Adaptive noise suppression dynamic shift quantization modulation strength signal separation method and system

Abstract

The application discloses a dynamic shift quantization modulation strength and weakness signal separation method and system with adaptive noise suppression, relates to signal processing technology, and comprises the following steps: performing symbol extension on an input signal; adopting a specified sliding window to count the maximum value Amax and the minimum value Amin of the signal and dynamically calculating required parameters for the symbol-extended input signal; injecting a first disturbance into the symbol-extended input signal, and performing high-pass noise shaping on the signal after the injection; amplifying the shaped signal by shifting it to the left by m bits, and injecting a second disturbance into the signal after the amplification; performing low-pass noise shaping on the signal after the injection of the second disturbance; extracting a strong signal by shifting the low-pass noise-shaped signal to the right by n bits and then to the left by n bits, and separating a weak signal by subtraction; and recovering the order of magnitude of the strong signal component and the weak signal component. The application provides a non-cooperative mixed signal separation method and system combining dynamic shift quantization modulation, phased disturbance injection and time-domain adaptive noise shaping technology.

Description

Technical Field

[0001] This application relates to the field of signal processing technology, to non-cooperative signal processing under external radiation source detection systems, and particularly to an adaptive noise suppression method and system for separating strong and weak mixed signals by dynamic shift quantization modulation. Background Technology

[0002] In electromagnetic sensing scenarios, received signals typically consist of strong direct waves (directly from the transmitter) and weak echoes (reflected by the target or scattered by the environment). The dynamic range difference between the two exceeds 60 dB (e.g., the strong signal amplitude is ±32767, while the weak signal is only ±328). Strong direct waves account for over 90% of the signal energy, masking weak echoes and leading to the loss of target information. Weak echoes contain target characteristics (such as range, velocity, and angle), but their amplitude is extremely low, making them easily obscured by noise and quantization errors. For example, in external radiation source radars using broadcast / communication signals as illumination sources, the dynamic range difference between strong direct waves and weak target echoes in the received signal often exceeds 80 dB; in electronic reconnaissance, strong enemy jamming signals are mixed with weak friendly communication signals, making separation difficult using traditional methods.

[0003] Achieving efficient and high-fidelity separation of strong direct wave and weak echo signals is fundamental for subsequent effective demodulation, localization, identification, and tracking. Three common approaches are employed: First, the reference antenna method, which uses an independent reference antenna channel to acquire a "clean" strong direct wave signal and then eliminates the strong signal component through adaptive filtering or cancellation techniques to extract the weak echo. Second, the transform domain separation method, which transforms the signal to the frequency domain, time-frequency domain (such as short-time Fourier transform), or sparse representation domain, utilizing the differences in the distribution of strong and weak signals in different domains for separation. Third, the blind source separation method, based on independent component analysis (ICA) or principal component analysis (PCA), assumes that strong and weak signals are statistically independent and achieves separation through matrix decomposition.

[0004] Comparing the three methods: the reference antenna method has relatively high cancellation accuracy (SNR improvement > 20dB), but it relies on additional hardware, increasing system complexity, cost, and environmental sensitivity with poor adaptability. Multipath interference leads to phase inconsistency, which degrades cancellation performance. The transform domain separation method suppresses strong signal bands in the frequency domain or time-frequency domain (e.g., FFT zeroing). Frequency domain processing is intuitive and suitable for narrowband signals, but computation is complex, real-time performance is poor, and phase consistency is difficult to guarantee. The blind source separation method does not require prior information and has strong theoretical universality, but the assumptions are stringent (strong / weak signals may be related due to multipath effects). In real-world scenarios, its performance is unstable, and iterative convergence is slow with large delays (> 10ms).

[0005] Another method is time-domain adaptive filtering separation, including the LMS / NLMS and RLS algorithms. These methods utilize adaptive filters to suppress strong signals and extract weak signals by adjusting the filter coefficients. The RLS algorithm, relatively speaking, converges faster and is suitable for real-time processing. For stable mixed signals, the weak signal separation effect is acceptable, but for changing signals, the effect is poor, and the computational resources are very high.

[0006] Disadvantages and shortcomings of existing technologies Hardware dependency and insufficient adaptability to complex environments: Existing technologies rely on independent reference antennas, floating-point operations, or complex frequency domain transformations, resulting in high hardware costs, poor real-time performance, and performance degradation in scenarios with multipath interference and signal abrupt changes. For example, the reference antenna method requires additional hardware (increasing costs by more than 30%), and multipath interference causes phase mismatch; the reference antenna method requires the additional deployment of high-isolation antennas, and in complex terrain (such as urban multipath environments), when the antenna spacing error exceeds λ / 10 (one-tenth of the wavelength), the cancellation performance degrades by more than 10dB; the frequency domain separation method (FFT / STFT) is computationally complex (delay > 1ms), and cannot meet the real-time requirements of radar pulse level.

