Advanced Multi-Dimensional Signal Processing for Wireless Systems

The thriving development of wireless communications calls for innovative and advanced signal processing techniques targeting at an enhanced performance in terms of reliability, throughput, robustness, efficiency, flexibility, etc.. This thesis addresses such a compelling demand and presents new and intriguing progress towards fulfilling it. We mainly concentrate on two advanced multi-dimensional signal processing challenges for wireless systems that have attracted tremendous research attention in recent years, multi-carrier Multiple-Input Multiple-Output (MIMO) systems and multi-dimensional harmonic retrieval. As the key technologies of wireless communications, the numerous benefits of MIMO and multi-carrier modulation, e.g., boosting the data rate and improving the link reliability, have long been identified and have ignited great research interest. In particular, the Orthogonal Frequency Division Multiplexing (OFDM)-based multi-user MIMO downlink with Space-Division Multiple Access (SDMA) combines the twofold advantages of MIMO and multi-carrier modulation. It is the essential element of IEEE 802.11ac and will also be crucial for the fifth generation of wireless communication systems (5G). Although past investigations on scheduling and precoding design for multi-user MIMO downlink systems have been fruitful, new advances are desired that exploit the multi-carrier nature of the system in a more efficient manner or aim at a higher spectral efficiency. On the other hand, a Filter Bank-based Multi-Carrier modulation (FBMC) featuring a well-concentrated spectrum and thus a low out-of-band radiation is regarded as a promising alternative multi-carrier scheme to OFDM for an effective utilization of spectrum fragments, e.g., in 5G or broadband Professional Mobile Radio (PMR) networks. Unfortunately, the existing transmit-receive processing schemes for FBMC-based MIMO systems suffer from limitations in several aspects, e.g., with respect to the number of supported receive antennas (dimensionality constraint) and channel frequency selectivity. The forms of MIMO settings that have been investigated are still limited to single-user MIMO and simplified multi-user MIMO systems. More advanced techniques are therefore demanded to alleviate the constraints imposed on the state-of-the-art. More sophisticated MIMO scenarios are yet to be explored to further corroborate the benefits of FBMC. In the context of multi-dimensional harmonic retrieval, it has been demonstrated that a higher estimation accuracy can be achieved by using tensors to preserve and exploit the multidimensional nature of the data, e.g., for model order estimation and subspace estimation. Crucial pending topics include how to further incorporate statistical robustness and how to handle time-varying scenarios in an adaptive manner. In Part I of this thesis, we first present an efficient and flexible transmission strategy for OFDM-based multi-user MIMO downlink systems. It consists of a spatial scheduling scheme, efficient multi-carrier ProSched (EMC-ProSched with an effective scheduling metric tailored for multi-carrier systems and two new precoding algorithms, linear precoding-based geometric mean decomposition (LP-GMD) and low complexity coordinated beamforming (LoCCoBF). These two new precoding schemes can be flexibly chosen according to the dimensions of the system. We also develop a system-level simulator where the parameters for the link-to-system level interface can be calibrated according to a certain standardization framework, e.g., IEEE 802.11ac. Numerical results show that the proposed transmission strategy, apart from guaranteeing the scheduling fairness and a small signaling overhead, achieves a much higher throughput than the state-of-the-art and requires a lower complexity. The remainder of Part I is dedicated to Filter Bank-based Multi-Carrier with Offset Quadrature Amplitude Modulation (FBMC/OQAM)-based MIMO systems. We begin with a thorough overview of FBMC. Then we present new transmit-receive processing techniques for FBMC/OQAM-based MIMO settings ranging from the single-user MIMO case to the Coordinated Multi-Point (CoMP) downlink considering various degrees of channel frequency selectivity. The use of widely linear processing is first investigated. A two-step receiver is designed for FBMC/OQAM-based point-to-point MIMO systems with low frequency selective channels. It exhibits a significant performance superiority over the linear MMSE receiver. The rationale in this two-step receiver