An oscilloscope noise floor dynamic cancellation method and device based on a virtual reference channel

Abstract

This invention discloses a method and apparatus for dynamic noise cancellation of an oscilloscope based on a virtual reference channel, relating to the field of electronic measurement instrument technology. A virtual reference channel is constructed using an idle channel or a built-in calibration input of the oscilloscope to acquire the noise floor signal under the current system state in real time. A noise transfer function model is established between the virtual reference channel and the main measurement channel. Based on the noise transfer function model and the noise floor signal, an adaptive filtering algorithm is used to generate a cancellation signal that is in phase and of equal amplitude with the noise floor of the main channel. The cancellation signal is subtracted from the digital waveform data of the main measurement channel in real time to obtain the purified target signal. This application utilizes the inherent multi-channel isomorphic front-end architecture of digital oscilloscopes and the high correlation of noise floor between channels to dynamically cancel the oscilloscope's own noise floor in real time through digital signal processing, achieving a significant improvement in the signal-to-noise ratio and effective bit depth of small-signal measurements.

Description

Technical Field

[0001] This invention relates to the field of electronic measuring instrument technology, specifically to a method and apparatus for dynamic noise cancellation of oscilloscopes based on a virtual reference channel. Background Technology

[0002] The vertical resolution and measurement accuracy of a digital oscilloscope are limited by the system's floor noise. Floor noise primarily originates from thermal noise of the front-end amplifier, power supply ripple, clock jitter, and electromagnetic interference introduced by PCB layout. When measuring minute signals (such as millivolt-level power supply ripple or weak reflections from high-speed links), if the oscilloscope's floor noise level approaches or exceeds the signal amplitude, the effective signal will be overwhelmed, leading to distorted measurement results and a significant decrease in the system's effective bit width (ENOB).

[0003] Existing methods for reducing oscilloscope noise mainly include: 1) using lower noise amplifier chips and more precise power supply designs, but this will significantly increase hardware costs and is limited by the limits of semiconductor technology, approaching the thermal noise limit; 2) enabling "Hi-Res Mode" in the software to reduce noise through digital filtering and oversampling, but this will sacrifice the oscilloscope's real-time bandwidth and sampling rate, reducing bandwidth by more than 50%; 3) using an external low-noise preamplifier, but this increases the complexity of the test system and is prone to introducing connection errors and frequency response unevenness.

[0004] Therefore, how to further suppress background noise and improve the signal-to-noise ratio and ENOB of small signal measurements based on the existing oscilloscope hardware architecture has become a technical problem that urgently needs to be solved in the field of oscilloscope R&D. Summary of the Invention

[0005] The purpose of this invention is to provide a method and apparatus for dynamic noise cancellation of oscilloscopes based on virtual reference channels. By utilizing the inherent multi-channel isomorphic front-end architecture of digital oscilloscopes and the high correlation of noise floor between channels, the method can dynamically cancel the noise floor of the oscilloscope itself in real time through digital signal processing without increasing additional hardware costs, relying on external signal sources or known waveforms, or affecting the measurement bandwidth. This results in a significant improvement in the signal-to-noise ratio of small signal measurements and the effective number of bits (ENOB) of the system.

[0006] To achieve the above objectives, this application provides the following solution: On one hand, the present invention provides a method for dynamic cancellation of oscilloscope noise floor based on a virtual reference channel, specifically including the following steps: S1. Obtain the occupancy status of each channel of the oscilloscope, determine the virtual reference channel and the main measurement channel based on the occupancy status of each channel, and collect the background noise signal of the virtual reference channel in real time. S2. Establish a noise transfer function model between the virtual reference channel and the main measurement channel. The noise transfer function model is used to characterize the amplitude attenuation and phase delay characteristics of the background noise signal propagating from the virtual reference channel to the main measurement channel. S3. Input the background noise signal into the adaptive filter to generate the cancellation signal. The initial coefficients of the adaptive filter are obtained by transforming the noise transfer function model. S4. The cancellation signal is subtracted from the digital waveform data input from the main measurement channel in real time to obtain the purified target signal.

[0007] In some specific technical solutions, the process of determining the virtual reference channel is as follows: S11. Monitor the insertion detection signal of the BNC connector of each channel of the oscilloscope, and determine the occupancy status of each channel based on the insertion detection signal; S12. Determine the channels in the occupied state as the main measurement channels, and the channels in the idle state as the virtual reference channels. Disconnect the connection with the external BNC through the input relay of the virtual reference channel, short-circuit the input terminal of the virtual reference channel to the matching load or keep it at high impedance and ground, and configure it as an idle channel mode. S13. If all channels are occupied, use the calibration signal port on the front panel of the oscilloscope to switch the calibration signal port to the reference channel preamplifier via the internal relay of the oscilloscope and configure it as the built-in calibration input mode.

[0008] In some specific technical solutions, the process of constructing the noise transfer function model is as follows: S21. During the oscilloscope initialization or self-test phase, short-circuit the input terminals of the main measurement channel and the virtual reference channel to the matching load simultaneously. S22. During the oscilloscope calibration stage, a broadband pseudo-random noise signal with a known spectrum is simultaneously injected into the main measurement channel and the virtual reference channel through the auxiliary signal output port of the oscilloscope. S23. Collect the response time-domain data of the main measurement channel and the virtual reference channel respectively, and calculate the initial noise transfer function model through frequency domain division or cross-correlation analysis; S24. Monitor the operating parameters of the oscilloscope in real time. When the operating parameters change, dynamically correct the parameters of the initial noise transfer function model according to the pre-stored three-dimensional dynamic compensation lookup table to obtain the final noise transfer function model.

[0009] In some specific technical solutions, the specific process of calculating the initial noise transfer function model through frequency domain division is as follows: S231. After windowing the response time-domain data of the main measurement channel and the virtual reference channel, perform FFT calculation to obtain the frequency domain vector X_1 of the main measurement channel and the frequency domain vector X_ref of the virtual reference channel. S232. Calculate the cross power spectral density between the main measurement channel and the virtual reference channel, as well as the self-power spectral density of the virtual reference channel, based on the frequency domain vectors X_1 and X_ref. S233. The frequency domain ratio of the cross power spectral density to the self power spectral density is smoothed and filtered to obtain the initial noise transfer function model.

[0010] In some specific technical solutions, the operating parameters include vertical range G, bandwidth setting B, and operating environment temperature T. The three-dimensional dynamic compensation lookup table includes temperature-gain correlation group, range-gain relationship group, and bandwidth correction corresponding group. When the ambient temperature T drifts, if the temperature drift exceeds the first preset threshold, the corresponding compensation coefficient is located in the temperature-gain association group in the three-dimensional dynamic compensation lookup table according to the temperature drift amount to correct the gain and phase of the initial noise transfer function model. When the vertical range G is switched, causing a change in the gain of the front-end amplifier inside the oscilloscope, the gain of the initial noise transfer function model is corrected by finding the gain correction coefficient of the corresponding frequency point in the range-gain relationship group according to the amount of gain change. When the bandwidth setting B changes, causing the amplitude-frequency characteristics of the anti-aliasing filter inside the oscilloscope to change, the currently set bandwidth is obtained. Based on the currently set bandwidth, the corresponding transfer function correction vector is found in the bandwidth correction corresponding group of the three-dimensional dynamic compensation lookup table. The transfer function correction vector is superimposed on the initial noise transfer function model to obtain the final noise transfer function model.