[0007] The quantization noise's effect on weak signals: When the amplitude of a weak signal is close to the quantization noise level (e.g., a quantization step size of 1 in a 16-bit ADC), traditional methods amplify the noise simultaneously when amplifying the weak signal, resulting in insufficient improvement in the signal-to-noise ratio (SNR) after separation. For example, RLS filtering can only suppress noise by about 15dB, and the weak signal is still submerged.

[0008] The conflict between dynamic range and real-time performance: Signal dynamic range fluctuates drastically (e.g., radar echoes change instantaneously by 40-70 dB), and traditional adaptive algorithms (such as LMS / RLS) require iterative convergence (delay > 200 μs), making it impossible to balance real-time performance and accuracy. For example, fixed-parameter designs cannot adapt to sudden changes in dynamic range (e.g., a sudden appearance of a target causing a jump in signal amplitude); real-time parameter adjustment algorithms (such as sliding window statistics) have high computational complexity and are difficult to deploy in low-power hardware.

[0009] The problem of wasting resources on strong signals: Existing methods focus on "suppressing strong signals," which leads to the loss of phase and amplitude information of the separated strong signals, making them unusable for subsequent tasks such as synchronization demodulation and reference signal generation. For example, strong signals cannot be reused after being canceled or zeroed in the frequency domain; phase truncation errors (such as setting low-level bits to zero in a right shift operation) cause distortion of strong signals. Summary of the Invention

[0010] This application provides an adaptive noise-suppressing method and system for separating strong and weak signals using dynamic shift quantization modulation (DSQM), which proposes a non-cooperative mixed signal separation method and system that combines dynamic shift quantization modulation (DSQM), staged perturbation injection, and time-domain adaptive noise shaping techniques.

[0011] This application provides an adaptive noise suppression method for separating strong and weak signals using dynamic shift quantization modulation, including: Sign extension of the input signal; For the symbol-spread input signal, the maximum value Amax and minimum value Amin of the signal are statistically analyzed using a specified sliding window. The shift parameters m, n, and k required for subsequent shift modulation, as well as the parameters required for shaping, are dynamically calculated. Furthermore, a first perturbation is injected into the symbol-spread signal. After injection, high-pass noise shaping is performed, and the high-pass filter for high-pass noise shaping adopts a first-order feedback structure. The shaped signal is shifted left by m bits and amplified, and a second perturbation is injected after amplification. ; The signal after the second disturbance is injected is subjected to low-pass noise shaping. The low-pass filter for low-pass noise shaping adopts a first-order feedback structure. For a low-pass noise shaped signal, a shift modulation method of shifting right by n bits and then left by n bits is used to extract the strong signal component, and a subtraction operation is performed between the low-pass noise shaped signal and the extracted strong signal to separate the weak signal component. The strong signal component and the weak signal component are shifted to the right by m bits to restore the magnitude.

[0012] Symmetrical and asymmetric truncation are performed on strong and weak signals of the recovery magnitude, respectively, to restore the bit width output.

[0013] This application provides a dynamic shift quantization modulation non-cooperative strong / weak signal separation system based on staged perturbation and adaptive noise shaping, including: The sign extension module is configured to extend the sign of the input signal; The dynamic parameter calculation module uses a specified sliding window to statistically analyze the maximum value Amax and minimum value Amin of the input signal after sign expansion, and dynamically calculates the required shift parameters. The phased perturbation injection module injects the first perturbation after sign expansion and after left-shift amplification, respectively. Second disturbance ; Noise shaping module, including high-pass filter ( ) and low-pass filter ( After injecting the first disturbance d1, high-pass noise shaping is performed. The high-pass filter for high-pass noise shaping adopts a first-order feedback structure. After injecting the second disturbance, low-pass noise shaping is performed. The low-pass filter for low-pass noise shaping adopts a first-order feedback structure. The signal separation module is used to extract strong signals from low-pass noise shaped signals by right shifting n bits and then left shifting n bits, and to separate weak signals by subtraction. The magnitude recovery module is used to recover the magnitude of strong and weak signal components.

