High-Speed Multibit Delta-Sigma ADCs with On-Line Digital Error Correction

Multibit delta-sigma modulators are widely used in recent analog-to-digital converter (ADC) implementations [1, Chapter 8]. Due to their improved stability, they can achieve the same high signal-to-noise ratio (SNR) with lower order modulator and lower oversampling ratio (OSR), i.e. higher speed, than their singlebit counterparts. Also, the use of a multibit feedback relaxes slew-rate and settling-time requirements for the opamps used in the integrators [2], [3]. However, single-loop multibit delta-sigma ADCs impose stringinent requirements for the feedback multibit digital-to-analog converter (DAC), which should provide the same linearity as the overall converter. Several techniques has been used to address the linearity of the multibit DAC. Mismatch shaping randomizes the nonlinearity error, and high-pass filters the resulting uncorrelated error (“noise”) signal [1, Section 8.3.3]. Although this technique is very popular, it is effective at high OSRs only, and it requires fairly large chip area, especially for quantizers larger than 5 bits. Calibration techniques were successfully implemented, and impressive measured performances were obtained [3], [4]. However, the previous calibration techniques work off line, so they may not follow the enviromental changes. Recently, a background calibration technique was presented [5], but which requires high OSR (Note: it requires high OSR as it was presented; NLTFd(z) filter is missing, see details later). An analog alternative, i.e. analog mismatch corrected switched-capacitor DAC, shows a promising solution to the problem [6], [7]. This report presents an on-line digital correction method for multibit delta-sigma ADCs. It originates from [2], [3], but it works on line, and it is effective at low-OSR, i.e. high-speed, applications also. While the mismatch-shaping technique spreads and shapes the nonlinearity error (“dirt”), this digital correction cancels it out (“vacuum-cleaning” versus “sweeping”), so it is more effective. A practical implementation using the proposed technique may achieve a spurious-free dynamic range (SFDR) of over 100 dB for OSR = 4, based on available simulation [8] results.