessing unit has a built-in 16-bit analog-to-digital conversion chip, an 8th-order anti-aliasing Butterworth filter circuit, and an amplitude normalization circuit to convert the analog Doppler signal into a digital Doppler signal. Specifically, after filtering out out-of-band interference exceeding the Nyquist frequency, the amplitude of the Doppler signal is normalized to obtain a digital Doppler signal with a bit width matching the subsequent wavelet decomposition processing.

[0050] Next, the conditioned digital Doppler signal is fed into the Doppler signal multi-scale wavelet decomposition hardware unit and the noise reduction effect quantification evaluation and parameter closed-loop optimization unit, respectively. In this embodiment, the Doppler signal multi-scale wavelet decomposition hardware unit adopts a pipelined cascaded FIR digital filter array, mainly using the Coiflet series wavelet basis, specifically selecting the coif5 wavelet basis to perform 4-level wavelet decomposition on the digital Doppler signal, outputting the first... High-frequency detail components of the layer and low-frequency approximation components, where By using multi-scale decomposition, time-frequency localization analysis of non-stationary Doppler signals is achieved, effectively separating high-frequency background noise from low-frequency core frequency shift information, thus providing a foundation for subsequent layered noise reduction processing.

[0051] Then, the Doppler background noise statistical characteristic estimation unit reads the high-frequency detail components obtained from the first layer decomposition. The module incorporates a hardware sorting network and a fixed-point parallel hardware computing unit. First, the sorting network extracts the median of the absolute values ​​of the coefficients. Then, the median absolute deviation statistical method is used to calculate the statistical standard deviation of the signal's background noise. The calculation formula is as follows:

[0052]

[0053] in, The first wavelet decomposition yields the... High-frequency detail components, The median extraction operation implemented by the hardware sorting circuit calculates the statistical standard deviation of the background noise. The data is transmitted in real time to the hierarchical adaptive threshold hardware calculation and nonlinear correction unit, providing basic parameters for subsequent hierarchical adaptive threshold generation.

[0054] Median extraction is a core preliminary step in the median absolute deviation statistical method. It involves extracting the median value from a discrete numerical sequence after sorting it in ascending / descending order. If the sequence length is odd, the (N+1) / 2th value after sorting is taken as the median; if the sequence length is even, the average of the N / 2th and N / 2+1th values ​​is taken as the median. Because Doppler signals have low signal-to-noise ratios and are susceptible to pulse spikes, and because median extraction is insensitive to abnormal pulse noise and accurately reflects the true statistical level of background noise, median extraction is chosen over mean extraction. This avoids interference from effective signal components with noise intensity estimation results and is easier to implement using pure digital hardware circuits, meeting real-time processing requirements.

[0055] In this embodiment, the specific hardware execution process of the median extraction operation is as follows:

[0056] Step 1: Take the absolute value of each of the 512 high-frequency detail components output from the first wavelet decomposition layer and latch them into a 16-bit wide input register group to form a discrete numerical sequence of length 512.

[0057] The second step: The sequence is sorted in ascending order by a fully pipelined parallel sorting circuit, with cascaded pairwise comparators and multiplexers. The entire sequence can be sorted in a single clock cycle.

[0058] Step 3: Using a hardware address decoding circuit, select the values ​​of the 256th and 257th bits of the sorted registers, take the average of the two as the median of the sequence, and output it synchronously to the subsequent fixed-point parallel hardware arithmetic unit to complete the calculation of the statistical standard deviation of the background noise.

