Hi Paul, VLF,

The observed effect is clearly a funtion of the temperature of the chip, which reaches abt 50 °C (human estimation) when in operation!

Meanwhile i managed to stream audio data from the Raspi+Octo using vtvorbis (which is BTW running over a terminal only, no GUI installed!) to another Linux-computer via SSH and also to an Icecast2 server, so i can put it to SpecLab to get a scrolling spectrogram, more or less 'live'. So now i am able to search for the strange behaviour of the soundcard.

In the middle of the spectrogram, see attachment, showing CH1(top) and CH2 (bottom) i sprayed some cooling spray on the chip. The dark vertical lines (unwanted effect) disappear, the whole noise level drops on both channels. When the higher case temperature recovers, the lines come back again. When spraying onto other components on the board, i can't see the lines disappearing.

Putting a small heat sink on the chip reduces the 'dark line' intensity. Probably it can be fully avoided by using proper heat sinks. The effect maybe happens d to the Affenhitze in my office, about 25 °C.

This all applies to 96 kS/s here.

But it is also possible to run the card at 48 kS/s. Then, the CPU load is lower, the temperature is lower and the effect is gone, on all channels! And the noise seems to be even lower, with some frequency response though, whereas the soundcard noise has been different on different channels at 96 kS/s, which is another strange effect!
So i conclude that either cooling of the chip helps, or running the card at 48 kS/s helps.

I want to decide to run the card at 48 kS/s. Or would you say that this will lead to a significant inaccuracy of the timing/precision?
During my last ULF recordings i run the card at 24 kS/s and was quite happy with the results.

BTW, meanwhile i am using a more efficient DC/DC converter which has about 95% efficiency.
Incredible!: Now, when running the Raspi + the GPS module and recording two antenna inputs to an USB stick (up to nearly 4 days, for portable postprocessable experiments), the power consumption is 13.9 V * 170 mA = 2.4 W !!!
The first time it allows to record from 2 antennas (e.g. two loops or a loop and an E field antenna) to form an optimised cardioid antenna by software. The recording process is locked to an atomic standard and accurately resampled to the expected sample rate.

Looks like it is time for a real test soon!

73, Stefan