Radio frequency fingerprint extraction method for TDMA interphone
By obtaining the IQ data of the TDMA intercom, calculating the RSSI, window-type intercepting the power waveform diagram and cleaning the data, combined with the deep learning network model, the problems of low accuracy of RF fingerprint recognition and high algorithm complexity in the TDMA duplex mode intercom are solved. A higher recognition accuracy and convenient feature extraction process are achieved.
Patent Information
- Application Number
- CN202410854344.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-28
- Publication Date
- 2024-11-08
AI Technical Summary
The existing RF fingerprint recognition technology has low recognition accuracy and high algorithm complexity in TDMA duplex intercom, making it difficult to achieve effective feature extraction and recognition.
By obtaining the IQ data of the TDMA intercom, calculating the RSSI of the received signal, window-type intercepting the power waveform diagram of the steady-state segment, cleaning the data, and using a deep learning network model for feature extraction and recognition.
It significantly improves the recognition accuracy, reduces the need for synchronous work, makes the RF fingerprint feature extraction process more convenient, and increases the number of samples used by deep learning network models, improving the training accuracy.
Smart Images

Figure CN118916779A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of wireless communication and signal processing, in particular to a radio frequency fingerprint extraction method for a TDMA intercom. Background Art
[0002] Radio Frequency Fingerprinting (RFF) uses the unique properties of the radio signal generated by the hardware differences of the radiator to achieve individual identification. These hardware differences will not affect the communication or detection function of the radiator, but will leave a unique "fingerprint" on the radio signal. Currently, TDMA duplex mode walkie-talkies can use RF fingerprinting technology for feature extraction and identification.
[0003] Existing RF fingerprint recognition technologies are mainly divided into three categories: 1) IQ data simple preprocessing + complex preprocessing + expert feature transformation + network model structure; 2) IQ data simple preprocessing + complex preprocessing + network model structure; 3) IQ data simple preprocessing + network model structure. Although these methods are theoretically applicable to TDMA duplex mode walkie-talkies, they have some obvious shortcomings because they are not specially designed for them.
[0004] First, although these methods are theoretically applicable to TDMA duplex mode walkie-talkies, they are not specially designed for them, so the recognition accuracy is low and it is difficult to achieve the expected effect. Secondly, the existing methods have high algorithm complexity and high computational cost, which makes them difficult to operate in practical applications.
[0005] Therefore, it is urgent to develop a radio frequency fingerprint extraction method specifically for TDMA duplex mode walkie-talkies to overcome the shortcomings of the existing technology. Summary of the invention
[0006] In view of the shortcomings of the prior art, the present invention provides a radio frequency fingerprint extraction method for TDMA walkie-talkies, which solves the problems of low recognition accuracy and high algorithm complexity of the existing radio frequency fingerprint recognition method for TDMA duplex mode walkie-talkies.
[0007] To achieve the above objectives, the present invention is implemented by the following technical scheme: a method for extracting radio frequency fingerprints for a TDMA walkie-talkie, comprising the following steps:
[0008] S1. Get the IQ data of the TDMA intercom;
[0009] S2, calculate the RSSI of the received signal;
[0010] S3, window-type interception of the power waveform of the steady-state section;
[0011] S4, performing data cleaning on the intercepted waveform graph;
[0012] S5. Use deep learning network model for feature extraction and recognition.
[0013] Preferably, the step S1 specifically includes the following steps:
[0014] S1.1. Record the TDMA intercom signal as a wav file;
[0015] S1.2. Use Matlab program to read IQ data in wav format;
[0016] S1.3. Save the read IQ data as a two-dimensional array.
[0017] Preferably, the calculation formula of RSSI in step S2 is:
[0018] RSSI = I 2 +Q 2 , where I and Q represent the real and imaginary parts of the IQ data, respectively.
[0019] Preferably, the step S3 specifically includes the following steps:
[0020] S3.1, set the threshold of RSS I;
[0021] S3.2, intercepting the waveform diagram of the part where the RSS I value is greater than the threshold;
[0022] S3.3. Save the captured waveform.
[0023] Preferably, the step S4 specifically includes the following steps:
[0024] S4.1, calculating the arithmetic mean of the RGB value of each point in the waveform;
[0025] S4.2. Delete the waveforms whose RGB average values are greater than 253.
