Nonlinear signal processing for visible light communication

Abstract

Recently many research efforts have been directed towards communication in visible light spectrum (visible light communication (VLC)) keeping in mind bandwidth hungry future 5G and beyond systems. However, the performance of VLC systems is degraded by channel-impairments like inter-symbol interference (ISI) and inherent device impairments like light emitting diode (LED) nonlinearity which limits its overall throughput. To mitigate these impairments, and to maximize throughput, adaptive signal processing techniques are proposed in this work which do not need explicit knowledge of the VLC channel nonlinearity. In other words, online algorithms are proposed in this work, which adapts, learns and inverts the channel impairments. This adaptive learning can be useful in tracking scenarios when there is aging in the front-end devices/hardware. Additionally, there are savings in the manufacturing cost of the optical system hardware as these proposed techniques relax the minimum required tolerance in manufacturing front-end hardware. Finally, reliability of the overall VLC link increases as these signal processing techniques help in maintaining the overall signal to noise ratio in the presence of instantaneous outages. This thesis aims at developing signal processing techniques for boosting the throughput of the VLC system. In order to boost the achievable data rate in VLC system, the first work in this thesis proposes a Chebyshev polynomial based nonlinear pre-distortion technique for mitigating LED nonlinearity, and ISI of the VLC channel. Better bit error rate (BER) characteristics were found upon the use of Chebyshev predistortion as compared to popular linear normalized least mean squares (NLMS) based pre-distortion. The results obtained in this work may be useful for benchmarking with respect to a “perfect” scenario. However, the performance gains achieved in this work rely on perfect knowledge of the detected symbols at the receiver to be relayed to the transmitted symbols in the uplink which is not feasible in general. The second work focuses onopen-loop VLC system in which the problem of improving the throughput of the VLC system is solved by open-loop post-distortion at the receiver. A novel sparse novelty criterion based kernel minimum symbol error rate (KMSER) post-distorter (or equalizer) based on reproducing kernel Hilbert space (RKHS) techniques is proposed in which savings incomputational complexity and superior BER characteristics are demonstrated. This work establishes the RKHS based post-distorter to be a better alternative for post-distortion as compared to Volterra post-distorters. The third work explores further techniques to bring the computational complexity of the sparse RKHS-based post-distorter down further by proposing fixed-budget based dictionary pruning criterion. It is demonstrated from this work that same BER performance is achieved by fixed-budget KMSER over impaired VLC channels with lower dictionary size as compared to novelty-criterion based KMSER technique. Exact mathematical insights on the transient performance of the dynamics of learning curves of fixed budget-KMSER are drawn in this work. The next work revisits the problem of post-distortion over VLC channels by exploring unsupervised techniques for post-distortion. In this work, the proposed multi-stage clustering based Hammerstein post-distorter is compared with the widely used modified cascaded-MMA (MCMMA)- based Volterra post-distorter. Superior convergence characteristics are obtained with the use of multi-stage Hammerstein based post-distortion. The steady-state mean squared error (MSE) characteristics of the multi-stage post-distorter is analyzed mathematically and many insights from the analyzed steady-state behavior are derived. Finally, a multiple-input multiple-output (MIMO)-multi-user scenario is considered with non-orthogonal multiple access (NOMA) being the multiple-access technique. A special NOMA-scenario is considered where all users have correlated channel matrices (and hence similar channel conditions). A novel precoding technique and its corresponding power-allocation strategy is derived. This is the first effort in direction of NOMA in VLC being extended to arbitrary number of users with similar channel conditions. Based on these precoding techniques, expression for BER is derived for square-quadrature amplitude modulation (QAM) and insights are provided. Above all, the signal processing algorithms provided in this thesis are aimed at maximizing the overall throughput of the VLC system in presence of impairments for 5G and beyond communication systems.