Manufacturing method of OFDM communication signal data set based on FPGA and programmable interception

Abstract

The invention discloses an OFDM (Orthogonal Frequency Division Multiplexing) communication signal data set manufacturing method based on an FPGA (Field Programmable Gate Array) and programmable interception, solves the problems of few tag dimensions and lack of an OFDM communication system of the existing communication signal data set, and belongs to the field of computer vision. Comprising the following steps: adopting FPGA equipment as a transmitter, a cooperative end receiver and a non-cooperative detection receiver; the channel simulator simulates a channel type, multipath information and Doppler frequency offset required by non-cooperative detection and reception, and respectively sends obtained OFDM communication signals to the cooperative end receiver and the non-cooperative detection and reception machine; using a Kalman filter to clean and retain the intercepted OFDM communication signal meeting the amplitude threshold, calculating the signal power of which the score meets the task requirement, and adding Gaussian white noise to obtain a target OFDM communication signal; the method comprises the following steps: storing an OFDM communication signal data set in a pickle format through python; according to the invention, the production of the OFDM communication signal data set is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of computer vision and relates to a method for producing an OFDM communication signal data set based on FPGA and programmable interception in a modulation carrier system. Background Art

[0002] In communication systems, because baseband signals are difficult to communicate over long distances and are easily interfered with, in the process of wireless communication transmission, the radio frequency needs to up-convert the baseband signal to convert it into a high-frequency digital signal; the receiving end needs to down-convert the signal through the corresponding modulation method to restore the original baseband signal and realize the transmission and reception of the baseband signal.

[0003] In cooperative communication, the sender and receiver obtain prior information about the required modulation style through the communication protocol agreed upon by both parties, and "handshake" the front-end signals of the communication in order to demodulate the data.

[0004] In non-cooperative communication, the receiving end cannot determine the signal modulation method. Therefore, it must analyze the signal characteristics to determine its modulation method for subsequent signal demodulation operations and applications. However, in today's 6G communication, communication systems face the challenge of low latency and high speed. Therefore, the appropriate use of the automatic modulation identification function in non-cooperative communication can alleviate this challenge to a certain extent.

[0005] In the existing technology, communication signal modulation recognition methods are divided into two types: model-driven modulation recognition methods and data-driven modulation recognition methods. However, in today's real-world environments for communication signal modulation recognition, the scenarios are characterized by strong noise, a wide variety of modulation types, and limited energy supply. Model-driven modulation recognition methods rely too heavily on a priori design and cannot meet these conditions. Data-driven modulation recognition methods, due to the use of deep learning, require a large amount of data for model training to achieve the function of signal feature extraction. Therefore, in deep learning-based communication signal modulation recognition, the signal dataset plays a key role in the implementation of signal modulation recognition.

[0006] Since O'shea published Convolutional Radio Modulation Recognition Networks in 2016, he conducted GNU-Radio simulations of Shakespeare works in ASCII format using centralized digital modulation data in 2017. He used whitening randomization to ensure that characters and bits appeared with equal probability. He publicly released three datasets: RML2016.10a, RML2016.10b, and RML2016.10c. He also published RML2018.01a in 2019. In 2020, Tekbiyik's team proposed the HisarMod.2019 dataset. The performance indicators of the above five datasets are shown in Table 1:

[0007] Table 1

[0008] Dataset name Modulation method Total sample size Sampling point Single sample number RML2016.10a 11 modulation modes: 8PSK, BPSK, CPFSK, GFSK, PAM4, 16QAM, 64QAM, QPSK, AM-DSB, AM-SSB, WBFM 220,000 128 Single modulation mode for each SNR: 1000 RML2016.10b 10 modulation modes: 8PSK, BPSK, CPFSK, GFSK, PAM4,16QAM, 64QAM, QPSK, AM-DSB, WBFM 1.2 million 128 Single modulation mode for each SNR: 6000 RML2016.10c 11 modulation modes: 8PSK, BPSK, CPFSK, GFSK, PAM4,16QAM, 64QAM, QPSK, AM-DSB, AM-SSB, WBFM 160,000 128 Single modulation mode for each SNR: 800 RML2018.01a 24 modulation modes: 32PSK, 16APSK, 32QAM, FM, GMSK, 32APSK, OQPSK, 8ASK, BPSK, 8PSK, AM-SSB-SC, 4ASK, 16PSK, 64APSK, 128QAM, 128APSK, AM-DSB-SC, AM-SSB-WC, 64QAM, QPSK, 256QAM, AM-DSB-WC, OOK, 16QAM 2.55 million 1024 Each signal-to-noise ratio single modulation mode: 4250 HisarMod2019 26 modulation modes: 32PSK, 16APSK, 32QAM, FM, GMSK, 32APSK, OQPSK, 8ASK, BPSK, 8PSK, AM-SSB-SC, 4ASK, 16PSK, 64APSK, 128QAM, 128APSK, AM-DSB-SC, AM-SSB-WC, 64QAM, QPSK, 256QAM, AM-DSB-WC, OOK, 16QAM, 256QAM, 16FSK 780,000 1024 Single modulation mode for each SNR: 1500

[0009] The aforementioned datasets were released relatively early, allowing them to be adapted to the rapidly evolving communication systems and signals. For example, Publication No. CN116016069A, Invention Title: A Modulation Recognition System Based on Transfer Learning, Employing Simulated Digital Modulation Signals, Including BPSK, 8PSK, PAM4, PAM8, 16QAM, and 64QAM, a total of six modulation signals. However, the six types of digital modulation in this system share the same communication system as the digital modulation methods in the aforementioned public datasets, all of which are single-carrier communication modes, lacking attention to the OFDM communication system currently dominant in the 6G Internet of Things. At the same time, with the continuous development of wireless communications, various communication systems are constantly being updated. The LTE communication protocol, represented by the OFDM communication system, has become the dominant system in today's communication systems. In the future 6G communication field, which features low latency and high speed, high-order modulation (i.e., N-QAM, N ≥ 64), where a single symbol carries more than six bits of information, will gradually be used in various fields. The patent discloses that "analog channel processing, adding samples to a temporary array variable, adjusting the number of modulated signal types, the termination signal-to-noise ratio, the number of generated samples, the signal type, and the signal-to-noise ratio, adding the adjusted samples to the temporary array variable, and writing the temporary array variable to an HDF5 file to generate a source domain modulated signal dataset." However, the signal processing method described above suffers from the problem of not cleaning the signal dataset. When controlling the total signal-to-noise ratio of the signal under complex channels, there is a possibility that the simulated channel will be distorted to a certain extent. Therefore, the uncleaned dataset method proposed in the patent results in insufficient dataset creation efficiency.