[0026] Preferably, the deep learning network model is trained and classified by a Lenet network model. Preferably, the specific implementation of the window-type intercepted power waveform diagram includes the following pseudo code:
[0027]
[0028]
[0029] Preferably, the specific implementation of the data cleaning includes the following pseudo code:
[0030]
[0031] The present invention provides a method for extracting radio frequency fingerprints for TDMA intercoms. It has the following beneficial effects:
[0032] 1. The present invention recorded the signals of 4 TDMA duplex mode walkie-talkies of two models by the method of the present invention, intercepted the power waveform by window, and conducted training experiments using the Lenet network model. The recognition accuracy was 84% before cleaning, and the accuracy increased to 86% after cleaning, showing the effectiveness of the data cleaning step.
[0033] 2. The present invention effectively extracts the radio frequency fingerprint features of TDMA duplex mode walkie-talkies by making power time domain graphs as training samples for the neural network. Experiments have shown that the accuracy of the power time domain graphs intercepted by the method of the present invention is as high as 86%, which significantly improves the accuracy of recognition.
[0034] 3. The present invention uses the steady-state segment of the signal to extract RF fingerprint features, which reduces the need for synchronous work and makes the RF fingerprint feature extraction process more convenient. In addition, through the method of windowed interception of the power time domain diagram, within a certain signal duration, the number of samples used by the deep learning network model can be greatly increased, thereby improving the accuracy of training. BRIEF DESCRIPTION OF THE DRAWINGS
[0035] Figure 1 is a flow chart of the present invention;
[0036] Figure 2 A schematic diagram of the identification method process of the present invention;
[0037] Figure 3 Schematic diagram of the full signal link of the present invention;
[0038] Figure 4 It is a schematic diagram of the signal time period of the present invention;
[0039] Figure 5 It is a time domain diagram of the signal power of a TDMA duplex mode intercom of the present invention;
[0040] Figure 6 This is a single power waveform diagram of a TDMA duplex mode intercom of the present invention;
[0041] Figure 7 This is a single power waveform diagram of another TDMA duplex mode intercom of the present invention;
[0042] Figure 8 Schematic diagram of the irregular square wave of the present invention. DETAILED DESCRIPTION
[0043] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments 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 creative work are within the scope of protection of the present invention.
[0044] Example:
[0045] Please see attached Figure 1 -Attached Figure 2 and attached Figure 8 The embodiment of the present invention provides a method for extracting radio frequency fingerprints for a TDMA intercom, comprising the following steps:
[0046] S1. Get the IQ data of the TDMA intercom;
[0047] S2. Calculate the RSS I of the received signal;
[0048] S3, window-type interception of the power waveform of the steady-state section;
[0049] S4, performing data cleaning on the intercepted waveform graph;
[0050] S5. Use deep learning network model for feature extraction and recognition.
[0051] Specifically, the present invention uses RSS I to implement the radio frequency fingerprint recognition of TDMA duplex mode walkie-talkies, which mainly includes the main processes of reading IQ data, window-type interception of power waveforms, data cleaning, model extraction of radio frequency fingerprint features, and identification / classification. Currently, radio frequency fingerprint recognition includes two major processes: data preprocessing and classification or recognition through network models. The data preprocessing mainly includes three stages: reading IQ data, window-type interception of power waveforms, and data cleaning. The specific implementation steps are as follows:
[0052] 1) Read IQ data:
[0053] The method of the present invention uses IQ data in wav format, which can be read out by Matlab program. Each piece of IQ data read out by wav is saved as a two-dimensional array.
[0054] 2) Window-type interception of power waveform:
[0055] pass Figure 5It can be seen that after each WAV format IQ data signal enters the steady-state segment, the RSSI calculated by the formula presents a square wave-like shape. The RSSI in the low power value is close to 0 (most of the values are 0, but not all are 0). By intercepting the signal in the steady-state segment and setting a certain threshold (such as RSSI>0.001), the part above the threshold can be extracted.
[0056] The experiment of Experiment 3 proves that all square waves can save the time domain diagram of success rate by this method.
