Intelligent signal waveform analysis method for handheld device

Abstract

The invention provides a signal waveform intelligent analysis method for handheld equipment, belongs to the technical field of deep learning, and aims to solve the problems that the conventional portable measurement equipment is limited in function, lacks intelligent identification capability and only supports basic spectrum measurement. The method comprises the following steps: step 1, receiving an IQ data stream by using the handheld equipment; step 2, preprocessing the received IQ data stream; step 3, training a deep learning model by using the preprocessed IQ data stream; and step 4, obtaining a signal waveform monitoring identification result based on the trained deep learning model. According to the invention, the robust and reliable signal identification function is realized by using the strong feature learning capability of deep learning. Deep learning and time-frequency analysis are adopted, and the limitation of traditional Fourier transform is broken through. A communication modulation signal, a frequency hopping signal and a radar signal can be detected in real time, and the signal identification precision gt is improved; 95%.

Description

Technical Field

[0001] The present invention relates to an intelligent signal waveform analysis method for handheld devices and belongs to the technical field of deep learning. Background Art

[0002] Existing signal analyzers usually rely on Fourier transform for signal processing and analysis, and have limitations in processing certain complex non-periodic signals and non-stationary signals. For a finite-length non-periodic signal, during truncation, the spectrum undergoes convolution, resulting in the appearance of false components in the spectrum and causing spectral leakage, which affects the frequency resolution and the overall shape of the spectrum, making the analysis results inaccurate. For non-stationary signals, the product of the time-domain window and the frequency-domain window is constant and greater than or equal to one-half. This means that it is impossible to simultaneously obtain high time resolution and frequency resolution. Signal analyzers usually need to process a large amount of data and perform complex mathematical operations and signal processing, and usually rely on high-performance computers for data analysis or use professional data processing software to assist in the analysis. They consume a large amount of power, and the data processing and analysis process takes a long time with high latency. Especially when processing large-scale data sets or performing real-time analysis, they not only cannot meet the real-time requirements of mobile scenarios but also cause cost waste. Traditional desktop spectrometers or waveform analyzers usually have a large volume and weight, are not convenient to carry and use mobilely, and are difficult to deploy in mobile application environments such as on-site, vehicle-mounted, and airborne. This poses a challenge for applications that require rapid measurements on-site. Although modern spectrometers and waveform analyzers are more compact and lightweight in design, their portability still needs to be improved compared to portable measurement devices. Moreover, existing portable devices have limited functions and lack intelligent recognition capabilities, and only support basic spectrum measurements. Therefore, the present invention proposes an intelligent signal waveform analysis method that combines deep learning and embedded computing, and focuses on deploying intelligent analysis algorithms on lightweight handheld devices to achieve a lightweight, accurate, efficient, and intelligent signal waveform cognitive analysis system. Summary of the Invention

[0003] The present invention aims to solve the problems that existing portable measurement devices have limited functions, lack intelligent recognition capabilities, and only support basic spectrum measurements, and further proposes an intelligent signal waveform analysis method for handheld devices.

[0004] The technical solution adopted by the present invention to solve the above problems is as follows: The present invention includes the following steps:

[0005] Step 1: Use a handheld device to receive IQ data streams;

[0006] Step 2: Preprocess the received IQ data streams;

[0007] Step 3: Use the preprocessed IQ data streams to train a deep learning model;

[0008] Step 4: Obtain the signal waveform monitoring and recognition results based on the trained deep learning model.

[0009] Further, step 2 specifically includes:

[0010] Step 2.1: Normalize the received IQ data stream.

[0011] Step 2.2: Frame the normalized IQ data stream.

[0012] Step 2.3: Convert the framed IQ data into a three-dimensional tensor, where the three-dimensional tensor is the number of signal samples × time steps × 2, and the last dimension '2' corresponds to the real and imaginary parts of the IQ data stream respectively, completing the preprocessing of the IQ data stream and outputting the IQ data stream after format conversion.

[0013] Further, step 2.2 specifically includes:

[0014] Divide the continuous IQ data stream after normalization processing into equal-length segments according to a time window. During the framing process, a 10% repetition rate is adopted, and the length of each frame is divided into 2400 sampling points.

[0015] Further, in step 3, the deep learning model adopts a CNN + LSTM structure for learning time series features and distinguishing input signals of different modulation types, where CNN is a convolutional neural network and LSTM is a long short-term memory network.

[0016] Further, the training steps in step 3 include:

[0017] Step 3.1: Configure the API training environment and set the deep learning architecture.