[0059] Furthermore, hierarchical adaptive threshold hardware computation and nonlinear correction units, such as Figure 2 The diagram shows the internal data flow of the hierarchical adaptive threshold hardware calculation unit. This unit includes a decomposition level index register, a hierarchical threshold attenuation coefficient pre-calculation read-only memory mapping table, a statistical term pre-calculation unit, and a fixed-point multiplication and division unit. It can calculate and output the current layer's adaptive threshold in real time based on the Doppler background noise statistical standard deviation, the length of the current layer's wavelet decomposition frequency domain coefficients, and the decomposition level index, providing parameters for the nonlinear compensation correction of subsequent wavelet decomposition frequency domain coefficients. This addresses the issue of Doppler signal noise increasing with the number of decomposition levels. The attenuation distribution characteristics are used to match a unique hierarchical adaptive threshold for each layer of decomposition coefficients. The calculation formula is as follows:

[0060]

[0061] In this embodiment, the number of decomposition layers Corresponding coefficient length The effect decreases exponentially with increasing number of floors. Attenuation adjustment coefficient The initial value is 0.1, and it is dynamically adjusted within the range of 0.02 to 0.2 based on the quantitative evaluation of the noise reduction effect and the feedback from the parameter closed-loop optimization unit; a layered threshold attenuation coefficient is introduced. This allows the threshold to decrease smoothly with increasing decomposition levels, ensuring sufficient suppression of noise in the high-frequency coefficients of the first layer while preventing excessive filtering of effective low-frequency signals in deeper layers. To reduce computational complexity and improve processing speed, different levels of decomposition are used for different thresholds. The values ​​are pre-stored in the pre-calculated parameter read-only storage mapping table and can be directly accessed through the layer index, which meets the requirements of real-time processing.

[0062] Furthermore, the nonlinear correction processing unit corrects the threshold of the wavelet decomposition frequency domain coefficients. The specific process is as follows: The amplitude and sign of the wavelet decomposition frequency domain coefficients are separated. A hardware comparator compares the coefficient amplitude with the corresponding hierarchical adaptive threshold. If the amplitude is less than or equal to the threshold, 0 is output directly, completing noise filtering. If the amplitude is greater than the threshold, soft thresholding and nonlinear compensation processing are performed simultaneously. The soft thresholding process calculates the basic difference using a fixed-point subtractor. ,in, Wavelet decomposition frequency domain coefficients amplitude, For the first Layered adaptive thresholding; nonlinear compensation is achieved by introducing a hyperbolic tangent square operator to synchronously start calculating the compensation term, and calculating the nonlinear compensation adjustment coefficient based on the statistical standard deviation of background noise. In this embodiment Take 0.2, then calculate the normalized variable, using the pre-stored hyperbolic tangent square value. The pre-computed parameters are stored in a read-only memory mapping table to obtain the mapping result, which is then compared with the first... Layer-based adaptive threshold Multiplying the values ​​yields the compensation term. Adding the basic difference to the compensation term gives the corrected amplitude, which is then synthesized with the original symbol to output the final denoised wavelet decomposition frequency domain coefficients.

[0063] The formula for correcting the frequency domain coefficients of wavelet decomposition based on soft thresholding and nonlinear compensation is shown below:

[0064]

[0065] in, These are the frequency domain coefficients of the original wavelet decomposition. Wavelet decomposition frequency domain coefficients amplitude, For the first Layer-specific adaptive threshold, The nonlinear compensation adjustment coefficient is adjusted in real time based on the statistical standard deviation of the background noise. The calculation formula is as follows: , This is a preset adjustable parameter, with a value range of 0.05 to 0.5. For symbol extraction function, The hyperbolic tangent function is a continuously differentiable nonlinear smoothing operator with a range of . Its square operator The range of values ​​is It is used to achieve nonlinear smoothing compensation of the frequency domain coefficients of wavelet decomposition.

[0066] In this embodiment, both the hierarchical threshold attenuation coefficient lookup table and the hyperbolic tangent square lookup table adopt a 16-bit address width and a 16-bit fixed-point data precision, and the pre-calculation range covers the full range of the input variables; the pipelined cascaded FIR digital filter array for multi-scale wavelet decomposition adopts a 16-order tap design, with coefficient quantization precision of 16-bit fixed-point numbers, and adopts a fully pipelined structure, completing the first-level decomposition operation in a single clock cycle; the signal buffer of the noise reduction effect quantization evaluation and parameter closed-loop optimization unit adopts a 32-bit wide dual-port RAM with a buffer depth of 1024 points, matching the sampling length of a single frame signal, and achieving precise alignment of the sampling points of the original signal and the noise-reduced signal through the frame synchronization signal.