[0057] The method of the present invention intercepts the power time domain graph of IQ data in a windowed manner, and the number of saved graphs is related to the length of the duplex mode time. However, under certain signal duration, the number of samples that can be used by the subsequent deep learning network model can be greatly increased, thereby improving the training accuracy.
[0058] 3) Data cleaning:
[0059] In actual experiments, it is found that the RSSI waveform calculated from each IQ data is not strictly in the form of a square wave, and many irregular square waves are randomly generated in the middle. Figure 8 Shown
[0060] At the same time, there are also problems with the selection of thresholds, because it cannot be guaranteed that the RSSI value will be absolutely lower than the threshold during the trough period. However, raising the threshold will result in incomplete square waves and may lose many features. Therefore, under this reason, a power time domain graph with very few points and almost blank will be generated. This is the defect of simply taking square waves according to the threshold. At the same time, experiments have shown that retaining these blank graphs and irregular square wave graphs will lead to a decrease in recognition accuracy. Therefore, data cleaning is required.
[0061] Data cleaning is to create a threshold according to the number of non-white points in an RSSI, and the power time domain graphs below a certain number are deleted. The calculation method is to calculate the arithmetic mean of the RGB values of the entire image respectively. If the average values of R, G, and B are all greater than 253, the image is deleted.
[0062] The threshold value of the specific average value can be adjusted according to the shape of the irregular square wave. If there are more points, the threshold can be appropriately reduced.
[0063] 4) Feature extraction and classification effects.
[0064] The S1 step specifically includes the following steps:
[0065] S1.1. Record the TDMA intercom signal as a wav file;
[0066] S1.2. Use Matlab program to read IQ data in wav format;
[0067] S1.3. Save the read IQ data as a two-dimensional array.
[0068] The calculation formula of RSS I in step S2 is:
[0069] RSSI = I 2 +Q 2 , where I and Q represent the real and imaginary parts of the IQ data, respectively.
[0070] The S3 step specifically includes the following steps:
[0071] S3.1, set the threshold of RSS I;
[0072] S3.2, intercepting the waveform diagram of the part where the RSS I value is greater than the threshold;
[0073] S3.3. Save the captured waveform.
[0074] The specific implementation of the window-type intercepted power waveform includes the following pseudo code:
[0075]
[0076]
[0077] The S4 step specifically includes the following steps:
[0078] S4.1, calculating the arithmetic mean of the RGB value of each point in the waveform;
[0079] S4.2. Delete the waveforms whose RGB average values are greater than 253.
[0080] The specific implementation of data cleaning includes the following pseudo code:
[0081]
[0082] The deep learning network model is trained and classified through the Lenet network model.
[0083] Experiment 1: Please see the attached Figure 3 -Attached Figure 7
[0084] Experimental design: Explore the application of RF fingerprints and extract RF fingerprint features through the power time domain diagram of the signal
[0085] Purpose:
[0086] 1. Discuss the application scenarios and actual effects of RF fingerprint in TDMA duplex mode walkie-talkies.
[0087] 2. Verify the effectiveness of extracting RF fingerprint features through the power time domain diagram of the signal.
[0088] Experimental content:
[0089] 1. Discuss the application scenarios and actual effects of RF fingerprint in TDMA duplex mode walkie-talkies:
[0090] By decomposing the transmission link of the TDMA duplex mode walkie-talkie signal, we know that the signal data we can get comes from the receiver. The signal received by the receiver can be written as follows, see formula 1:
[0091] y(t)=G(h(t)*F k (x(t)))+n(t), where h(t) represents the influence of the channel on the signal. The signal is convolved with the transmitted signal through the channel and the receiver outputs y(t). The noise caused by hardware or other reasons is represented by n(t).
[0092] from Figure 3 From the signal chain dimension shown, the reasons for the RF fingerprint of TDMA duplex mode walkie-talkies are: carrier frequency offset (CFO) caused by the crystal oscillator (ωtx, Φtx(t)); I, Q differences (gItx≠gQtx) and phase differences (θtx) caused by the mixer; differences caused by DAC and filters; temperature drift differences; nonlinear factors of the power amplifier; different antennas, etc.
