Marc P., Filberto C., Erik v R., Daniel S., Dave B., Richard M., Robert S.
A first look at channels in the test stand found ~10 drifting lines (in 0-1000 Hz band) on each empty ADC channel (O3 16-bit ADC boards), with coherences between empty channels reaching above 0.1 for these lines (see Figure 1). The channels were on two cards connected to AA chassis. There were two extra ADC cards to imitate the DCPD I/O chassis. After finding the lines, we moved or shut off computers and other potential sources, but the lines persisted until we started pulling ADC cards.
While the ADC sampling is clocked externally in our setup, the boards have ~50 MHz clocks for their microprocessors. A spectrum analyzer connected to the power supply card showed peaks around 50 MHz with spacings that were consistent with the beat frequencies we were seeing in the ADC channels (see Figure 2). For further confirmation, I was also able to remove a ~10 Hz peak from the ADC channels by removing one of the two cards that had their 50 MHz clocks spaced apart by ~10 Hz. Finally, I lowered one of the ADC clock frequencies by touching the top of the clock chip with an insulated finger.
Marc found that the new cards we are purchasing could be bought with different clock frequencies: http://www.generalstandards.com/specs/66_18ai32ssc1m_spec_110216.pdf. We could space the ADC card frequencies in each I/O chassis by greater than 10 kHz. The current ADC cards also have the option of different master clock frequencies. There would, of course, be some cost to this. But I think that something like this should be done, at least for the most important channels.
I also investigated magnetic coupling to the test stand ADC channels. The magnetic coupling was too small for any magnetic fields measured inside the I/O chassis to produce features on the ADC channels.
The I/O chassis has the new timing board and the new synched power supply card. I found no large fields or wndering lines associated with these boards.
The next step is to investigate transients and lines that might be problematic for searches, with help from DetChar.