[0067] Finally, the wavelet decomposition frequency domain coefficients after nonlinear compensation correction are sent to the wavelet decomposition frequency domain coefficient inverse transform hardware reconstruction unit. The wavelet decomposition frequency domain coefficient inverse transform hardware reconstruction unit adopts an inverse wavelet transform pipeline cascaded FIR digital filter array corresponding to the decomposition process to perform inverse transform on the high-frequency detail components and low-frequency approximate components of each layer after denoising, directly outputting the denoised Doppler signal, which is directly sent to the subsequent frequency shift calculation unit to complete the accurate measurement of parameters such as underwater flow velocity and industrial rotation speed. At the same time, the denoised Doppler signal is input to the noise reduction effect quantification evaluation and parameter closed-loop optimization unit.

[0068] The noise reduction effect quantification evaluation and parameter closed-loop optimization unit is the filter's closed-loop optimization unit. It has a built-in dual-port signal buffer, fixed-point hardware arithmetic unit, and optimization decision logic, and is used to quantify the noise reduction effect. The specific working process is as follows:

[0069] The first step is to synchronously buffer the original digital Doppler signal output by the analog-to-digital conversion and Doppler signal preprocessing unit and the noise-reduced Doppler signal output by the wavelet decomposition frequency domain coefficient inverse transform hardware reconstruction unit, and ensure that the sampling points of the two sets of signals are accurately aligned through the frame synchronization signal.

[0070] The second step involves the hardware computing unit performing parallel hardware calculations of the signal-to-noise ratio (SNR) and root mean square error (RMSE) based on two sets of synchronization signals. In this embodiment, the preset target SNR threshold is 20 dB, the upper limit of the RMS error is 0.05, and the lower limit of the RMS error is 0.02.

[0071] The third step is to optimize the decision unit by executing threshold optimization logic based on the calculation results: when the signal-to-noise ratio (SNR) is <20dB and the root mean square error (RMSE) is >0.05, the attenuation adjustment coefficient is lowered by a preset step size of 0.02 to reduce the threshold of the deep decomposition level and prevent the effective low-frequency signal from being over-filtered; when the SNR is >20dB and the RMSE is <0.02, the attenuation adjustment coefficient is raised by a preset step size of 0.02 to increase the threshold of the high-frequency decomposition level and further suppress residual noise. After optimization, the attenuation adjustment coefficient is updated in real time to the parameter register of the hierarchical adaptive threshold hardware calculation and nonlinear correction unit, and the threshold is dynamically adjusted in a closed loop to optimize the noise reduction performance. The preset step size is an adjustable fixed value pre-stored in the parameter register, with a value range of 0.01~0.05 and a preferred value of 0.02. The adjustment range of the attenuation adjustment coefficient is limited to 0.02~0.2. The preset target threshold range of the signal-to-noise ratio is 18dB~25dB. The preset upper limit of MSE is 0.04~0.06 and the preset lower limit of MSE is 0.01~0.03.

[0072] In this embodiment, after the noise reduction effect is quantitatively evaluated and the parameter closed-loop optimization unit performs closed-loop optimization, the input noisy Doppler signal with a signal-to-noise ratio of 15dB is used, such as Figure 3 As shown, the output signal-to-noise ratio can reach up to 25.8dB, and the root mean square error can be as low as 0.021. Compared with the noise reduction filter with fixed parameters in open loop, the noise reduction performance is further improved, and it can adaptively adapt to the Doppler signal processing requirements under different noise environments.