[0093] From the perspective of time, the RF fingerprint characteristics of the signal exist in the instantaneous period of the signal, that is, the time period when the signal is turned on or the system is adjusted, and also in the steady-state period of the signal, that is, the time period when the signal sends content. Figure 4 shown.
[0094] Exploration results: Through the above experimental process, it can be seen that the present invention uses the steady-state segment of the signal, which can greatly reduce the synchronization work and make the RF fingerprint feature extraction more convenient.
[0095] 2. Verify the effectiveness of extracting RF fingerprint features through the power time domain diagram of the signal:
[0096] At present, most receivers process signals using IQ data. The original IQ signal we obtain contains almost all the RF fingerprint features. IQ data is time domain data with the time domain characteristics of the signal. IQ data can be used to calculate the power time domain strength of the signal, that is, RSSI. The calculation method is shown in Formula 2:
[0097] RSSI = I 2 +Q 2 , where I and Q represent the real and imaginary parts of the IQ data, respectively.
[0098] Use Matlab to draw it as follows Figure 5,like Figure 5 As shown in the figure, after the signal enters the steady-state section, the power waveform presents a square wave-like shape, which is a more obvious feature of the TDMA duplex mode. However, if a single square wave is intercepted, the signal power in this working section is not stable, and its difference has a certain regularity.
[0099] Exploration results: Figure 6 , 7 As shown in the figure, the comparison shows that there is a clear difference between the waveforms of the two walkie-talkies at the beginning of the single power waveform. Figure 6 , 7 The transmission contents of the two walkie-talkies are the same, and this difference has nothing to do with the signal content; at the same time, the receivers are the same, and the transmission and reception environments are the same, so this difference can be considered as a radio frequency fingerprint feature. And this difference is not just the one in the above two figures, there are many similar differences. Therefore, we can extract the radio frequency fingerprint feature through the power time domain diagram of the signal.
[0100] In summary:
[0101] 1. Generation and analysis of RF fingerprint features:
[0102] The RF fingerprint characteristics of TDMA duplex mode walkie-talkies are generated by the combined effects of multiple hardware and signal transmission link factors.
[0103] The steady-state segment of the signal provides a stable and simplified feature extraction window, helping to reduce the complexity of the synchronization task.
[0104] 2. Application of power time domain diagram in feature extraction:
[0105] The RF fingerprint features can be effectively extracted through the power time domain diagram. The power waveforms of different walkie-talkies show significant differences in specific sections. These differences can be used as RF fingerprint features for identification.
[0106] The experiment proves the effectiveness of extracting RF fingerprint features through power time domain diagram, which provides a reliable method for individual identification of TDMA duplex mode walkie-talkies.
[0107] Experiment 2:
[0108] Experimental design: Research on RF fingerprint extraction method based on TDMA walkie-talkie
[0109] Purpose:
[0110] Verify the RF fingerprint extraction method based on TDMA walkie-talkie, analyze the process of data preprocessing, feature extraction, data cleaning and model training through experiments, and evaluate its recognition and classification accuracy.
[0111] Experimental Materials:
[0112] equipment:
[0113] 4 different models of TDMA duplex mode walkie-talkies
[0114] Signal Receiver
[0115] Computer with Matlab and neural network training software (such as TensorFlow or PyTorch) installed
[0116] software:
[0117] Mat l ab
[0118] TensorFlow or PyTorch
[0119] Experimental steps:
[0120] 1. Data collection:
[0121] Use a signal receiver to receive signals from 4 TDMA walkie-talkies of different models and save them as IQ data files in wav format.
[0122] 2. Data preprocessing:
[0123] Read IQ data: Use Matlab to read the IQ data in each wav file and save it in a two-dimensional array format.
[0124] Calculate RSS I: For the read IQ data, use the formula RSSI = I 2 +Q 2 Calculate the signal power at each sampling point.
[0125] Windowed power waveform capture: By setting a threshold (such as RSSI>0.001), the signal in the steady-state segment is extracted and the power waveform is saved.
[0126] The specific implementation of the window-type intercepted power waveform includes the following pseudo code:
[0127]
[0128]
[0129] 3. Data cleaning:
[0130] Perform data cleaning on the generated power time domain graph and delete irregular square waves and blank graphs.
