Accelerating Model Adaptation of Multi-Metric Digital Predistortion for RF Power Amplifiers Using Composited Quadratic Loss Function

The conventional model extraction of digital predistortion (DPD) for RF power amplifiers is usually based on minimizing errors in the time domain, such as the normalized mean square error. In the forthcoming systems, not only time-variant, but also multi-metric requirements in both the time and frequency domains will be posed, such as error vector magnitude and spectrum mask. Correspondingly, the DPD coefficients must be fast adapted, so as to satisfy the multiple requirements in a timely manner. In this paper, a composited quadratic loss function is proposed, so that the DPD coefficients can converge to the desired values to meet the multi-metric requirements with fewer iterations, when using the quadratic simultaneous perturbation stochastic approximation (SPSA) algorithm. The validation on a sequential load modulated balanced amplifier indicates that the proposed method is promising to be employed in the emerging scenarios.

I. Introduction

The state-of-the-art or forthcoming communication standards, such as 5G New Radio and IEEE 802.11 be, will introduce multiple performance requirements, e.g., the error vector magnitude (EVM) in the time domain and spectrum mask in the frequency domain [1], [2]. The multiple requirements may change from time to time depending on the real-time system conditions, e.g., lower modulation will be employed when the channel quality deteriorates, and spectrum mask requirement can be relaxed when no device or service is using the adjacent channel. Consequently, at the transmitter side, or more specifically, for the power amplifier (PA), multi - metric requirements will be posed. It is known that digital predistortion (DPD) is used to keep PA operating with high efficiency without losing linearity, as a result, multiple requirements will also be posed on DPD metrics, such as the normalized means square error (NMSE) and the adjacent channel power ratio (ACPR).