Claims

1. An adaptive Doppler signal denoising filter, characterized by, The filter is integrated into an FPGA or ASIC hardware circuit, including a sensor interface impedance matching and electrical isolation unit, an analog-to-digital conversion and Doppler signal preprocessing unit, a Doppler signal multi-scale wavelet decomposition hardware unit, a Doppler background noise statistical characteristic estimation unit, a hierarchical adaptive threshold hardware calculation and nonlinear correction unit, a wavelet decomposition frequency domain coefficient inverse transform hardware reconstruction unit, a noise reduction effect quantitative evaluation and parameter closed-loop optimization unit, and a power management module to power the filter. The input terminal of the sensor interface impedance matching and electrical isolation unit receives the externally input analog Doppler signal, and the output terminal of the sensor interface impedance matching and electrical isolation unit is connected to the input terminal of the analog-to-digital conversion and Doppler signal preprocessing unit to complete the impedance matching and electrical isolation of the Doppler signal. The first output of the analog-to-digital conversion and Doppler signal preprocessing unit is connected to the input of the Doppler signal multi-scale wavelet decomposition hardware unit, and the second output of the analog-to-digital conversion and Doppler signal preprocessing unit is connected to the first input of the noise reduction effect quantization evaluation and parameter closed-loop optimization unit. This unit is used to convert the analog Doppler signal into a digital signal, complete anti-aliasing filtering and amplitude normalization processing, and output a standardized digital Doppler signal. The first output of the Doppler signal multi-scale wavelet decomposition hardware unit is connected to the input of the Doppler background noise statistical characteristic estimation unit, and the second output of the Doppler signal multi-scale wavelet decomposition hardware unit is connected to the first input of the hierarchical adaptive threshold hardware calculation and nonlinear correction unit. The Coiflet wavelet basis is used to perform 3 to 5 layers of wavelet decomposition on the input digital Doppler signal, which is used to output the wavelet decomposition frequency domain coefficients and decomposition layer index corresponding to each layer. The output of the Doppler background noise statistical characteristic estimation unit is connected to the second input of the hierarchical adaptive threshold hardware calculation and nonlinear correction unit. It has a built-in hardware sorting circuit and a fixed-point parallel hardware operation unit, which is used to calculate the background noise statistical standard deviation of the Doppler signal based on the high-frequency detail components obtained by the first layer wavelet decomposition and through the median absolute deviation statistical method. The feedback input of the hierarchical adaptive threshold hardware calculation and nonlinear correction unit is connected to the output of the noise reduction effect quantification evaluation and parameter closed-loop optimization unit. The output of the hierarchical adaptive threshold hardware calculation and nonlinear correction unit is connected to the input of the wavelet decomposition frequency domain coefficient inverse transform hardware reconstruction unit. It is used to generate the hierarchical adaptive threshold of the corresponding decomposition level according to the current decomposition level, the length of the current layer wavelet decomposition frequency domain coefficient, the statistical standard deviation of Doppler background noise, and the feedback optimization parameters of the noise reduction effect quantification evaluation and parameter closed-loop optimization unit, and to perform nonlinear compensation correction processing on the nonlinear wavelet coefficients of the wavelet decomposition frequency domain coefficients, that is, to correct the wavelet decomposition frequency domain coefficients based on the hierarchical adaptive threshold. The formula for calculating the hierarchical adaptive threshold is: in, The statistical standard deviation of background noise. This is the attenuation adjustment coefficient, with an initial value of 0.1 and an adjustment range of 0.02 to 0.

