A Software Level Calibration Based on Bayesian Regression for a Successive Stochastic Approximation Analog-to-Digital...

Area of Science:

• Electrical Engineering
• Signal Processing
• Machine Learning

Background:

• A novel low-power, high-precision successive stochastic approximation analog-to-digital converter (ADC) has been developed.
• This ADC produces two distinct outputs requiring a software-based error correction method for a unified high-precision output.

Purpose of the Study:

• To propose a practical software-level error correction method for the successive stochastic approximation ADC.
• To enhance the precision of the ADC by effectively combining its dual outputs.

Main Methods:

• A Bayesian regression approach with incremental learning is proposed.
• Data points are successively selected based on the uncertainty of their predicted total output.
• Uncertainty is estimated using the upper bound of standard deviations from Bayesian predictive distributions within data blocks.

Main Results:

• Numerical experiments demonstrate the effectiveness of the proposed error correction method.
• The method successfully combines dual outputs to achieve higher precision.

Conclusions:

• The proposed Bayesian regression with incremental learning offers a practical and effective solution for error correction in successive stochastic approximation ADCs.
• This method enhances ADC performance by adaptively refining data selection based on predictive uncertainty.