[0010] Therefore, combining the above patents with existing public datasets, the following defects exist:

[0011] First, they are all single-carrier communication modes, which limits most communication signal modulation recognition research and makes it difficult to study and model multi-subcarrier communication signals;

[0012] Second, the system uses a communication signal generated by GNR-Radio software and then adds Gaussian white noise in Matlab to reduce the signal-to-noise ratio. This system is essentially a software simulation, and the generated signal is not actually transmitted or received in real time by a radio frequency transmitter or receiver, nor does it simulate the non-cooperative scenario used in modulation recognition. The interference noise is all generated from Gaussian white noise, and there is a lack of simulation of real-world LTE channel standards, such as multipath effects and Doppler frequency offset, or of actual communication signal transmission and reception conditions. Consequently, the proposed model is based on pure laboratory-generated signals, making it difficult to apply in the real world.

[0013] Third, in reality, communication signals are collected and represented as single sequence signals in IQ format. However, in the above dataset, signal samples are saved in the form of two-dimensional sequences + labels, making it impossible to pre-process the received IQ sequences before labeling, i.e., segment, save, label, and evaluate them. At the same time, because their labels are all modulation types and signal-to-noise ratios, subsequent researchers will face difficulties in carrying out multi-modulation type aliasing resolution. Summary of the Invention

[0014] In order to solve the technical problems of the current communication signal dataset with few label dimensions and the lack of OFDM communication system in the dataset, the present invention proposes a method for producing an OFDM communication signal dataset based on FPGA and programmable interception, in which an FPGA device with built-in Laview software is used as a transmitter, a cooperative receiver and a non-cooperative receiver respectively; the transmitter sends the LTE communication signal to the wireless channel, and after simulating the channel type, multipath information and Doppler frequency deviation required for non-cooperative detection through a channel simulator, the obtained OFDM communication signal is sent to the cooperative receiver and the non-cooperative receiver respectively; the non-cooperative receiver records the OFDM communication signal in the time domain and stores it. The system stores the non-cooperative signal in IQ format; uses the Kalman filter to clean the non-cooperative signal to obtain the target non-cooperative signal; retains the sequence of the target non-cooperative signal whose amplitude meets the amplitude threshold to obtain the intercepted OFDM communication signal; calculates the power of the intercepted OFDM communication signal whose score meets the task requirements, generates it by the signal-to-noise ratio and adds Gaussian white noise to obtain the target OFDM communication signal; uses Python to segment, reorganize, cut, sample, add preset labels and store it as samples in pickle format, and after testing, stores it as an OFDM communication signal dataset, which solves the problems of the current communication signal dataset with few label dimensions and the lack of OFDM communication system in the dataset.

[0015] The purpose of the present invention is specifically achieved through the following technical solutions:

[0016] The present invention discloses a method for producing an OFDM communication signal data set based on FPGA and programmable interception, using an FPGA device with built-in Laview software as a transmitter, a cooperative end receiver, and a non-cooperative receiver. The method includes:

[0017] In step 1, the transmitter sends the LTE communication signal into the wireless channel. After simulating the channel type, multipath information, and Doppler frequency deviation required for non-cooperative detection through a channel simulator, the resulting OFDM communication signal is sent to the cooperative receiver and the non-cooperative receiver respectively.

[0018] Step 2: The non-cooperative receiver records the OFDM communication signal in the time domain and stores it as a non-cooperative signal in IQ format; a Kalman filter is used to clean the non-cooperative signal to obtain the target non-cooperative signal;

[0019] Step 3: The cooperative end receiver records the OFDM communication signal and stores it as a cooperative signal in IQ format; an amplitude threshold is set based on the amplitude of the cooperative signal; and a sequence of the target non-cooperative signal whose amplitude meets the amplitude threshold is retained to obtain the intercepted OFDM communication signal;

[0020] Step 4: construct a difference metric function based on the statistical characteristics of the cooperative signal and the intercepted OFDM communication signal; construct a score function through the difference metric function; calculate the power of the intercepted OFDM communication signal whose score meets the task requirements, generate it from the signal-to-noise ratio and add Gaussian white noise to obtain the target OFDM communication signal;

[0021] Step 5: Use the numpy function library in Python to split the target OFDM communication signal into I sequence and Q sequence, reassemble the I sequence and Q sequence in the order of reception, cut them into sub-blocks based on the time interval and sampling rate, sample each sub-block, add preset labels, and store them as pickle format samples. The pickle format samples with correct storage format are stored as the OFDM communication signal dataset.

[0022] In step 1, the channel simulator simulates the channel type, multipath information, and Doppler frequency deviation required for non-cooperative detection in the following manner:

[0023] The channel simulator of the wireless channel uses a preset channel standard to linearly superimpose the channels corresponding to the LTE communication signal. By setting the transmission delay and Doppler frequency deviation values, it simulates the multipath effect and Doppler frequency deviation in the transmission channel of non-cooperative detection in a real environment to obtain the OFDM communication signal.

[0024] In step 2, the Kalman filter is used to clean the non-cooperative signal to obtain the target non-cooperative signal. The method includes:

[0025] S10, predict the error covariance of the non-cooperative signal through the hidden state to obtain the observation matrix of the Kalman filter;

[0026] S11, calculate the gain of the Kalman filter based on the observation matrix and output the predicted state at the current moment;

[0027] S12, constructing an update equation based on the updated observation gradient, updating the observation matrix, and obtaining the target non-cooperative signal after cleaning the non-cooperative signal using a Kalman filter.