[0011] In some specific technical solutions, the process of generating the cancellation signal is as follows: S31. Perform an inverse FFT transform on the noise transfer function model to convert it into a time-domain impulse response, which is used as the initial coefficients of the adaptive filter. S32. Divide the background noise signal into segments according to a fixed number of sampling frequency points L to obtain a frame of data, with the data overlap between adjacent frames being L / 2 frequency points. Perform FFT calculation on each frame of data to obtain frequency domain vector data. S33. Input the frequency domain vector data into the adaptive filter for frequency domain filtering, and then perform IFFT processing to obtain the time domain output signal. S34. Take the last half L / 2 data of the time-domain output signal as the effective output to obtain the cancellation signal.

[0012] In some specific technical solutions, the process of checking and updating the initial coefficients of the adaptive filter is also included. The specific process is as follows: Obtain the current bandwidth setting of the oscilloscope and get the frequency range corresponding to the current bandwidth setting; The initial coefficients of the adaptive filter are updated within the frequency range using the following method: The difference between the cancellation signal and the digital waveform data is calculated to obtain the error signal. The error signal is then padded with zeros to N data points and subjected to FFT transformation to obtain the correction signal E(n). The initial coefficients W(n) are updated based on the corrected signal E(n) and the background noise signal N_ref(n). The updated initial coefficients W(n+1) are: W(n+1) = W(n) + μ·E(n)·N_ref(n) / (||N_ref(n)||² + ε) Where μ represents the update step size of the adaptive filter, and ε represents the zero-factor elimination.

[0013] In some specific technical solutions, the adaptive filter adopts the normalized LMS algorithm or the recursive least squares (RLS) algorithm.

[0014] In some specific technical solutions, the method for constructing the three-dimensional dynamic compensation lookup table is as follows: Place the oscilloscope in a temperature chamber and set several test temperature points in fixed steps within a preset temperature range. At each test temperature point, iterate through all vertical range settings and all bandwidth settings to form several combinations of operating parameters. Each combination of operating parameters includes a test temperature point, a vertical range setting, and a bandwidth setting. Record the noise transfer function model generated under each combination of operating parameters. Using one combination of operating parameters as a reference, calculate the correction coefficients relative to the reference for each of the remaining combinations of operating parameters. The correction coefficients include gain and phase. By binding the temperature, vertical range, and bandwidth settings under each combination of operating parameters to correction coefficients, we obtain temperature-gain correlation groups, range-gain relationship groups, and bandwidth correction corresponding groups. The temperature-gain correlation group, range-gain relationship group, and bandwidth correction corresponding group under each combination of operating parameters are stored as a record in the three-dimensional dynamic compensation lookup table.

[0015] Secondly, this application provides an oscilloscope noise floor dynamic cancellation device based on a virtual reference channel, comprising: The reference signal acquisition module is used to obtain the occupancy status of each channel of the oscilloscope, determine the virtual reference channel and the main measurement channel based on the occupancy status of each channel, and acquire the background noise signal of the virtual reference channel in real time. The transfer function modeling module is used to establish a noise transfer function model between the virtual reference channel and the main measurement channel. The noise transfer function model is used to characterize the amplitude attenuation and phase delay characteristics of the background noise signal propagating from the virtual reference channel to the main measurement channel. The adaptive filtering cancellation module is used to input the background noise signal into the adaptive filter to generate a cancellation signal. The initial coefficients of the adaptive filter are obtained by conversion from the noise transfer function model. The cancellation signal is then subtracted from the digital waveform data input from the main measurement channel in real time to obtain the purified target signal.

[0016] The concept of this invention is as follows: The inherent noise floor of a digital oscilloscope primarily originates from its front-end amplifier thermal noise, power supply ripple, ADC aperture jitter and quantization noise, clock phase noise, and PCB crosstalk. At GHz bandwidths, the thermal noise of the 50Ω terminating resistor (approximately 4.0 μVrms) dominates, forming the broadband noise floor. Existing methods for reducing oscilloscope noise mainly involve improving hardware design or sacrificing the oscilloscope's real-time bandwidth and sampling rate to use high-resolution acquisition modes. These solutions not only rely on external signal sources but also increase hardware costs.

[0017] In this application, we consider the following common characteristics of multi-channel digital oscilloscope architecture: all channels share the same power distribution network, the same ADC sampling clock source and clock distribution buffer, the same internal temperature field environment, and the front-end amplifiers of each channel use the same model chip, with PCB traces designed according to the principle of equal length. These common characteristics determine the high correlation between the noise floor of each channel. Measured multi-channel noise correlation coefficients typically reach above 0.85 (under the same range and bandwidth settings), and some architectures (such as dual-channel designs sharing an ADC chip) even have correlation coefficients exceeding 0.92. It is precisely because of this inherent inter-channel noise correlation that the noise signals of each channel of the digital oscilloscope originate from the same noise source through different but highly similar propagation paths, thus establishing a deterministic transfer function relationship between the channels.

[0018] Based on the aforementioned shared characteristics, this application configures the oscilloscope's idle input channels or built-in calibration ports as virtual reference channels. It acquires in real-time the instrument's internal background noise (thermal noise, power supply ripple, clock jitter), which is highly correlated with the main measurement channel, as a background noise signal (reference signal) for noise cancellation. By establishing a noise transfer function model between the virtual reference channel and the main measurement channel, it models the amplitude attenuation and phase delay caused by PCB layout differences between channels. Using this as the initial value for the filter, an adaptive filtering algorithm is initiated to generate a cancellation signal that is in phase and has the same amplitude as the main channel's background noise, which is then subtracted from the digital waveform data in real-time. Specifically, this application also includes a three-dimensional dynamic compensation mechanism for the oscilloscope—responding in real-time to changes in the vertical range G, bandwidth B, and temperature T, correcting the transfer function through a lookup table. Unlike existing vector sensor methods that require independent physical reference sensors, sound insulation shielding, and external signal sources, this application achieves a significant improvement in the signal-to-noise ratio and system effective number of bits (ENOB) for small signal measurements by dynamically canceling the oscilloscope's own noise floor in real time through digital signal processing without increasing additional hardware costs, relying on external signal sources or known waveforms, or affecting the measurement bandwidth.

[0019] The beneficial effects of this invention are as follows: This invention utilizes the oscilloscope's own idle hardware resources (such as unused channels or calibration ports) as a virtual reference channel for noise reference source input, eliminating the need for additional low-noise sensors, external signal sources, or complex circuits. The closed-loop architecture of a noise transfer function model and adaptive filtering can track changes in noise characteristics (including thermal drift, aging, and electromagnetic environment changes) in real time, and can track and cancel system noise floor variations that change with environment and measurement range, significantly reducing the oscilloscope's noise floor. This makes it possible to measure weak signals at low vertical ranges. The initial coefficients of the adaptive filter are generated by the noise transfer function model rather than randomly initialized, improving the convergence speed of the adaptive filter. This invention can reduce the oscilloscope's noise floor by 4-7 dB (typical observable noise reduction) without sacrificing bandwidth, and increase the system's effective bit width (ENOB) by 1.2-1.6 bits. The signal-to-noise ratio improvement is particularly significant in low vertical range (mV level) small-signal measurement scenarios. It effectively improves the oscilloscope's effective bit width (ENOB) and dynamic range, enhancing measurement accuracy in applications such as power integrity testing and high-speed signal eye diagram analysis. Furthermore, the algorithm has low overhead and is not dependent on specific signal types or measurement scenarios—the virtual reference channel functions correctly regardless of whether the measured signal is a periodic signal, transient pulse, or random noise. This method can be directly applied to all digital oscilloscopes with multi-channel architectures, mixed-signal oscilloscopes, and any multi-channel homogeneous front-end broadband sampling instruments. It is easily integrated into existing DSP architectures. Attached Figure Description

[0020] Figure 1 A flowchart of an oscilloscope noise floor dynamic cancellation method based on a virtual reference channel provided in an embodiment of the present invention; Figure 2 The figure shows the simulation results of the noise power spectral density comparison before and after cancellation provided in the embodiments of the present invention. Figure 3 This is a simulation graph of the inter-channel noise correlation coefficient versus temperature provided in an embodiment of the present invention. Figure 4 Simulation diagrams comparing the noise cancellation effects of dynamic compensation and no compensation at different temperatures, provided for embodiments of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the present invention or its application or use.

