Abstract
This article presents the design and analysis of a millimeter-wave (mmWave) four-way Doherty power amplifier (PA), aiming to enhance the PA energy efficiency when amplifying orthogonal frequency-division multiplexing (OFDM)-based 5G new radio (NR) signals with a 10–12-dB peak-to-average power ratio (PAPR). We first introduce a systematic approach to extending a conventional two-way Doherty PA to $N$ ways, followed by a new transformer-based $N$ -way Doherty network synthesis flow. The proposed network synthesis achieves $N$ -way Doherty load modulation using ( $N-1$ ) transformers, one fewer transformer and thus lower loss than conventional designs. In addition, it enables the desired impedance transformation from $R_{\text{ANT}}$ to $R_{\text{OPT}}$ and effectively absorbs the parasitic capacitance of the power cells. Along with the Doherty network, we also introduce a high-speed adaptive biasing circuit, addressing the modulation bandwidth bottleneck in prior Doherty PA demonstrations. As proof of concept, a four-way Doherty PA prototype is implemented in the 47-GHz 5G band (band n262) using the GlobalFoundries 45-nm CMOS silicon-on-insulator (SOI) process. It achieves 24.0-dBm saturated power ( $P_{\text{SAT}}$ ), 23.7-dBm output 1-dB compression point ( $P_{1\,\text{dB}}$ ), 26.8% peak power-added efficiency ( $\text{PAE}_{\text{PEAK}}$ ), 26.3% PAE at $P_{1\,\text{dB}}$ ( $\text{PAE}_{1\,\text{dB}}$ ), 21.7% PAE at 6-dB back-off ( $\text{PAE}_{6\,\text{dB}}$ ), and 13.1% PAE at 12-dB back-off ( $\text{PAE}_{12\,\text{dB}}$ ), demonstrating state-of-the-art performance. In the modulation tests, the PA achieves 14.1-dBm average output power ( $P_{\text{avg}}$ ) and 13.7% average efficiency ( $\text{PAE}_{\text{avg}}$ ) when amplifying a 2000-MHz 5G NR 64-QAM OFDM signal. To the best of our knowledge, this is the first silicon PA demonstration of 2000-MHz channel modulation bandwidth for 5G NR OFDM along with back-off efficiency enhancement up to 12-dB back-off.

This publication has 52 references indexed in Scilit: