Dr Zhihe Lu
Academic and research departmentsFaculty of Engineering and Physical Sciences, Department of Electrical and Electronic Engineering, Centre for Vision, Speech and Signal Processing (CVSSP).
This paper presents a novel frequency-domain energy detection scheme based on extreme statistics for robust sensing of OFDM sources in the low SNR region. The basic idea is to exploit the frequency diversity gain inherited by frequency selective channels with the aid of extreme statistics of the differential energy spectral density (ESD). Thanks to the differential stage the proposed spectrum sensing is robust to noise uncertainty problem. The low computational complexity requirement of the proposed technique makes it suitable for even machine-to-machine sensing. Analytical performance analysis is performed in terms of two classical metrics, i.e. probability of detection and probability of false alarm. The computer simulations carried out further show that the proposed scheme outperforms energy detection and second order cyclostationarity based approach for up to 10 dB gain in the low SNR range. © 2011 IEEE.
Spectrum sensing is one of key enabling techniques to advanced radio technologies such as cognitive radios and ALOHA. This paper presents a novel non-cooperative spectrum sensing approach that can achieve a good trade-off between latency, reliability and computational complexity. Our major idea is to exploit the first-order cyclostationarity of the primary user's signal to reduce the noise-uncertainty problem inherent in the conventional energy detection approach. It is shown that the proposed approach is suitable for detecting the primary user's activity in the interweave paradigm of cognitive spectrum sharing, while the active primary user is periodically sending training sequence. Computer simulations are carried out for the typical IEEE 802.11g system. It is observed that the proposed approach outperforms both the energy detection and the second-order cyclostationarity approach when the observation period is more than 10 frames corresponding to 0.56 ms. ©2010 IEEE.
A common strategy adopted by existing state-of-the-art unsupervised domain adaptation (UDA) methods is to employ two classifiers to identify the misaligned local regions between source and target domain. Following the 'wisdom of the crowd' principle, one has to ask: why stop at two? Indeed, we find that using more classifiers leads to better performance, but also introduces more model parameters, therefore risking overfitting. In this paper, we introduce a novel method called STochastic clAssifieRs (STAR) for addressing this problem. Instead of representing one classifier as a weight vector, STAR models it as a Gaussian distribution with its variance representing the inter-classifier discrepancy. With STAR, we can now sample an arbitrary number of classifiers from the distribution, whilst keeping the model size the same as having two classifiers. Extensive experiments demonstrate that a variety of existing UDA methods can greatly benefit from STAR and achieve the state-of-the-art performance on both image classification and semantic segmentation tasks.