[0022] Furthermore, for clarity and brevity, descriptions of well-known structures, functions, and configurations may have been omitted. Those skilled in the art will recognize that various changes and modifications can be made to the examples described herein without departing from the spirit and scope of this disclosure.

[0023] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0024] Example 1 like Figure 1 As shown, this embodiment provides a method for dynamic noise cancellation of an oscilloscope based on a virtual reference channel. This method can be implemented in the oscilloscope's FPGA + digital signal processing (DSP) chip or post-processing software, and specifically includes the following steps: S1. Obtain the occupancy status of each channel of the oscilloscope, determine the virtual reference channel and the main measurement channel based on the occupancy status of each channel, and collect the background noise signal of the virtual reference channel in real time. The noise floor signal includes thermal noise of the oscilloscope front-end amplifier, power distribution network ripple and switching noise, ADC aperture jitter and quantization noise, clock distribution phase noise and PCB crosstalk electromagnetic interference. The noise floor signal N_ref(n) inside the oscilloscope instrument is highly correlated with the noise floor of the main measurement channel, with a correlation coefficient of not less than 0.85. The oscilloscope in this embodiment can be a digital oscilloscope, a mixed-signal oscilloscope, or any broadband sampling instrument with a multi-channel isomorphic front-end architecture.

[0025] Specifically, the process of determining the virtual reference channel is as follows: S11. Monitor the insertion detection signal of the BNC connector of each channel of the oscilloscope, and determine the occupancy status of each channel based on the insertion detection signal; S12. Determine the channels in the occupied state as the main measurement channels and the channels in the idle state as the virtual reference channels. Disconnect the connection with the external BNC through the input relay of the virtual reference channel. Short-circuit the input terminal of the virtual reference channel to the matching load or keep it at high impedance and ground. The gain of its front-end amplifier is consistent with or proportional to that of the main measurement channel to ensure that the acquired noise floor signal N_ref(n) is highly correlated with the noise floor of the main channel and configured as an idle channel mode. S13. If all channels are occupied, use the calibration signal port on the front panel of the oscilloscope to switch the calibration signal port to the reference channel preamplifier via the internal relay of the oscilloscope and configure it as the built-in calibration input mode.

[0026] In applications where the oscilloscope has multiple input channels (e.g., 4 channels: CH1-CH4), if the user only uses some channels (e.g., CH1 and CH2), the system automatically configures the idle channel (e.g., CH3) into "idle channel mode." In this mode, the input relay of CH3 disconnects the external BNC connection, internally shorting its input terminal to a precision 50-ohm matched load (or maintaining high impedance and shorting to ground according to user settings). At this time, the signal acquired by CH3 is the noise floor N_ref(n) generated by the current environment and the instrument itself. Since CH3 shares the same ADC clock buffer, the same power distribution network, and the same internal temperature field as CH1 / CH2, and the front-end amplifiers are from the same model and batch, the correlation coefficient between N_ref(n) and the noise floor of CH1 can reach 0.88 (typical value at 25℃) in actual measurements. The noise acquired by CH3 has a very strong correlation with the noise floor of the currently operating CH1 / CH2 (correlation coefficient can reach above 0.85).

[0027] The defined virtual reference channel supports three configuration modes, which automatically switch according to the oscilloscope's current channel occupancy and user needs: (1) Idle Channel Mode: When the oscilloscope has four channels (CH1-CH4) and the user only uses CH1 and CH2, the system automatically configures CH3 as a virtual reference channel, with both CH1 and CH2 serving as the main measurement channels. The signal from CH3 is used as the reference signal for cancellation. The input of CH3 is connected to a 50Ω matching resistor or a high-impedance input via a software-controlled analog switch, depending on the current measurement scenario. The system monitors the insertion detection signal of the BNC connector for each channel in real time to obtain the usage status of each channel. When it detects that CH3 is connected to an external probe by the user, indicating that CH3 is activated by the user, the system automatically switches to CH4 or the built-in calibration input. This mode is the default operating mode of this application.

[0028] (2) Built-in Calibration Input Mode: Utilizing the oscilloscope's existing calibration signal port (input CAL IN or output CAL OUT), which is typically connected to a high-precision internal reference, this mode switches to the reference channel preamplifier via an internal relay. Configure CALIN in input mode and switch to the reference channel preamplifier via the internal relay. Noise correlation in this mode can reach over 0.90 because the CAL port shares part of the power plane and ground plane with the main channel, resulting in a typical noise correlation above 0.90. This mode is suitable for scenarios where all channels are occupied by the user.

[0029] (3) External Injection Mode: Under high-precision measurement requirements, during the calibration phase, a broadband pseudo-random noise signal (-30dBm, bandwidth covering the full bandwidth of the oscilloscope) or a low-level white noise signal (such as -40dBm AWGN) with a known spectrum is simultaneously injected into the main measurement channel and the virtual reference channel through the auxiliary signal output port of the oscilloscope. An initial transfer function model is established by correlation analysis or frequency domain division. This mode is suitable for oscilloscope architectures with high inter-channel isolation (greater than 60dB). This further reduces the impact of inter-channel isolation differences on modeling accuracy.

[0030] S2. Establish a noise transfer function model between the virtual reference channel and the main measurement channel. The noise transfer function model is used to characterize the amplitude attenuation and phase delay characteristics of the background noise signal propagating from the virtual reference channel to the main measurement channel. Due to slight differences in PCB layout, component batches, and trace lengths between the virtual reference channel (CH3) and the main measurement channel (CH1), the noise floor will experience different amplitude attenuation and phase delay when propagating to the two channels. To accurately cancel this out, a mapping relationship between the two must be established. During oscilloscope power-on self-test or when the user manually triggers "noise floor calibration," all channel inputs are shorted to a matched load, and pure noise data for a period of time is acquired. The time-domain signal is converted to a frequency-domain signal using Fast Fourier Transform (FFT), and the cross-power spectral density and the reference channel's self-power spectral density are calculated. Based on the frequency domain ratio of the cross-power spectral density to the reference channel's self-power spectral density from the synchronously acquired data of the two channels, the frequency response function H(f) is obtained as the noise transfer function model. This H(f) includes the gain factor and phase shift for each frequency component.

[0031] Furthermore, the construction of the noise transfer function model H(f) can be performed automatically periodically to track parameter drift caused by system aging. The specific method is as follows: S21. During the oscilloscope initialization or self-test phase, short-circuit the input terminals of the main measurement channel and the virtual reference channel to the matched load (such as a 50-ohm termination). S22. During the oscilloscope calibration phase, a broadband pseudo-random noise signal (test excitation signal or ambient noise) with a known spectrum is simultaneously injected into the main measurement channel and the virtual reference channel through the auxiliary signal output port of the oscilloscope. For example, the generated test excitation signal x(n) = A·sin(2πf_0 t) is generated, where f_0 is the frequency point from 1kHz to the current bandwidth limit (e.g., 100MHz), A represents the amplitude, and the step frequency is 1 / 1024 of the current sampling rate.