[0014] The truncation output module is used to symmetrically truncate strong signals and asymmetrically truncate weak signals respectively, ensuring the recovery of the bit width.

[0015] This application proposes a non-cooperative hybrid signal separation method and system that combines dynamic shift quantization modulation (DSQM), staged perturbation injection, and time-domain adaptive noise shaping techniques. It is applicable to high dynamic range (>90dB), low latency (<100ns), and high-precision real-time signal processing scenarios such as external radiation source radar, 5G communication anti-interference, and medical ultrasound imaging.

[0016] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0017] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a basic flowchart illustrating the dynamic shift quantization modulation strong-weak mixed signal separation method according to an embodiment of this application. Figure 2 This is a schematic diagram of the architecture and data flow of the dynamic shift quantization modulation strong-weak mixed signal separation system according to an embodiment of this application; Figure 3 This is a schematic diagram of dynamic parameter calculation and binding perturbation / shaping of a dynamic shift quantization modulation strong and weak mixed signal separation system according to an embodiment of this application. Detailed Implementation

[0018] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0019] This application provides an improved non-cooperative signal separation method and system based on Dynamic Shift Quantization Modulation (DSQM), belonging to the field of digital signal processing technology. The method of this application employs staged perturbation injection (after symbol expansion). After perturbation and left shift amplification This paper describes a collaborative design of perturbation and adaptive noise shaping (high-pass / low-pass filtering), combined with a shaping coefficient control strategy linked to dynamic parameters (mn), to achieve high-precision phase-preserving separation under extreme dynamic range (>90dB) in FPGA hardware architecture. Core innovations include: 1) staged noise spectrum shaping technology to improve the signal-to-noise ratio (SNR) of weak signals to over 70dB; 2) a hardware optimization architecture based on shift addition, with resource increment <15% (total usage 670 LUTs) and latency ≤90ns@100MHz. It is applicable to radar, communication, medical imaging, and other scenarios. Specifically, this application provides a dynamic shift quantization modulation non-cooperative strong-weak mixed signal separation method combining staged adaptive perturbation injection and noise shaping technology, such as... Figure 1 As shown, it includes the following steps: In step S101, the input signal is sign-extended. In some embodiments, sign-extending the input signal includes reducing the bit width from B... in Expand to =2× The high-order bit is filled with the sign bit.

[0020] In step S102, for the symbol-expanded input signal, the maximum value Amax and minimum value Amin of the signal are statistically analyzed using a specified sliding window, and the parameters required for shift modulation, perturbation, and shaping are dynamically calculated, such as the shift parameters m, n, k required for shift modulation. / And, to the symbol-spread signal, inject the first perturbation. After injection, high-pass noise shaping is performed, and the high-pass filter for high-pass noise shaping adopts a first-order feedback structure.

[0021] In step S103, the shaped signal is shifted left by m bits and amplified, and a second perturbation is injected after amplification. .

[0022] In step S104, the signal after the second disturbance is injected is subjected to low-pass noise shaping, and the low-pass filter for low-pass noise shaping adopts a first-order feedback structure.

[0023] In step S105, the strong signal is extracted from the low-pass noise shaped signal by a shift modulation method of shifting right by n bits and then left by n bits, and the weak signal is separated by subtracting the extracted strong signal from the low-pass noise shaped signal.

[0024] In step S106, the strong signal component and the weak signal component are shifted right by m bits to restore their magnitude.

[0025] In step S107, symmetric truncation and asymmetric truncation are performed on the strong signal and the weak signal of the restored magnitude respectively to restore the bit width output. Specifically, the weak signal truncation retains negative details, and the strong signal truncation is limited within the original bit width range.

[0026] In some embodiments, the parameters required for dynamically calculating shift modulation, perturbation and shaping include: Right shift number: n=floor(log2(Amax / max(Amin,1))); Signal-to-noise ratio: SNRdB=20log10(Amax / Amin); Adaptive shift number: when SNRdB≥10dB, k=ceil(SNRdB / 3); when SNRdB<10dB, k=2; Left shift number: m=clamp(n+k,3,B), where B=Bext-Bin.

[0027] In some embodiments, the amplitude of the injected first perturbation d1 is |d1|=clamp(2^(m-n),1,Amin / 8); The injected second perturbation Amplitude is | |=clamp(2^(m-n),1,Amin<<m / 8), where Amax and Amin are the maximum value and the minimum value of the sliding window statistical signal respectively.

[0028] In some embodiments, the coefficient of the high-pass filter is: =0.5+0.12*(m-n), ranging from 0.5 to 0.9; The coefficient of the low-pass filter is: =0.3+0.08*(m-n), ranging from 0.3 to 0.8.

