Deep Learning-Driven Black-Box Doherty Power Amplifier with Pixelated Output Combiner and Extended Efficiency Range

Abstract

This article presents a deep learning-driven inverse design methodology for Doherty power amplifiers (PA) with multi-port pixelated output combiner networks. A deep convolutional neural network (CNN) is developed and trained as an electromagnetic (EM) surrogate model to accurately and rapidly predict the S-parameters of pixelated passive networks. By leveraging the CNN-based surrogate model within a blackbox Doherty framework and a genetic algorithm (GA)-based optimizer, we effectively synthesize complex Doherty combiners that enable an extended back-off efficiency range using fully symmetrical devices. As a proof of concept, we designed and fabricated two Doherty PA prototypes incorporating three-port pixelated combiners, implemented with GaN HEMT transistors. In measurements, both prototypes demonstrate a maximum drain efficiency exceeding 74% and deliver an output power surpassing 44.1 dBm at 2.75 GHz. Furthermore, a measured drain efficiency above 52% is maintained at the 9-dB back-off power level for both prototypes at the same frequency. To evaluate linearity and efficiency under realistic signal conditions, both prototypes are tested using a 20-MHz 5G new radio (NR)-like waveform exhibiting a peak-to-average power ratio (PAPR) of 9.0 dB. After applying digital predistortion (DPD), each design achieves an average power added efficiency (PAE) above 51%, while maintaining an adjacent channel leakage ratio (ACLR) better than -60.8 dBc.

Figures (12)

• Figure 1: Overview of the proposed deep learning–driven approach for the synthesis and design of black-box Doherty PAs with symmetrical size and extended back-off efficiency range. (a) Analytical black-box approach for deriving combiner parameters from load-pull data. (b) Deep learning–based top-down approach employing a pixelated Doherty combiner layout to explore the full design space.
• Figure 2: Illustration of the black-box technique under different driving conditions. The output power is normalized to the corresponding saturated output power for each current‑scaling ratio. (a) Normalized current drive profiles as a function of input voltage. (b) Theoretical performance comparison for different drive profiles (Cases I–III). (c) Corresponding current ratios and back-off levels for each case.
• Figure 3: Architecture of the trained deep CNN with residual connections. The input is a binary $15 \times 15$ matrix representing the pixelated EM layout structure, and the output is the predicted real and imaginary components of the corresponding S-parameters across the frequency range of interest.
• Figure 4: The employed pixelated Doherty combiner networks, where the feed locations for the main (port $1$), auxiliary (port $2$), and output (port $3$) are placed at the center of each corresponding edge of the $15$-pixel array. The output port $3$ is connected to the load ($R_{\mathrm{L}}$), while port $4$ is left open.
• Figure 5: The two synthesized pixelated Doherty combiner networks, (a) and (b), along with their EM-simulated S-parameter results compared to the corresponding responses predicted by the deep learning approach, over the frequency range of $2.55$–$2.95~\mathrm{GHz}$.