Methods for detecting and suppressing RF ADC spurious emissions

Abstract

This invention discloses a method for detecting and suppressing spurious signals in a radio frequency (RF) ADC, comprising: performing a fixed-point FFT spectral analysis on the signal acquired by the RF ADC to obtain the original spectral analysis result; obtaining the input signal and spurious signal based on the original spectral analysis result; determining a fixed-frequency spurious signal based on the obtained spurious signal; performing fixed-frequency spurious compensation on the acquired signal; extracting nonlinear spurious features; using the extracted nonlinear spurious features to generate training data to train a one-dimensional convolutional neural network model; inputting the signal to be suppressed into the model; and the model outputting a spurious suppression signal. This method for detecting and suppressing RF ADC spurious signals can automatically suppress signals using a trained model, greatly improving efficiency.

Description

Technical Field

[0001] This invention belongs to the field of radio frequency communication and signal processing technology, and specifically relates to a method for detecting and suppressing spurious emissions from a radio frequency ADC. Background Technology

[0002] In radio frequency (RF) communication systems, the RF analog-to-digital converter (ADC) is a critical component used to convert RF signals into digital signals for further processing. However, during the signal acquisition process of the RF ADC, spurious signals are often introduced, including fixed-frequency spurious signals and nonlinear spurious signals. Fixed-frequency spurious signals are typically caused by external environmental noise and inherent system noise, while nonlinear spurious signals are mainly caused by the nonlinear characteristics of the ADC (such as offset error, gain error, and crossover error). The presence of these spurious signals limits the spurious-free dynamic range (SFDR) of the data acquisition system, thereby degrading the overall system performance.

[0003] Existing spurious suppression methods typically only address specific types of spurious signals and have limited effectiveness. For example, some methods reduce spurious signal generation through hardware design optimization, but these methods are costly and difficult to completely eliminate spurious signals. Other methods employ digital signal processing techniques for spurious signal suppression, but these methods often require complex algorithms and substantial computational resources, and are poorly adaptable to different types of spurious signals. Summary of the Invention

[0004] This invention provides a method for detecting and suppressing spurious emissions from a radio frequency ADC, which at least solves one of the aforementioned technical problems. Specifically, the technical solution is as follows:

[0005] A method for detecting and suppressing spurious emissions from a radio frequency ADC includes the following steps:

[0006] The original spectrum analysis results are obtained by performing a fixed-point FFT spectrum analysis on the signal acquired by the RF ADC;

[0007] The input signal and spurious signal are obtained based on the original spectrum analysis results;

[0008] Determine the fixed-frequency spurious signal based on the obtained spurious signal;

[0009] Fixed-frequency spurious compensation is performed on the acquired signals;

[0010] Extracting nonlinear spurious features;

[0011] The extracted nonlinear spurious features are used to generate training data to train a one-dimensional convolutional neural network model. The signal to be suppressed is input into the model, and the model outputs a spurious suppression signal.

[0012] Furthermore, the original spectrum analysis results include the input signal, fixed-frequency spurious signals, and nonlinear spurious signals.

[0013] Furthermore, the specific method for obtaining the input signal and spurious signal based on the original spectrum analysis results is as follows:

[0014] A signal with the same frequency and amplitude as the original spectrum analysis result is generated using a signal source. The signal is then sequentially acquired by an RF ADC and subjected to a fixed-point FFT spectrum analysis. The analysis result is compared with the original spectrum analysis result. The signal with the highest consistency is determined as the input signal, and the remaining signals are determined as spurious signals.

[0015] Furthermore, the specific method for determining fixed-frequency spurious signals based on the obtained spurious signals is as follows:

[0016] Further analysis of the identified spurious signals involves generating a series of signals at different frequencies using a signal source, sequentially acquiring them through an RF ADC, and performing a fixed-point FFT spectrum analysis. The signal with the same frequency and amplitude is identified as the fixed-frequency spurious signal.

[0017] Furthermore, the specific method for performing fixed-frequency spurious compensation on the acquired signal is as follows:

[0018] Based on the spectrum analysis results, the frequency of the fixed-frequency spurious signal is extracted and its amplitude is calculated. The phase of the input frequency is calculated, the compensation signal is determined, and the compensation data is obtained by substituting the sampling time point. The fixed-frequency spurious signal compensation is completed by subtracting the compensation data from the signal acquired by the RF ADC.

[0019] Furthermore, the specific method for extracting nonlinear spurious features is as follows:

[0020] Determine the frequency of the input signal, calculate the possible nonlinear spurious frequencies, compare the actual spurious frequencies to determine the causes of the nonlinear spurious signals, and extract the nonlinear spurious characteristics.

