An adaptive demodulation method for a digital oscilloscope

Abstract

The application discloses a kind of adaptive demodulation methods of digital oscilloscope, first train GRU neural network in GPU offline and generate coefficient matrix, then coefficient matrix is loaded to the FPGA that deployed GRU neural network, then through the acquisition system of digital oscilloscope, the signal to be demodulated is collected, the sampling data is multiplied with coefficient matrix, and then the modulation classification corresponding to the signal to be demodulated is determined;Finally, according to the modulation classification result of the signal to be demodulated, the corresponding modulation mode is selected, and the adaptive demodulation of the signal to be demodulated is completed, if classification error phenomenon occurs in the demodulation process, then GPU online incremental learning is carried out, so as to provide the accuracy of demodulation.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of digital oscilloscopes, and more specifically relates to an adaptive demodulation method for a digital oscilloscope. BACKGROUND

[0002] In modern communication systems, signals undergo various changes and disturbances. Channel conditions, such as multipath effects, signal attenuation, interference, etc., all affect the reception of signals. With the development of wireless communication technology, more complex signal processing techniques are needed to cope with the diversity of signals and the uncertainty of channels. The complexity and dynamics of the channel are also increasing. In early communication systems, fixed modulation and demodulation schemes are usually used, which means that even if the channel conditions change, the modulation method and demodulation strategy of the system remain unchanged. However, fixed demodulation schemes will lead to decreased communication efficiency and increased error rate when facing problems such as unstable signal strength, frequency-selective fading, time-varying interference, etc.

[0003] Traditional analog and digital demodulation techniques are relatively simple in design, but are usually affected by noise and interference. In addition, traditional demodulation usually does not support multiplexing technology, and due to the accuracy of channel conditions or estimation algorithms, there is a deviation between the estimated parameters and the actual parameters, which will affect the output results of signal demodulation. In a non-cooperative communication environment, these problems are particularly prominent, because there is no pre-defined protocol to assist in adapting to these changes, and the signal characteristics need to be estimated for blind processing demodulation. Cooperative communication provides some mechanisms to improve signal quality, but also requires the system to have higher flexibility and adaptability, and the user needs to manually change the parameter settings of the signal. It can be seen that traditional demodulation techniques have been unable to meet current needs, and in order to more flexibly and universally process modulated signals and improve the reliability and efficiency of communication, a demodulation method that can automatically identify and adapt to these changes is needed. SUMMARY

[0004] The purpose of the present application is to overcome the shortcomings of the prior art and provide an adaptive demodulation method for a digital oscilloscope, which selects the most suitable demodulation scheme according to real-time channel information, thereby improving spectral utilization, reducing interference, enhancing the reliability of communication, and improving the quality of blind demodulation of modulated signals and the accuracy of parameter estimation.

[0005] To achieve the above-mentioned application purpose, the adaptive demodulation method for a digital oscilloscope comprises the following steps:

[0006] (1) training a GRU neural network containing M neurons offline in a GPU and generating a coefficient matrix P;

[0007] (2) deploying the GRU neural network in the FPGA and loading the coefficient matrix P;

[0008] (3), sampling the signal to be demodulated and inputting to the FPGA;

[0009] (4), estimating the parameters of the sampled signal through the parameter estimation module;

[0010] (5), determining the modulation classification of the signal to be demodulated;

[0011] (6), selecting the corresponding modulation mode according to the modulation classification result of the signal to be demodulated and demodulating.

[0012] The application aims to achieve the following:

[0013] The adaptive demodulation method of the digital oscilloscope of the application first trains the GRU neural network offline in the GPU and generates the coefficient matrix, then loads the coefficient matrix to the FPGA where the GRU neural network is deployed, then collects the signal to be demodulated through the acquisition system of the digital oscilloscope, multiplies the sampling data with the coefficient matrix, and further determines the modulation classification corresponding to the signal to be demodulated; finally, according to the modulation classification result of the signal to be demodulated, the corresponding modulation mode is selected, and the adaptive demodulation of the signal to be demodulated is completed, and if classification error occurs during demodulation, online incremental learning of the GPU is performed, thereby providing the accuracy of demodulation.

[0014] Meanwhile, the adaptive demodulation method of the digital oscilloscope of the application also has the following beneficial effects:

[0015] (1), the application solves the problem of low accuracy of signal demodulation and parameter estimation under uncertain signal format and parameters; in addition, the adaptive demodulation technology not only can handle the diversity of signals and the uncertainty of channels, but also can adjust the demodulation strategy in real time to optimize the performance of the communication link.