[0131] The specific implementation of data cleaning includes the following pseudo code:
[0132]
[0133] 4. Feature extraction and model training:
[0134] Feature extraction: Use a deep learning network (such as LeNet) to extract and train features on the cleaned power time domain graph.
[0135] Loading the image dataset
[0136] Build and train the model: Use a deep learning model (such as LeNet) for training to extract RF fingerprint features.
[0137] Experimental process:
[0138] Step 1: Data collection and preprocessing
[0139] Each walkie-talkie sends a signal with the same content and obtains IQ data through the receiver.
[0140] The data were collected 10 times to ensure the adequacy and diversity of the data.
[0141] Step 2: RSS I calculation and window interception
[0142] Calculate RSSI for each segment of IQ data, extract the steady-state segment and save the power waveform.
[0143] Each captured square wave power graph is saved separately for later use.
[0144] Step 3: Data cleaning
[0145] Check all saved power time domain graphs and delete irregular waveforms and blank graphs to ensure image quality for training.
[0146] The power time domain diagram after cleaning should clearly show the signal characteristics.
[0147] Step 4: Model training and testing
[0148] The cleaned power time domain graph is used as training data and input into the deep learning model for training.
[0149] After training is complete, the model is tested using unseen data to evaluate its recognition accuracy.
[0150] Step 5: Result Analysis
[0151] Compare the recognition accuracy of different models (logistic regression, decision tree, support vector machine, neural network).
[0152] Analyze the changes in model recognition accuracy before and after cleaning and evaluate the importance of data cleaning.
[0153] Experimental results and analysis:
[0154] Table 1: Comparison of accuracy
[0155]
[0156] Comparison of accuracy:
[0157] Before cleaning: The LeNet network model was used for training, and the recognition accuracy was 84%.
[0158] After cleaning: The recognition accuracy rate increased to 86%, proving the positive impact of data cleaning on model training.
[0159] Table 2: Model comparison
[0160]
[0161] Model comparison:
[0162] Using the spectrogram for training, the accuracy was 6%.
[0163] Using time-frequency graphs for training, the accuracy rate was 31%.
[0164] The power time domain diagram intercepted by the method of this patent was used for training, with an accuracy rate of 86%, which greatly improved the recognition accuracy.
[0165] Table 3: Characteristic stability
[0166]
[0167] Feature stability:
[0168] The differences in the power waveforms of different walkie-talkies in the steady-state section were analyzed to verify the uniqueness and stability of the RF fingerprint features.
[0169] Experimental conclusion:
[0170] The TDMA walkie-talkie radio frequency fingerprint extraction method based on RSS I can effectively improve the accuracy of recognition and classification.
[0171] The data cleaning process significantly improved the training effect and accuracy of the model.
[0172] Note:
[0173] Ensure the consistency of the experimental environment, including factors such as signal receiver and environmental noise.
[0174] The experiment was repeated several times to verify the stability and reliability of the results.
[0175] Experiment 3:
[0176] Experimental design: Windowed power waveform capture to extract RF fingerprint features
[0177] Purpose:
[0178] By intercepting the steady-state segment of the TDMA duplex mode walkie-talkie signal and setting a certain threshold, the power waveform (RSS I) is extracted to verify the effectiveness of this method in extracting RF fingerprint features.
[0179] Experimental Materials:
[0180] Multiple TDMA duplex mode walkie-talkies (at least two)
[0181] RF signal receiver
[0182] Matlab software
[0183] Data processing and analysis computers
[0184] Experimental steps:
[0185] 1. Preparation
[0186] Equipment preparation:
[0187] Prepare two or more TDMA duplex mode walkie-talkies and ensure they are in normal working condition.
[0188] Configure the RF signal receiver to ensure that it can correctly receive and record the intercom signal.
[0189] Environment setup:
[0190] Choose an experimental environment with less interference and minimize interference from external radio signals.
[0191] Ensure that the temperature, humidity and other conditions in the experimental environment are relatively stable.
[0192] 2. Data Collection
[0193] Signal transmission:
[0194] Use each radio to send the same signal under the same conditions.
[0195] The receiver receives and records the IQ data of each radio.
[0196] Data Records:
[0197] Import the received IQ data into the computer to ensure that the data is complete and correct.