[0029] In some embodiments, the high-pass filter is implemented by shift addition. For example, in some examples, =0.85, the formula is: y=x+(prev_diff-(prev_diff>>3)-(prev_diff>>5)), where prev_diff=prev_y-prev_x.

[0030] In some embodiments, the dynamic calculation is implemented based on a parallel comparator tree, and the Amax / Amin statistics of a 256-point window are completed in a single cycle.

[0031] In some embodiments, the method further comprises: for the signal of the restored magnitude, using symmetric saturation truncation for the strong signal component and asymmetric truncation for the weak signal to output.

[0032] This application also proposes an adaptive noise suppression dynamic shift quantization modulation strong-weak mixed signal separation system, including: The sign extension module is configured to extend the sign of the input signal; The dynamic parameter calculation module uses a specified sliding window to statistically analyze the maximum value Amax and minimum value Amin of the input signal after sign expansion, and dynamically calculates the required parameters. The phased perturbation injection module injects the first perturbation d1 and the second perturbation d2 after sign expansion and left shift amplification, respectively. Noise shaping module, including high-pass filter ( ) and low-pass filter ( After injecting the first disturbance d1, high-pass noise shaping is performed. The high-pass filter for high-pass noise shaping adopts a first-order feedback structure. After injecting the second disturbance, low-pass noise shaping is performed. The low-pass filter for low-pass noise shaping adopts a first-order feedback structure. The signal separation module is used to extract strong signals from low-pass noise shaped signals by right shifting n bits and then left shifting n bits, and to separate weak signals by subtraction. The magnitude recovery module is used to recover the magnitude of strong and weak signal components.

[0033] The truncation output module is used to symmetrically truncate strong signals and asymmetrically truncate weak signals respectively, ensuring the recovery of the bit width.

[0034] In a specific example, the system of this application is as follows: like Figure 2 As shown, the data flow is as follows: A [Input Signal] --> B [Signal Extension] --> C [d1 Perturbation Injection] --> D [Noise Shaping 1 (High-Pass)] --> E [Left-Shift Amplification] --> F [d2 Perturbation Injection] --> G [Noise Shaping 2 (Low-Pass)]: G --> H [Strong Signal Extraction] --> I [Scale Recovery and Truncation Output] G --> J [Weak Signal Separation] --> K [Scale Recovery and Output Truncation] Control flow: B [Symbol Expansion] --> Parameters m / n / k [Dynamic Parameter Calculation] This application is aimed at and Stage noise characteristics, stage-by-stage matching and directional noise shaping (high-pass / low-pass).

[0035] Noise Shaping 1 (NS1): Location of action: After perturbation injection, shift left before amplification Target noise: Original quantization noise (already quantized) (Perturbation whitening) Strategy: Shift noise energy to the lower frequency range of the signal (e.g., 0~0.2Fs) to reduce the impact on the high-frequency details of weak signals. Filter Design: First-order high-pass feedback filter y[n]=x[n]+ ⋅(y[n-1]-x[n-1]) Noise Shaping 2 (NS2): Location of action: After perturbation injection, before strong signal extraction Target noise: Amplified and truncated noise (introduced by right-shift-left-shift operation) Strategy: Shift noise energy to higher frequency bands of the signal (e.g., 0.8Fs~Fs) to avoid interfering with the low-frequency components of strong signals. Filter Design: First-order low-pass feedback filter y[n]=x[n]+ ⋅(y[n-1]-x[n-1]).

[0036] The precise synergy effect of staged perturbation injection is shown in the table below: against / The noise characteristic differentiation is directly suppressed in the time domain, which is precise, has low resource consumption, and high phase fidelity.

[0037] A phased first-order filter design maximizes the balance between hardware efficiency and noise suppression, and allows for flexible dynamic parameter binding, such as... Figure 3 As shown. Shaping coefficient ( / Deep binding with mn enables adaptive spectrum shaping. Physical meaning of the derived parameter mn (left-shifted protection bits) The weak signal amplification factor, 2^(mn), directly determines the visibility of the weak signal in quantization noise.

[0038] Effective bit width retention: The larger mn is, the more effective bits are retained for weak signals, and the lower the quantization noise floor.

[0039] Where: n, right shift bits, n=floor(log2(Amax / max(Amin,1))); k, SNR adaptive shift bits, k=ceil(SNR_dB / 3), SNR_dB=20log10(Amax / Amin); m, left shift bits, m=clamp(n+k,3,B) (B is the overflow protection bits).

