A UAV identification system and identification method based on PYNQ

CN114553364BActive Publication Date: 2024-11-01GUILIN UNIV OF ELECTRONIC TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202210189825.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-02-28
Publication Date
2024-11-01
Estimated Expiration
2042-02-28

AI Technical Summary

Technical Problem

The existing deep learning-based RF fingerprint recognition method is difficult to achieve real-time, accurate and efficient drone identity recognition in low-power and low-cost embedded scenarios. Traditional security protocols have potential forgery risks and cannot effectively identify unregistered or Unlicensed drone.

Method used

Using a PYNQ-based drone identification system, the PYNQ-Z2 FPGA development board, signal receiving device, LCD display and SD card is used to accelerate the preprocessing and feature extraction of the IP core through preprocessing IP cores and neural networks, using three The order accumulation amount is used as the radio frequency fingerprint feature and is recognized through the convolutional neural network to realize real-time identity recognition.

Benefits of technology

It realizes stable identity recognition under complex channel conditions, with an identification rate of more than 90%, low system power consumption, suitable for a variety of application scenarios, high practical value and low hardware cost, and can quickly identify and verify the identity of the drone.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114553364B_ABST
    Figure CN114553364B_ABST
Patent Text Reader

Abstract

The present invention discloses a drone identification system and identification method based on PYNQ. The drone identification system adopts a signal receiving device to receive the drone signal to be identified, transmits the signal to PYNQ‑Z2 through an RJ45 interface, extracts the third-order cumulant from the received radio frequency signal as the unique radio frequency fingerprint feature of each drone, and classifies the radio frequency fingerprint features based on the obtained drone radio frequency fingerprint features by using a neural network, and uses a neural network acceleration IP core to perform accelerated calculation, so that the identity of the drone that needs to be authenticated can be recognized and authenticated in real time.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to wireless communication physical layer security technology, and in particular to a PYNQ-based drone identification system and identification method. Background Art

[0002] As the drone industry gradually becomes cheaper, more like a toy and more popular, we often see drones flying without registration or permission in many large-scale events. Once these drones lose control and crash, they will cause great casualties and property losses, greatly interfere with the normal flight of civil aviation flights, and may also cause some malicious incidents.

[0003] In order to eliminate the security risks brought by illegal drones, a powerful and efficient method is needed to identify and manage these illegal drones. Due to the openness of wireless media, the traditional drone identification method based on security protocols has hidden dangers. Radio frequency fingerprint recognition based on the physical layer has the advantage that its features are difficult to forge and can effectively and correctly identify the identity of the drone.

[0004] RF fingerprint refers to the slight hardware differences between devices caused by the tolerance of the production process of the internal electronic components of the signal sending device. These hardware differences cause the transmitted signal to have certain slight changes compared to the ideal signal. These slight changes do not affect the normal communication between devices, but give each signal a unique fingerprint feature.

[0005] The third-order cumulant is a statistical description of the signal in the time domain. When a non-Gaussian signal to be observed is superimposed with additive colored noise, the third-order cumulant of the superimposed signal is equal to the third-order cumulant of the signal to be observed. Therefore, when the third-order cumulant is used as a radio frequency fingerprint feature, it can effectively eliminate the impact of noise.

[0006] At present, the RF fingerprint recognition method based on deep learning technology mainly uses convolutional neural networks or its variants. Most convolutional neural networks basically run in a GPU environment. Although GPU can achieve real-time processing, its high power consumption and high cost make it difficult to meet the application requirements of some low-power and low-cost embedded scenarios. Therefore, it is of great practical significance to be able to develop a set of RF fingerprint recognition systems with fast speed, high accuracy and low power consumption. Summary of the invention

[0007] The purpose of the present invention is to provide a drone identification system and identification method based on PYNQ to address the deficiencies of the prior art. The system can identify and authenticate the drone in real time, and the method has high identification accuracy.