[0028] In S10, the method of obtaining the observation matrix of the Kalman filter by predicting the error covariance of the non-cooperative signal through the hidden state is:

[0029] ;

[0030] Where y t is the hidden state, H t is the observation matrix; x t is the true value of the Kalman filter at time t; v t is the observation noise, v t Gaussian white noise with covariance Rt; t is the time variable;

[0031] In S11, the gain of the Kalman filter is calculated based on the observation matrix, and the calculation method for outputting the predicted state at the current moment is:

[0032] ;

[0033] ;

[0034] Where G t is the gain of the Kalman filter; P t is the state covariance matrix; Indicates a value of 10 -6 The constant parameter, T is the transpose; x t+1 is the predicted state at the current moment, that is, the output of the Kalman filter; H t x t is x t The observed value, y t -H t x t is the error between the observed value and the predicted observed value;

[0035] In S12, the update equation is:

[0036] ;

[0037] Where H t-1 is the observation matrix at time t-1, α is the learning step size, which is used to control the update rate, v t-1 is the observation noise at time t-1, which represents the update rate; g(x t-1 ) represents the updated observation gradient, x t-1 is the true value of the Kalman filter at time t-1.

[0038] In S12, the calculation method for updating the observation gradient is:

[0039] ;in,

[0040] ;

[0041] Where, Represents the observation function, j represents the imaginary part, N represents the total length of the signal, k is the discrete time variable, r(k) is the discrete time signal sequence, r is the discrete signal sequence function expression, k+N represents the offset of the sequence index, and e is a natural constant.

[0042] In step 3, the amplitude is calculated as:

[0043] ;

[0044] Where, is the amplitude, i represents the sequence number of the signal, I is the in-phase component of the signal, and Q is the quadrature component of the signal; the shape of the signal is: [number of samples, u, sample length]; where u = 0 represents the in-phase component I of the signal; u = 1 represents the quadrature component Q of the signal.

[0045] In step 4, the statistical characteristics include: amplitude mean, amplitude standard deviation, amplitude minimum, amplitude maximum, amplitude skewness and amplitude kurtosis; where,

[0046] The calculation method for the amplitude average is:

[0047] ;

[0048] The calculation method for the amplitude standard deviation is:

[0049] ;

[0050] The calculation method for the minimum amplitude is:

[0051] ;

[0052] The maximum amplitude is calculated as:

[0053] ;

[0054] The calculation method of amplitude skewness is:

[0055] ;

[0056] The calculation method of amplitude kurtosis is:

[0057] ;

[0058] Where, represents the average amplitude of the signal, Indicates the standard deviation of the signal amplitude; Indicates the minimum amplitude, represents the maximum amplitude; Sk represents the amplitude skewness; Kur represents the amplitude kurtosis.

[0059] In step 4, the difference measurement function is:

[0060] ;

[0061] The score function constructed based on the difference measurement function is:

[0062] ;

[0063] Where, d(f u real , f u ideal ) is the difference metric function value; f u real Represents the statistical characteristics of the intercepted OFDM communication signal, f u ideal is the statistical property of the cooperative signal, is a positive parameter; Q is the score, u is the weight number of the statistical characteristic, U represents the total weight of the statistical characteristic, w u is the weight of the statistical feature.

[0064] In step 4, the power of the intercepted OFDM communication signal whose score meets the task requirements is calculated. After generating the target OFDM communication signal by the signal-to-noise ratio and adding Gaussian white noise, the method is as follows:

[0065] ;

[0066] ;

[0067] Where, is power, The square of the amplitude of the intercepted OFDM communication signal indicating that the score meets the task requirements; is the signal-to-noise ratio, SNR=[20, 18, 16, 14, 12, 10, 8, 6, 4, 2, 0, -2, -4, -6, -8,-10]; The noise power is set to generate Gaussian white noise, which is then added to the intercepted OFDM communication signal whose score meets the task requirements to obtain the target OFDM communication signal.

[0068] In step 5, the target OFDM communication signal is divided into I sequence and Q sequence by using the numpy function library in Python. The I sequence and Q sequence are reassembled in the order of reception, and then cut into sub-blocks based on the time interval and sampling rate. After sampling and adding preset labels to each sub-block, the samples are stored as pickle format samples. The method for storing the pickle format samples with correct storage format as the OFDM communication signal dataset is as follows:

[0069] S20, construct the stable region [s, b] of the target OFDM communication signal:

[0070] [s, b] = find_stable_region ,

[0071] Where s is the starting index of the stable region, b is the ending index of the stable region, find_stable_region is the reading function in the numpy function library; D is the target OFDM communication signal, represents the sampling rate, The number of seconds to skip at the beginning, The number of seconds to skip at the end to remove unstable sequences;

[0072] S21, dividing the target OFDM communication signal in the stable region into an I sequence and a Q sequence: recombining the I sequence and the Q sequence into recombined samples in a receiving order, each recombined sample including a real part and an imaginary part;

[0073]

[0074]

[0075] Where, I[n] is the I sequence, which is the imaginary data of the recombined sample, Q[n] is the Q sequence, which is the real data of the recombined sample; D[2n + s] is the even position data in the stable region, D[2n + s + 1] is the odd position data in the stable region; n is the sampling sequence number;

[0076] S22, based on time interval The recombined samples are cut into sub-blocks by the sampling rate, where the sequence length L of the sub-block is: ; The number of sub-blocks B is: ; The sub-block range is: , is the sub-block number, ;

[0077] S23, after calculating the number of target samples to be skipped, random sampling or sequential sampling is used to extract a specified number of target samples in each sub-block. , , is the total length of the sub-block; the calculation method for the target number of samples to be skipped is:

[0078] ;

[0079] ;

[0080] Where, Indicates the target number of samples to skip at the beginning, Indicates the target number of samples to skip at the end;

[0081] Random sampling location for:

[0082]

[0083] Sequential sampling positions for:

[0084]

[0085] Where, is the sampling point position of each sub-block, Uniform is the function for generating uniformly distributed random numbers, p k ~Uniform means Obey uniformly distributed random numbers, is the sampling step length;

[0086] S24, using integer division to calculate the maximum number of target samples to be collected, and correcting the specified number based on the maximum number of samples to obtain the target number; performing a consistency check to ensure that the lengths of the real data and the imaginary data in the target samples collected are consistent, and then storing the samples in an IQ format original file;

[0087] S25, use the save_pickle function to add preset labels to the original sample file in IQ format and save it in pickle format;

[0088] S26, using the pickle.load function to read the labels and target numbers of the sample original file in pickle format, using the for 1 in range statement to print out the storage format of the sample original file in pickle format, and compare it with the test format, and use the sample original file in pickle format that passes the comparison as the OFDM communication signal dataset.