[0016] (2), the application can more flexibly realize adaptive demodulation, achieve initial offline training, and then realize online incremental learning and demodulation. The system can gradually increase the types of signals that can be processed through online incremental learning; in the case of classification error, the current signal characteristics can be extracted through online incremental learning, the parameters can be fine-tuned, and the probability of subsequent same classification error can be reduced. Finally, the convenience, practicality, accuracy and reliability are continuously improved with the increase of use time.

[0017] (3), the application can be applied to cooperative communication and non-cooperative communication fields, does not need the user to customize the parameter setting of demodulation, realizes good man-machine interaction performance, can intelligently demodulate different modulation signal inputs, and improves the reliability and efficiency of communication.

[0018] (4), the application completes an adaptive demodulation system of a modulated signal under a large amount of data training and storage resource conditions in a GPU, meets the demand of modern measurement technology for intelligent technology, and has certain guiding significance for the field of modulated signal communication.

[0019] (5), the application can demodulate both analog modulated signals and digital modulated signals;

[0020] (6), the application trains a neural network by a GPU, and finally deploys the neural network in an FPGA. Compared with a neural network realized by software, the processing speed of the FPGA is faster by more than 100 times, and the real-time performance is stronger. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is a principle diagram of an adaptive demodulation method of a digital oscilloscope of the application;

[0022] Figure 2 is a sampling signal analog demodulation block diagram;

[0023] Figure 3 is a sampling signal digital demodulation block diagram;

[0024] Figure 4 is an error vector magnitude EVM schematic diagram;

[0025] Figure 5 is a 16-QAM demodulation result diagram. DETAILED DESCRIPTION

[0026] The specific embodiments of the application will be described below in conjunction with the accompanying drawings, so that those skilled in the art can better understand the application. It should be particularly noted that in the following description, when the detailed description of known functions and designs may dilute the main content of the application, these descriptions will be omitted here.

[0027] EMBODIMENT

[0028] In this embodiment, as shown in the application, an adaptive demodulation method of a digital oscilloscope, includes the following steps: Figure 1

[0029] (1), a GRU neural network containing M neurons is trained offline in a GPU, the GRU neural network completes the classification of N kinds of modulated signals every batch, thereby outputting a coefficient matrix P with a size of MxN, wherein the element p ij in the i-th row and j-th column of the matrix P is the neural weight coefficient of the i-th neuron about the j-th modulation type;

[0030] In this embodiment, the number of neurons in the GRU neural network is M=100, and the classification of 20 kinds of modulated signals is completed every batch.​

[0031] (2), the GPU loads the signal template library, the signal template library is composed of known n kinds of modulation signals with different formats, different parameters and different noises, and N-n kinds of random signals; the signal template library is input to the trained GRU neural network to obtain a coefficient matrix P; then the coefficient matrix P is sent to the CPU and then deployed to the FPGA by the CPU;

[0032] In the embodiment, the signal template library includes 11 kinds of known modulation formats of modulation signals, including 8 digital modulation signals (BPSK, 8PSK, QPSK, CPFSK, GFSK, 16-QAM, 64-QAM, 4-PAM) and 3 analog modulation signals (AM-DSB, AM-SSB, WBFM); 9 kinds of random signals are modulation signals of undetermined modulation type;

[0033] (3), as shown in Figure 1 , the demodulated signal S(t) is input to the digital oscilloscope, and the sampling signal with a length of M is obtained through ADC sampling, and then the sampling signal is input to the FPGA;

[0034] In the embodiment, the demodulated signal can be an analog modulation signal and a digital modulation signal, wherein the analog modulation signal can only be a periodic signal, and the baseband signal thereof is one of a sine wave, a square wave and a triangular wave.

[0035] (4), the parameters of the sampling signal are estimated by the parameter estimation module in the FPGA, including: carrier frequency f c , signal bandwidth B, symbol rate R, signal amplitude A, signal frequency f1 and roll-off factor a;

[0036] (5), the FPGA multiplies the sampling signal with a length of M by the coefficient matrix P to obtain a 1*N probability matrix, each element in the probability matrix representing the probability that the demodulated signal belongs to each modulation type signal in the signal template library, and then the maximum probability value is taken as the modulation classification corresponding to the demodulated signal;

[0037] (6), according to the modulation classification result of the demodulated signal, the corresponding modulation mode is selected, if the demodulated signal is an analog modulation signal, step (7) is entered; if the demodulated signal is a digital modulation signal, step (8) is entered; otherwise, step (9) is entered;

[0038] (7), the sampling signal is analog demodulated, and the specific demodulation process is shown in Figure 2 ;

[0039] (7.1), in the analog demodulation module, the sampling signal is multiplied by the carrier frequency f c, and then low-pass filtered through a low-pass filter with a bandwidth of B, so as to demodulate a baseband signal X = {x1, x2, …, xM} containing M points, wherein xi represents a quantized value of an i-th sampling point. i ,…,x M},x i represents a quantized value of an i-th sampling point.