[0032] S23. Collect the response time-domain data of the main measurement channel and the virtual reference channel respectively. During collection, N=1024 points are collected at each frequency point, with a sampling rate f_s ≥ 2.56 × bandwidth. Calculate the initial noise transfer function model H_0(f) through frequency domain division or cross-correlation analysis. The specific process of calculating the initial noise transfer function model through frequency domain division is as follows: S231. After windowing the time-domain response data of the main measurement channel and the virtual reference channel (using the Blackman-Harris window to reduce spectral leakage), perform FFT calculation to obtain the frequency domain vector X_1 of the main measurement channel and the frequency domain vector X_ref of the virtual reference channel. S232. Calculate the cross power spectral density between the main measurement channel and the virtual reference channel, as well as the self-power spectral density of the virtual reference channel, based on the frequency domain vectors X_1 and X_ref. S233. The frequency domain ratio of the cross power spectral density to the self power spectral density is smoothed and filtered to obtain the initial noise transfer function model.

[0033] S24. Monitor the operating parameters of the oscilloscope in real time. When the operating parameters change, dynamically correct the parameters of the initial noise transfer function model according to the pre-stored three-dimensional dynamic compensation lookup table to adapt to the changes in noise characteristics under different operating conditions and obtain the final noise transfer function model.

[0034] Specifically, the operating parameters include vertical range G, bandwidth setting B, and operating ambient temperature T. The three-dimensional dynamic compensation lookup table includes temperature-gain correlation group, range-gain relationship group, and bandwidth correction corresponding group. When the ambient temperature T drifts, if the temperature drift exceeds the first preset threshold, the corresponding compensation coefficient is located in the temperature-gain association group in the three-dimensional dynamic compensation lookup table according to the temperature drift amount to correct the gain and phase of the initial noise transfer function model. When the vertical range G is switched, causing a change in the gain of the front-end amplifier inside the oscilloscope, the gain of the initial noise transfer function model is corrected by finding the gain correction coefficient of the corresponding frequency point in the range-gain relationship group according to the amount of gain change. When the bandwidth setting B changes, causing the amplitude-frequency characteristics of the anti-aliasing filter inside the oscilloscope to change, the currently set bandwidth is obtained. Based on the currently set bandwidth, the corresponding transfer function correction vector is found in the bandwidth correction corresponding group of the three-dimensional dynamic compensation lookup table. The transfer function correction vector is superimposed on the initial noise transfer function model to obtain the final noise transfer function model.

[0035] Because the front-end noise characteristics of an oscilloscope are not static, the noise characteristics of the oscilloscope's front-end amplifier change with the switching of the vertical range (V / div), the activation and deactivation of the bandwidth limiting filter, and changes in ambient temperature. For example, when the user rotates the vertical knob to change the range (e.g., from 1V / div to 10mV / div), the gain stage of the front-end amplifier switches, causing a change in the thermal noise floor. Furthermore, increased internal instrument temperature also increases resistive thermal noise. Therefore, it is necessary to monitor these operating parameters (T, G, B) in real time and obtain the corresponding compensation coefficients based on a pre-stored lookup table (LUT) or polynomial fitting formula. This allows for dynamic correction of the gain and phase of the noise transfer function model H(f), and real-time updates to the step size μ of the adaptive filter or the reference gain of the transfer function H(f). For instance, when the range is reduced by a factor of 10, the noise amplification factor increases, and the system correspondingly increases the gain weight of the cancellation signal to ensure that the cancellation depth is always maintained at an optimal level (typically reducing the noise floor by 3-6dB). These corrections ensure consistent cancellation performance across the entire range and temperature range.

[0036] S3. Input the background noise signal into the adaptive filter to generate the cancellation signal. The initial coefficients of the adaptive filter are obtained by transforming the noise transfer function model. In the actual measurement process, the acquired noise floor N_ref(n) is input into an adaptive FIR filter. The initial coefficients of this filter are obtained by inverse FFT transformation of H(f) obtained in step S2, rather than random initialization, thus significantly shortening the convergence time. The output of the filter is the predicted main channel noise floor N_cancel(n), which is the cancellation signal that needs to be canceled from the main measurement channel, resulting in the cleaned signal S_clean(n). To prevent the introduction of new quantization noise or ringing effects during the cancellation process, this embodiment uses a band-limited LMS algorithm, which performs the cancellation operation only within the bandwidth currently set by the oscilloscope (such as when the 20MHz bandwidth limit is enabled). That is, the adaptive filtering only performs band-limiting processing within the bandwidth currently set by the oscilloscope, effectively suppressing computational instability in the high-frequency band.

[0037] Specifically, the process of generating the cancellation signal N_cancel(n) is as follows: S31. Perform an inverse FFT transform on the noise transfer function model to convert it into a time-domain impulse response, which is used as the initial coefficients of the adaptive filter. S32. Divide the background noise signal into segments according to a fixed number of sampling frequency points L to obtain a frame of data, with the data overlap between adjacent frames being L / 2 frequency points. Perform FFT calculation on each frame of data to obtain frequency domain vector data. S33. Input the frequency domain vector data into the adaptive filter for frequency domain filtering, and then perform IFFT processing to obtain the time domain output signal. S34. Take the last half L / 2 data of the time-domain output signal as the effective output to obtain the cancellation signal.

[0038] S4. Subtract the cancellation signal N_cancel(n) from the digital waveform data S_main(n) input from the main measurement channel in real time to obtain the purified target signal S_clean(n): S_clean(n) = S_main(n) - N_cancel(n) Specifically, the adaptive filter employs either the normalized LMS algorithm or the recursive least squares (RLS) algorithm. Its core idea is to use the reference signal N_ref(n) as input, generate a cancellation signal through an adjustable-weight FIR filter, and use the digital waveform data S_main(n) input from the main measurement channel as an estimate of the desired signal (which is pure noise when there is no external signal input). (1) Normalized LMS (NLMS) algorithm: filter order M=64, step size factor μ=0.01, μ is adaptively adjusted according to residual noise power to ensure fast convergence when noise characteristics change abruptly, and maintain low offset in steady state, which can be adaptively adjusted according to signal power.

[0039] The weight update formula is: W(n+1) = W(n) + μ·E(n)·N_ref(n) / (||N_ref(n)||² + ε), where ε=0.001 is a small constant to prevent the denominator from being zero. This algorithm achieves a balance between computational complexity and convergence speed, making it suitable for real-time implementation.

[0040] (2) Recursive Least Squares (RLS) algorithm: using an exponentially weighted forgetting factor λ=0.99, and an initial covariance matrix P(0)=δ (-1)·I (δ=0.01), where I represents the identity matrix. This algorithm has a fast convergence speed (approximately 1 / 10 of the iterations of NLMS), but its computational cost is relatively high (O(M²)). It is suitable for scenarios requiring high convergence speed, such as when rapidly switching oscilloscope ranges.

[0041] In practice, the filter is implemented in the frequency domain: an N-point FFT (N=256) is performed on the reference signal N_ref(n) and the main channel signal S_main(n) respectively, and the filtered output Y(n)=W(n)·N_ref(n) is calculated in the frequency domain, where W(n) are the frequency domain filter coefficients. Frequency domain implementation can reduce the computational complexity to O(NlogN).

[0042] In some specific technical solutions, the process of checking and updating the initial coefficients of the adaptive filter is also included. The specific process is as follows: Obtain the current bandwidth setting of the oscilloscope and get the frequency range corresponding to the current bandwidth setting; The initial coefficients of the adaptive filter are updated within the frequency range to minimize the mean square error of the output signal. The update method is as follows: The difference between the cancellation signal and the digital waveform data is calculated to obtain the error signal. The error signal is then padded with zeros to N data points and subjected to FFT transformation to obtain the correction signal E(n). The initial coefficients W(n) are updated based on the corrected signal E(n) and the background noise signal N_ref(n). The updated initial coefficients W(n+1) are: W(n+1) = W(n) + μ·E(n)·N_ref(n) / (||N_ref(n)||² + ε) Here, μ represents the update step size of the adaptive filter, and ε represents the zero-factor elimination. Through continuous iteration, the mean square value of E(n) is minimized, thereby achieving precise cancellation of the noise floor.

