Novel techniques for multi-bit oversampled data conversion are described. State-of-the-art oversampled data converters are analyzed, leading to the conclusion that their performance is limited mainly by low-resolution signal representation. To increase the resolution, high-performance, high-resolution internal D/A converters are required. Unit-element mismatch-shaping D/A converters are analyzed, and the concept of mismatch-shaping is generalized to include scaled-element D/A converters. Several types of scaled-element mismatch-shaping D/A converters are proposed. Simulations show that, when implemented in a standard CMOS technology, they can be designed to yield 100 dB performance at 10 times oversampling. The proposed scaled-element mismatch-shaping D/A converters are well suited for use as the feedback stage in oversampled delta-sigma quantizers. It is, however, not easy to make full use of their potential, because that requires a high-resolution loop quantizer which introduces only a small delay. Generally, it is not acceptable to design the loop quantizer as a high-resolution flash quantizer because they require a large chip area and high power consumption. Pipeline techniques are proposed to circumvent this problem. This way, the delta-sigma quantizer's feedback signal is obtained by a multiple-stage quantization, where the loop quantizer (low-resolution and minimum-delay) implements only the last-stage quantization. Hence, high-speed, high-resolutiondelta-sigma quantization is feasible without using complex circuitry. An improved version of the MASH topology is also proposed. A delta-sigma quantizer is used to quantize the input signal into an oversampled digital representation of low-to-moderate resolution. The delta-sigma quantizer'struncation error is estimated either directly, or as the first-order difference of the output signal from the loop filter's first integrator stage. This technique avoids the need for accurate matching of analog and digital filters that characterizes the MASH topology, and it preserves the signal-band suppression of quantization errors. Simulations show that quantizers of this type can yield 100 dB performance at 10 times oversampling. There are no requirements for high-resolution flash quantizers or other hard-to-implement circuitry.
Main Research Area:
Department of Information Technology, Technical University of Denmark, 1999