Oversampling digital-to-analog conversion

Oversampling digital-to-analog converters employing sigma-delta modulation noise shaping and single-bit quantization are attractive for use in digital audio applications because of their relaxed reconstruction filtering requirements and their tolerance of component mismatch. In such a converter, a multibit digital input is digitally encoded into a single-bit sequence and then converted into a two-level analog signal. However, the use of a two-level D/A interface results in a large amount of out-of-band quantization noise that typically must be attenuated by a carefully designed analog reconstruction filter.;This dissertation examines several aspects of oversampling digital-to-analog conversion. The noise shaping operation is first studied within the context of the design of second-order and third-order digital noise shapers. The imperfections associated with the conversion of discrete-time analog signals to continuous time are then analyzed. A means of simplifying the reconstruction filter design through the use of a current-mode semi-digital finite-impulse response (FIR) filter is presented. Lastly, a current-to-voltage converter employing a CMOS operational amplifier and a polysilicon feedback resistor is described.;The main emphasis of this dissertation is the implementation of an oversampling D/A converter wherein a current-mode semi-digital FIR filter is used to implement a multi-level digital-to-analog interface that attenuates the out-of-band quantization noise without requiring stringent component matching. An experimental implementation of the converter achieves a dynamic range of 94 dB and 72 dB attenuation of out-of-band quantization noise for a baseband of 20 kHz. The prototype converter, which consists of a linear interpolator, a second-order noise shaper and a 128-tap semi-digital FIR filter, dissipates 59 mW from a 5-V supply and occupies an active area of 3.0...