Looks like I cracked it, turned the brightness up on the oscilloscope and there was the noise. a 100N cured it :-) Any comments apart from Doooooooo....

I had problems with A/D noise (also D/A) in several different designs. I used different (combinations of) techniques as time evolved, I found that noise without the USB lead was too high and much worse with it connected, but found each of the following helped so tend to now apply these techniques generally:

1. adding a clamp on ferrite at each end of the USB helped enormously.

2. I also found that good HF decoupling across the supply improved the noise situation in general.

3. I don't recall the values used but a ferrite bead and low value capacitor hf capacitor at the A/D inputs also reduced the noise.

4. Finally when I'm only taking values at a low frequency I take multiple readings and use the average.

For the DAC, I never got the output to have the low noise I required, it seems to be the internal chip clock feed through. My solution was to use an external SPI DAC (Microchip) and not have anything else on that bus. The noise can then be incredibly low as there is only bus activity in a short burst when changing the value.

Finally a good motherboard PCB layout with an earth plane under the module enhances the performance over what a breadboard will deliver.