As the market for drones grows rapidly in recent years, drones are getting smaller, faster, and smarter because of artificial intelligence. As a result, serious problems such as privacy violations and terrorism are rapidly increasing as well as problems such as crashes and falls of drones. To cope with these threats, it is necessary to develop a radar capable of detecting small drones.
In this paper, I propose frequency modulated continuous wave (FMCW) radar systems for small drone detection. Compared to pulse radar, FMCW radar is excellent in price, peak power, and resolution. Also, in the manner of operation, the pulse radar has a limitation on the minimum detection range, while the FMCW radar has no limitation. However, due to the way the FMCW radar emits signals continuously, the leakage signal from the transmitter to the receiver causes several problems. Among them, the problem that the phase noise of the leakage signal seriously degrades the SNR (Signal to Noise Ratio) in the FMCW radar affects the direct detection. In this paper, the FMCW radar systems are designed and produced to perform the analyses on actual signals and noises, and the phase noise of leakage signal is especially analyzed in detail. Then, based on the analyses on the phase noise, novel techniques to lower the noise floor are proposed, resulting in successful detection of small drones that are difficult to detect due to the small radar cross section (RCS). In addition, the results of successfully classifying small drones based on deep learning along with the proposed techniques are shown.
More specific research in this paper are as follows. First, I propose the stationary point concentration (SPC) technique that greatly increases the SNR by concentrating the phase noise of the leakage signal, which appears as a voltage or current noise, at the stationary point on a sinusoidal signal. The theory and realization method of the proposed SPC technique are presented, its performance is predicted based on the simulation results, and finally, the SPC technique is applied to the produced FMCW radar to verify its performance. In the experiments of verification, commercial drones, DJI Inspire 1 and DJI Spark, a small palm-sized drone, were used. Experimental results show that the degree of improvement predicted by simulations is almost matched with the actual degree of improvement and that the SPC technique significantly lowers the noise floor up to about 10.5 dB and at least about 5.2 dB so that improves SNR for small drones. The proposed SPC technique is an original and novel leakage mitigation method that is completely different from the traditional approach, it has the advantage of being realized only by digital signal processing through the strategic frequency planning and oversampling without additional hardware. Moreover, by compensating for delays within radar hardware systems, more accurate distance information of targets can be obtained.
Additionally, through detection experiments using a little-finger sized micro-drone, CHERSON CX-10A, it is shown that the signal of micro-drone buried in the noise floor is detected successfully after the SPC technique is applied. This experiment is an unprecedented ultra-small drone detection experiment, and it is significant in that it has set a new record for ultra-small drone detection by applying the SPC technique.
Second, for detecting small moving drones, the SPC technique is extended to a range-Doppler (r-D) map, the result of the two-dimensional (2-D) Fourier transform. In the r-D map, the leakage signal has a problem that increases the entire two-dimensional noise floor. In addition, in practical radar systems, distorted velocity information of small moving drones are measured in the r-D map due to the problem of the Doppler shift in the radar system. In this study, the theoretical analyses on the causes of these problems are presented and verified experimentally. Then, the theory and experiments verify that the SPC technique not only reduces the 2-D noise floor but also resolves the problem of the Doppler shift by making radar coherent through the phase calibration effect. As a result of applying the SPC technique to a problematic radar system where the 2-D noise floor is raised by the leakage signal and the Doppler frequency changes at every measurement, it is shown that the SPC technique significantly lowers the 2-D noise floor up to about 23.3 dB and at least about 6.7 dB, and resolves the problem of randomly changed Doppler frequency. Therefore, the SNR of small moving drone, DJI Spark, is improved and its accurate velocity information can be obtained. Thanks to the aforementioned effects, the SPC technique can be used not only for simple target detection, but also for other radar applications such as coherent integration, static clutter suppression, motion detection/identification, micro-Doppler signature (MDS) classification, meteorological radar, synthetic aperture radar (SAR), and inverse SAR (ISAR).
