DAC resolution

[Terry Michaels]

Message text written by "Fraser, Colin J"
>The accuracy of 1/2 LSB is the quantisation error for a continuous signal.
If the steps in the output levels you require match the steps in the DAC
output, you don't have any problem with errors. 
The spec for the DAC-08 quotes non-linearity of +/-0.1% of FS - this is the
relevant error in this case.

Colin f

<

Hi Colin:

+/- 0.1% of full scale, which is 128 notes, is +/- one-eighth of a note. 
That is a huge amount of potential error.  I say potential error because,
as Paul Schreiber pointed out, the 1/2 LSB error is a maximum spec, most
times a DAC will have less than this amount of error.  You may by chance
get an 8 bit part with errors that are inaudible.  Also, this refers to a
maximum deviation from a straight line plot of the conversion relationship.
 Usually that type of error will result a range of pitches that will all be
flat or sharp, because the straight line relationship might actually be a
slight curve.  The note to note deviation because of an incorrect step size
by the DAC (differential nonlinearity)  will usually be small, but most
DACs are guaranteed only to have a one bit step that is greater than zero. 
This means, worst case, two adjacent notes could have nearly the same
pitch, or they could have a difference of nearly two semitones.  The
advantage of using 8 bits of a 12 bit DAC for this application is the
maximum differential linearity, or note to note error, is 16 times smaller,
because the 12 bit DAC has 16 times more resolution, and the maximum
differential nonlinearity will be only 1/16th of a semitone.  The maximum
deviation of a straight line over 128 notes will also be 1/16th of the
maximum deviation of an 8 bit DAC.

Terry Michaels