I realised that the input and output caps couldn't be the
cause of the temperature dependence: the signal and reference
frequencies are so far from the corner frequency.

It looks like the problem is a defect in the ebsynth software.
The input and output buffer delays change slightly as the sample
rate varies with temperature.  The program isn't correcting
for the delay changes.

Hopefully have that fixed shortly once I figure out a way
to correct it without introducing some slowly accumulating
numerical error.

> 6 synchronous sampled inputs, that's the perfect system
> for the tree.

Due April, so they say.  I hope it doesn't need an RPi 3
because the power budget then has to allow for the unused
built-in Bluetooth and WiFi.  I wonder if it A/D converts
6 inputs in unison, or does them sequentially.  It might
do them in stereo pairs.  We'll have to wait and see.

--
Paul Nicholson
--