Digital predistortion for power amplifier linearization in sparse multi-user systems

Doctoral thesis, 2026

To meet demands on higher data rates, ubiquitous coverage and a good quality of service in wireless communications, the trend in construction of wireless transmitters is towards multiple-input, multiple-output (MIMO) designs. In the transmitter, the power amplifier (PA) is traditionally the most power consuming component. These PAs are always designed with a trade-off between power consumption and nonlinearity, from which nonlinear distortion can arise that deteriorates performance if not addressed.

In this thesis, digital pre-distortion (DPD) adaptions in various multi-user linearization scenarios are explored. With MIMO and multi-user transmitters, traditional single-input, single-output (SISO) techniques for linearization are often superfluous and inflexible in their computational complexity, and thereby excessive in their energy consumption. With the additional degrees of freedom, and oftentimes sparsity, available in multi-user and MIMO transmitters, new techniques can achieve sufficient linearization at a much lower computational cost.

In Paper A, sparse concurrent multi-band transmission using a technique known as frequency relocation is studied, expanding upon previous research by considering PA gain and phase variations through linear pre-equalization filters. The combination of techniques allows for sample rate reductions in the digital-to-analog converters, resulting in reduced power consumption, whilst achieving comparable results to conventional SISO linearization. 

In Paper B, concurrent multi-beam transmission in MIMO systems is explored. It is then shown that for a sparse beam space, i.e. when there are few beams to be transmitted compared to the number of transmit branches, a dimensionality reduction can be performed in a in a so-called virtual array, allowing few DPDs to linearize many transmit branches. 

In Paper C, spatial sparsity is utilized for so-called spatial and frequency distortion shaping. This paper presents a technique, and a new metric, that for a fixed DPD complexity allows for improving linearization performance in selected spatial regions and for particular frequencies, at the expense of shaping nonlinear distortion to end up in directions deemed less important.

Using the techniques proposed in these papers, good communication quality can be maintained whilst operating PAs more efficiently and whilst scaling to more difficult multi-user cases.