[0021] Furthermore, by analyzing the spurious frequency positions introduced by offset error and gain error, the formation law of nonlinear spurious signals is derived, and the characteristics of nonlinear spurious signals, including the frequency, amplitude, and phase information of the spurious signals, are extracted.

[0022] Furthermore, when training a one-dimensional convolutional neural network model, the loss function uses frequency-domain weighted mean square error, including time-domain mean square error loss, frequency-domain loss, spectral convergence loss, and regularization loss.

[0023] Furthermore, when training a one-dimensional convolutional neural network model, the Adabound algorithm is used to improve the problems of gradient vanishing and overfitting.

[0024] The advantage of this invention lies in the fact that the provided method for detecting and suppressing RF ADC spurious signals can automatically suppress spurious signals through a trained model, which greatly improves efficiency. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 A schematic diagram illustrating a method for detecting and suppressing spurious emissions from a radio frequency ADC provided by the present invention;

[0027] Figure 2 This is a schematic diagram of input signal analysis and spurious detection provided by the present invention;

[0028] Figure 3 A schematic diagram of fixed-frequency spurious emission detection provided by the present invention;

[0029] Figure 4 This is a schematic diagram illustrating the training of the 1DCNN model provided by the present invention. Detailed Implementation

[0030] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0031] like Figure 1 The diagram illustrates a method for detecting and suppressing spurious emissions from a radio frequency ADC according to this application, comprising the following steps:

[0032] S1: Perform a fixed-point FFT spectrum analysis on the signal acquired by the RF ADC to obtain the original spectrum analysis results.

[0033] Specifically, such as Figure 2 As shown, an unknown signal is input to an RF ADC, and the acquired data is subjected to a fixed-point FFT spectrum analysis to obtain a spectrum containing several signals x1(t), x2(t), ..., x n (t) The original spectral analysis results of frequency and amplitude, which include the input signal, fixed-frequency spurious signals and nonlinear spurious signals.

[0034] S2: Obtain the input signal and spurious signal based on the original spectrum analysis results.

[0035] Specifically, in the embodiments of this application, the method for obtaining the input signal and spurious signal based on the original spectrum analysis results is as follows:

[0036] A signal with the same frequency and amplitude as the original spectrum analysis result is generated using a signal source. The signal is then sequentially acquired by an RF ADC and subjected to a fixed-point FFT spectrum analysis. The analysis result is compared with the original spectrum analysis result. The signal with the highest consistency is determined as the input signal, and the remaining signals are determined as spurious signals.

[0037] S3: Determine the fixed-frequency spurious signal based on the obtained spurious signal.

[0038] In the embodiments of this application, the specific method for determining fixed-frequency spurious signals based on the obtained spurious signals is as follows:

[0039] Further analysis of the identified spurious signals involves generating a series of signals at different frequencies using a signal source, sequentially acquiring them through an RF ADC, and performing a fixed-point FFT spectrum analysis. The signal with the same frequency and amplitude is identified as the fixed-frequency spurious signal.

[0040] like Figure 3 As shown, where s n (t) represents signals of different frequencies generated by the signal source. After passing through the RF ADC in sequence, n different signals are acquired. A fixed-point FFT spectrum analysis is performed on these signals to find the signal with the same frequency and amplitude, which is the fixed-frequency spurious u(t).

[0041] S4: Perform fixed-frequency spurious compensation on the acquired signal.

[0042] In the embodiments of this application, the specific method for performing fixed-frequency spurious compensation on the acquired signal is as follows:

[0043] Based on the spectrum analysis results, the frequency of the fixed-frequency spurious signal is extracted and its amplitude is calculated. The phase of the input frequency is calculated, the compensation signal is determined, and the compensation data is obtained by substituting the sampling time point. The fixed-frequency spurious signal compensation is completed by subtracting the compensation data from the signal acquired by the RF ADC.

[0044] Specifically, after determining the frequency and amplitude of u(t), the phase is unified, and the value U(k) of u(t) at each sampling time point is calculated based on the sampling frequency. The difference between the data acquired by the RF ADC and U(t) is then calculated to complete the fixed-frequency spurious compensation. The specific steps are as follows:

[0045] Based on the spectral analysis results, the frequency f of the fixed-frequency spurious u(t) is extracted. u And calculate its amplitude A u ; Calculate the input frequency f in The phase φ; determine the compensation signal u cSubstituting (t) into the sampling time point yields the compensated data U. c (k) subtracts the ADC data from the compensation data and truncates the output to complete fixed-frequency spurious compensation.