[0089] The beneficial effects of the present invention are:

[0090] 1. Compared with the single-carrier signals in the public dataset, there is a lack of simulation for the widely used OFDM communication system in modern times. At the same time, the simulated low-order digital modulation cannot meet the actual needs of modern communication for low latency and high speed. The present invention uses the LabVIEW Communications Suite framework in the built-in Laview software through FPGA devices to realize the generation of high-order modulated LTE communication signals in the communication system, making it more suitable for modern wireless communication needs.

[0091] 2. Regarding the traditional data set preparation method based on GNU-Radio, the gradient descent simulation of the signal-to-noise ratio is performed by adding Gaussian white noise through Matlab, which lacks the simulation of realistic LTE channel standards such as multipath effects and Doppler frequency deviation, and the simulation of actual communication signal transmission and reception conditions. The present invention utilizes LabVIEW software, calls its LabVIEW Communications Suite framework, encodes the communication content, and sends it through an FPGA. The KSW-TECH channel simulator is used to simulate the channel type, multipath information, and Doppler frequency deviation required for non-cooperative detection. From software simulation to hardware simulation, the simulation of multipath effects and Doppler frequency deviation is realized, and the non-cooperative signals in real scenarios are simulated to be close to the real data, realizing the implementation of signal radio frequency transmission.

[0092] 3. Use the Kalman filter to introduce a timing prediction mechanism to achieve cleaning preprocessing of non-cooperative signals, further suppress individual sampling errors in the signal sequence caused by IQ imbalance, propagation multipath effects, etc., and realize the correction of some signal sampling points.

[0093] 4. The constructed scoring function evaluates the signal after passing through the Kalman filter. The degree of modification of the Kalman filter is evaluated based on the score, which is determined by the quality and credibility of the reference signal data set.

[0094] 5. The present invention uses a dataset creation program programmed in Python to segment, reorganize, cut, sample, add preset labels to IQ signals, and then store them as pickle format samples. This achieves the adjustability of data dimensions and adopts a detection and error reporting mechanism to better preserve the true characteristics of the data. It is suitable for large-scale datasets and is more adaptable to the production of datasets with more sampling points.

[0095] 6. The OFDM communication signal dataset generated in this work is an ASCII-formatted OFDM signal with a 16 MHz bandwidth, 300 subcarriers, and a center carrier frequency of 3.18 GHz. It is used to simulate real-world LTE communication signals. Compared to the five datasets mentioned in Table 1, this dataset is an OFDM communication signal with 11 digital modulation types, enabling non-cooperative signal detection in simulated real-world communication environments. Furthermore, compared to the IQ dual-channel signal format saved by the purchased USRP device, which only provides two-dimensional label determination (modulation type and signal-to-noise ratio), the dataset is unprocessed and unsuitable for dataset training standards in the field of artificial intelligence deep learning. The dataset constructed in this invention converts the output IQ file into a three-dimensional label determination (modulation type, transmitter, and signal-to-noise ratio) format in PKL format, which facilitates deep learning training. This invention achieves dataset integration and is more conducive to research in the field of cognitive radio. BRIEF DESCRIPTION OF THE DRAWINGS

[0096] The present invention will be described in further detail below with reference to the accompanying drawings and examples.

[0097] Figure 1 This is a schematic diagram of an OFDM communication signal modulation recognition data set production system based on FPGA and programmable interception provided by an example of the present invention.

[0098] Figure 2 This is a schematic diagram of the cooperative communication transmitter signal situation provided by an example of the present invention.

[0099] Figure 3 This is a schematic diagram of the signal status of the cooperation end receiver provided by an example of the present invention.

[0100] Figure 4 This is a time-frequency conversion diagram of the non-cooperative detection signal effect provided by an example of the present invention.

[0101] Figure 5 This is a schematic diagram of the signal situation of a non-cooperative receiver provided by an example of the present invention.

[0102] Figure 6 This is a schematic diagram of the data set production structure provided by an example of the present invention.

[0103] Figure 7This is a schematic diagram of the interface for implementing data interception and labeling using Python, provided by an example of the present invention.

[0104] Figure 8 It is a graph showing the comparison results of experimental data provided by the examples of the present invention. DETAILED DESCRIPTION

[0105] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0106] like Figure 6 As shown, the embodiment of the present invention provides a method for producing an OFDM communication signal data set based on FPGA and programmable interception, such as Figure 1 As shown, an FPGA device with built-in Laview software is used as a transmitter, a cooperative receiver, and a non-cooperative receiver. The method includes:

[0107] In step 1, the transmitter sends the LTE communication signal into the wireless channel. After simulating the channel type, multipath information, and Doppler frequency deviation required for non-cooperative detection through a channel simulator, the resulting OFDM communication signal is sent to the cooperative receiver and the non-cooperative receiver respectively.

[0108] In the embodiment of the present invention, the FPGA devices of the transmitter and the cooperative end receiver use the National Instrument's NI USRP-2953R signal transmitter, calling the LabVIEW Communications Suite framework in the built-in Laview software, using the DSP module package, and through the NI Labview software + USRP hardware, realize the generation of high-order modulation for the communication system, making it more suitable for modern wireless communication needs; the transmitter is as follows Figure 2 As shown, the cooperative end receiver is as follows Figure 3 As shown in the figure, the constellation diagram, signal reception diagram and spectrum analysis function of the cooperative receiver are used to ensure the successful establishment and effective implementation of the signal link; the FPGA device of the non-cooperative detection machine uses the National Instrument NI PXle-5644 signal receiver, as shown in the figure. Figure 4 and Figure 5 As shown; the channel simulator uses the KSW-TECH channel simulator.

[0109] In step 2, the non-cooperative receiver records the OFDM communication signal in the time domain and stores it as a non-cooperative signal in IQ format; uses a Kalman filter to clean the non-cooperative signal to obtain the target non-cooperative signal; and removes unstable and non-equilibrium signal samples.