is that the intrinsic interference is first mitigated to facilitate the exploitation of the non-circularity residing in the signals. It sheds light upon further studies on widely linear processing for FBMC/OQAM-based systems. Moreover, two coordinated beamforming algorithms are devised for FBMC/OQAM-based point-to-point MIMO systems to relieve the dimensionality constraint of existing schemes that the number of transmit antennas must be larger than the number of receive antennas. The channel on each subcarrier is assumed to be flat fading, which is categorized as the class of intermediate frequency selective channels. With the Channel State Information at the Transmitter (CSIT) known, the precoder designed based on a Zero Forcing (ZF) criterion or the maximization of the Signal-to-Leakage-plus-Noise-Ratio (SLNR) is jointly and iteratively computed with the receiver, leading to an effective mitigation of the intrinsic interference inherent in FBMC/OQAM-based systems. The benefits of the coordinated beamforming concept are successfully translated into the FBMC/OQAM-based multi-user MIMO downlink and the CoMP downlink. Three intrinsic interference mitigating coordinated beamforming (IIM-CBF) schemes are developed. The first two IIM-CBF schemes are proposed for FBMC/OQAM-based multi-user MIMO downlink settings with different dimensions and are able to effectively suppress the Multi-User Interference (MUI) as well as the intrinsic interference. A novel FBMC/OQAM-based CoMP concept is established via the third IIM-CBF scheme which enables the joint transmission of adjacent cells to the cell edge users to combat the strong interference as well as the heavy path loss and to boost the cell edge throughput. The performance of the proposed algorithms is evaluated via extensive numerical simulations. Their convergence behavior is studied, and the complexity issue is also addressed. In addition, the stronger resilience of FBMC over OFDM against frequency misalignments is demonstrated. Furthermore, we cover the case of highly frequency selective channels and provide solutions to the very challenging task of suppressing the MUI, the Inter-Symbol Interference (ISI), as well as the Inter-Carrier Interference (ICI) and supporting per-user multi-stream transmissions. Several design criteria of the multi-tap precoders are devised including the Mean Squared Error (MSE) minimization as well as the Signal-to-Leakage-Ratio (SLR) and SLNR maximization. By rendering a larger computational load at the base station, only single-tap spatial receive filters are required at the user terminals with a weaker computational capability, which enhances the applicability of the proposed schemes in real-world multi-user MIMO downlink systems. Part II focuses on the context of multi-dimensional harmonic retrieval. We demonstrate the incorporation of statistical robustness into multi-dimensional model order estimation schemes by substituting the sample covariance matrices of the unfoldings of the measurement tensor with robust covariance estimates. It is observed that in the presence of a very severe contamination of the measurements due to brief sensor failures, the robustified tensor-based model order estimation schemes lead to a satisfactory estimation accuracy. This philosophy of introducing statistical robustness also inspires robust versions of parameter estimation algorithms. Last but not the least, we present a generic framework for Tensor-based subspace Tracking via Kronecker-structured projections (TeTraKron) for time-varying multi-dimensional harmonic retrieval problems. It allows to extend arbitrary matrix-based subspace tracking schemes to track the tensor-based subspace estimate in an elegant and efficient manner. By including forward-backward-averaging, we show that TeTraKron can also be employed to devise real-valued tensor-based subspace tracking algorithms. Taking a few matrix-based subspace tracking approaches as an example, a remarkable improvement of the tracking accuracy is observed in case of the TeTraKron-based tensor extensions. The performance of ESPRIT-type parameter estimation schemes is also assessed where the subspace estimates obtained by the proposed TeTraKron-based subspace tracking algorithms are used. We observe that Tensor-ESPRIT combined with a tensor-based subspace tracking scheme significantly outperforms the combination of standard ESPRIT and the corresponding matrix-based subspace tracking method. These results open the way for robust multi-dimensional big data signal processing applications in time-varying environments.