[0046] S5: Extract nonlinear spurious features.

[0047] In the embodiments of this application, the specific method for extracting nonlinear spurious features is as follows:

[0048] Determine the frequency of the input signal, calculate the possible nonlinear spurious frequencies, compare the actual spurious frequencies to determine the causes of the nonlinear spurious signals, and extract the nonlinear spurious characteristics.

[0049] In the embodiments of this application, by analyzing the spurious frequency positions introduced by offset error and gain error, the formation law of nonlinear spurious is derived, and the characteristics of nonlinear spurious are extracted, including the frequency, amplitude, and phase information of the spurious.

[0050] Understandably, once the frequency of the input signal is determined, in addition to easily obtaining the frequency locations of harmonic spurious signals, the frequency locations of spurious signals introduced by mismatch and gain mismatch can also be derived. By comparing the theoretical and actual spurious signal locations, the formation patterns of these nonlinear spurious signals can be identified, and their characteristics can be extracted for subsequent elimination.

[0051] Offset error is a DC component introduced into the output signal of a monolithic ADC due to factors such as signal path inconsistency, chip temperature, and stress. By definition, for a system containing M interleaved ADCs, its output in the time domain can be expressed as:

[0052]

[0053] Among them O k Let T represent the offset error of the k-th ADC, and T be the system sampling period. Taking a Fourier transform of the above equation, we get:

[0054]

[0055] in,

[0056]

[0057] Therefore, the location of the spurious points introduced by the misalignment error is:

[0058]

[0059] Gain error arises during quantization due to the difference in slope between the actual output signal and the ideal signal. Therefore, for a system with M interleaved ADCs and a sampling period of T, the time-domain expression for the signal with gain mismatch is:

[0060]

[0061] By using Fourier transform, the expression of gain mismatch in the time domain is transformed into its expression in the frequency domain:

[0062]

[0063] Let the spectrum of the ideal output x(t) be X(jω). Convolving it with the gain mismatch G(jω) yields the expression for the actual output:

[0064]

[0065] Therefore, the location of the spurious points introduced by the gain error is:

[0066]

[0067] In summary, after determining the input signal frequency, the possible nonlinear spurious frequencies of the RF ADC can be calculated using the following formula. The following formula represents the frequency of the nth harmonic spurious signal; changing the variable k can locate the aliased harmonic spurious signals to different Nyquist zones.

[0068] nf in ±kf s k = 0, 1, 2, ...

[0069] Offset error will produce spurious signals at the frequency of the following equation:

[0070]

[0071] Gain and phase errors introduce spurious signals at the frequency of the following equation:

[0072]

[0073] Furthermore, the nonlinearity and mismatch error of interleaved ADCs work together to introduce spurious signals at the location shown in the following equation, where M is the number of interleaved ADCs, and HD n It is an nth-order harmonic stray.

[0074]

[0075] By comparing the calculated potential nonlinear spurious frequencies of the RF ADC with the spurious frequencies obtained after spectrum analysis, the cause of each nonlinear spur can be determined.

[0076] S6: Use the extracted nonlinear spurious features to generate training data to train a one-dimensional convolutional neural network model. Input the signal to be suppressed into the model, and the model outputs a spurious suppression signal.

[0077] like Figure 4 As shown, the initial data (ADC signal with fixed-frequency spurious compensation) is used for model training after data augmentation. The training speed is improved by optimizing the algorithm, and the Batch Normalization (BN) layer is used to enhance the expressive power and prevent gradient vanishing and overfitting. After inputting the initial data to be processed into the trained model, the model can output spurious-suppressed data predicted by the model.

[0078] In addition to the ADC signal (initial data) compensated for spurious emissions at a fixed frequency, additional data highlighting nonlinear spurious characteristics is input. A clean signal x(t) is set according to the input signal frequency, and a spurious signal n(t) is set according to the frequency points of the aforementioned nonlinear spurious emissions. A 5% random offset is added to the spurious frequencies, and appropriate broadband noise n'(t) is added, with the amplitude of the interference signal following a Gaussian distribution, to obtain the data used for supervised training.

[0079] y(t)=x(t)+n1(t)+n2(t)+…+n′(t)

[0080] In the 1DCNN network architecture, the first part consists of a 3×1 kernel convolutional layer and a 3×1 pooling layer, mainly used for extracting short-term features from the signal. The middle multi-scale fusion framework contains convolutional kernels of sizes 3×1, 5×1, and 7×1, and each layer has three identical sets of convolutional layers, batch normalization (BN) layers, activation layers, and pooling layers. The loss function uses frequency-domain weighted mean squared error (MSE), and the Adabound algorithm is used to improve the model's gradient vanishing and overfitting issues.