[0008] The technical solution for achieving the purpose of the present invention is:

[0009] A PYNQ-based drone identification system comprises a PYNQ-Z2 and a signal receiving device connected to the PYNQ-Z2, a liquid crystal display LCD and an SD card, wherein the PYNQ-Z2 and the signal receiving device are connected by a twisted pair cable, wherein the PYNQ-Z2 is an FPGA development board based on a dual-core Cortex-A9 processor, wherein the FPGA development board is provided with an APSoC chip, an AXI data transmission bus, a DDR3 memory controller, an RJ45 interface, an HDMI Out interface and an SD Card interface, wherein the DDR3 memory controller is connected to the ASPSoC chip and the SDCard interface, the ASPSoC chip is connected to the RJ45 interface and the HDMI Out interface, the RJ45 interface is connected to the twisted pair cable, the HDMI Out interface is connected to the HDMI cable, and the SD Card interface is connected to the SD Card interface. Card interface is connected to SD card, wherein the APSoC chip is provided with a programmable logic PL end and a processing system PS end, the processing system PS end is connected to the DDR3 memory controller, the AXI data transmission bus is connected to the programmable logic PL end and the processing system PS end, the programmable logic PL end is provided with a data transmission module, a preprocessing IP core, a neural network acceleration IP core and an HDMI high-definition display module, the data transmission module is connected to the AXI data transmission bus for accessing the drone signal stream data; the preprocessing IP core is connected to the AXI data transmission bus for extracting the effective part of the received signal and generating parallel accelerated calculation of the third-order cumulative amount; the neural network acceleration IP core is connected to the AXI data transmission bus for parallel accelerated calculation of the neural network; the HDMI high-definition display module is connected to the HDMI Out interface for outputting the drone recognition result, and the processing system PS end is provided with a drone signal preprocessing module and a drone signal recognition module. The signal receiving device is used to collect drone signals.

[0010] The liquid crystal display LCD is used to display the drone recognition result.

[0011] The twisted pair is used for data transmission between the signal receiving device and PYNQ-Z2.

[0012] The SD card is used to store the operating system of PYNQ-Z2, the data model recognized by the drone, and the executable code.

[0013] A PYNQ-based drone identification method includes the above-mentioned PYNQ-based drone identification system, and the method includes the following steps:

[0014] 1) Signal reception and preprocessing: A signal receiving device is used to receive the control signal of the drone. The signal receiving device transmits the signal to the DDR of PYNQ through the RJ45 interface, and then transmits it to the programmable logic PL end of PYNQ through the AXI data transmission bus. A preprocessing IP core is used in the programmable logic PL end to calculate the average power of the invalid area signal containing only the noise signal and the average power of the valid area signal through the adaptive threshold method, where the valid area signal contains the noise signal and the drone signal; then, the average power value of the valid area signal is used as the power threshold, and the signal below the threshold is regarded as the invalid area signal and removed, leaving the valid area signal;

[0015] 2) Extraction of radio frequency fingerprint features of the signal: Using the method described in step 1), respectively collect the effective area signals of each drone to be identified, normalize the power of the effective area signals to avoid interference from different drone signal powers, and use the indirect estimation method to calculate the third-order cumulant of each drone signal to be identified and store it in the DDR;

[0016] 3) Training a neural network for signal recognition: A convolutional neural network is constructed using the Tensorflow platform. The convolutional neural network is equipped with an input layer, a convolution layer, a pooling layer, a fully connected layer, and an output layer. The input layer is used to input the third-order cumulants of the signal to be recognized, and the output layer is used to output the recognition result of the drone. The method in step 2) is used to take the third-order cumulants of X types of X drones to be recognized as the training set to train the neural network.

[0017] 4) Signal classification and recognition: Import the neural network trained in step 3) into the SD card, then read it from the SD card into the DDR, send the RF fingerprint features to be recognized into the trained neural network, start the neural network acceleration IP core for calculation, and cyclically execute the forward derivation calculation of the convolution layer, pooling layer and fully connected layer in the neural network to obtain the signal recognition result, and write the result back to the DDR after calculation;

[0018] 5) The signal recognition result obtained in step 4) is output to the liquid crystal display LCD through the HDMI high-definition display module.

