Low-complexity Algorithms for Digital Predistortion and Digital Self-interference Cancellation

Judging from the massive impact wireless communications made during the past decades, there is little doubt that it will still be one of the major foundations of contemporary society also in the future. In the context of modern radio networks, two main goals are currently pursued. First, a more efficient usage of the extremely scarce radio spectrum, with the aim of really exploiting the usable frequency ranges. Second, an improved cost-effective trade-off of the physical transmitters, with the aim of minimizing the size and cost of the involved circuitry, and also to facilitate more power-efficient solutions which will reduce the overall expenses of wireless networks and also aid in the reduction of carbon emissions. In order to improve the system spectral efficiency and transmitter energy efficiency, digital predistortion (DPD) and digital self-interference (SI) cancellation techniques have been widely proposed in the literature, however, research in associated low-complexity solutions is more scarce. With reduced-complexity algorithms being a new trend in 5G and beyond networks, it becomes crucial to explore solutions which minimize the computational complexities that are involved. This claim is motivated by the following three trends. First, the utilized signal bandwidths are becoming wider and wider in emerging networks, thus also increasing DPD processing rates. Second, it is a clear new preference to have a smaller base station (BS) configuration, thus also the available power budget dedicated for front-end digital processing techniques is reduced. Third, many solutions require fast adaptation of the model coefficients, such as the SI channel in in-band full duplex (IBFD), or the DPD coefficient estimation in mmW beam-steered antenna array systems. To this end, this thesis firstly concentrates on providing several low-complexity modeling solutions which are applicable for both DPD and IBFD applications. In particular, four cascaded structures are proposed and combined with a complex injection-based spline-interpolation scheme, which is able to provide accurate modeling of the PA-induced nonlinearity. Additional memory effects are modeled by elementary linear time-invariant adaptive filters. The cascaded structures also incorporate efficient gradient-descent-based algorithms to adaptively update the model parameters. Furthermore, a look-up table-based memory polynomial (MP) structure is pro- posed, which alternatively models the memory through a MP-like parallel branched structure. This approach provides richer modeling capabilities, and it does not re- quire any physical knowledge of the system under study. In the context of DPD, three reduced-complexity signed closed-loop techniques are devised to reduce the complexity of the learning path. Additionally, different inverse covariance matrix estimation methods are presented in the context of a self-orthogonalized learning system, which reduce its computational complexity even further. Secondly, this thesis presents exhaustive and comprehensive RF verification and validation, which, together with detailed complexity analyses, allow assessment of the performance-complexity trade-offs of the proposed methods. In the context of DPD, the proposed solutions are tested with two different frequency range 1 (FR- 1) and frequency range 2 (FR-2) RF measurement environments, considering also different off-the-shelf PA systems. In all cases, it is shown how the proposed methods enhance the energy efficiency of the transmitter, while keeping the unwanted in- band and out-of-band distortion under the levels specified in the latest 3GPP 5G NR Release 15. In the context of IBFD, corresponding evaluations are carried out with two actual real-life IBFD prototypes, showing that the proposed techniques achieve a similar cancellation to other state-of-the-art techniques, regardless of the drastic complexity reductions. Altogether, the strong mathematical foundations of the developed solutions, together with the obtained results, show that more efficient yet reliable transmitters can be achieved through the proposed DPD methods, allowing for the minimization of cost and size of the circuitry involved, and improving the energy efficiency of the transmitter. Additionally, this thesis verifies the commercial feasibility of IBFD, considering that the SI is suppressed. SI suppression is achieved by the proposed methods, which bring IBFD one step closer to commercially feasible implementations