[0040] (7.2), using a KNN algorithm to classify the demodulated baseband signal to obtain a waveform type of the baseband signal;

[0041] In this embodiment, the KNN algorithm has a sample signal library containing sample sequences of complete waveforms of sine waves, square waves and triangular waves. By calculating the Euclidean distances between the baseband signal and each sample signal, the smallest Euclidean distance is obtained, and the waveform of the baseband signal is classified as the sample signal corresponding to the smallest Euclidean distance.

[0042] (7.3), the CPU generates a reference waveform Y = {y1, y2, …, yM} containing M points according to the signal amplitude A, the signal frequency f1 and the waveform type obtained by KNN classification, wherein yi represents a quantized value of an i-th reference point. i ,…,y M},y i represents a quantized value of an i-th reference point.

[0043] (7.4), the Manhattan distance between the demodulated baseband signal and the reference waveform is calculated:

[0044]

[0045] Curve similarity is used to measure the degree of similarity between two or more curves. Manhattan distance is a commonly used curve similarity, which calculates the sum of absolute differences between two vectors (or curves), considers the differences between dimensions, and is suitable for curve data with multiple dimensions.

[0046] (7.5), comparing the Manhattan distance MD with a set threshold value, if MD is less than the set threshold value, it is determined that the demodulation and classification result is correct, and the processing is ended; otherwise, step (7.6) is entered.

[0047] (7.6), GPU online incremental learning;

[0048] (7.6.1), controlling the host computer to display the sampling signal, and identifying the modulation type of the sampling signal through artificial retrieval;

[0049] (7.6.2), judging whether the identified modulation type is a known modulation type in the signal template library, if the modulation type is not a known modulation type, adding a known modulation signal to the signal template library, i.e. the known types of modulation signals in the signal template library are changed from n to n+1; if the modulation type is a known modulation type, no new modulation type is added;

[0050] (7.6.3), the GPU reloads the signal template library, then inputs to the trained GRU neural network to obtain a new coefficient matrix P', and then sends the new coefficient matrix P' to the CPU and redeploys it to the FPGA, and replaces the original coefficient matrix P, completing the GPU online incremental learning;

[0051] (8), sampling signal digital demodulation, the specific demodulation process is shown in Figure 3 ;

[0052] (8.1), the sampling signal is digitally mixed with a carrier with a carrier frequency of f c to obtain sum and difference frequency signals, and then low-pass filtering is performed through a low-pass filter with a bandwidth of B to reserve the difference frequency signal;

[0053] The coefficients of the root-raised cosine filter are generated according to the roll-off factor α, and then the root-raised cosine filter is used to perform matched filtering on the difference frequency signal to obtain the baseband signal;

[0054] The relationship between the coefficients of the root-raised cosine filter and the roll-off factor is as follows:

[0055]

[0056] Where T R is the symbol period, which is the inverse of the symbol rate R; g R (t) is the continuous-time expression of the root-raised cosine filter coefficient; in practical applications, it also needs to be discretized and truncated according to the actual situation.

[0057] (8.2), resample the baseband signal to make the sampling rate 4 times of R; then perform timing synchronization on the resampled signal, and then perform carrier synchronization to correct the carrier frequency offset and carrier phase offset, to obtain two corrected I / Q baseband signals, and send them to the CPU;

[0058] (8.3), obtain the binary sequence in the I / Q baseband signal respectively, and then generate I ref / Q ref two-way reference signals;

[0059] (8.4), calculate the root mean square E RMS of the error vector magnitude EVM:

[0060]

[0061] wherein E(i) is the vector difference of the i-th sampling point, I(i), Q(i) are the quantized values of the i-th sampling point in the I / Q baseband signal, I ref (i), Q ref (i) are the quantized values of the i-th sampling point in the I / Q baseband signal, I ref / Q ref the quantized values of the i-th reference point in the reference signal;

[0062] In the present embodiment, the EVM is one of the most widely used parameters in the current vector signal analysis, as shown in the following formula (1) : Figure 4 The EVM reflects the amplitude of the vector difference between the measured signal and the ideal signal, and can comprehensively reflect the quality of the received signal. The EVM measurement is independent of the specific modulation format. Defects in any link of the modulated signal will be reflected on the EVM. Therefore, the EVM is called the total error of the vector signal analysis. The smaller the EVM is, the smaller the probability of misjudgment of the transmitted signal is, which indicates that the demodulation result of the system is correct.

