A low computation adaptive blind mismatch correction for time-interleaved ADCs

We introduce a computationally very efficient technique for adaptive blind correction of M-channel time-interleaved analog-to-digital converters (TIADC). Under wide-sense stationary (WSS) and bandlimited input assumption, gain and timing errors are estimated in the digital domain by detecting shift-dependence of TIADC output autocorrelation. Input signal integrity is preserved since there is no analog input pre-processing. Gain mismatch is digitally corrected, but sampling time mismatches are compensated in the analog domain by adjusting individual clock timing offsets. This mixed-domain technique takes advantage of each domain to dramatically reduce computational requirement on the digital side. As a by-product, convergence is guaranteed under mild conditions with arbitrary number of channels. Experimental demonstration is performed by a proof-of-concept, M=4 400-MSPS real-time setup. After 300 iterations, mismatch spurs are suppressed by more than 40-dB, down to ∼80dB below the signal at ∼171MHz. This is the first proposal and demonstration of low-computation blind technique with guaranteed convergence.