To preserve the effective resolution of the individual converters in time-interleaved analog-to-digital converters (TIADCs), channel mismatches should be carefully corrected. In this paper, the proposed method is concentrated on achieving adaptive blind calibration for nonlinearity mismatches in twochannel TIADCs. It utilizes a least-mean square (LMS) algorithm as well as a polynomial-represented nonlinearity model. We assume a certain degree of oversampling for the overall converter to estimate and compensate for the mismatch errors. The calibration accuracy and convergence rate are investigated and evaluated through behavioral-level simulations.
I. INTRODUCTIONTo satisfy the fast-growing requirement of sampling system, Black and Hodges proposed a sampling architecture which consists of multiple parallel converters operating in a circulatory manner, i.e., the time-interleaved ADCs (TIADCs) [1]. However the effective resolution of the sub-ADC is unfortunately not maintained in the overall sampling system due to mismatches between the parallel channels. During the last decades, much effort has been attracted on designing the correction methods for linear mismatches (e.g., DC-offset, gain, and time-skew mismatches) [2]- [5]. Whereas, only a few contributions have addressed the calibration method for nonlinearity mismatches [6]-[8], which can be easily introduced by the imperfection of the analog front-end circuits as well as the differential and integral nonlinearity (DNL and INL) of the sub-ADCs. In [6], a compensation method was proposed by using a randomization strategy, where the additional converters would increase the hardware cost. Training signals could be adopted to identify the mismatch errors for the post compensation [7], [8], which however would interrupt the normal operation and is incapable of tracking the mismatch variation.Motivated by the method in [9], we use a least-mean square (LMS) algorithm to blindly calibrate nonlinearity mismatch errors in two-channel TIADCs. In this paper, a certain degree of oversampling is assumed for the overall interleaved converter rather than for each individual converter. It means that by using the equivalent computational resource, this method can improve the bandwidth utilization twice compared to the method in [9]. Further, this method can save the amount of multiplications due to that it directly calibrates the mismatch errors rather than successively estimate and compensate.The outline of this paper is as follows. In Section II, we analyze the two-channel TIADC model with nonlinearity