[0110] In step three, the cooperative receiver records the OFDM communication signal and stores it as a cooperative signal in IQ format. Based on the amplitude of the cooperative signal, an amplitude threshold is set. The sequences in the target non-cooperative signal whose amplitude meets the amplitude threshold are retained to obtain the intercepted OFDM communication signal. The unstable sequences generated by manual operation of the RF equipment at the start or end stage are removed.

[0111] Step 4: Construct a difference measurement function based on the statistical characteristics of the cooperative signal and the intercepted OFDM communication signal; construct a scoring function through the difference measurement function; calculate the power of the intercepted OFDM communication signal whose score meets the task requirements, and obtain the target OFDM communication signal after generating it from the signal-to-noise ratio and adding Gaussian white noise; the scoring function is used to evaluate the degree of signal processing and cleaning, and provide a basis for whether re-acquisition is needed.

[0112] Step 5: Use the numpy function library in Python to split the target OFDM communication signal into I sequence and Q sequence, reassemble the I sequence and Q sequence in the order of reception, cut them into sub-blocks based on the time interval and sampling rate, sample each sub-block and add preset labels to store them as pickle format samples, and store the pickle format samples with correct storage format as the OFDM communication signal dataset, as shown in the following example: Figure 7 shown.

[0113] In step 1, the channel simulator simulates the channel type, multipath information, and Doppler frequency deviation required for non-cooperative detection in the following manner:

[0114] The channel simulator of the wireless channel uses a preset channel standard to linearly superimpose the channels corresponding to the LTE communication signal. By setting the transmission delay and Doppler frequency deviation values, it simulates the multipath effect and Doppler frequency deviation in the transmission channel of non-cooperative detection in a real environment to obtain the OFDM communication signal.

[0115] In step 2, the Kalman filter is used to clean the non-cooperative signal to obtain the target non-cooperative signal. The method includes:

[0116] S10, predict the error covariance of the non-cooperative signal through the hidden state to obtain the observation matrix of the Kalman filter;

[0117] S11, calculate the gain of the Kalman filter based on the observation matrix and output the predicted state at the current moment;

[0118] S12, constructing an update equation based on the updated observation gradient, updating the observation matrix, and obtaining the target non-cooperative signal after cleaning the non-cooperative signal using a Kalman filter.

[0119] In S10, the method of obtaining the observation matrix of the Kalman filter by predicting the error covariance of the non-cooperative signal through the hidden state is:

[0120] ;

[0121] Where y t is the hidden state, H t The observation matrix is ​​used to correct the measured part. In the initial stage, it is set to a full 0 matrix, indicating that y t The connection between and observation space to minimize the prediction error of the model is the observation matrix that obtains the observation value from the state vector; t is the true value of the Kalman filter at time t, initially set to the input true value; v t is the observation noise, v t Gaussian white noise with covariance Rt, Rt represents the uncertainty of measurement, specifically the observation noise covariance matrix; t is the time variable;

[0122] In S11, the gain of the Kalman filter is calculated based on the observation matrix, and the calculation method for outputting the predicted state at the current moment is:

[0123] ;

[0124] ;

[0125] Where G t is the gain of the Kalman filter; P t is the state covariance matrix, which represents the prediction error, specifically the prediction error covariance matrix; Indicates a value of 10 -6 A constant parameter to prevent the denominator from being 0, T is the transpose; x t+1 is the predicted state at the current moment, that is, the output of the Kalman filter; H t x t is x t The observed value, y t -H t x t is the error between the observed value and the predicted observed value;

[0126] In S12, the update equation is:

[0127] ;

[0128] Where Ht-1 is the observation matrix at time t-1, α is the learning step size, which is used to control the update rate, v t-1 is the observation noise at time t-1, represents the update rate, and is set to 0.005 in the embodiment of the present invention; g(x t-1 ) represents the updated observation gradient, x t-1 is the true value of the Kalman filter at time t-1. This implements signal cleaning preprocessing to reduce multipath effects and unbalanced samples in the sequence.

[0129] In S12, the calculation method for updating the observation gradient is:

[0130] ;in,

[0131] ;

[0132] Where, Represents the observation function, j represents the imaginary part, N represents the total length of the signal, k is the discrete time variable, r(k) is the discrete time signal sequence, r is the discrete signal sequence function expression, k+N represents the offset of the sequence index, and e is a natural constant.

[0133] In step 3, the amplitude is calculated as:

[0134] ;

[0135] Where, is the amplitude, i represents the sequence number of the signal, I is the in-phase component of the signal, and Q is the quadrature component of the signal; the shape of the signal is: [number of samples, u, sample length]; where u = 0 represents the in-phase component I of the signal; u = 1 represents the quadrature component Q of the signal.

[0136] Preferably, set the minimum amplitude threshold min_power_threshold=10 -4 , filter out samples with too small amplitude (probably noise), then set the maximum amplitude threshold max_power_threshold=10.0 to filter out samples with too large amplitude (probably abnormal signals or interference), and exclude samples containing NaN values.

[0137] In step 4, the statistical characteristics include: amplitude mean, amplitude standard deviation, amplitude minimum, amplitude maximum, amplitude skewness and amplitude kurtosis; where,

[0138] The calculation method for the amplitude average is:

[0139] ;

[0140] The calculation method for the amplitude standard deviation is:

[0141] ;

[0142] The calculation method for the minimum amplitude is:

[0143] ;

[0144] The maximum amplitude is calculated as:

[0145] ;

[0146] The calculation method of amplitude skewness is:

[0147] ;

[0148] The calculation method of amplitude kurtosis is:

[0149] ;

[0150] Where, represents the average amplitude of the signal, Indicates the standard deviation of the signal amplitude; Indicates the minimum amplitude, represents the maximum amplitude; Sk represents the amplitude skewness; Kur represents the amplitude kurtosis.