[0040] The coefficient increment is proportional to mn, ensuring an improvement of each unit mn. / The increases of 0.12 and 0.08 respectively resulted in a significant improvement in noise transfer efficiency.

[0041] Range constraints: (to ensure the stability and validity of coefficients) Lower limit 0.5: Noise floor reduction of 15 dB @ 0.1 Fs; Upper limit 0.9: To avoid excessive high-frequency suppression leading to signal distortion (THD<2%); Lower limit 0.3 → Suppression 18dB@0.1Fs; Upper limit 0.8: to prevent excessive attenuation of low frequencies from affecting the integrity of strong signals (THD<2%).

[0042] 0.12 ≈ 1 / 8 ≈ right shift 3 bits, 0.08 ≈ 1 / 16 ≈ right shift 4 bits, supports shift and addition.

[0043] This application also proposes a computational optimization implementation method. Coefficient formula decomposition calculation: =0.5 + (mn>>3) + (mn>>4) → Approximately 0.5+0.12*(mn) (error <1.2%); =0.3 + (mn>>4) → Approximately 0.3+0.08*(mn) (error <1.5%).

[0044] Resource consumption: Each coefficient calculation requires 2 shifters + 1 adder (≈40 LUTs).

[0045] Implementation steps: Step 1: The dynamic parameter calculation module updates m / n / k in real time. Step 2: / Bind to mn and optimize the shift-add module in the FPGA to reduce calculation errors; The beneficial effects are shown in the table below: The specific technical implementation of the system in this application embodiment can be adopted in the following ways: Hardware architecture (FPGA implementation) Symbol extension module: Function: 16-bit → 32-bit sign extension; Resources: 20 LUTs, 1 cycle delay (10ns@100MHz).

[0046] Dynamic parameter calculation module: Submodule: Parallel comparator tree (single-cycle extreme value statistics, 128 LUTs); Logarithmic computation (priority encoder + lookup table, 50 LUTs).

[0047] Total resources: 178 LUTs, latency 3 cycles.

[0048] Noise shaping module: High-pass filter: ( =0.85 optimized implementation).

[0049] Resources: 45 LUTs, 1 cycle delay.

[0050] Low-pass filter: ( =0.75 optimized implementation).

[0051] Resources: 45 LUTs, 1 cycle delay.

[0052] Example 1: Separation of radar signals from external radiation sources Input signals: Strong direct wave: Amax=32767 (16-bit), weak target echo: Amin=1024 (SNR_dB=30dB).

[0053] Parameter calculation: mn = 10; =0.5 + 0.12 * 10 = 0.9 =0.3+0.08*10=0.8.

[0054] Effect: The weak signal SNR increased from 30dB to 75dB (an improvement of 45dB), with a phase error of 0.8°. Resource usage: 670 LUTs, latency: 90ns@100MHz.

[0055] Example 2: 5G communication interference resistance Input signals: Strong interference: Amax=30000 (neighboring base station), weak user signal: Amin=600 (SNR_dB=34dB).

[0056] Parameter calculation: mn = 10; =0.5 + 0.12 * 10 = 0.9 =0.3+0.08*10=0.8.

[0057] Effect: The bit error rate changed from 1e-3 to 1e-6, with a dynamic range of 90dB. Phase error 0.7°, supports real-time demodulation (delay 90ns) This application achieves significant improvements in weak signal signal-to-noise ratio (SNR improvement ≥15dB) and phase fidelity (error ≤0.9°) by employing a synergistic optimization approach of "staged perturbation injection + time-domain noise shaping," while maintaining the lightweight nature (<700 LUTs) and low latency (<100ns) of the DSQM method and system. The noise shaping coefficients are dynamically bound in real-time, further reducing hardware complexity and making the application suitable for high-precision scenarios such as radar, communication, and medical imaging.

[0058] It should be noted that, in the embodiments of this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0059] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0060] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0061] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims. All of these forms are within the protection scope of this application.

