Digital Predistortion Solutions for Large Antenna Array Transmitters

Power amplifiers (PAs) are the most power-hungry components in base stations (BSs), and thus they significantly contribute to the overall energy consumption and power efficiency of cellular networks. Unfortunately, PAs need to operate close to their compression or saturation region in order to be power efficient, which heavily distorts the transmit signals and other users or systems operating in adjacent frequency bands. Since such distortions can significantly compromise the spectral efficiency of networks, digital predistortion (DPD) solutions are commonly adopted at transmitters (TXs) to compensate for the PA-induced distortions. DPD is a well established and known technology in the context of legacy narrowband transmitters with a few antennas, however, the fifth generation of mobile communications (5G) introduces a number of technical developments that require a top-to-bottom redesign of traditional digital predistorters. Particularly, massive multiple-input-multiple- output (mMIMO) is a technique that makes it possible to multiplex a large number of users in the spatial domain, so that more information can be squeezed into the already limited and congested electromagnetic spectrum. Furthermore, it allows a more efficient utilization of the radiated energy, by synthesising very-sharp beams that focus the electromagnetic waves in the directions of interest, while minimizing interference elsewhere. However, mMIMO, in its large variety of topologies, imposes many new challenges to the design and implementation of DPD solutions, which if not addressed properly, can seriously compromise the efficient utilization of PAs in these transmitter architectures, leading to increased carbon footprint, to increased costs for network operators, and to more complex BS designs. This thesis presents a number of contributions in the field of digital predistortion, particularly for large antenna array or mMIMO transmitters, to address some of the major challenges encountered in their linearization, so that spectral and energy-efficient transmitters can be utilized in the present and future generation of cellular networks. The first contribution of this thesis adds to the foundation of the fundamental understanding of the effective linearization of phased-array and hybrid MIMO transmitters. To that end, mathematical models of the radiated nonlinear distortion are derived, followed by developing efficient beam-based predistortion and learning principles under different beamforming configurations and propagation scenarios. Particularly, it is shown that the composite over-the-air (OTA) combined distortion in the main-beam(s) can be effectively minimized by means of one-dimensional digital predistorters. Closed-loop-based DPD structures are adopted, which are more resilient to hardware impairments in the observation receivers and to crosstalk, together with reduced complexity gradient-based adaptive algorithms. Spatial-domain analyses of the radiated distortions are provided, revealing the rationale behind the cancellation of nonlinear distortion for every particular scenario. Comprehensive numerical experiments are provided, substantiating the different concepts. The second contribution is the development of a number of more advanced digital predistortion techniques that allow the linearization of practical active mmWave phased-array transmitters subject to challenging impairments. These solutions particularly deal with the compensation of strong and beam-dependent nonlinearities, while favouring reduced complexity processing. To that end, a closed-loop decorrelation-based piece-wise DPD, a multi-beam oriented neural-network-based DPD, and reduced complexity frequency-domain DPD processing were proposed. Extensive radio-frequency (RF) measurements demonstrate that the different solutions can provide excellent linearization performance, while dealing with very challenging hardware impairments in an efficient manner. The third contribution of this thesis is the development of stream-/beam-level predistorters for fully-digital MIMO architectures. Different DPD structures are proposed to enable the linearization of such MIMO transmitters through a single and common predistorter. By doing so, the need for deploying a dedicated DPD per PA/antenna unit is avoided, thus allowing for reduced complexity linearization of the whole transmitter. Different solutions are proposed for single-carrier and orthogonal frequency division multiplexing (OFDM) based transmitters, with single-user and multi-user transmission. Overall, the proposed methods and results provide clear advancements in the DPD technology area, beyond the current state-of-the-art