[0151] In actual operation, Figures 2 to 5 Comparing the waveforms, it is obvious that there is a significant difference between the cooperative signal and the intercepted OFDM communication signal. Therefore, a scoring function constructed based on the difference metric function is used to evaluate the effect comparison between the cooperative signal and the intercepted OFDM communication signal. In step 4, the constructed difference metric function is:

[0152] ;

[0153] The score function constructed based on the difference measurement function is:

[0154] ;

[0155] Where, d(f u real , f u ideal ) is the difference metric function value; f u real Represents the statistical characteristics of the intercepted OFDM communication signal, f u ideal is the statistical property of the cooperative signal, It is a positive parameter used to avoid the denominator being zero. Q is the score, u is the weight number of the statistical feature, U represents the total weight of the statistical feature, and w uThis is the weight of the statistical characteristics. This metric can appropriately give larger values ​​to cases with larger deviations. The scoring function takes into account characteristics such as signal amplitude, phase, standard deviation of I and Q channels, maximum and minimum values, skewness, kurtosis, and average power.

[0156] In step 4, the power of the intercepted OFDM communication signal whose score meets the task requirements is calculated. After generating the target OFDM communication signal by the signal-to-noise ratio and adding Gaussian white noise, the method is as follows:

[0157] ;

[0158] ;

[0159] Where, is power, The square of the amplitude of the intercepted OFDM communication signal indicating that the score meets the task requirements; is the signal-to-noise ratio, SNR=[20, 18, 16, 14, 12, 10, 8, 6, 4, 2, 0, -2, -4, -6, -8,-10]; The noise power is set to generate Gaussian white noise, which is then added to the intercepted OFDM communication signal whose score meets the task requirements to obtain the target OFDM communication signal.

[0160] In step 5, the target OFDM communication signal is divided into I sequence and Q sequence by using the numpy function library in Python. The I sequence and Q sequence are reassembled in the order of reception, and then cut into sub-blocks based on the time interval and sampling rate. After sampling and adding preset labels to each sub-block, the samples are stored as pickle format samples. The method for storing the pickle format samples with correct storage format as the OFDM communication signal dataset is as follows:

[0161] S20, construct the stable region [s, b] of the target OFDM communication signal:

[0162] [s, b] = find_stable_region ,

[0163] Where s is the starting index of the stable region, b is the ending index of the stable region, find_stable_region is the reading function in the numpy function library; D is the target OFDM communication signal, represents the sampling rate, The number of seconds to skip at the beginning, The number of seconds to skip at the end is used to remove unstable sequences. For example, skip 0.2 seconds at the beginning and 0.2 seconds at the end to skip unstable time periods at the beginning and end of the data and remove unstable sequences.

[0164] S21, dividing the target OFDM communication signal in the stable region into an I sequence and a Q sequence: recombining the I sequence and the Q sequence into recombined samples in a receiving order, each recombined sample including a real part and an imaginary part;

[0165]

[0166]

[0167] Where, I[n] is the I sequence, which is the imaginary data of the recombined sample, Q[n] is the Q sequence, which is the real data of the recombined sample; D[2n + s] is the even position data in the stable region, D[2n + s + 1] is the odd position data in the stable region; n is the sampling sequence number;

[0168] S22, based on time interval The recombined samples are cut into sub-blocks by the sampling rate, where the sequence length L of the sub-block is: ; The number of sub-blocks B is: ; The sub-block range is: , is the sub-block number, ;

[0169] S23, after calculating the number of target samples to be skipped, random sampling or sequential sampling is used to extract a specified number of target samples in each sub-block. , , is the total length of the sub-block; the calculation method for the target number of samples to be skipped is:

[0170] ;

[0171] ;

[0172] Where, Indicates the target number of samples to skip at the beginning, Indicates the target number of samples to skip at the end;

[0173] Random sampling location for:

[0174]

[0175] Sequential sampling positions for:

[0176]

[0177] Where, is the sampling point position of each sub-block, Uniform is the function for generating uniformly distributed random numbers, p k ~Uniform means Obey uniformly distributed random numbers, is the sampling step size; S24, uses integer division to calculate the maximum number of target samples to be collected, specifically:

[0178] max_samples = real_data_length / / length; Based on the real data length real_data_length (the imaginary data length is equal to it) within the determined stable interval and the length of each sample length, integer division ( / / ) is used to calculate the maximum number of target samples that can be extracted, max_samples.

[0179] Correct the specified number based on the maximum number of samples to obtain the target number. That is, if the specified number is greater than the maximum number of samples, use an if statement to adjust the sample number parameter sample_num and automatically reduce the sample size to max_samples to avoid unreasonable sample size requests.

[0180] After ensuring the consistency of the lengths of the real and imaginary data in the target number of target samples collected through consistency checking, the samples are stored as original files in IQ format;

[0181] S25, use the save_pickle function to add preset tags to the original sample file in IQ format and save it in pickle format; the specific steps are as follows:

[0182] After adding preset labels to the original sample files in IQ format using the file.split function combined with the for I in range loop, the os.path.join function is used to combine the directory path and file name specified by the pkl_dir parameter into a complete file path for easy storage. Finally, the pickle.dump function is used to serialize the IQ samples and save them in pickle format as a dictionary by saving the labels to the specified path. At the same time, the with statement is used to ensure that the file is properly closed after use in the correct path. Even if an exception occurs during the writing process, the file is ensured to be properly closed to avoid resource leaks.

[0183] S26, using the pickle.load function to read the labels and target numbers of the sample original file in pickle format, using the for 1 in range statement to print out the storage format of the sample original file in pickle format, and compare it with the test format, and use the sample original file in pickle format that passes the comparison as the OFDM communication signal dataset.

[0184] In order to explain the technical solution of the present invention in detail, a specific example is now provided:

[0185] By controlling the same recording time and the same number of sampling points, the network is trained using the classic CNN+SE attention neural network recognition network to verify the effectiveness of the dataset. The verification results are as follows: Figure 8 and as shown in Table 2.

[0186] Table 2

[0187] Preparation method Number of single label samples Number of sampling points Dataset file size Random intercept 6000 1024 6.1G Random interception + amplitude detection 4000 1024 4.2G Random interception + amplitude detection + power spectrum plus noise 3000 1024 3.8G Random interception + amplitude detection + power spectrum plus noise + scoring evaluation 2000 1024 3.3G

[0188] Table 2 shows the results of improving the network's recognition rate for the dataset by increasing the number of data samples, so that the difference in recognition accuracy between them is around 0.5%. The effectiveness of the dataset is evaluated by comparing the number of single label samples and the dataset file size of different datasets.