Third, the advanced-SPC (A-SPC) technique, which improves the realization method of the SPC technique is proposed. In realizing the SPC technique, the A-SPC technique introduces a quadrature demodulator so that it enables the SPC technique to be implemented without the strategic frequency planning and oversampling, which were required by the original SPC technique. This allows all the effects of the SPC technique to be obtained while eliminating the limitation in frequency planning and reducing the costs of analog to digital converter (ADC) and memory that can occur due to oversampling. Besides, the A-SPC technique can obtain the aforementioned effects of the SPC technique regardless of the shape of the phase noise, so it can be applied in all kinds of FMCW radar systems regardless of the radar architecture. The results of experiments for the verification of the A-SPC technique show that the A-SPC technique improves the SNR additionally as well as expected effects. In addition, if the sampling frequency for the A-SPC technique is the same as the oversampling frequency in the SPC technique, it is possible to obtain a maximum unambiguous range of two times more than in the SPC technique. Moreover, it is proved that while the SPC technique can only be applied in the heterodyne FMCW radar, the A-SPC technique can be applied well in both the heterodyne FMCW radar, which has no range correlation effect, and the homodyne FMCW radar, which has the range correlation effect.
Forth, using the A-SPC technique, MDS images originated from the propellers of small drones are extracted, and the small drones are classified based on the self-designed Convolutional Neural Network (CNN). The A-SPC technique makes the MDS of small drones more vivid, thereby showing the features of MDS of each drone more clear so that it proves the classification rate of the small drones can be improved. Besides, unlike existing methods that mainly use the transfer learning with the well-known networks such as GoogLeNet, applying the A-SPC technique can achieve a high level of classification rate even with a light CNN using much fewer network parameters.
Fifth, the A-SPC technique is applied to SAR to improve the quality of a SAR image. After building automobile SAR (AutoSAR), which uses a car as a moving platform for SAR, and conducting SAR experiments, it is verified that the quality of the SAR image is improved when the SAR signal processing is performed with the A-SPC technique.
Additionally, I analyze the uncorrelated phase noise (UPN), which occurs in the heterodyne FMCW radar, and based on the analyses, I propose a strategic decision method for parameter values in FMCW radar for low noise floor over middle-long range. A formula for the degree of improvement due to the proposed decision method is devised, and the performance of the proposed method is verified with the experiments.
Through these studies in this paper, I verify that the proposed core technique, the SPC/A-SPC technique, greatly contributes to the successful detection and classification of small drones that pose serious threats to both civil and military recently. In addition, by applying the SPC/A-SPC technique to MDS extraction and SAR to improve their image quality, I verify that the proposed SPC/A-SPC technique has sufficient potential as a source technology for the FMCW radar.
- Park, Seong-Ookresearcher; 박성욱researcher
- 한국과학기술원 :전기및전자공학부,
- Issue Date
학위논문(박사) - 한국과학기술원 : 전기및전자공학부, 2020.8,[vii, 125 p. :]
advanced-stationary point concentration (A-SPC) technique▼aconvolutional neural network (CNN)▼adata processing▼adeep learning▼adigital signal processing(DSP)▼adoppler shift▼afrequency modulated continuous wave (FMCW) radar▼afrequency planning▼aheterodyne architecture▼ahomodyne architecture▼aleakage▼aleakage mitigation▼amicro-Doppler▼amicro-drone detection▼anoise floor▼aoversampling▼aphase calibration▼aphase noise▼aradar system design▼arange correlation effect▼arange-Doppler (r-D) map▼asignal to noise ratio (SNR)▼asmall drone detection and classification▼astationary point concentration (SPC) technique▼asynthetic aperture radar (SAR)▼auncorrelated phase noise(UPN); 거리-도플러 맵▼a고급 정류점 집중 기법▼a노이즈 층▼a누설▼a누설 완화; 디지털 신호 처리▼a데이터 처리▼a도플러 천이▼a딥러닝▼a레이다 시스템 설계▼a마이크로-도플러▼a범위 상관 효과▼a비상관 위상 잡음▼a소형 드론 탐지 및 식별▼a신호 대 잡음 비▼a오버샘플링▼a위상 보정▼a위상 잡음▼a이차원 노이즈 층▼a정류점 집중 기법▼a주파수 계획▼a주파수 변조 연속파 레이다▼a초소형드론탐지▼a합성 개구 레이다▼a합성곱 신경망▼a헤테로다인 구조▼a호모다인 구조
- Appears in Collection