2. This represents the current decomposition level. For the first The actual length of the frequency domain coefficients in layer wavelet decomposition. For the first The layer has a fixed natural logarithm value. This is the attenuation coefficient for the stratified threshold. The calculation formula for the nonlinear compensation and correction processing of the nonlinear wavelet coefficients is as follows: in, These are the frequency domain coefficients of the original wavelet decomposition. The magnitude of the frequency domain coefficients of the original wavelet decomposition is given. For the first Layer-specific adaptive threshold, For nonlinear compensation adjustment coefficients, For symbol extraction function, The function is a hyperbolic tangent function, and the nonlinear compensation adjustment coefficient is... Adjusted in real time based on the statistical standard deviation of background noise, the calculation formula is as follows: , The preset adjustable parameter has a value range of 0.05~0.5, with a preferred value of 0.2; the hyperbolic tangent square value is pre-stored in a 16-bit wide pre-calculated parameter read-only memory mapping table, and is read in a single clock cycle through address indexing; The wavelet decomposition frequency domain coefficient inverse transform hardware reconstruction unit adopts an inverse wavelet transform pipelined FIR digital filter array corresponding to the multi-scale wavelet decomposition hardware unit of Doppler signal. Its first output terminal is connected to the second input terminal of the noise reduction effect quantification evaluation and parameter closed-loop optimization unit. The second output terminal outputs the final noise-reduced Doppler signal, which is used to perform wavelet inverse transform on the wavelet decomposition frequency domain coefficients after wavelet coefficient nonlinear compensation correction, and output the noise-reduced Doppler signal. The output of the noise reduction effect quantification evaluation and parameter closed-loop optimization unit is connected to the feedback input of the hierarchical adaptive threshold hardware calculation and nonlinear correction unit. It is used to calculate the signal-to-noise ratio and root mean square error based on the original digital Doppler signal output by the analog-to-digital conversion and Doppler signal preprocessing unit and the noise-reduced signal output by the wavelet decomposition frequency domain coefficient inverse transform hardware reconstruction unit. With the optimization objectives of maximizing the Doppler signal-to-noise ratio and minimizing the amplitude error of the reconstructed signal, the attenuation adjustment coefficient adjustment value is generated and fed back to the hierarchical adaptive threshold hardware calculation and nonlinear correction unit in a closed loop to realize the dynamic optimization of the threshold.

2. The adaptive Doppler signal noise reduction filter according to claim 1, characterized in that, The sensor interface impedance matching and electrical isolation unit is an impedance matching interface circuit with a built-in TVS electrostatic protection unit and a magnetic coupling electrical isolation unit. It is used to match the output characteristics of the Doppler sensor with the input requirements of the analog-to-digital conversion and Doppler signal preprocessing unit to ensure the complete transmission of the Doppler signal.

3. The adaptive Doppler signal noise reduction filter according to claim 1, characterized in that, The analog-to-digital conversion and Doppler signal preprocessing unit incorporates a 16-bit or higher analog-to-digital conversion chip, an 8th-order anti-aliasing filter circuit, and an amplitude normalization circuit. It completes the analog-to-digital conversion of the analog Doppler signal at a preset sampling rate, filters out out-of-band interference signals, and outputs a standardized digital Doppler signal with a bit width that matches the subsequent processing. The preset sampling rate is set based on the highest effective frequency of the Doppler signal in the tested scene and the number of multi-scale wavelet decomposition layers, with a value range of 1kHz to 20MHz. It must satisfy the Nyquist sampling theorem and match the frequency band division requirements of wavelet decomposition.

4. The adaptive Doppler signal noise reduction filter according to claim 1, characterized in that, The hardware unit for multi-scale wavelet decomposition of the Doppler signal adopts a pipelined cascaded FIR digital filter array. The coif5 wavelet basis is selected to perform 4-level wavelet decomposition on the digital Doppler signal. The pipelined cascaded FIR digital filter array adopts a 16-order tap design, and the coefficient quantization accuracy is 16-bit fixed-point.

5. An adaptive Doppler signal noise reduction filter according to claim 1, characterized in that, The Doppler background noise statistical characteristic estimation unit incorporates a hardware sorting circuit and a fixed-point parallel hardware arithmetic unit. Based on the high-frequency detail components obtained from the first-level wavelet decomposition, it calculates the statistical standard deviation of the signal's background noise using the median absolute deviation statistical method. The calculation formula is as follows: in, The background noise statistical standard deviation of the Doppler signal. The first wavelet decomposition yields the... High-frequency detail components The median extraction function sorts the input discrete numerical sequence in ascending order using a hardware sorting circuit and extracts the value at the middle position of the sequence. The hardware sorting circuit is a fully pipelined parallel digital logic circuit, consisting of cascaded pairwise numerical comparators and multiplexers, and completes sorting and median output in a single clock cycle.

