of Microprocessors

DIGITAL-TO-ANALOG AND ANALOG-TO-DIGITAL CONVERSION 377
dB, respecrively. What this actually amounts to is a piecewise linear approximation
of a true exponential curve.
Segmented DACs using unequal segment slopes can also be used to
make a piecewise linear-approximate exponential curve. The DAC-86 is a 7­
bit plus sign unit that does this using a 4-bit fraction, a 3-bit exponent, and
a base of 2. Its application is in digital telephone systems and occasionally
speech synthesizers. To the author's knowledge, nobody has yet manufactured
a high-fidelity (12 to 16 bits) exponential DAC module.
Which Is Best?
The logical question, then, is: Which technique is best for highfidelity
audio? Using a true 16-bit DAC is unmatched for convenience and
does the best job possible on 16-bit sample data. It is expensive, however,
and may require maintenance. Actually, the other techniques were developed
years ago when a true 16-bit DAC was considered an "impossible dream."
Next in convenience is the use of any of the several "designed-for-audio"
16-bit converters currently on the market. While noise and distortion at high
signal levels fall somewhat short of ideal 16-bit performance, it is doubtful if
anyone could ever hear the difference. Architectures range from brute-force
ladders with the differential linearity errors ignored to segmented units to
sign-magnitude units. Burr-Brown, for example, offers both a ladder
(PCM-51) and a segmented (PCM-52) IG-bit audio DAC in a 24-pin IC
package for about $40. Analog Devices has a nice 40-pin CMOS segmented
DAC (AD7546) in the same price range that will be used later in an example
audio DAC circuit. Analogic sells a somewhat more expensive signmagnitude
hybrid module that probably comes closer to true 16-bit
performance than any other "shortcut" unit currently available.
Floating-point and exponential DACs have the lowest potential cost for
a wide-dynamic-range audio DAC. They are also suitable for expansion
beyond the dynamic range of a true 16-bit unit if other audio components
can be improved enough to make it worthwhile. However, the gaincontrolled
amplifier settling time and overall coding complexity make it a
difficult technique to implement with discrete logic.
Reducing Distortion
As was mentioned earlier, glitching of rhe DAC can contribute to
distortion that is unrelated to the resolution of the DAC itself. Actually,
there would be no problem if the glitch magnitude and polarity were independent
of the DAC output. Unfortunately, however, the glitch magnitude
depends heavily on both the individual DAC steps involved and their combination.
Typically, the largest glitch is experienced when the most significant
bit of the DAC changes during the rransition. Proportionally smaller glitches