Multistage cascaded models for power amplifier digital predistortion : implementation and measurements

Abstract

The power amplifers (PAs) in basestations amplify and drive the signal to transmitting antenna. Most often the number of PAs in one basestation is significant. Massive multiple-input-multiple-output (MIMO) is a technique introduced in the fifth generation of mobile communications (5G) which enhances the spectral efficiency by enabling more simultaneous spatial streams between the basestation and user. Furthermore, the massive MIMO concept enables beamforming of the transmitted signal, in which the transmitter can shape the beam and focus the radiated energy to desired direction. There are vast range of massive MIMO topologies, including phased-array, hybrid-beamforming, and fully digital topologies. Moreover, the downside of massive MIMO transmitter is that large number of both antennas and PAs are required.

Significant portion of the basestation’s overall energy consumption comes from the PAs solely. Therefore, there’s a great interest in the industry to enhance efficiency of the PAs. For the PA to be power efficient, it needs to be operated in compression close to the saturation point, which in turn introduces undesired distortions to the transmitted signal, causing degradation of transmitted signal quality and poor spectral efficiency. To cope with the drawback of PA-induced distortions, digital predistortion (DPD) solution can be employed at the transmitter chain to compensate for the distortions induced by the PA. Signifcant factor to the linearization performance of the DPD is the choice of the behavioural model that the DPD implements. The behavioural model is mathematical mapping that tries to capture input-output relations of the PA. The more accurate the DPD model is, the better the linearization performance of the DPD will be. Unfortunately, in practical applications the DPD model has hardware resource limitations which limit the model size, and hence the modeling accuracy.

This thesis studies linearization performance of cascaded DPD models and proposes a way to identify the model coefficients and incorporate the additional stage to the existing adaptive DPD model. The linearization performance of generalized memory polynomial (GMP) model cascaded with linear finite impulse response (FIR) filter is demonstrated for Doherty laterally-diffused metal-oxide semiconductor (LDMOS) PA where single stage GMP DPD model lacks modeling accuracy, which can be seen as a inadequate performance in terms of adjacent channel power ratio (ACPR). The PA has 1805-1880 MHz bandwidth and the experiment is carried out utilizing entire 75 MHz bandwidth of the PA. The signal used through out the thesis was continuously allocated 5G multicarrier signal having 75 MHz composite bandwidth aggregated from three carrier signals each having bandwidth of 25 MHz. It is shown that linearization performance of GMP model cascaded with FIR filter was better than what was achievable with single stage GMP DPD model. The cascaded model was able to improve the measured ACPR by up to 4.44 dB when compared to the single stage GMP DPD model.