6. The adaptive Doppler signal noise reduction filter according to claim 1, characterized in that, The hierarchical adaptive threshold hardware calculation and nonlinear correction unit includes a decomposition hierarchical index register, a hierarchical threshold attenuation coefficient lookup table unit, a statistical item pre-calculation unit, a threshold calculation hardware circuit, and a nonlinear correction processing unit. The decomposition level index register is used to store the current decomposition level index and call the corresponding logarithmic value in the pre-calculated parameter read-only storage mapping table, and combine it with the attenuation adjustment coefficient to complete the threshold calculation. The layered threshold decay coefficient lookup table unit is a pre-calculated parameter read-only storage mapping table, used to pre-store the fixed natural logarithmic values ​​corresponding to different decomposition layers. The statistical item pre-calculation unit is used to output the corresponding statistical item based on the current layer number index; The threshold calculation hardware circuit includes a multiplier and a divider, used to receive the background noise statistical standard deviation, statistical terms and damping attenuation coefficient, and output the hierarchical adaptive threshold corresponding to the decomposition level.

7. The adaptive Doppler signal noise reduction filter according to claim 1, characterized in that, The inverse wavelet transform pipelined cascaded FIR digital filter array is perfectly matched with the filter parameters of the Doppler signal multi-scale wavelet decomposition hardware unit. It adopts a fully pipelined structure and completes the first-level reconstruction operation in a single clock cycle.

8. An adaptive Doppler signal noise reduction filter according to claim 1, characterized in that, The power management module is a linear regulated power supply circuit with an output ripple of less than 10mV, providing multi-channel isolated power supply for all modules.

9. An adaptive Doppler signal noise reduction filter according to claim 1, characterized in that, The first input of the noise reduction effect quantification evaluation and parameter closed-loop optimization unit receives the original digital Doppler signal output from the analog-to-digital conversion and Doppler signal preprocessing unit, and the second input receives the noise-reduced Doppler signal output from the wavelet decomposition frequency domain coefficient inverse transform hardware reconstruction unit. Based on the two input signals, the signal-to-noise ratio (SNR) and root mean square error (RMSE) are calculated to generate a dynamic adjustment value for the hierarchical threshold attenuation coefficient, which is then output to the hierarchical adaptive threshold hardware calculation and nonlinear correction unit. The formulas for calculating the SNR and RMS are as follows: in, The original digital Doppler signal output by the analog-to-digital conversion and Doppler signal preprocessing unit; The noise-reduced Doppler signal is the output of the wavelet decomposition frequency domain coefficient inverse transform hardware reconstruction unit; N is the number of sampling points of a single frame signal, and the number of sampling points of a single frame signal is not less than 1024 points.

10. An adaptive Doppler signal noise reduction filter according to claim 9, characterized in that, The noise reduction effect quantification evaluation and parameter closed-loop optimization unit aims to maximize the Doppler signal-to-noise ratio and minimize the reconstructed signal amplitude error. The dynamic adjustment value of the layered threshold attenuation coefficient is used to adjust the attenuation adjustment coefficient in the layered adaptive threshold hardware calculation and nonlinear correction unit. The preset step size is 0.02, the preset target threshold for signal-to-noise ratio is 20dB, the upper limit of root mean square error is 0.05, and the lower limit of root mean square error is 0.

02. The optimization logic is as follows: when the signal-to-noise ratio is lower than the preset target threshold and the root mean square error is higher than the preset upper limit, the attenuation adjustment coefficient is lowered by the preset step size to reduce the threshold of the deep decomposition level, so as to retain more effective low-frequency signals. When the signal-to-noise ratio is higher than the preset target threshold and the root mean square error is lower than the preset lower limit, the attenuation adjustment coefficient is increased by a preset step size to improve the threshold of the high-frequency decomposition level, thereby enhancing the noise suppression capability and realizing closed-loop dynamic optimization of the threshold.

Images