[0043] The three-dimensional dynamic compensation lookup table (LUT(G, T, B)) was established through a full-temperature, full-range calibration experiment before the oscilloscope left the factory. The calibration procedure is as follows: Place the oscilloscope in a temperature chamber and set several test temperature points within a preset temperature range (e.g., -10℃ to +70℃) in fixed steps (e.g., 5℃). Assuming 17 test temperature points, at each test temperature point, iterate through all vertical range settings (typical values ​​are 1mV / div, 2mV / div, 5mV / div, 10mV / div, 20mV / div, 50mV / div, 100mV / div, 200mV / div, 500mV / div, 1V / div, 2V / div, 5V / div, 10V / div, a total of 13 ranges) and all bandwidth settings (including full bandwidth, 20MHz band-limited, 250MHz band-limited, etc., typical values, a total of 3-8 ranges), forming several operating condition parameter combinations Pi(G, T, B). Each operating condition parameter combination includes a test temperature point, a vertical range setting, and a bandwidth setting. Record the noise transfer function model H_i{G,T,B}(f) generated under each operating condition parameter combination Pi(G,T,B). Taking one of the operating condition parameter combinations as the benchmark, for example, choosing H_0{G,T,B}(f) == H_{1V / div, 25℃, full bandwidth} as the benchmark, calculate the correction coefficient relative to the benchmark for each of the remaining operating condition parameter combinations: ΔA(f)= | H_i{G,T,B}(f)| / | H_0{G,T,B}(f)|, where ΔA(f) includes the gain correction factor ΔA, the temperature correction factor ΔA_T, the bandwidth correction factor ΔA_B, and the phase correction coefficient Δφ(f) = angle(H_i{G,T,B}(f)) - angle(H_0{G,T,B}(f)).

[0044] By binding the temperature, vertical range, and bandwidth settings under each combination of operating parameters to correction coefficients, we obtain temperature-gain correlation groups, range-gain relationship groups, and bandwidth correction corresponding groups. The temperature-gain correlation group, range-gain relationship group, and bandwidth correction corresponding group under each combination of operating parameters are stored as a record in the three-dimensional dynamic compensation lookup table.

[0045] Experimental results show that the gain correction factor ΔA changes most significantly with G—when switching from 1V / div to 1mV / div, the gain changes by approximately 1000 times, corresponding to a noise gain correction factor ΔA of approximately 31.6 (i.e., a 30dB increase in gain leads to a 30dB increase in noise floor). The temperature correction factor ΔA_T changes by approximately 1.8dB (approximately 1.23 times) within the range of -10℃ to +70℃, and the bandwidth correction factor ΔA_B is approximately 0.14 times the full bandwidth when 20MHz band limiting is enabled. A three-dimensional lookup table is stored in a 32×16×8 matrix (G is 32 points rounded down from 13 levels, T is 16 points rounded down from 17 levels, and B is 8 points from levels 3 to 8), and the correction factors at non-nominal parameter points are obtained through linear interpolation. The temperature sensor (located near the front-end amplifier, accuracy ±0.5℃), the vertical range register value G, and the bandwidth setting B are read at 100ms intervals. When a change in any of the parameters G, T, or B is detected, the corresponding ΔA (gain correction) and Δφ (phase correction) are read from a pre-calibrated 32×16×8 (G×T×B) three-dimensional lookup table. The amplitude response of H(f) is updated multiplicatively, the phase response is updated additively, and the filter coefficients are regenerated. (1) Temperature compensation: Three NTC thermistors are placed in key locations inside the oscilloscope (front-end amplifier, ADC, power supply module) with an accuracy of ±0.5℃. A temperature-gain correlation group LUT_T is established: when the temperature change ΔT>2℃, the system automatically calls the compensation coefficient K_T(ΔT)=1+α·ΔT in LUT_T, where α=-0.02% / ℃ (typical value). At the same time, the phase is corrected: φ_T(ΔT)=β·ΔT, β=-0.1° / ℃.

[0046] (2) Range compensation: A range-gain relationship group LUT_G is established. When the vertical range switches from G_0 to G_1, the gain change of the front-end amplifier is ΔG=20·log10(G_1 / G_0) dB. The system finds the gain correction coefficient K_G(f, G_1) of the corresponding frequency point according to the pre-stored LUT_G. The corrected transfer function is H_corrected(f)=H_0(f)·K_G(f, G_1)·K_T(ΔT).

[0047] (3) Bandwidth compensation: When the bandwidth limitation state B switches from full bandwidth to 20MHz, the anti-aliasing filter inside the system will change the amplitude-frequency characteristics. The transfer function correction vector ΔH_B(f) under different bandwidth settings is pre-stored to form the bandwidth correction corresponding group, which is superimposed on the current model in real time: H_final(f)=H_corrected(f)+ΔH_B(f, B).

[0048] Understandably, the self-calibration process of the noise transfer function model is automatically triggered in three situations: the oscilloscope's first power-on run, the user manually triggering the noise floor calibration, or more than 24 hours since the last calibration. The self-calibration process includes the following detailed calculation steps: Step 1: Disconnect the external connection of the input relays of the main measurement channel CH1 and the virtual reference channel CH3, and switch to the internal 50Ω matched load terminal (accuracy ±0.05%).

[0049] Step 2: After waiting for the analog switch to stabilize (typical time 10ms), simultaneously acquire dual-channel data for 10ms each at a sampling rate of 2.5GS / s, for a total of 25M sampling points.

[0050] Step 3: Divide the acquired long data sequence into N=488 data blocks, each block containing M=1024 sampling points, with 50% overlap between blocks. After applying a Hanning window (to reduce spectral leakage) to each block, perform an M-point FFT to obtain the frequency domain vector X_1 of the main measurement channel and the frequency domain vector X_ref of the virtual reference channel.

[0051] Step 4: Calculate the cross power spectrum P_xy(f) = X_1·X_ref for each data block and the self power spectrum P_xx(f) = |X_ref|² for the reference channel.

[0052] Step 5: Take the set average of all N data blocks to obtain the average cross power spectrum S_xy(f) and the average self power spectrum S_xx(f), and calculate the initial transfer function H_0(f) = S_xy(f) / S_xx(f). The frequency resolution is Δf = 2.5GHz / 1024 ≈ 2.44MHz.

[0053] Step 6: Perform amplitude smoothing (5th-order median filter + 3-point moving average) and phase unwrapping on H_0(f) to ensure phase continuity.

[0054] Step 7: Transform H_0(f) into a 256th-order time-domain impulse response W(n) using IFFT and store it in the initial coefficient register of the adaptive filter.

[0055] Step 8: Save H_0(f) and h(n) to non-volatile memory simultaneously as the initial reference for the next startup.

[0056] To meet the real-time processing requirements at GHz sampling rates, adaptive filtering employs an overlap-save method in the frequency domain to reduce computational complexity. Specifically, the background noise signal N_ref(n) is segmented into frames of L=512 points each, with adjacent frames overlapping by 256 points. An FFT is performed on each frame to obtain the frequency domain vector X(n). After initialization in the frequency domain (directly set by H(f), the filter coefficients W(n) are updated in each frame according to the following steps: Step 1: Frequency domain filtering: Y(n) = W(n) · X(n).

[0057] Step 2: Obtain the time-domain output y(n) using IFFT, and take the second half of 256 points as the valid output N_cancel(n).

[0058] Step 3: Calculate the error signal e(n) = S_main(n) - N_cancel(n), pad it with zeros to 512 points, and then use FFT to obtain E(n).

[0059] Step 4: Frequency domain NLMS update: W(n+1) = W(n) + μ · E(n) · X*(n) / (|X(n)|² +ε), where μ=0.1, ε=1e-6, and * denotes complex conjugate.