[0063] (8.5), comparing the root mean square E RMS with the size of the set threshold value, if E RMS is smaller than the set threshold value, it is determined that the demodulation and classification results are correct, and the processing is ended; otherwise, step (8.6) is entered;

[0064] (8.6), GPU online incremental learning;

[0065] (8.6.1), controlling the host computer to display the sampling signal, and identifying the modulation type of the sampling signal through an artificial retrieval method;

[0066] (8.6.2), if the modulation type is not a known modulation type, adding a known modulation signal to the signal template library, i.e. the number of known modulation signals in the signal template library is changed from n to n+1; if the modulation type is a known modulation type, no new modulation type is added;

[0067] (8.6.3), the GPU reloads the signal template library, and then inputs to the trained GRU neural network to obtain a new coefficient matrix P', and then sends the new coefficient matrix P' to the CPU and redeploy it to the FPGA, and replace the original coefficient matrix P, to complete the GPU online incremental learning;

[0068] (9), online incremental learning of the signal of the undetermined modulation type;

[0069] (9.1), controlling the host computer to display the sampling signal, and identifying the modulation type of the sampling signal through an artificial retrieval method;

[0070] (9.2), increase the signal template library with a known modulation signal, i.e. the known type of modulation signal in the signal template library is changed from n to n+1;

[0071] (9.3), the GPU reloads the signal template library, then inputs to the trained GRU neural network to obtain a new coefficient matrix P', and then sends the new coefficient matrix P' to the CPU and redeploy it to the FPGA, and replace the original coefficient matrix P, complete the GPU online incremental learning.

[0072] Example simulation

[0073] In this embodiment, a 16-QAM modulation signal is input into the system, after being sampled by the ADC, it is input into the FPGA. The FPGA estimates its carrier frequency f c , bandwidth B, symbol rate R and roll-off factor a, and classifies it at the same time, the classification result is 16-QAM, which belongs to digital modulation. Therefore, the f c , B, R, a are used for digital demodulation next. And the demodulation result is output to the CPU module, after comparison and evaluation by the CPU module, the constellation diagram of the 16-QAM demodulation result of the adaptive demodulation system is as Figure 5 shown, the horizontal coordinate is I, the vertical coordinate is Q, and the center of the 16 "+"s represents the position of the 16-QAM ideal constellation point, and the coordinates (i, q) of the points around it each correspond to a pair of demodulated I / Q baseband signals. From the figure, it can be seen that the EVM vector deviation value is small, and the demodulation is successful.

[0074] Although the above describes the illustrative specific embodiments of the present application in order to facilitate the understanding of the present application by those skilled in the art, it should be clear that the present application is not limited to the scope of the specific embodiments, and for those skilled in the art, as long as various changes are within the spirit and scope of the present application defined and determined by the appended claims, all the inventions utilizing the concept of the present application are included in the protection.

