On Some Signal Processing Algorithms Applicable to Cognitive Radio Communication

Abstract

Convergence of Wireless communication and Internet has leveraged for mammoth expansion of wireless radio access techniques leading to overcrowding of spectrum. Additionally there is need for large bandwidth to accommodate high data rate demand. In spite of intensive research and development efforts scarcity of available radio spectrum is a critical limitation. Opportunistic dynamic spectrum access (DSA) and cognitive radio (CR) techniques have received increasing attention as potential solutions to spectrum shortage issue in 5G wireless communication systems. Recently, spectrum sensing and identification have been considered critical functionalities to accommodate DSA. Spectrum sensing, enables CRs to identify spectral holes without interfering with licensed primary user maintaining quality of service of secondary user. On other hand, expanding wireless communication usage has rendered for radio access techniques(RATs) with various modulation, antenna systems, hardware architectures and channel scenario. Hence, there is a requirement to investigate robust spectrum sensing and radio access identification methods. This thesis, focuses on two important aspects of spectrum sensing. One is identifying the occupancy of spectrum and other is differentiate different users occupying the spectrum on basis of modulation type and RATs classification. Hence, by this process efficient management of interference mitigation and DSA is possible in the emerging heterogeneous networks. Realization of both the subtask leads to addressing various challenges. One of the important challenge is quickly identifying occupancy of spectrum without any priori information about primary user by using very few samples of data. Next traditional spectrum sensing algorithms become unsuitable when real time wireless communication signals passed through harsh channel impairment in presence of hardware imperfections due to non-stationary nature of signal. Last challenge is related to radio signal classification in which there is no single optimal solution to classify all the types modulation techniques and RATs signals in emerging next generation wireless communication systems. This thesis address the first challenge by proposing a blind eigenvalue based signal processing solutions that work under sample starving environment to enhance spectral detection efficiency. Two modified eigenvalue based spectrum sensing (SS) techniques called corrected John’s test (CJT) and higher order eigenvalue-moment ratio (HO-EMR) based techniques are proposed. Further to increase robustness of sensing a dual stage SS is adopted as per IEEE 802.22 CR standard, with the use of energy detection and EMR based techniques. Dual SS possess the capability to estimate the noise variance and enhance prediction accuracy by combining superior features of both detection algorithms. To counter the dynamic nature of emerging radio signal, time frequency based spectrum sensing is proposed. These have been historically used for non-stationary radar signal detection. Various TF distribution based SS are also proposed and comparative analysis carried out. Later, to make the CR intelligent, applicability of combination TF distribution and recent trends of machine learning in form of deep learning techniques are used for modern modulation detection and radio access techniques (RAT) identification. Deep learning based network are data driven classifiers and they avoid need for manual expert feature design, which is a necessity in traditional feature based classifiers. Their performance are analyzed in terms of classification accuracy with recently proposed DL architectures. Overall thesis proposes novel spectrum sensing and radio signal classification algorithms that assist in imbibing cognitive capability among emerging next generation wireless networks.