Signal reconstruction in distributed sampling systems with application to time-interleaved A/D converters

Abstract

High-speed distributed sampling systems, such as time-interleaved analog-to-digital converters, are becoming increasing popular for many modern applications. Calibration is a serious challenge in the design of such systems. Indeed, component variations and layout constraints lead to gain mismatches and timing misalignment. Previous methods for calibration have used training signals or expensive circuitry to overcome these problems. In this work, we investigate alternative approaches for signal recovery. In particular, we develop blind calibration techniques that focus on the estimation of the associated unknown gain and timing offset parameters, from which the calibrated signal is reconstructed. Estimation algorithms are considered for both deterministic and random input signal models. For deterministic signals, a least-squares estimation of the input is examined; for random signals, Maximum Likelihood (ML) parameter estimates are obtained via the Expectation-Maximize (EM) algorithm. Tradeoffs in reconstruction quality between varying calibration times, oversampling factors, and number of converters are developed. Overall, the tradeoffs achieved achieved appear promising for practical applications.