...
Looked at the signal ground of the 2m licor inside the Licor box, with the oscilloscope probe ground connected to power ground. Saw hair-like spikes with a range of +- 1V at the bit frequency of the signal. By shorting signal ground to power ground we could get rid of these spikes.
...
Up'd the sampling frequency of the 2m licor in port 7 to 20Hz, 19200 baud. Saw an increase in "spurious interrupts" which did not diminish when we shorted the signal and power grounds together, so those spikes do not seem to be the issue.
Swapped dsm chassis boxes, replacing box #1 with box #8. Also swapped PC104 stacks. This configuration did not work well at all, with data_stats showing bad data rates. Swapped original PC104 stack back, and still saw problem. Eventually noticed that the input from the licor on port 17 had lots of jibberish. Moved this to port 20 and it cleaned up and the data_stats issues went awaylooked OK. So now believe the data_stats this sampling issue was due to a bad port 17 on the new box and not an issue with the pc104 stack.
...
The problem with sampling Licors at 20Hz seems to be a data system issue, in that it cannot keep up with the total serial interrupt load from the sensors on the tower. Based on what we saw on the scope, it is not an issue with the quality of RS232 signals. Since it didn't change when we swapped system enclosures and the PC104 stacks, it is not related to interface panels, the CPU or the serial cards. The power to the Licor's Licors looks good.
Our simulation of this configuration in the lab did not include 5 TRHs (1 sample/sec), the barometer (1/sec) or the GPS (2 samples/sec). The addition of these inputs seems to break the camel's back.