Early Stopping Criterion for Recursive Least Squares Training of Behavioural Models

Area of Science:

• Electrical Engineering
• Signal Processing
• Wireless Communications

Background:

• Rapid evolution of wireless communications necessitates complex radio frequency (RF) transceiver architectures.
• Power amplifiers (PAs) are critical components in RF transceivers, exhibiting nonlinear behavior that causes signal distortion.
• Behavioral models are essential for simulating the performance of nonlinear PAs, but adaptive training methods like Recursive Least Squares (RLS) can suffer from coefficient instability.

Purpose of the Study:

• To develop a computationally efficient technique for detecting the onset of instability during adaptive RLS training of PA behavioral models.
• To provide a mechanism to cease training dynamically when instability is detected, thereby avoiding model convergence issues.
• To enhance the reliability and accuracy of behavioral models for nonlinear power amplifiers in advanced wireless systems.

Main Methods:

• A novel technique is proposed to monitor the autocorrelation function update during adaptive RLS training.
• The method detects the onset of coefficient instability without modifying the core RLS algorithm.
• The technique's effectiveness is validated through experimental testing with various modulation schemes.

Main Results:

• The proposed technique successfully identifies the point of instability in RLS training for PA behavioral models.
• It enables timely cessation of training, preventing the development of unstable model coefficients.
• Experimental validation confirmed the technique's efficacy across LTE OFDM, 5G-NR, DVB-S2X, and WCDMA signals.

Conclusions:

• The presented method offers a computationally efficient and non-intrusive way to manage instability in adaptive RLS training for PA behavioral models.
• This approach contributes to more robust and reliable RF transceiver design for current and future wireless communication standards.
• The technique ensures the integrity of dynamic memory polynomial-based behavioral models, crucial for accurate performance simulation.