Power Amplifier Behavioral Modeling Using Bio-Inspired Multi-Resolution Dynamic Neural Network

Power amplifiers (PAs) are key components in modern consumer wireless electronics, where compact design, high power efficiency, and excellent linearity are required to guarantee transmission quality under stringent cost and size constraints. However, the nonlinear and time-varying characteristics of PAs, especially in wideband 5G and emerging 6G systems, pose significant challenges for accurate behavioral modeling. Traditional data-driven models often lack robustness and adaptability when deployed in resource-limited consumer devices. To address these limitations, this paper proposes a bio-inspired multi-resolution dynamic neural network (MRDNN) for efficient and precise PA behavioral modeling. Drawing inspiration from biological adaptive perception, the MRDNN integrates a multi-resolution feature fusion mechanism to jointly capture short-term nonlinearities and long-term memory effects, together with a liquid-state dynamic core that adapts to time-varying signal environments. This hierarchical architecture achieves high modeling accuracy with low computational complexity, enabling hardware-efficient deployment in real-world communication systems. Experimental results on multiple PA datasets demonstrate that the MRDNN outperforms state-of-the-art neural models, improving the adjacent channel error power ratio (ACEPR) by up to 3.5 dB. These findings highlight the promise of biologically inspired dynamic learning for next-generation energy-efficient and intelligent RF systems in consumer electronics. The source code is publicly available at GitHub.