[0189] Figure 8 In the test, by ensuring that the classifier for the same recognition is the CNN-SC attention mechanism, the difference in recognition rate is guaranteed to be around 0.5%, and the quality of the signal data set is measured by the number of samples; among them, random interception: 71.65%; random interception + amplitude detection: 71.07%; random interception + amplitude detection + power noise: 71.42%; random interception + amplitude detection + power noise + score evaluation: 71.87%. Figure 8 It can be seen that the present invention can effectively improve the quality of the dataset and achieve a similar recognition rate with fewer samples.

[0190] The beneficial effects of the embodiments of the present invention are:

[0191] 1. Compared with the single-carrier signals in the public dataset, there is a lack of simulation for the widely used OFDM communication system in modern times. At the same time, the simulated low-order digital modulation cannot meet the actual needs of modern communication for low latency and high speed. The present invention uses the LabVIEW Communications Suite framework in the built-in Laview software through FPGA devices to realize the generation of high-order modulated LTE communication signals in the communication system, making it more suitable for modern wireless communication needs.

[0192] 2. Regarding the traditional data set preparation method based on GNU-Radio, the gradient descent simulation of the signal-to-noise ratio is performed by adding Gaussian white noise through Matlab, which lacks the simulation of realistic LTE channel standards such as multipath effects and Doppler frequency deviation, and the simulation of actual communication signal transmission and reception conditions. The present invention utilizes LabVIEW software, calls its LabVIEW Communications Suite framework, encodes the communication content, and sends it through an FPGA. The KSW-TECH channel simulator is used to simulate the channel type, multipath information, and Doppler frequency deviation required for non-cooperative detection. From software simulation to hardware simulation, the simulation of multipath effects and Doppler frequency deviation is realized, and the non-cooperative signals in real scenarios are simulated to be close to the real data, realizing the implementation of signal radio frequency transmission.

[0193] 3. Use the Kalman filter to introduce a timing prediction mechanism to achieve cleaning preprocessing of non-cooperative signals, further suppress individual sampling errors in the signal sequence caused by IQ imbalance, propagation multipath effects, etc., and realize the correction of some signal sampling points.

[0194] 4. The constructed scoring function evaluates the signal after passing through the Kalman filter. The degree of modification of the Kalman filter is evaluated based on the score, which is determined by the quality and credibility of the reference signal data set.

[0195] 5. The present invention uses a dataset creation program programmed in Python to segment, reorganize, cut, sample, add preset labels to IQ signals, and then store them as pickle format samples. This achieves the adjustability of data dimensions and adopts a detection and error reporting mechanism to better preserve the true characteristics of the data. It is suitable for large-scale datasets and is more adaptable to the production of datasets with more sampling points.

[0196] 6. The OFDM communication signal dataset generated in this work is an ASCII-formatted OFDM signal with a 16 MHz bandwidth, 300 subcarriers, and a center carrier frequency of 3.18 GHz. It is used to simulate real-world LTE communication signals. Compared to the five datasets mentioned in Table 1, this dataset is an OFDM communication signal with 11 digital modulation types, enabling non-cooperative signal detection in simulated real-world communication environments. Furthermore, compared to the IQ dual-channel signal format saved by the purchased USRP device, which only provides two-dimensional label determination (modulation type and signal-to-noise ratio), the dataset is unprocessed and unsuitable for dataset training standards in the field of artificial intelligence deep learning. The dataset constructed in this invention converts the output IQ file into a three-dimensional label determination (modulation type, transmitter, and signal-to-noise ratio) format in PKL format, which facilitates deep learning training. This invention achieves dataset integration and is more conducive to research in the field of cognitive radio.

[0197] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A method for producing an OFDM communication signal data set based on FPGA and programmable interception, characterized in that: An FPGA device with built-in Laview software is used as a transmitter, a cooperative receiver, and a non-cooperative receiver. The method includes: In step 1, the transmitter sends the LTE communication signal into the wireless channel. After simulating the channel type, multipath information, and Doppler frequency deviation required for non-cooperative detection through a channel simulator, the resulting OFDM communication signal is sent to the cooperative receiver and the non-cooperative receiver respectively. Step 2: The non-cooperative receiver records the OFDM communication signal in the time domain and stores it as a non-cooperative signal in IQ format; a Kalman filter is used to clean the non-cooperative signal to obtain the target non-cooperative signal; Step 3: The cooperative end receiver records the OFDM communication signal and stores it as a cooperative signal in IQ format; an amplitude threshold is set based on the amplitude of the cooperative signal; and a sequence of the target non-cooperative signal whose amplitude meets the amplitude threshold is retained to obtain the intercepted OFDM communication signal; Step 4: construct a difference metric function based on the statistical characteristics of the cooperative signal and the intercepted OFDM communication signal; construct a score function through the difference metric function; calculate the power of the intercepted OFDM communication signal whose score meets the task requirements, generate it from the signal-to-noise ratio and add Gaussian white noise to obtain the target OFDM communication signal; Step 5: Use the numpy function library in Python to split the target OFDM communication signal into I sequence and Q sequence, reassemble the I sequence and Q sequence in the order of reception, cut them into sub-blocks based on the time interval and sampling rate, sample each sub-block, add preset labels, and store them as pickle format samples. The pickle format samples with correct storage format are stored as the OFDM communication signal dataset.

2. The method according to claim 1, wherein In step 1, the channel simulator simulates the channel type, multipath information, and Doppler frequency deviation required for non-cooperative detection in the following manner: The channel simulator of the wireless channel uses a preset channel standard to linearly superimpose the channels corresponding to the LTE communication signal. By setting the transmission delay and Doppler frequency deviation values, it simulates the multipath effect and Doppler frequency deviation in the transmission channel of non-cooperative detection in a real environment to obtain the OFDM communication signal.

3. The method according to claim 2, wherein In step 2, the Kalman filter is used to clean the non-cooperative signal to obtain the target non-cooperative signal. The method includes: S10, predict the error covariance of the non-cooperative signal through the hidden state to obtain the observation matrix of the Kalman filter; S11, calculate the gain of the Kalman filter based on the observation matrix and output the predicted state at the current moment; S12, constructing an update equation based on the updated observation gradient, updating the observation matrix, and obtaining the target non-cooperative signal after cleaning the non-cooperative signal using a Kalman filter.

