Digital adaptive calibration of data converters using independent component analysis

Abstract

The theory and practice of applying a neural network model and learning algorithm—Independent Component Analysis (ICA)—to the online adaptive calibration of analog-to-digital converters (ADCs) is covered in this chapter. Exploiting the independence between the input signal and an injected pseudorandom bit sequence (PRBS), the technique attempts to blindly separate the two in the digital conversion output, and while doing so, an equivalent model of the ADC non-idealities is identified, resulting in the subsequent linearization of the conversion process. The ICA framework offers new signal-processing insights into the widely used correlation-based error-parameter identification method for the background calibration of multistage ADCs. In addition, it provides a useful technique to minimize the analog overhead associated with the calibration by simultaneously identifying multiple model parameters using a single PRBS, improving the efficiency and potentially the application regime of the online calibration approach for data converters.