[0019] The third-order cumulant described in step 2) is calculated using the indirect estimation method, and the process is as follows:

[0020] 3-1) Divide the N data in the original signal into K segments, each segment contains M samples;

[0021] 3-2) Let x (k) (0),x (k) (1),......,x (k)(M-1) is the kth segment of data, where k = 1, ..., K, and the estimated value of the third-order cumulant of each segment is calculated:

[0022]

[0023] 3-3) Take the average value of the third-order cumulants of all segments as the third-order cumulant estimate of the entire observation data set, that is,

[0024]

[0025] Among them, i and j in formulas (1) and (2) are the delay amounts of the signal, i = -M, ..., M, j = -M, ..., M, M1 in formula (1) is the maximum value among 0, -i, and -j, and M2 in formula (1) is the minimum value among M-1, M-1-i, and M-1-j.

[0026] Compared with the prior art, the beneficial effects of this technical solution are:

[0027] 1) This technical solution performs drone identification based on the drone's radio frequency fingerprint characteristics. This method is different from the existing drone identification algorithm and has the characteristics of not requiring prior knowledge of the drone's radio frequency signal and strong applicability;

[0028] 2) This technical solution uses the third-order cumulative amount as the RF fingerprint feature, eliminating the part that is easily affected by noise signals, has good stability, and can stably identify the drone under complex channel conditions;

[0029] 3) This technical solution adopts a lightweight neural network to reduce the calculation amount of the model while ensuring the recognition rate. When the signal-to-noise ratio is greater than 20dB, the recognition rate is greater than 90%;

[0030] 4) This technical solution uses the parallel processing capability of the FPGA part of the PYNQ-Z2 chip, so that the system can identify the identity of the drone in a short time interval, which has high practical value;

[0031] 5) This technical solution realizes a relatively small drone detection system with low total power consumption and high portability, which is suitable for most application scenarios;

[0032] 6) The hardware price of this technical solution has obvious advantages compared to PC.

[0033] This system can identify and authenticate the identity of a drone in real time, with high recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a schematic diagram of the structure diagram in the embodiment;

[0035] Figure 2 Schematic diagram of the method flow in the embodiment. DETAILED DESCRIPTION

[0036] The content of the present invention is further described below in conjunction with the drawings and embodiments, but the present invention is not limited thereto.

[0037] Example:

[0038] Reference Figure 1 A PYNQ-based drone identification system includes a PYNQ-Z2 and a signal receiving device connected to the PYNQ-Z2, a liquid crystal display LCD and an SD card. The PYNQ-Z2 and the signal receiving device are connected by a twisted pair cable. The PYNQ-Z2 is an FPGA development board based on a dual-core Cortex-A9 processor. The FPGA development board is provided with an APSoC chip, an AXI data transmission bus, a DDR3 memory controller, an RJ45 interface, an HDMI Out interface, and an SD Card interface. The DDR3 memory controller is connected to the ASPSoC chip and the SD Card interface, the ASPSoC chip is connected to the RJ45 interface and the HDMI Out interface, the RJ45 interface is connected to the twisted pair cable, the HDMI Out interface is connected to the HDMI cable, and the SD Card interface is connected to SD card, wherein the APSoC chip is provided with a programmable logic PL end and a processing system PS end, the processing system PS end is connected to the DDR3 memory controller, the AXI data transmission bus is connected with the programmable logic PL end and the processing system PS end, the programmable logic PL end is provided with a data transmission module, a preprocessing IP core, a neural network acceleration IP core and an HDMI high-definition display module, the data transmission module is connected to the AXI data transmission bus for accessing the drone signal stream data; the preprocessing IP core is connected to the AXI data transmission bus for extracting the effective part of the received signal and generating parallel accelerated calculation of the third-order cumulative amount; the neural network acceleration IP core is connected to the AXI data transmission bus for parallel accelerated calculation of the neural network; the HDMI high-definition display module is connected to the HDMI Out interface for outputting the drone recognition result, and the processing system PS end is provided with a drone signal preprocessing module and a drone signal recognition module.