Claims

1. A method for separating strong and weak signals in dynamic shift quantization modulation with adaptive noise suppression, characterized in that, comprising: Sign-extend the input signal to increase the bit width from B in Extend to B ext =2×B in High-order fill sign bit; For the sign-extended signal, a specified sliding window is used to count the maximum value Amax and the minimum value Amin of the signal, and parameters required for shift modulation, perturbation and shaping are dynamically calculated; and a first perturbation d1 is injected into the sign-extended signal, and high-pass noise shaping is performed after the injection, wherein a high-pass filter for the high-pass noise shaping adopts a first-order feedback structure; shifting the shaped signal left by m bits for amplification, and injecting a second perturbation d2 after the amplification; performing low-pass noise shaping on the signal after the second perturbation is injected, wherein a low-pass filter for the low-pass noise shaping adopts a first-order feedback structure; extracting a strong signal from the signal subjected to the low-pass noise shaping through a shift modulation method of right-shifting by n bits and then left-shifting by n bits, and performing a subtraction operation on the low-pass noise shaped signal and the strong signal to separate a weak signal; shifting the strong signal component and the weak signal component right by m bits to restore the magnitude; performing symmetric truncation and asymmetric truncation on the strong signal and the weak signal with restored magnitude respectively to recover the bit width for output; dynamically calculating the parameters required for shift modulation, perturbation and shaping comprises: number of right shift bits: n=floor(log2(Amax / max(Amin,1))); signal-to-noise ratio: SNRdB=20log10(Amax / Amin); adaptive shift bit number: when SNRdB≥10dB, k=ceil(SNRdB / 3); when SNRdB<10dB, k=2; Left shift bits: m = clamp(n + k, 3, B), where B = B ext -B in B represents the number of bits for overflow protection.

2. The method for separating strong and weak signals by dynamic shift quantization modulation as described in claim 1, characterized in that, the amplitude of the injected first perturbation d1 is |d1|=clamp(2^(m-n),1,Amin / 8); the amplitude of the injected second perturbation d2 is |d2|=clamp(2^(m-n),1,Amin<<m / 8), wherein Amax and Amin are respectively the maximum value and the minimum value of the sliding window statistical signal.

3. The dynamic shift quantization modulation strong and weak signal separation method according to claim 1, characterized in that, the coefficient of the high-pass filter is: α1=0.5+0.12*(m-n), with a range of 0.5~0.9; the coefficient of the low-pass filter is: α2=0.3+0.08*(m-n), with a range of 0.3~0.

8.

4. The dynamic shift quantization modulation strong / weak signal separation method as described in claim 3, characterized in that, The high-pass filter is implemented by shift addition.

5. The method for separating strong and weak signals by dynamic shift quantization modulation as described in claim 1, characterized in that, The dynamic calculation is implemented based on a parallel comparator tree, and the Amax / Amin statistics of a 256-point window are completed in a single cycle.

6. The method for separating strong and weak signals by dynamic shift quantization modulation as described in claim 1, characterized in that, Performing symmetric truncation and asymmetric truncation on the strong signal and the weak signal with restored magnitude respectively comprises: For the signals with restored magnitude, a symmetric saturation truncation is adopted for the strong signal component, and an asymmetric truncation is adopted for the weak signal for output.

7. A dynamic shift quantization modulation strong / weak signal separation system with adaptive noise suppression, characterized in that, comprising: The sign extension module is configured to sign-extend the input signal, increasing the bit width from B... in Extend to B ext =2×B in High-order fill sign bit; a dynamic parameter calculation module, which uses a specified sliding window to count the maximum value Amax and the minimum value Amin of the sign-extended input signal, and dynamically calculates required parameters; a staged perturbation injection module, which injects a first perturbation d1 and a second perturbation d2 after sign extension and after left shift amplification respectively; The noise shaping module includes a high-pass filter and a low-pass filter. After the first disturbance d1 is injected, high-pass noise shaping is performed. The high-pass filter for high-pass noise shaping adopts a first-order feedback structure. The module also performs low-pass noise shaping on the signal after the second disturbance is injected. The low-pass filter for low-pass noise shaping adopts a first-order feedback structure. The signal separation module is used to extract strong signals from low-pass noise shaped signals by right shifting n bits and then left shifting n bits, and to separate weak signals by subtraction. The magnitude recovery module is used to recover the magnitude of strong and weak signal components. The truncation output module is used to perform symmetrical and asymmetrical truncation on strong and weak signals respectively, ensuring the recovery of bit width; The parameters required for dynamic calculation of shift modulation, perturbation, and shaping include: Right shift by bits: n = floor(log2(Amax / max(Amin,1))); Signal-to-noise ratio: SNRdB = 20log10(Amax / Amin); Adaptive shifter: when SNRdB ≥ 10dB, k = ceil(SNRdB / 3); when SNRdB < 10dB, k = 2; Left shift bits: m = clamp(n + k, 3, B), where B = B ext -B in B represents the number of bits for overflow protection.

Images