Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

OpenDPD: An Open-Source End-to-End Learning & Benchmarking Framework for Wideband Power Amplifier Modeling and Digital Pre-Distortion

Yizhuo Wu, Gagan Deep Singh, Mohammadreza Beikmirza, Leo C. N. de Vreede, Morteza Alavi, Chang Gao

TL;DR

OpenDPD delivers an open-source end-to-end framework for rapid exploration and standardized benchmarking of digital pre-distortion methods for wideband power amplifiers. It introduces an end-to-end learning approach that backpropagates through a neural PA model to train a DPD, avoiding the need for a commutativity assumption in indirect learning architectures and optimizing $L_{PA}$ and $L_{CAS}$ under a target gain $G$. The Dense GRU–DPD (DGRU-DPD) architecture with a feature extractor achieves state-of-the-art ACPR and EVM on a 200 MHz OFDM DPA_200MHz benchmark, outperforming GMP, LSTM, GRU, VDLSTM, and RVTDCNN baselines. The public availability of OpenDPD code, datasets, and pretrained models aims to enhance reproducibility and foster broader contributions, with plans to incorporate additional PA datasets and backbones.

Abstract

With the rise in communication capacity, deep neural networks (DNN) for digital pre-distortion (DPD) to correct non-linearity in wideband power amplifiers (PAs) have become prominent. Yet, there is a void in open-source and measurement-setup-independent platforms for fast DPD exploration and objective DPD model comparison. This paper presents an open-source framework, OpenDPD, crafted in PyTorch, with an associated dataset for PA modeling and DPD learning. We introduce a Dense Gated Recurrent Unit (DGRU)-DPD, trained via a novel end-to-end learning architecture, outperforming previous DPD models on a digital PA (DPA) in the new digital transmitter (DTX) architecture with unconventional transfer characteristics compared to analog PAs. Measurements show our DGRU-DPD achieves an ACPR of -44.69/-44.47 dBc and an EVM of -35.22 dB for 200 MHz OFDM signals. OpenDPD code, datasets, and documentation are publicly available at https://github.com/lab-emi/OpenDPD.

OpenDPD: An Open-Source End-to-End Learning & Benchmarking Framework for Wideband Power Amplifier Modeling and Digital Pre-Distortion

TL;DR

OpenDPD delivers an open-source end-to-end framework for rapid exploration and standardized benchmarking of digital pre-distortion methods for wideband power amplifiers. It introduces an end-to-end learning approach that backpropagates through a neural PA model to train a DPD, avoiding the need for a commutativity assumption in indirect learning architectures and optimizing and under a target gain . The Dense GRU–DPD (DGRU-DPD) architecture with a feature extractor achieves state-of-the-art ACPR and EVM on a 200 MHz OFDM DPA_200MHz benchmark, outperforming GMP, LSTM, GRU, VDLSTM, and RVTDCNN baselines. The public availability of OpenDPD code, datasets, and pretrained models aims to enhance reproducibility and foster broader contributions, with plans to incorporate additional PA datasets and backbones.

Abstract

With the rise in communication capacity, deep neural networks (DNN) for digital pre-distortion (DPD) to correct non-linearity in wideband power amplifiers (PAs) have become prominent. Yet, there is a void in open-source and measurement-setup-independent platforms for fast DPD exploration and objective DPD model comparison. This paper presents an open-source framework, OpenDPD, crafted in PyTorch, with an associated dataset for PA modeling and DPD learning. We introduce a Dense Gated Recurrent Unit (DGRU)-DPD, trained via a novel end-to-end learning architecture, outperforming previous DPD models on a digital PA (DPA) in the new digital transmitter (DTX) architecture with unconventional transfer characteristics compared to analog PAs. Measurements show our DGRU-DPD achieves an ACPR of -44.69/-44.47 dBc and an EVM of -35.22 dB for 200 MHz OFDM signals. OpenDPD code, datasets, and documentation are publicly available at https://github.com/lab-emi/OpenDPD.
Paper Structure (11 sections, 3 equations, 5 figures, 1 table)

This paper contains 11 sections, 3 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. The OpenDPD Framework
  3. E2E Learning
  4. Benchmarking
  5. Dataset
  6. Comparison between DPD Models
  7. Dense GRU-based DPD
  8. Experimental Results
  9. Experimental Setup, Dataset and Training
  10. Results and Discussion
  11. Conclusion

Figures (5)

  • Figure 1: Generalized E2E learning architecture. With a DPA, an RF-DAC will be used instead of the DAC, Up-converter, and analog PA.
  • Figure 2: The DGRU-DPD architecture.
  • Figure 3: Setup for dataset acquisition and DPD performance measurement.
  • Figure 4: Training and measurement results on 200 MHz 10-channel$\times$20 MHz OFDM signals from the DPA_200MHz validation set. Each curve represents the best performance of each algorithm over 5 random seeds. (a) The 500-parameter PA modeling NMSE over training epochs. The gold PA model of each algorithm is saved at the lowest NMSE; (b) The 500-parameter DPD learning SIM-ACPR over training epochs. The gold DPD model of each algorithm is saved at the lowest averaged SIM-ACPR; (c) Measured ACPR vs. real-valued model parameters; (d) The 500-parameter DPD learning SIM-EVM over training epochs; (e) Measured EVM vs. real-valued model parameters.
  • Figure 5: Measured spectrum and constellation map on the 200 MHz Signal.