[0018] Step 3.2: Set the optimization parameters during the training process.

[0019] Step 3.3: Set the network structure of the deep learning model, and divide the preprocessed IQ data stream into a training set and a test set according to a ratio of 7:3.

[0020] Step 3.4: Use the training set to train the deep learning model. After each training is completed, input the validation set for analysis. If the accuracy of the analysis result does not meet the preset value, improve the network. If the recognition detection is not completed, deepen the network layer and repeat the training of the network's optimization parameters until the accuracy of the analysis result meets the preset value.

[0021] The beneficial effects of the present invention are:

[0022] 1) The present invention can achieve efficient and intelligent analysis, surpassing traditional signal processing methods. By leveraging the powerful feature learning ability of deep learning, the present invention realizes a robust and reliable signal recognition function. Adopting deep learning + time-frequency analysis, it breaks through the limitations of traditional Fourier transform. It can detect communication modulation signals, frequency hopping signals, and radar signals in real time, improving the signal recognition accuracy, with the recognition accuracy > 95%.

[0023] 2) The present invention innovatively integrates signal preprocessing and representation and signal intelligent recognition software technology with a lightweight embedded handheld signal analyzer hardware platform, facilitating use and expansion. It has completed a low-latency, high-efficiency, and systematic intelligent waveform analysis platform. Compared with traditional desktop spectrometers, the selected device is small in size and low in power consumption, suitable for rapid on-site deployment.

[0024] 3) It has multi-scenario applicability: The intelligent, portable, and efficient signal waveform analysis method proposed by the present invention, based on artificial intelligence and embedded computing, breaks through the limitations of traditional signal analysis and is widely applicable to fields such as wireless communication, radar detection, security monitoring, 5G / 6G monitoring, radar signal analysis, and communication signal detection, having important commercial application value and application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a flow chart of a signal waveform intelligent analysis method for handheld devices provided by the present invention;

[0026] Figure 2 It is a flow chart of offline data recognition based on an embedded Linux platform provided by the present invention;

[0027] Figure 3 It is a schematic diagram of intelligent signal waveform analysis based on an embedded Linux platform provided by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0028] Combined with Figures 1-3 to illustrate this embodiment. As Figure 1 shown, the steps of a signal waveform intelligent analysis method for handheld devices described in this embodiment include:

[0029] S1: Use a handheld device to receive IQ data stream;

[0030] S2: Preprocess the received IQ data stream;

[0031] S201: Normalization processing: Reduce the influence of noise and improve signal stability;

[0032] S202: Frame processing: The continuous signal is segmented into equal-length segments according to a time window, and the length of each frame is 2400 sampling points. When framing, a 10% overlap rate is adopted to ensure the extraction of local features of the time series while avoiding information loss.

[0033] S203: Format conversion: Convert the IQ data into a three-dimensional tensor (number of samples × time step × 2), where the last dimension corresponds to the real part and the imaginary part respectively. For example, a data set containing 1500 samples will be converted into a tensor with a shape of (1500, 2400, 2) as the input to the network.

[0034] S3: Use the preprocessed IQ data stream to train the deep learning model;

[0035] In this embodiment, the deep learning model adopts a CNN + LSTM structure to achieve temporal feature learning and distinguish different modulation types, such as AM, FM, QPSK, BPSK, etc.;

[0036] S301: Configure the API training environment and set the deep learning architecture;

[0037] S302: Set the optimization parameters during training;

[0038] S303: Set the network structure of the deep learning model, randomly divide the preprocessed IQ data stream into a training set and a test set in a ratio of 7:3, and use the training set to train the deep neural model;

[0039] S304: After each training, input the validation set for analysis. If the accuracy of the analysis result does not meet the preset value, improve the network. If the recognition detection is not completed, deepen the network layer and repeat the training of the network's optimization parameters until the accuracy of the analysis result meets the preset value.

[0040] Based on the above method, this embodiment builds a system and realizes it in software: Deploy the AI model using Miniforge3 + TensorFlow Lite to optimize the inference speed, support Python + C++ hybrid development, and improve the embedded computing performance; Hardware architecture: Adopt the i.MX 8M processor (Cortex-A53, dual-core 1.2GHz), with low power consumption and high computing power, and use USB 3.0 high-speed data transmission to ensure real-time signal processing ability.

