Low-Complexity Digital Predistortion for 5G Massive MIMO and Handset Transmitters

The demand for new wireless communication systems to support high mobility and low latency necessitates a rethink of the architecture of wireless communication systems as well as the design of their key components. This thesis presents several novel techniques to solve the major challenges in digital predistortion (DPD) for millimeter wave multi-input multi-output (MIMO) and handset transmitters to lower the hardware cost and computational complexity of the fifth generation (5G) communication systems. The first part of the thesis focuses on the architecture of the MIMO DPD solution for 5G transmitters. To extract DPD model coefficients, a feedback data acquisition path is required. In conventional single-input single-output (SISO) systems, the output is usually acquired directly from the power amplifier (PA) with a coupler. In massive MIMO systems, the number of RF chains is large. Using dedicated feedback paths for each PA separately is not feasible. To lower the hardware cost, a novel data acquisition scheme is proposed to obtain the output signals in far field over the air (OTA) using a single antenna and feedback loop, and then reconstruct the output of each PA. Simulation and experimental results demonstrate that the proposed OTA data acquisition can accurately reconstruct the output of each PA in the MIMO systems and the DPD solutions derived from the reconstructed data can successfully linearize the nonlinear MIMO transmitters. In the multi-user scenario, the nonlinearity of the transmitters varies with the movement of user equipments (UEs), and the DPD model coefficients need to be updated accordingly. To meet the requirement of high mobility, the complexity of the system update must be low. In the second part of the thesis, we present a new DPD system, where DPD model can be updated fast and accurately without capturing PA output or applying costly model extraction algorithms. In the proposed method, nonlinear characteristics of the PA are encoded into low-dimensional PA features using feature extraction algorithms. To identify DPD model coefficients, PA features are extracted first and the DPD model coefficients are then generated directly by DPD generator with PA features. Experimental results show that the proposed DPD solution can linearize PA with very low complexity compared to that using the conventional solutions. Finally, the focus shifts to handset transmitters. Conventionally, DPD is usually deployed for high power base stations. With the continuously increasing bandwidth, DPD may also be required for handset PAs in 5G communication systems. Different from the models used for base stations, DPD model for handset PAs must have very low complexity because of the stringent power budget limit. At the same time, the tolerance for load mismatch must also be considered. The third part of the thesis analyzes the characteristics of handset PAs with load mismatch and introduces a low-complexity DPD model based on magnitude-selective affine (MSA) function. Experimental results demonstrate that the extended MSA (EMSA) model shows better linearization performance while keeping much lower complexity than the conventional DPD models.