[0060] Step 5: Band-limited processing: W(n) is only allowed to be updated within the frequency range corresponding to the current bandwidth setting B; W(n) outside the range is set to zero. For example, when B = 20MHz, only the coefficients in the 2.44MHz to 20MHz band are updated; DC and coefficients above 20MHz retain their initial values ​​or are set to zero.

[0061] The frequency domain implementation reduces the computational cost of each iteration from O(L²) in the time domain to O(L·log2L), resulting in a computational efficiency improvement of approximately 45 times when L=512.

[0062] Example 2 This application provides an oscilloscope noise floor dynamic cancellation device based on a virtual reference channel, comprising: The reference signal acquisition module is used to obtain the occupancy status of each channel of the oscilloscope, determine the virtual reference channel and the main measurement channel based on the occupancy status of each channel, and acquire the background noise signal of the virtual reference channel in real time. Specifically, the reference signal acquisition module is connected to the input relays of each channel of the oscilloscope. A software-controlled analog switch matrix selects an idle channel or the CAL IN port as a virtual reference channel. This module includes a 50Ω terminating resistor network and a high-impedance buffer to provide the reference channel with the same input impedance as the main channel. The output is a synchronously sampled digital sequence N_ref(n), precisely clock-aligned with the main channel ADC data stream.

[0063] The transfer function modeling module is used to establish a noise transfer function model between the virtual reference channel and the main measurement channel. The noise transfer function model is used to characterize the amplitude attenuation and phase delay characteristics of the background noise signal propagating from the virtual reference channel to the main measurement channel. The transfer function modeling module comprises a self-calibration control subunit and a frequency domain calculation engine. The self-calibration control subunit manages the calibration timing—controlling the input relay to switch to the matched load, triggering dual-channel synchronous acquisition, and calling the frequency domain calculation engine to perform FFT / cross-spectrum / division and smoothing filtering. The outputs are H(f) and h(n), stored in the frequency domain coefficient RAM and time domain coefficient RAM, respectively.

[0064] The adaptive filtering cancellation module is used to input the background noise signal into an adaptive filter to generate a cancellation signal. The initial coefficients of the adaptive filter are obtained by transforming the noise transfer function model. The cancellation signal is then subtracted from the digital waveform data input from the main measurement channel in real time to obtain the purified target signal. The adaptive filtering cancellation module performs signal subtraction operations in the frequency domain or time domain and supports multi-channel parallel processing to eliminate noise introduced by inter-channel crosstalk.

[0065] The adaptive filtering cancellation module is implemented based on an FPGA pipeline architecture and includes an FFT engine, a complex multiplier array, an NLMS update unit, and an IFFT engine. It employs an overlap-preservation method to perform adaptive filtering in the frequency domain, with a processing latency of 2048 clock cycles per 512-point frame (approximately 8.2 μs at a 250MHz FPGA clock, far below the target of 1 μs). The outputs are N_cancel(n) and S_clean(n).

[0066] The dynamic update module is used to dynamically correct the parameters of the noise transfer function model based on changes in environmental parameters and range settings.

[0067] The dynamic update module reads the temperature sensor (I²C interface, accuracy ±0.5℃), vertical range register value G, and bandwidth status register value B at 100ms intervals. It maintains a DDR memory for a 32×16×8 three-dimensional lookup table and a linear interpolation engine. When a parameter change is detected, the updated H(f) coefficients are written to the coefficient RAM of the adaptive filter cancellation module within 50ms via the AXI bus.

[0068] The above modules are integrated through a heterogeneous computing platform of FPGA and digital signal processor (DSP). The FPGA is responsible for real-time pipelined acquisition and frequency domain transformation of front-end data, while the DSP is responsible for adaptive filter coefficient iteration and lookup table interpolation operations. This heterogeneous architecture supports multi-channel parallel processing, with a single-channel processing latency of less than 1 microsecond, meeting the real-time processing requirements at gigahertz sampling rates. The entire device does not change the existing hardware structure of the oscilloscope; all functions are implemented through firmware upgrades. Specifically: (1) The FPGA adopts the Xilinx Kintex-7 series to realize a high-speed data acquisition interface (connected to 4-channel ADC, sampling rate 1GS / s), performs FFT / IFFT transformation (using pipelined FFT architecture, delay less than 1μs), and implements frequency domain multiplier. The FPGA internally builds a direct memory access (DMA) channel to transmit the acquired data to the DSP.

[0069] (2) DSP module (Texas Instruments C6000 series): Runs adaptive filtering algorithms (NLMS or RLS) and performs frequency domain filtering with a complexity of O(N log N). The DSP connects to the FPGA through the EMIF interface, with a data exchange delay of less than 10μs. It has a built-in temperature sensor interface (I2C bus) and range detection circuit. (3) Memory hierarchy: On-chip SRAM (64KB) stores filter coefficients and temporary data; off-chip DDR3 SDRAM (512MB) stores transfer function models and calibration data; EEPROM (4KB) stores factory calibration parameters and user configuration.

[0070] (4) Multi-channel parallel processing: The device supports simultaneous cancellation of 4 channels, with each channel running an adaptive filter instance independently (4 in total) and sharing the same reference channel data. Through time division multiplexing (TDM) technology, 4-channel parallel processing is achieved on a single DSP with a total processing delay of less than 5μs.

[0071] To verify the technical effectiveness of this application and to make a quantitative comparison with existing technologies, a complete simulation model of multi-channel noise of a digital oscilloscope was established on the simulation platform MATLAB / Simulink R2024a, supplemented by Python 3.11 for data post-processing. The simulation settings, experimental conditions, and result data are described in detail below.

[0072] Oscilloscope model parameters: 4-channel digital oscilloscope, 1GHz analog bandwidth, 2.5GS / s sampling rate, 8-bit ADC quantization resolution, input impedance 50Ω. The model accurately includes the following noise sources: ① Front-end amplifier thermal noise – Johnson-Nyquist noise from a 50Ω terminating resistor (approximately 0.91nV / √Hz at PSD=4kTR, T=298K), amplified by the amplifier gain stage (approximately 400 times gain at 1mV / div) to an output noise density of approximately 0.36μV / √Hz; ② Power supply ripple – 100kHz fundamental and 2nd to 7th harmonics, with amplitudes of 200μVpp, 80μVpp, and 5th harmonics respectively. 0μVpp, 30μVpp, 15μVpp, 8μVpp; ③ ADC aperture jitter – σ_jitter=0.5ps, introducing approximately 0.15mVrms of noise at a 50MHz input; ④ Clock phase noise – -120dBc / Hz at a 1kHz offset, -135dBc / Hz at a 10kHz offset; ⑤ Quantization noise – 8-bit quantization, LSB=0.8mV (at 5V full scale), quantization noise approximately 0.23mVrms. All channels share the same clock source and power supply model. The only random differences between channels are in the front-end amplifier gain tolerance (±0.5%) and PCB trace parasitic parameters (±0.2pF capacitor deviation, ±0.5nH inductor deviation).

[0073] Test signal: 10mVpp sine wave, frequency 50MHz, close to the noise floor level of the oscilloscope's low vertical range (5mV / div), used to evaluate the improvement effect of signal-to-noise ratio in small signal measurements.

[0074] Design the following four sets of comparative experiments, with each set repeated 10 times and the average value taken: Experiment A (Uncancelled Baseline): CH1 is used to measure a 50MHz test signal alone. The CN121207306A vector sensor method is not applicable (no reference sensor). The CH3 signal is not used as a reference input. Record the output noise power spectrum, signal-to-noise ratio (SNR), and effective number of bits (ENOB) of CH1.