[0039] The signal receiving device is used to collect drone signals.

[0040] The liquid crystal display LCD is used to display the drone recognition result.

[0041] The twisted pair is used for data transmission between the signal receiving device and PYNQ-Z2.

[0042] The SD card is used to store the operating system of PYNQ-Z2, the data model recognized by the drone, and the executable code.

[0043] like Figure 2 As shown, a PYNQ-based drone identification method includes the above-mentioned PYNQ-based drone identification system, and the method includes the following steps:

[0044] 1) Signal reception and preprocessing: A signal receiving device is used to receive the control signal of the drone. The signal receiving device transmits the signal to the DDR of PYNQ through the RJ45 interface, and then transmits it to the programmable logic PL end of PYNQ through the AXI data transmission bus. A preprocessing IP core is used in the programmable logic PL end to calculate the average power of the invalid area signal containing only the noise signal and the average power of the valid area signal through the adaptive threshold method, where the valid area signal contains the noise signal and the drone signal; then, the average power value of the valid area signal is used as the power threshold, and the signal below the threshold is regarded as the invalid area signal and removed, leaving the valid area signal;

[0045] 2) Extraction of radio frequency fingerprint features of the signal: Using the method described in step 1), respectively collect the effective area signals of each drone to be identified, normalize the power of the effective area signals to avoid interference from different drone signal powers, and use the indirect estimation method to calculate the third-order cumulant of each drone signal to be identified and store it in the DDR;

[0046] 3) Training a neural network for signal recognition: A convolutional neural network is constructed using the Tensorflow platform. The convolutional neural network is equipped with an input layer, a convolution layer, a pooling layer, a fully connected layer, and an output layer. The input layer is used to input the third-order cumulants of the signal to be recognized, and the output layer is used to output the recognition result of the drone. The method in step 2) is used to take the third-order cumulants of X types of X drones to be recognized as the training set to train the neural network.

[0047] 4) Signal classification and recognition: Import the neural network trained in step 3) into the SD card, then read it from the SD card into the DDR, send the RF fingerprint features to be recognized into the trained neural network, start the neural network acceleration IP core for calculation, and cyclically execute the forward derivation calculation of the convolution layer, pooling layer and fully connected layer in the neural network to obtain the signal recognition result, and write the result back to the DDR after calculation;

[0048] 5) The signal recognition result obtained in step 4) is output to the liquid crystal display LCD through the HDMI high-definition display module.

[0049] The third-order cumulant described in step 2) is calculated using the indirect estimation method, and the process is as follows:

[0050] 3-1) Divide the N data in the original signal into K segments, each segment contains M samples;

[0051] 3-2) Let x (k) (0),x (k) (1),......,x (k) (M-1) is the kth segment of data, where k = 1, ..., K, and the estimated value of the third-order cumulant of each segment is calculated:

[0052]

[0053] 3-3) Take the average value of the third-order cumulants of all segments as the third-order cumulant estimate of the entire observation data set, that is,

[0054]

[0055] Among them, i and j in formula (1) and (2) are the delay of the signal, i=-M,...,M, j=-M,...,M, M1 in formula (1) is the maximum value among 0, -i, and -j, and M2 in formula (1) is the minimum value among M-1, M-1-i, and M-1-j. In this example, PYNQ-Z2 is an FPGA development board based on a dual-core Cortex-A9 processor developed by Xilinx; the signal device is USRP, model E310, and the ADC / DAC chip is AD9361; the SD card storage format is FAT32, and the storage capacity is 64GB.