[0081] As shown in the table below, consider the following loss term, where L MSE L spec L SC L sparse These are the time-domain mean square error loss, frequency-domain loss, spectral convergence loss, and regularization loss, respectively.

[0082] Table 1 Optimization Objectives for Each Loss Item

[0083] Loss Items Optimization Objective <![CDATA[L MSE ]]> Accurate reconstruction of time-domain waveforms ensures overall signal fidelity. <![CDATA[L spec ]]> The focus is on suppressing spurious frequency bands and eliminating specific frequency interference. <![CDATA[L SC ]]> Spectral energy distribution matching to avoid excessive suppression of useful signals <![CDATA[L sparse ]]> Sparsification of residual signals reduces erroneous modifications to non-spurious regions.

[0084] The formulas for calculating each loss item are as follows:

[0085]

[0086] Where x i For the i-th sampling point of the training data, Let i be the i-th sampling point of the network prediction signal.

[0087]

[0088] Where K is the number of time frames, F is the number of frequency points, and W(f) is the weighting coefficient.

[0089]

[0090] Where ||·|| represents the Frobenius norm, which forces a matching of the spectral energy distribution.

[0091] L sparse Apply regularization to the residual signal:

[0092]

[0093] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the above embodiments do not limit the present invention in any way, and all technical solutions obtained by equivalent substitution or equivalent transformation fall within the protection scope of the present invention.

Claims

1. A method for detecting and suppressing spurious emissions from a radio frequency ADC, characterized in that, Includes the following steps: The original spectrum analysis results are obtained by performing a fixed-point FFT spectrum analysis on the signal acquired by the RF ADC; The input signal and spurious signal are obtained based on the original spectrum analysis results; Determine the fixed-frequency spurious signal based on the obtained spurious signal; Fixed-frequency spurious compensation is performed on the acquired signals; Extracting nonlinear spurious features; The extracted nonlinear spurious features are used to generate training data to train a one-dimensional convolutional neural network model. The signal to be suppressed is input into the model, and the model outputs a spurious suppression signal. The specific method for extracting nonlinear spurious features is as follows: Determine the frequency of the input signal, calculate the possible nonlinear spurious frequencies, compare the actual spurious frequencies to determine the causes of the nonlinear spurious signals, and extract the nonlinear spurious characteristics. By analyzing the spurious frequency positions introduced by offset error and gain error, the formation law of nonlinear spurious is derived, and the characteristics of nonlinear spurious are extracted, including the frequency, amplitude and phase information of the spurious. When training a one-dimensional convolutional neural network model, the loss function uses frequency-domain weighted mean square error, which includes time-domain mean square error loss, frequency domain loss, spectral convergence loss and regularization loss. When training a one-dimensional convolutional neural network model, the Adabound algorithm is used to improve the problems of gradient vanishing and overfitting.

2. The method for detecting and suppressing RF ADC spurious emissions according to claim 1, characterized in that, The original spectrum analysis results include the input signal, fixed-frequency spurious signals, and nonlinear spurious signals.

3. The method for detecting and suppressing RF ADC spurious emissions according to claim 2, characterized in that, The specific method for obtaining the input signal and spurious signal based on the original spectrum analysis results is as follows: A signal with the same frequency and amplitude as the original spectrum analysis result is generated using a signal source. The signal is then sequentially acquired by an RF ADC and subjected to a fixed-point FFT spectrum analysis. The analysis result is compared with the original spectrum analysis result. The signal with the highest consistency is determined as the input signal, and the remaining signals are determined as spurious signals.

4. The method for detecting and suppressing RF ADC spurious emissions according to claim 3, characterized in that, The specific method for determining fixed-frequency spurious signals based on the obtained spurious signals is as follows: Further analysis of the identified spurious signals involves generating a series of signals at different frequencies using a signal source, sequentially acquiring them through an RF ADC, and performing a fixed-point FFT spectrum analysis. The signal with the same frequency and amplitude is identified as the fixed-frequency spurious signal.

5. The method for detecting and suppressing RF ADC spurious emissions according to claim 3, characterized in that, The specific method for fixed-frequency spurious compensation of the acquired signal is as follows: Based on the spectrum analysis results, the frequency of the fixed-frequency spurious signal is extracted and its amplitude is calculated. The phase of the input frequency is calculated, the compensation signal is determined, and the compensation data is obtained by substituting the sampling time point. The fixed-frequency spurious signal compensation is completed by subtracting the compensation data from the signal acquired by the RF ADC.

Images