[0075] Experiment B (Method of this application—Static calibration, without dynamic compensation): CH1 is the main channel (measuring noisy signals), and CH3 is the virtual reference channel (50Ω grounded termination, acquiring background noise). A self-calibration to establish H_0(f) is performed at startup at 25℃, and adaptive filter cancellation is activated. The temperature is increased from 25℃ to 60℃ at a rate of 1℃ / min (simulating the temperature rise of an oscilloscope during long-term operation), without temperature compensation. The degradation trend of the cancellation effect is recorded.

[0076] Experiment C (Method of this application—Full-function, including dynamic compensation): Same hardware configuration as Experiment B, but with full dynamic compensation mechanism enabled, including real-time T / G / B monitoring and LUT correction. The compensation effect and stability were recorded across the entire temperature range.

[0077] Experiment D (Comparison—Hi-Res Mode): Using the oscilloscope's oversampling mode (4x oversampling + digital filtering), the bandwidth was reduced to approximately 250MHz. The degree of noise reduction and bandwidth cost were recorded.

[0078] like Figure 2 As shown, a comparison of the noise power spectral density before and after cancellation is presented after performing the method of this application at 25℃. From Figure 2 It can be seen that, across the entire frequency band from 100kHz to 1GHz, the residual noise power spectral density after cancellation using the method described in this application is reduced by 4-7dB compared to the original noise floor. Specifically: in the low-frequency band (<1MHz), the power supply ripple discrete spectral lines (100kHz fundamental and its harmonics) are effectively suppressed—the 100kHz spectral line decreases from -45dBm to -52dBm (power supply ripple suppression of 7dB), and the second harmonic decreases from -52dBm to -58dBm (a reduction of 6dB); in the mid-frequency band (1MHz-500MHz), the broadband thermal noise floor decreases from -52dBm / Hz to -57dBm / Hz (a reduction of 5dB); and in the high-frequency band (500MHz-1GHz), the reduction is approximately 4dB. The cancellation effect in the high-frequency band is slightly lower than that in the low-frequency band, mainly due to the slightly increased phase mismatch between channels at high frequencies caused by differences in PCB parasitic parameters, but the noise reduction of more than 4dB is still significant.

[0079] Using a 50MHz, 10mVpp test signal as a reference, the quantization results at a 1GHz bandwidth and a 5mV / div setting are as follows: Experiment A (uncancelled baseline): SNR = 12.3dB, ENOB = 5.2 bits (measured at 50MHz input).

[0080] Experiment C (method of this application, 25℃): SNR = 18.1dB, ENOB = 6.8 bits. Signal-to-noise ratio improved by 5.8dB, effective number of bits increased by 1.6 bits.

[0081] Experiment D (Hi-Res mode, 4x oversampling): SNR = 16.5dB, ENOB = 6.2 bits. However, the bandwidth decreased from 1GHz to approximately 250MHz.

[0082] The above results show that the method of this application achieves better noise suppression effect than Hi-Res mode while maintaining full bandwidth (1 GHz), and the effective bit depth is 0.6 bits higher than that of Hi-Res mode.

[0083] like Figure 4 As shown, a comparison of the noise cancellation effects of Experiment B (without dynamic compensation) and Experiment C (with dynamic compensation) during the heating process from 25℃ to 60℃ is presented, demonstrating the effect of temperature drift and dynamic compensation. Under the constant temperature condition of 25℃, the initial cancellation effects of Experiment B and Experiment C are consistent, both showing a noise reduction of 4.8dB. From Figure 4 It can be seen that as the temperature increases: Experiment B (uncompensated): The noise reduction monotonically decreased from 4.8 dB at 25°C to 1.5 dB at 60°C (a decrease of 3.3 dB). The main reason is that the noise figure of the front-end amplifier increases with temperature (approximately 0.015 dB / °C), causing a change in the noise floor. The static transfer function model cannot track this change, leading to a gradual increase in the mismatch between the offset signal and the actual noise in the main channel.

[0084] Experiment C (with compensation): The noise reduction remained between 4.2-5.0 dB within the temperature range of 25℃ to 60℃, with the lowest point (55℃) at 4.2 dB, a decrease of only 0.6 dB from 4.8 dB at 25℃. LUT temperature compensation effectively corrected the gain drift and phase shift caused by temperature changes.

[0085] The comparison shows that the three-dimensional dynamic compensation mechanism reduces the degradation of the compensation effect from 3.3dB to 0.6dB over a wide temperature range, while maintaining more than 75% of the compensation depth.

[0086] like Figure 3 The curves showing the noise correlation coefficient between CH1 and CH3 as a function of temperature are illustrated. The results show that the correlation coefficient is 0.88 at 25℃ (within a 1mV / div range); at 60℃, the correlation coefficient increases to 0.92. The reason for the increased correlation with temperature is that at low temperatures, power supply ripple and 1 / f noise account for a higher proportion of the total noise (the inter-channel correlation of these noises is slightly lower), while at high temperatures, thermal noise dominates—due to the physical homogeneity of the thermal noise sources (50Ω terminating resistor), the inter-channel correlation of thermal noise is higher. The correlation coefficient in the range of 0.85–0.92 provides a good basis for noise cancellation in this application.

[0087] Table 1 summarizes the quantitative comparison results between the method of this application and existing noise reduction techniques in multiple dimensions: Table 1. Comparison of Noise Reduction Effects Note: The "Noise Reduction" in Table 1 is a typical test value under a 1GHz bandwidth and 5mV / div setting. The actual effect may vary depending on the oscilloscope model and specific range settings.

[0088] Based on the above simulation experiments, the following conclusions can be drawn: (1) The oscilloscope noise floor dynamic cancellation method based on virtual reference channel proposed in this application can achieve a noise reduction of 4-7dB and an ENOB improvement of 1.2-1.6 bits under 1GHz full bandwidth conditions. The effect is better than the Hi-Res mode at the cost of bandwidth, and no additional hardware cost is required.

[0089] (2) The three-dimensional dynamic compensation mechanism (T / G / B) is the key guarantee for the feasibility of this method in the oscilloscope scenario. Without compensation, temperature drift causes the compensation effect to degrade from 4.8dB to 1.5dB, while after compensation, it is maintained above 4.2dB. This mechanism is a technical feature that vector sensor methods have not addressed at all.

[0090] (3) The high noise correlation (0.85-0.92) between multiple channels of the oscilloscope verifies the physical basis of the concept of "virtual reference channel", that is, the signal collected by the idle channel is not a noise sample but a noise "copy" with a definite transfer function relationship.

[0091] (4) The comparative data in Table 1 clearly shows that the method of this application differs substantially from that of the other in terms of application field (oscilloscope or underwater acoustic vector sensor), reference channel construction method (virtual idle channel or independent physical sensor), noise floor signal generation logic (time domain point-by-point subtraction or frequency domain spectrum estimation), dynamic compensation mechanism (with or without) and real-time requirements (sub-microsecond or millisecond level).

[0092] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Based on the technical essence of the present invention, any simple modifications, equivalent substitutions, and improvements made to the above embodiments within the spirit and principles of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for dynamic noise cancellation of an oscilloscope based on a virtual reference channel, characterized in that, Specifically, the following steps are included: S1. Obtain the occupancy status of each channel of the oscilloscope, determine the virtual reference channel and the main measurement channel based on the occupancy status of each channel, and collect the background noise signal of the virtual reference channel in real time. S2. Establish a noise transfer function model between the virtual reference channel and the main measurement channel. The noise transfer function model is used to characterize the amplitude attenuation and phase delay characteristics of the background noise signal propagating from the virtual reference channel to the main measurement channel. S3. Input the background noise signal into the adaptive filter to generate the cancellation signal. The initial coefficients of the adaptive filter are obtained by transforming the noise transfer function model. S4. The cancellation signal is subtracted from the digital waveform data input from the main measurement channel in real time to obtain the purified target signal.

