Blind and reference channel-based time interleaved ADC calibration schemes: a comparison

Abstract

Many digital background calibration techniques exist which correct for offset, gain, timing and bandwidth mismatches in time-interleaved (TI) ADCs. Some require an additional reference channel, whereas others are blind and rely on the presence of a band where no signal is present (usually around the Nyquist frequency) or exploit other properties of the input signal. Blind calibration techniques, which don’t use a reference channel, are suitable for correction of commercial TI-ADCs, or TI-ADC systems using commercial ADCs as channels. Techniques employing additional channels require a more complex layout (especially for the clock tree) and need an additional ADC, whose overhead cost is significant, especially for 2- or 4- channel TI-ADCs. However, we show that the estimation process is faster and more accurate when a reference channel is present, and many different error models can be used (exploiting different points in the accuracy / complexity trade-off).