| Summary: | 碩士 === 國立中興大學 === 電機工程學系所 === 107 === Ultrasound radar has been widely used in short range obstacle detection due to its simplicity and cost effectiveness in implementation. Conventional ultrasound radars use pulse transmission and adopt the principle "time of flight" to calculate distance. The detection quality is susceptible to noise and interference. FMCW (frequency modulation continuous waveform) radars, nonetheless, use the frequency offset between the transmitted and the received signals and can function properly under inferior SNR environments. However, to obtain the direction information of the objects, an array instead of one single transducer is needed to achieve the estimation.
This thesis develops an ultrasonic radar system for indoor, people rich environments 2) The system adopts an FMCW modulation scheme and employs an array configuration to perform distance and direction estimations simultaneously. The proposed system includes the following features: 1) tailored to the FMCW based systems 2) performing the estimations in the frequency domain, 3) capable of detecting multiple objects either with an identical distance from the radar or along the same incident directions, 4) mitigating the angle aliasing problem when a non-Nyquist spatially sampled transducer array is adopted. The system can thus construct obstacle maps of the surrounding environment. Because the detection principle of FMCW lies on the frequency offset between the transmitted and received signals, a Fast Fourier transform (FFT) is always required. The Direction of Arrival (DoA) can thus be performed in the frequency domain on a per frequency component basis. Because the reflection signals from two equally distanced objects co-exist in the same frequency component in an FMCW system, the DoA estimation should be able to distinguish them. A matching pursuit (MP) plus least square (LS) estimation scheme is thus developed to find the directions from a codebook with predefined steering vectors. In particular, the MP scheme finds the candidate vectors first and the LS scheme determines the best ones from them. Because no low-frequency (＜100kHz) ultrasonic transducer array devices are available, individual transducers are put together to form an array. But the size of the ultrasonic transducer is too big to make an array meeting the Nyquist spatial sampling criterion. This leads to an aliasing problem in detection. To mitigate the problem, we propose a new array configuration consisting of 6+2 transducers. Six transducers are placed collinearly to form a linear array with a 3λ/2 spacing. Two auxiliary transducers are put on the opposite sides of the linear array with a horizontal displacement of λ/2 and λ, respectively, to resolve the aliasing issue. The steering vectors of the codebook are redefined subject to this new array configuration and the proposed DoA estimation scheme can be equally applied.
Matlab simulations are conducted to verify the performance of the proposed scheme. We start with the simulations assuming a virtual linear array with aλ/2 spacing is available. After this verifications, simulations using the proposed 6+2 array configuration are conducted next.
There are two object models adopted in the simulations. The first model assumes that each object has only one reflection point, and the second one assumes nine reflection points are associated with the object. In addition, we assume the codebook contains steering vectors with an angular resolution of 5°. The simulations are conducted under different SNR settings with no interferences. The evaluation criteria include of the accuracy of selecting the best match steering vectors from the codebook and root mean square (RMS) of the estimation in degrees. The tolerance of the estimation error is set to 3°. The DoA estimation schemes under comparison include the conventional MUSIC, ESPRIT schemes and an approach based on orthogonal matching pursuit (OMP). The simulation results indicate the effectiveness of the proposed scheme. Taking the case of SNR setting to 10dB as an example, the estimation error rate of the proposed scheme is merely 4.67% while the RMS of estimation error is just 0.3523°. These numbers are better than the OMP based approach. The conventional MUSIC and SPIRIT schemes fail to distinguish objects with equal distances. As for the simulation results using the proposed 6+2 array configuration, the object model containing 9 reflection points is adopted. The estimation error rate is 1.33% and the RMS of estimation error is 3.5863°. Again, these numbers are better than the OMP based approach while the complexity reduction can be as high as 73.4%
|
|---|