Claims

1. A method for identifying unmanned aerial vehicles based on PYNQ, wherein the method is implemented by using a PYNQ-based unmanned aerial vehicle identification system, the PYNQ-based unmanned aerial vehicle identification system comprising a PYNQ-Z2 and a signal receiving device connected to the PYNQ-Z2, a liquid crystal display LCD and an SD card, the PYNQ-Z2 and the signal receiving device are connected by a twisted pair cable, the PYNQ-Z2 is an FPGA development board based on a dual-core Cortex-A9 processor, the FPGA development board is provided with an APSoC chip, an AXI data transmission bus, a DDR3 memory controller, an RJ45 interface, an HDMIOut interface, and an SD Card interface, the DDR3 memory controller is connected to the ASPSoC chip and the SD Card interface, the ASPSoC chip is connected to the RJ45 interface and the HDMIOut interface, the RJ45 interface is connected to the twisted pair cable, the HDMIOut interface is connected to the HDMI cable, and the SD Card interface is connected to the SD card, wherein: The APSoC chip is provided with a programmable logic PL end and a processing system PS end, the processing system PS end is connected to the DDR3 memory controller, the AXI data transmission bus connects the programmable logic PL end and the processing system PS end, the programmable logic PL end is provided with a data transmission module, a preprocessing IP core, a neural network acceleration IP core and an HDMI high-definition display module, and the data transmission module is connected to the AXI data transmission bus; The preprocessing IP core is connected to the AXI data transmission bus; the neural network acceleration IP core is connected to the AXI data transmission bus; the HDMI high-definition display module is connected to the HDMIOut interface, and the processing system PS end is provided with a drone signal preprocessing module and a drone signal recognition module, characterized in that the implementation of the method includes the following steps: 1) Signal reception and preprocessing: A signal receiving device is used to receive the control signal of the drone. The signal receiving device transmits the signal to the DDR of PYNQ through the RJ45 interface, and then transmits it to the programmable logic PL end of PYNQ through the AXI data transmission bus. A preprocessing IP core is used in the programmable logic PL end to calculate the average power of the invalid area signal containing only the noise signal and the average power of the valid area signal through the adaptive threshold method, where the valid area signal contains the noise signal and the drone signal; then, the average power value of the valid area signal is used as the power threshold, and the signal below the threshold is regarded as the invalid area signal and removed, leaving the valid area signal; 2) Extraction of radio frequency fingerprint features of the signal: Using the method described in step 1), respectively collect the effective area signals of each drone to be identified, normalize the power of the effective area signals, and use the indirect estimation method to calculate the third-order cumulative amount of each drone signal to be identified and store it in the DDR; 3) Training the neural network for signal recognition: The Tensorflow platform is used to build a convolutional neural network. The convolutional neural network has: input layer, convolution layer, pooling layer, fully connected layer, and output layer. The input layer is used to input the third-order cumulants of the signal to be identified, and the output layer is used to output the recognition result of the drone. The method in step 2) is used to take the X types of third-order cumulants of X drones to be identified as the training set to train the neural network; 4) Signal classification and recognition: Import the neural network trained in step 3) into the SD card, then read it from the SD card into the DDR, send the RF fingerprint features to be recognized into the trained neural network, start the neural network acceleration IP core for calculation, and cyclically execute the forward derivation calculation of the convolution layer, pooling layer and fully connected layer in the neural network to obtain the signal recognition result, and write the result back to the DDR after calculation; 5) Outputting the signal recognition result obtained in step 4) to a liquid crystal display LCD via an HDMI high-definition display module; The third-order cumulant described in step 2) is calculated using the indirect estimation method, and the process is as follows: 2-1) Divide the N data in the original signal into K segments, each segment contains M samples; 2-2) Let x (k) (0),x (k) (1),......,x (k) (M-1) is the kth segment of data, where k = 1, ..., K, and the estimated value of the third-order cumulant of each segment is calculated: 2-3) Take the average of the third-order cumulants of all segments as the third-order cumulant estimate of the entire observation data set, that is, Wherein, i and j in formula (1) and (2) are the delay amounts of the signal, i = -M, ..., M, j = -M, ..., M, In formula (1), M1 is the maximum value among 0, -i, and -j, and M2 in formula (1) is the minimum value among M-1, M-1-i, and M-1-j.