Claims

1. An adaptive demodulation method for a digital oscilloscope, characterized in that, Includes the following steps: (1) Train a GRU neural network with M neurons offline on a GPU. The GRU neural network classifies N kinds of modulation signals in each batch, and outputs a coefficient matrix P of size M×N, where the element p in the i-th row and j-th column of matrix P is... ij It is the neural weight coefficient of the i-th neuron with respect to the j-th modulation type; (2) The GPU loads the signal template library, which consists of n known modulated signals with different formats, parameters, and noise, as well as Nn random signals. The signal template library is input into the trained GRU neural network to obtain the coefficient matrix P. Then the coefficient matrix P is sent to the CPU and deployed to the FPGA by the CPU. (3) Input the signal to be demodulated into a digital oscilloscope, obtain a sampling signal of length M through ADC sampling, and then input the sampling signal into the FPGA; (4) Estimate the parameters of the sampled signal using the parameter estimation module in the FPGA, including: carrier frequency f c Signal bandwidth B, symbol rate R, signal amplitude A, signal frequency f1, and roll-off factor α; (5) The FPGA multiplies the sampled signal of length M with the coefficient matrix P to obtain a 1*N probability matrix. Each element in the probability matrix is ​​the probability that the signal to be demodulated belongs to each modulation type signal in the signal template library. Then, the maximum probability is taken as the modulation classification corresponding to the signal to be demodulated. (6) Select the corresponding modulation method according to the modulation classification of the signal to be demodulated. If the signal to be demodulated is an analog modulation signal, proceed to step (7); if the signal to be demodulated is a digital modulation signal, proceed to step (8); otherwise, proceed to step (9). (7) Analog demodulation of the sampled signal; (7.1) The analog demodulation module determines the frequency f of the sampled signal based on the carrier frequency f. c The signal bandwidth B is digitally down-converted and low-pass filtered to demodulate the baseband signal X = {x1, x2, ..., x...}, which contains M points. i ,…,x M }, x i This represents the amplitude of the i-th sampling point; (7.2) Use the KNN algorithm to classify the demodulated baseband signal and obtain the waveform type of the baseband signal; (7.3) The CPU generates a reference waveform Y = {y1, y2, ..., y3} containing M points based on the signal amplitude A, the signal frequency f1, and the waveform type obtained from KNN classification. i ,…,y M }, y i This represents the magnitude at the i-th reference point; (7.4) Calculate the Manhattan distance between the demodulated baseband signal and the reference waveform: (7.5) Compare the Manhattan distance (MD) with the set threshold. If the MD is less than the set threshold, the demodulation and classification results are considered correct and the process ends; otherwise, proceed to step (7.6). (7.6) GPU online incremental learning; (7.6.1) Control the host computer to display the sampling signal and identify the modulation type of the sampling signal through manual retrieval; (7.6.2) Determine whether the identified modulation type is a known modulation type in the signal template library. If the modulation type is not a known modulation type, add a known modulation signal to the signal template library, that is, the known types of modulation signals in the signal template library change from n to n+1; if the modulation type is a known modulation type, do not add a new modulation type. (7.6.3) The GPU reloads the signal template library and then inputs it into the trained GRU neural network to obtain a new coefficient matrix P'. The new coefficient matrix P' is then sent to the CPU and redeployed to the FPGA to replace the original coefficient matrix P, thus completing the GPU's online incremental learning. (8) Digital demodulation of the sampled signal; (8.1) The sampled signal and the carrier frequency f c The carrier wave is digitally mixed to obtain the sum frequency signal and the difference frequency signal, and then the difference frequency signal is retained by low-pass filtering with a bandwidth of B. The coefficients of the root-raised cosine filter are generated based on the roll-off factor α, and then the root-raised cosine filter is used to perform matched filtering on the difference frequency signal to obtain the baseband signal. (8.2) Resample the baseband signal to make the sampling rate 4 times R; then perform timing synchronization on the resampled signal, and then perform carrier synchronization to correct the carrier frequency offset and carrier phase offset to obtain two corrected I / Q baseband signals, and send them to the CPU. (8.3) Obtain the binary sequences in the I / Q baseband signals respectively, and then generate I. ref / Q ref Two reference signals; (8.4) Calculate the root mean square E of the error vector magnitude EVM RMS : Where E(i) is the vector difference of the i-th sampling point, and I(i) and Q(i) are the quantized values ​​of the i-th sampling point in the I / Q baseband signal, I ref (i), Q ref (i) is I ref / Q ref The quantized value of the i-th reference point in the reference signal; (8.5) Comparison of the root mean square magnitude of the error vector E RMS With respect to the set threshold size, if E RMS If the result is less than the set threshold, the demodulation and classification results are considered correct, and the processing ends; otherwise, proceed to step (8.6). (8.6) GPU online incremental learning; (8.6.1) Control the host computer to display the sampling signal and identify the modulation type of the sampling signal through manual retrieval; (8.6.2) Determine whether the identified modulation type is a known modulation type in the signal template library. If the modulation type is not a known modulation type, add a known modulation signal to the signal template library, that is, the known types of modulation signals in the signal template library change from n to n+1; if the modulation type is a known modulation type, do not add a new modulation type. (8.6.3) The GPU reloads the signal template library and then inputs it into the trained GRU neural network to obtain a new coefficient matrix P'. The new coefficient matrix P' is then sent to the CPU and redeployed to the FPGA to replace the original coefficient matrix P, thus completing the GPU's online incremental learning. (9) Online incremental learning of signals with undetermined modulation types; (9.1) Control the host computer to display the sampling signal and identify the modulation type of the sampling signal through manual retrieval; (9.2) Add a known modulation signal to the signal template library, that is, change the known types of modulation signals in the signal template library from n to n+1; (9.3) The GPU reloads the signal template library and then inputs it into the trained GRU neural network to obtain a new coefficient matrix P'. The new coefficient matrix P' is then sent to the CPU and redeployed to the FPGA to replace the original coefficient matrix P, thus completing the GPU's online incremental learning.

2. The adaptive demodulation method for a digital oscilloscope according to claim 1, characterized in that, The demodulated signal includes analog modulation signal and digital modulation signal. The analog modulation signal can only be a periodic signal, and its baseband signal is one of sine wave, square wave and triangular wave.

Images