[0198] 3. Data Processing and Analysis
[0199] Signal preprocessing:
[0200] The received IQ data is normalized and denoised to ensure data quality.
[0201] Matlab is used to process the data and calculate the signal power time domain intensity (RSSI). The calculation formula is: RSS I = I 2 +Q 2 , where I and Q represent the real and imaginary parts of the IQ data respectively. Windowed intercepted power waveform:
[0202] Write the following pseudo code to perform window interception of the power waveform and draw a power time domain graph:
[0203]
[0204] Feature extraction and analysis:
[0205] Compare the power waveforms of different walkie-talkies, analyze their differences, and extract RF fingerprint features.
[0206] Record the significant feature points and segments of each walkie-talkie in the power time domain diagram.
[0207] 4. Feature Verification
[0208] Repeat the experiment:
[0209] The experiment was repeated several times to ensure the consistency and stability of the extracted RF fingerprint features.
[0210] Repeat the above experiment in different time periods and environments to verify the stability of the RF fingerprint feature. Feature matching:
[0211] Establish a walkie-talkie radio frequency fingerprint feature database.
[0212] By comparing with the features in the database, the uniqueness and recognition accuracy of the RF fingerprint features are verified.
[0213] Experimental results and analysis
[0214] Table 4:
[0215]
[0216] Results and Discussion:
[0217] Data Analysis:
[0218] Under different experimental conditions, the radio frequency fingerprint characteristics of the walkie-talkie present a square wave shape in the steady-state section, and by setting the threshold, the significant power waveform segment can be effectively intercepted.
[0219] The power time domain diagrams of Walkie-Talkie 1 and Walkie-Talkie 2 show significant differences at the beginning of the waveform. These differences can be used as RF fingerprint features.
[0220] Results and Discussion:
[0221] Experimental verification shows that the window-type interception power waveform method can stably and accurately extract RF fingerprint features and has a high recognition accuracy.
[0222] This method showed good stability in different experimental environments and time periods, proving its feasibility and effectiveness.
[0223] in conclusion:
[0224] The application of window-type intercepted power waveform diagram in TDMA duplex mode walkie-talkies has significant effect and provides a reliable method for extracting and identifying RF fingerprint features.
[0225] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.
Claims
1. A radio frequency fingerprint extraction method for a TDMA intercom, characterized in that: The following steps are involved: S1. Get the IQ data of the TDMA intercom; S2, calculate the RSSI of the received signal; S3, window-type interception of the power waveform of the steady-state section; S4, performing data cleaning on the intercepted waveform graph; S5. Use deep learning network model for feature extraction and recognition.
2. A method for extracting radio frequency fingerprints for TDMA intercoms according to claim 1, characterized in that: The S1 step specifically includes the following steps: S1.
1. Record the TDMA intercom signal as a wav file; S1.2, use Matlab program to read IQ data in wav format; S1.
3. Save the read IQ data as a two-dimensional array.
3. The method for extracting radio frequency fingerprints for TDMA intercom according to claim 1, characterized in that: The calculation formula of RSSI in step S2 is: RSSI = I 2 +Q 2 , where I and Q represent the real and imaginary parts of the IQ data, respectively.
4. The method for extracting radio frequency fingerprints for TDMA intercoms according to claim 1, characterized in that: The S3 step specifically includes the following steps: S3.1, set the RSSI threshold; S3.2, intercept the part of the waveform where the RSSI value is greater than the threshold; S3.
3. Save the captured waveform.
5. The method for extracting radio frequency fingerprints for TDMA intercom according to claim 1, characterized in that: The S4 step specifically includes the following steps: S4.1, calculating the arithmetic mean of the RGB value of each point in the waveform; S4.
2. Delete the waveforms whose RGB average values are greater than 253.
6. The method for extracting radio frequency fingerprints for TDMA intercoms according to claim 1, characterized in that: The deep learning network model is trained and classified through the Lenet network model.
7. The method for extracting radio frequency fingerprints for TDMA intercoms according to claim 1, characterized in that: The specific implementation of the window-type intercepted power waveform diagram includes the following pseudo code:
8. The method for extracting radio frequency fingerprints for TDMA intercoms according to claim 1, characterized in that: The specific implementation of the data cleaning includes the following pseudo code:
