Abstract
Spectrum sensing is a signal processing tool utilised to identify the presence of signal in a noisy spectrum of interests. It is a traditional research topic in the area of signal processing for communications with wide applications in listen before talk communication protocols. Recently, spectrum sensing has received increasing interests in the area of mobile communications due to the development of emerging flexible networks. Although the local spectrum sensing has been quite well investigated in the literature, it is realised that the existing techniques can hardly achieve a reliable performance in low signal-to-noise-ratio (SNR) environments without introducing high computational complexity. Moreover, the requirement of short sensing delay also arises a challenge to the existing spectrum sensing techniques. Motivated by the above problems, the main contributions of this thesis is twofold: Firstly, a novel pilot-assisted spectrum sensing technique for orthogonal frequency-division multiplexing (OFDM) systems is proposed. The key idea is based upon the physical nature that subcarriers carrying pilots or payload data have different first-order and second-order statistical properties. These differences vanish when the spectrum of interest is unoccupied. Therefore, the decision of spectrum availability can be formed based upon these differences, which can be explored through employment of frequency-domain differential operations. The simulation result has shown that the proposed technique outperforms the conventional pilot-assisted technique up to 7 dB. Secondly, a novel blind spectrum sensing method is introduced to form fast and accurate decision of the spectrum usage with high sensitivity to weak signals. The key idea lies in the fundamental nature of white Gaussian process, i. e. , if the measured data is white Gaussian noise, it must have identical order-statistics in both the time and frequency domains, and vice versa. Therefore, the decision of spectrum usage is formed by means of measuring the likelihood of time-frequency domain order-statistics. The probability of false alarm is mathematically derived with the Log-Normal approximation which is shown to be very close to the real case. It is shown that the proposed method forms accurate decision with the measured data length to be shorter than one symbol duration and the signal-to-noise ratio to be as low as -20 dB. This result is better than the state-of-the-art by up to 8 dB. At last, it is said that the proposed spectrum sensing techniques can benefit the future flexible networks, and improve the overall efficiency of spectrum utilisation.