Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MP-DPD: Low-Complexity Mixed-Precision Neural Networks for Energy-Efficient Digital Predistortion of Wideband Power Amplifiers

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

TL;DR

This work tackles the high energy cost of digital pre-distortion (DPD) in wideband RF power amplifiers by introducing mixed-precision neural networks (MP-NNs) with quantized fixed-point parameters. A GRU-based DPD model is trained with quantization-aware training to balance accuracy and energy, retaining FP32 feature extraction while quantizing weights/activations to 8–16 bits. The MP-DPD achieves ACPR and EVM comparable to FP32 baselines on a 160 MHz, 1024-QAM OFDM signal, while delivering up to about 4–5× energy savings per inference and significant overall power reductions (e.g., 2.8× for certain configurations). These results, demonstrated on 7 nm/45 nm tech nodes with detailed energy metrics, suggest MP-DPD as a practical pathway to energy-efficient, scalable DPD for next-generation wireless systems, with open-source OpenDPD integration for reproducibility.

Abstract

Digital Pre-Distortion (DPD) enhances signal quality in wideband RF power amplifiers (PAs). As signal bandwidths expand in modern radio systems, DPD's energy consumption increasingly impacts overall system efficiency. Deep Neural Networks (DNNs) offer promising advancements in DPD, yet their high complexity hinders their practical deployment. This paper introduces open-source mixed-precision (MP) neural networks that employ quantized low-precision fixed-point parameters for energy-efficient DPD. This approach reduces computational complexity and memory footprint, thereby lowering power consumption without compromising linearization efficacy. Applied to a 160MHz-BW 1024-QAM OFDM signal from a digital RF PA, MP-DPD gives no performance loss against 32-bit floating-point precision DPDs, while achieving -43.75 (L)/-45.27 (R) dBc in Adjacent Channel Power Ratio (ACPR) and -38.72 dB in Error Vector Magnitude (EVM). A 16-bit fixed-point-precision MP-DPD enables a 2.8X reduction in estimated inference power. The PyTorch learning and testing code is publicly available at \url{https://github.com/lab-emi/OpenDPD}.

MP-DPD: Low-Complexity Mixed-Precision Neural Networks for Energy-Efficient Digital Predistortion of Wideband Power Amplifiers

TL;DR

This work tackles the high energy cost of digital pre-distortion (DPD) in wideband RF power amplifiers by introducing mixed-precision neural networks (MP-NNs) with quantized fixed-point parameters. A GRU-based DPD model is trained with quantization-aware training to balance accuracy and energy, retaining FP32 feature extraction while quantizing weights/activations to 8–16 bits. The MP-DPD achieves ACPR and EVM comparable to FP32 baselines on a 160 MHz, 1024-QAM OFDM signal, while delivering up to about 4–5× energy savings per inference and significant overall power reductions (e.g., 2.8× for certain configurations). These results, demonstrated on 7 nm/45 nm tech nodes with detailed energy metrics, suggest MP-DPD as a practical pathway to energy-efficient, scalable DPD for next-generation wireless systems, with open-source OpenDPD integration for reproducibility.

Abstract

Digital Pre-Distortion (DPD) enhances signal quality in wideband RF power amplifiers (PAs). As signal bandwidths expand in modern radio systems, DPD's energy consumption increasingly impacts overall system efficiency. Deep Neural Networks (DNNs) offer promising advancements in DPD, yet their high complexity hinders their practical deployment. This paper introduces open-source mixed-precision (MP) neural networks that employ quantized low-precision fixed-point parameters for energy-efficient DPD. This approach reduces computational complexity and memory footprint, thereby lowering power consumption without compromising linearization efficacy. Applied to a 160MHz-BW 1024-QAM OFDM signal from a digital RF PA, MP-DPD gives no performance loss against 32-bit floating-point precision DPDs, while achieving -43.75 (L)/-45.27 (R) dBc in Adjacent Channel Power Ratio (ACPR) and -38.72 dB in Error Vector Magnitude (EVM). A 16-bit fixed-point-precision MP-DPD enables a 2.8X reduction in estimated inference power. The PyTorch learning and testing code is publicly available at \url{https://github.com/lab-emi/OpenDPD}.
Paper Structure (10 sections, 5 equations, 4 figures, 1 table)

This paper contains 10 sections, 5 equations, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. The DPD Computing's Energy Problem
  3. Mixed-Precision Neural Networks DPD
  4. Gated Recurrent Unit-based DPD
  5. Mixed-Precision DPD
  6. Experimental Results
  7. Experimental Setup
  8. Results and Discussion
  9. Power Consumption Comparison to Prior Works
  10. Conclusion

Figures (4)

  • Figure 1: (a) The Von Neumann architecture with energy costs. (b) Operation and 8KB SRAM access energy in 45 nm horowitz20141 and 7 nm jouppi2021ten vs. precision.
  • Figure 2: Setup for dataset acquisition and DPD performance measurement.
  • Figure 3: Parameter scan of DPD models vs. (a) ACPR (left) (b) ACPR (right) (c) EVM; (d) EVM (left Y-axis) and energy per inference (right Y-axis) vs. precision.
  • Figure 4: Measured spectrum and constellation map on the 160 MHz signal.