[0041] The present invention innovatively integrates the signal preprocessing and representation and signal intelligent recognition software technology with a lightweight embedded handheld signal analyzer hardware platform, which is convenient for use and expansion. A low-latency, high-efficiency, and systematic intelligent waveform analysis platform is completed. Compared with traditional desktop spectrum analyzers, the selected device is small in size and low in power consumption, and is suitable for on-site rapid deployment.

[0042] After the system is built, as Figure 2 shown, extract the waveform information of the two paths of the intermediate frequency signal IQ and save it. Call the network model for waveform analysis, and build a deep learning environment (miniforge3) on the embedded Linux platform. According to the model requirements and the size of resources and data, input the preprocessed two-path IQ data into the loaded model and obtain the output result.

[0043] In this embodiment, the built system is experimentally verified. The experimental environment is laboratory & outdoor radio monitoring. The intelligent signal waveform analysis process during the experiment is shown in Table 1, and the analysis results are as Figure 3 shown. From Figure 3 the verification results, it can be obtained that the device signal classification accuracy of this embodiment is > 95%.

[0044] Through the above experimental steps, it can be obtained that the present invention can achieve efficient and intelligent analysis, exceeding traditional signal processing methods. The present invention utilizes the powerful feature learning ability of deep learning to achieve a robust and reliable signal recognition function. By adopting deep learning + time-frequency analysis, it breaks through the limitations of traditional Fourier transforms. It can detect communication modulation signals, frequency hopping signals, and radar signals in real time, improving the signal recognition accuracy.

[0045] Table 1

[0046]

[0047]

[0048] In summary, the intelligent, portable, and efficient signal waveform analysis method proposed by the present invention, based on artificial intelligence and embedded computing, breaks through the limitations of traditional signal analysis and is widely applicable to fields such as wireless communication, radar detection, security monitoring, 5G / 6G monitoring, radar signal analysis, and communication signal detection, having important commercial application value and application prospects.

[0049] The above are only the preferred embodiments of the present invention and do not impose any form of limitation on the present invention. Although the present invention has been disclosed above with the preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some changes or modifications to the above-disclosed technical content to obtain equivalent embodiments with equivalent changes. However, as long as it does not depart from the technical content of the present invention and is based on the technical essence of the present invention, any simple modifications, equivalent replacements, and improvements made to the above embodiments still fall within the protection scope of the technical solution of the present invention.

Claims

1. An intelligent signal waveform analysis method for handheld devices, characterized in that, The steps of the signal waveform intelligent analysis method for handheld devices include: Step 1: Receive IQ data stream using a handheld device; Step 2: Preprocess the received IQ data stream; Step 3: Use the preprocessed IQ data stream to train a deep learning model; Step 4: Obtain the signal waveform monitoring and recognition results based on the trained deep learning model.

2. The intelligent signal waveform analysis method for a handheld device according to claim 1, characterized in that, Step 2 specifically includes: Step 2.1: Normalize the received IQ data stream; Step 2.2: Perform frame splitting on the normalized IQ data stream; Step 2.3: Convert the frame-split IQ data into a three-dimensional tensor, where the three-dimensional tensor is the number of signal samples × time step × 2, and the last "2" in the last dimension corresponds to the real part and the imaginary part of the IQ data stream respectively, complete the preprocessing of the IQ data stream, and output the IQ data stream after format conversion.

3. An intelligent signal waveform analysis method for a handheld device according to claim 2, characterized in that Step 2.2 specifically includes: The continuous IQ data stream after normalization is divided into equal-length segments according to a time window. During the frame splitting process, a 10% repetition rate is adopted, and the length of each frame is divided into 2400 sampling points.

4. An intelligent signal waveform analysis method for handheld devices according to claim 1, characterized in that, In Step 3, the deep learning model adopts a CNN+LSTM structure for temporal feature learning to distinguish input signals of different modulation types, where CNN is a convolutional neural network and LSTM is a long short-term memory network.

5. A signal waveform intelligent analysis method for a handheld device according to claim 1, characterized in that, The training steps in Step 3 include: Step 3.1: Configure the API training environment and set the deep learning architecture; Step 3.2: Set the optimization parameters during the training process; Step 3.3: Set the network structure of the deep learning model, and randomly divide the preprocessed IQ data stream into a training set and a test set at a ratio of 7:3; Step 3.4: Use the training set to train the deep learning model. After each training is completed, input the validation set for analysis. If the accuracy of the analysis result does not meet the preset value, improve the network. If the recognition detection is not completed, deepen the network layer and repeat the training of the network optimization parameters until the accuracy of the analysis result meets the preset value.

Images