OFDM under Oscillator Phase Noise : Contributions to Analysis and Estimation Methods

Most modern transmitters and receivers involve an analog front-end unit and a digital back-end unit. The digital back-end is responsible for information processing which involves thefollowing: redundancy removal from information; information representation; improvinginformation resilience; and information correction. The analog front-end is responsible forinformation transmission and reception. The information processing algorithms developedand implemented in the digital back-end assume a linear and noiseless analog front-end which,in reality, is not the case. This renders some of the information processing algorithms to be lesseffective in practice. The focus of this thesis is on orthogonal frequency-division multiplexing(OFDM) systems under the influence of oscillator phase noise. OFDM is an efficientinformation representation technique used in many communication systems. On the otherhand, phase noise is one type of undesired noise that occurs in the oscillator device used in theanalog front-end. It arises due to the imperfect task of frequency conversion, performed by theoscillator device, between baseband and radio frequency. This thesis contributes to the areas of analysis and estimation in OFDM systems under theinfluence of oscillator phase noise. With regard to analysis, this thesis contributes by derivingthe channel capacity assuming a Gaussian input alphabet. The aim here is to show bothquantitatively and qualitatively the degradation in performance of the OFDM system in thepresence of phase noise. The analysis is conducted for phase noise processes that occur in bothfree-running and phase-locked loop based oscillators and also extended to include the effect of carrier frequency offset. With regard to estimation, two new phase noise estimation algorithmsare proposed in this thesis. In particular, these algorithms restrict the search space to a specific subset, where the desired phase noise parameter of interest is shown to lie. For example, in the first estimation method, possible subspaces in which the desired phase noise spectral vector may lie are used in the estimation step. In the second method, the geometry of the desired phase noise spectral vector is used in the estimation step. Specifically, this geometry corresponds to a non-convex set described by a set of quadratic forms that involve permutation matrices. By restricting the search space to this set, the accuracy of phase noise estimation can be improved.