2. The oscilloscope noise floor dynamic cancellation method based on a virtual reference channel according to claim 1, characterized in that, The specific process for determining the virtual reference channel is as follows: S11. Monitor the insertion detection signal of the BNC connector of each channel of the oscilloscope, and determine the occupancy status of each channel based on the insertion detection signal; S12. Determine the channels in the occupied state as the main measurement channels, and the channels in the occupied state as the idle channels as the virtual reference channels. Disconnect the connection with the external BNC interface through the input relay of the virtual reference channel, short-circuit the input terminal of the virtual reference channel to the matching load or keep it at high impedance and ground, and configure it as an idle channel mode. S13. If all channels are occupied, use the calibration signal port on the front panel of the oscilloscope to switch the calibration signal port to the reference channel preamplifier via the internal relay of the oscilloscope and configure it as the built-in calibration input mode.

3. The oscilloscope noise floor dynamic cancellation method based on a virtual reference channel according to claim 1, characterized in that, The process of constructing the noise transfer function model is as follows: S21. During the oscilloscope initialization or self-test phase, short-circuit the input terminals of the main measurement channel and the virtual reference channel to the matching load simultaneously. S22. During the oscilloscope calibration stage, a broadband pseudo-random noise signal with a known spectrum is simultaneously injected into the main measurement channel and the virtual reference channel through the auxiliary signal output port of the oscilloscope. S23. Collect the response time-domain data of the main measurement channel and the virtual reference channel respectively, and calculate the initial noise transfer function model through frequency domain division or cross-correlation analysis; S24. Monitor the operating parameters of the oscilloscope in real time. When the operating parameters change, dynamically correct the parameters of the initial noise transfer function model according to the pre-stored three-dimensional dynamic compensation lookup table to obtain the final noise transfer function model.

4. The oscilloscope noise floor dynamic cancellation method based on a virtual reference channel according to claim 3, characterized in that, The specific process for calculating the initial noise transfer function model using frequency domain division is as follows: S231. After windowing the response time-domain data of the main measurement channel and the virtual reference channel, perform FFT calculation to obtain the frequency domain vector X_1 of the main measurement channel and the frequency domain vector X_ref of the virtual reference channel. S232. Calculate the cross power spectral density between the main measurement channel and the virtual reference channel, as well as the self-power spectral density of the virtual reference channel, based on the frequency domain vectors X_1 and X_ref. S233. The frequency domain ratio of the cross power spectral density to the self power spectral density is smoothed and filtered to obtain the initial noise transfer function model.

5. The oscilloscope noise floor dynamic cancellation method based on a virtual reference channel according to claim 3, characterized in that, Operating parameters include vertical range G, bandwidth setting B, and operating ambient temperature T. The three-dimensional dynamic compensation lookup table includes temperature-gain correlation group, range-gain relationship group, and bandwidth correction corresponding group. When the ambient temperature T drifts, if the temperature drift exceeds the first preset threshold, the corresponding compensation coefficient is located in the temperature-gain association group in the three-dimensional dynamic compensation lookup table according to the temperature drift amount to correct the gain and phase of the initial noise transfer function model. When the vertical range G is switched, causing a change in the gain of the front-end amplifier inside the oscilloscope, the gain of the initial noise transfer function model is corrected by finding the gain correction coefficient of the corresponding frequency point in the range-gain relationship group according to the amount of gain change. When the bandwidth setting B changes, causing the amplitude-frequency characteristics of the anti-aliasing filter inside the oscilloscope to change, the currently set bandwidth is obtained. Based on the currently set bandwidth, the corresponding transfer function correction vector is found in the bandwidth correction corresponding group of the three-dimensional dynamic compensation lookup table. The transfer function correction vector is superimposed on the initial noise transfer function model to obtain the final noise transfer function model.

6. The oscilloscope noise floor dynamic cancellation method based on a virtual reference channel according to claim 1, characterized in that, The specific process of generating the cancellation signal is as follows: S31. Perform an inverse FFT transform on the noise transfer function model to convert it into a time-domain impulse response, which is used as the initial coefficients of the adaptive filter. S32. Divide the background noise signal into segments according to a fixed number of sampling frequency points L to obtain a frame of data, with the data overlap between adjacent frames being L / 2 frequency points. Perform FFT calculation on each frame of data to obtain frequency domain vector data. S33. Input the frequency domain vector data into the adaptive filter for frequency domain filtering, and then perform IFFT processing to obtain the time domain output signal. S34. Take the last half L / 2 data of the time-domain output signal as the effective output to obtain the cancellation signal.

7. The oscilloscope noise floor dynamic cancellation method based on a virtual reference channel according to claim 6, characterized in that, It also includes the process of checking and updating the initial coefficients of the adaptive filter, the specific process of which is as follows: Obtain the current bandwidth setting of the oscilloscope and get the frequency range corresponding to the current bandwidth setting; The initial coefficients of the adaptive filter are updated within the frequency range using the following method: The difference between the cancellation signal and the digital waveform data is calculated to obtain the error signal. The error signal is then padded with zeros to N data points and subjected to FFT transformation to obtain the correction signal E(n). The initial coefficients W(n) are updated based on the corrected signal E(n) and the background noise signal N_ref(n). The updated initial coefficients W(n+1) are: W(n+1) = W(n) + μ·E(n)·N_ref(n) / (||N_ref(n)||² + ε) Where μ represents the update step size of the adaptive filter, and ε represents the zero-factor elimination.

8. The oscilloscope noise floor dynamic cancellation method based on a virtual reference channel according to claim 1, characterized in that, The adaptive filter employs either the normalized LMS algorithm or the recursive least squares (RLS) algorithm.

9. The oscilloscope noise floor dynamic cancellation method based on a virtual reference channel according to claim 3, characterized in that, The method for constructing the three-dimensional dynamic compensation lookup table is as follows: Place the oscilloscope in a temperature chamber and set several test temperature points in fixed steps within a preset temperature range. At each test temperature point, iterate through all vertical range settings and all bandwidth settings to form several combinations of operating parameters. Each combination of operating parameters includes a test temperature point, a vertical range setting, and a bandwidth setting. Record the noise transfer function model generated under each combination of operating parameters. Using one combination of operating parameters as a reference, calculate the correction coefficients relative to the reference for each of the remaining combinations of operating parameters. The correction coefficients include gain and phase. By binding the temperature, vertical range, and bandwidth settings under each combination of operating parameters to correction coefficients, we obtain temperature-gain correlation groups, range-gain relationship groups, and bandwidth correction corresponding groups. The temperature-gain correlation group, range-gain relationship group, and bandwidth correction corresponding group under each combination of operating parameters are stored as a record in the three-dimensional dynamic compensation lookup table.

10. An oscilloscope noise floor dynamic cancellation device based on a virtual reference channel, characterized in that, include: The reference signal acquisition module is used to obtain the occupancy status of each channel of the oscilloscope, determine the virtual reference channel and the main measurement channel based on the occupancy status of each channel, and acquire the background noise signal of the virtual reference channel in real time. The transfer function modeling module is used to establish a noise transfer function model between the virtual reference channel and the main measurement channel. The noise transfer function model is used to characterize the amplitude attenuation and phase delay characteristics of the background noise signal propagating from the virtual reference channel to the main measurement channel. The adaptive filtering cancellation module is used to input the background noise signal into the adaptive filter to generate a cancellation signal. The initial coefficients of the adaptive filter are obtained by conversion from the noise transfer function model. The cancellation signal is then subtracted from the digital waveform data input from the main measurement channel in real time to obtain the purified target signal.

Images