The key elements in mixed-signal applications are analog-to-digital converters (ADCs) and digital-to-analog converters (DACs). Despite of the availability of analog functions in a ULP technology virtually all signal processing tasks can be done more efficiently with digital implementations. From the viewpoint of signal coding, digital signal processing is more power efficient than analog for a high signal-to-noise ratio (SNR): an analog signal needs a power proportional to the SNR, whereas only the logarithm of the SNR is required for a digitally encoded signal.
The main problem in ADCs and
DACs is the linearity of the
converter. This problem is much more severe in low-voltage
circuits because of the increased non-linearities due to the
smaller
.
Therefore, ADCs like successive-approximation, flash,
pipelined ADCs, and the like
will be difficult if not impossible to realize for
high-precision applications [28, pp289].
Similar barriers must be faced in the case of low-voltage DACs,
that use R-2R ladders, switched-current, charge redistribution
ADCs, etc. Although in some cases non-linearity can be minimized through
appropriate transistor sizing
it will be still more severe than for higher supply voltages.
For example, in switched-current DACs the switches and the current
mirrors can be sized as 2k, where k is the bit index. However,
this is affordable only for a small number of bits and it will not
alleviate the matching problem.
On the other hand, oversampled 1-bit ADCs and DACs [28] are perfect for realization as ULP circuits, because they do not require any linearity of the quantization element 5.1 and the requirements on the analog front-end reduce to a minimum. Some techniques, such as predictive ADCs, may require special circuit designs to avoid the need for high-precision analog filters or multibit DACs. The higher required speed of the digital part due to the oversampling is not a problem because this performance comes automatically with a digital ULP CMOS technology. Thus, speed is traded for precision.
The simplest oversampled 1-bit ADC is a first-order sigma-delta converter (cf. Section B, Fig. B.1). It uses an integrator as noise-shaping filter and a comparator for 1-bit quantization. Ideally, the SNR 5.2 improves at 9dB per octave oversampling. In a real circuit the achievable SNR is somewhat lower depending on the nonideal properties of the components (cf. Section B).