A novel technique for the digital background calibration of time-interleaved analog-to-digital converters is proposed. The technique corrects at the same time for both errors due to gain, offset and timing mismatches among the time-interleaved channels and errors due to nonlinearities in the channels, for instance due to capacitor mismatches in switched capacitor implementations. This feature, together with the use of the recursive least mean squares algorithm, makes the technique particularly fast (12 bits of accuracy can be achieved after about 4000 samples for a two-channel converter). The proposed calibration technique employs wideband differentiators, thus enabling digital background calibration of timing skews even with wideband input signals. Besides, undersampled differentiator filters are proposed, and it is shown that the technique is capable of calibrating undersampling converters by estimating the derivative of wideband input signals even outside the first Nyquist band.