4. The method according to claim 3, wherein In S10, the method of obtaining the observation matrix of the Kalman filter by predicting the error covariance of the non-cooperative signal through the hidden state is: ； Where y t is the hidden state, H t is the observation matrix; x t is the true value of the Kalman filter at time t; v t is the observation noise, v t Gaussian white noise with covariance Rt; t is the time variable; In S11, the gain of the Kalman filter is calculated based on the observation matrix, and the calculation method for outputting the predicted state at the current moment is: ； ； Where G t is the gain of the Kalman filter; P t is the state covariance matrix; Indicates a value of 10 -6 The constant parameter, T is the transpose; x t+1 is the predicted state at the current moment, that is, the output of the Kalman filter; H t x t is x t The observed value, y t -H t x t is the error between the observed value and the predicted observed value; In S12, the update equation is: ； Where H t-1 is the observation matrix at time t-1, α is the learning step size, which is used to control the update rate, v t-1 is the observation noise at time t-1, which represents the update rate; g(x t-1 ) represents the updated observation gradient, x t-1 is the true value of the Kalman filter at time t-1.

5. The method according to claim 4, wherein In S12, the calculation method for updating the observation gradient is: ;in, ； Where, Represents the observation function, j represents the imaginary part, N represents the total length of the signal, k is the discrete time variable, r(k) is the discrete time signal sequence, r is the discrete signal sequence function expression, k+N represents the offset of the sequence index, and e is a natural constant.

6. The method according to claim 5, wherein In step 3, the amplitude is calculated as: ； Where, is the amplitude, i represents the sequence number of the signal, I is the in-phase component of the signal, and Q is the quadrature component of the signal; the shape of the signal is: [number of samples, u, sample length]; where u = 0 represents the in-phase component I of the signal; u = 1 represents the quadrature component Q of the signal.

7. The method according to claim 6, wherein In step 4, the statistical characteristics include: amplitude mean, amplitude standard deviation, amplitude minimum, amplitude maximum, amplitude skewness and amplitude kurtosis; where, The calculation method for the amplitude average is: ； The calculation method for the amplitude standard deviation is: ； The calculation method for the minimum amplitude is: ； The maximum amplitude is calculated as: ； The calculation method of amplitude skewness is: ； The calculation method of amplitude kurtosis is: ； Where, represents the average amplitude of the signal, Indicates the standard deviation of the signal amplitude; Indicates the minimum amplitude, represents the maximum amplitude; Sk represents the amplitude skewness; Kur represents the amplitude kurtosis.

8. The method according to claim 7, wherein In step 4, the difference measurement function is: ； The score function constructed based on the difference measurement function is: ； Where, d(f u real , f u ideal ) is the difference metric function value; f u real Represents the statistical characteristics of the intercepted OFDM communication signal, f u ideal is the statistical property of the cooperative signal, is a positive parameter; Q is the score, u is the weight number of the statistical characteristic, U represents the total weight of the statistical characteristic, w u is the weight of the statistical feature.

9. The method according to claim 8, wherein In step 4, the power of the intercepted OFDM communication signal whose score meets the task requirements is calculated. After generating the target OFDM communication signal by the signal-to-noise ratio and adding Gaussian white noise, the method is as follows: ； ； Where, is power, The square of the amplitude of the intercepted OFDM communication signal indicating that the score meets the task requirements; is the signal-to-noise ratio, SNR=[20, 18, 16, 14, 12, 10, 8, 6, 4, 2, 0, -2, -4, -6, -8, -10]; The noise power is set to generate Gaussian white noise, which is then added to the intercepted OFDM communication signal whose score meets the task requirements to obtain the target OFDM communication signal.

10. The method according to claim 9, wherein In step 5, the target OFDM communication signal is divided into I sequence and Q sequence by using the numpy function library in Python. The I sequence and Q sequence are reassembled in the order of reception, and then cut into sub-blocks based on the time interval and sampling rate. After sampling and adding preset labels to each sub-block, the samples are stored as pickle format samples. The method for storing the pickle format samples with correct storage format as the OFDM communication signal dataset is as follows: S20, construct the stable region [s, b] of the target OFDM communication signal: [s, b] = find_stable_region , Where s is the starting index of the stable region, b is the ending index of the stable region, find_stable_region is the reading function in the numpy function library; D is the target OFDM communication signal, represents the sampling rate, The number of seconds to skip at the beginning, The number of seconds to skip at the end to remove unstable sequences; S21, dividing the target OFDM communication signal in the stable region into an I sequence and a Q sequence: recombining the I sequence and the Q sequence into recombined samples in a receiving order, each recombined sample including a real part and an imaginary part; Where, I[n] is the I sequence, which is the imaginary data of the recombined sample, Q[n] is the Q sequence, which is the real data of the recombined sample; D[2n + s] is the even position data in the stable region, D[2n + s + 1] is the odd position data in the stable region; n is the sampling sequence number; S22, based on time interval The recombined samples are cut into sub-blocks by the sampling rate, where the sequence length L of the sub-block is: ; The number of sub-blocks B is: ; The sub-block range is: , is the sub-block number, ; S23, after calculating the number of target samples to be skipped, random sampling or sequential sampling is used to extract a specified number of target samples in each sub-block. , , is the total length of the sub-block; the calculation method for the target number of samples to be skipped is: ； ； Where, Indicates the target number of samples to skip at the beginning, Indicates the target number of samples to skip at the end; Random sampling location for: Sequential sampling positions for: Where, is the sampling point position of each sub-block, Uniform is a function for generating uniformly distributed random numbers, ~Uniform means Obey uniformly distributed random numbers, is the sampling step length; S24, using integer division to calculate the maximum number of target samples to be collected, and correcting the specified number based on the maximum number of samples to obtain the target number; performing a consistency check to ensure that the lengths of the real data and the imaginary data in the target samples collected are consistent, and then storing the samples in an IQ format original file; S25, use the save_pickle function to add preset labels to the original sample file in IQ format and save it in pickle format; S26, using the pickle.load function to read the labels and target numbers of the sample original file in pickle format, using the for 1 in range statement to print out the storage format of the sample original file in pickle format, and compare it with the test format, and use the sample original file in pickle format that passes the comparison as the OFDM communication signal dataset.

Images