According to the manual for the 6361 the noise level is 17 to 315 uV rms, depending on range. The peak to peak noise is probably 5-10 times higher. SO your 0.5 mV noise with the inputs shorted may be the device internal noise.
I have explored your data and am convinced that you have some kind of signal aliased in there. I low pass filtered the time domain data before taking the FFT. Then I changed the filter cut off frequency while watching both the filtered output and its spectrum. The behavior is not easily described in words, but it has inconsistencies which make me think it contains an alias. As I said in a previous post, it is impossible to reconstruct the aliased signal or to separate it from the desired signal. You have no "desired signal" in that data. If it contained only random values the spectrum would be flat and the spectrum of the filtered signal would reflect the filter transfer function.
When you short the input terminals, do you just short AI0+ to AI0- or do you also connect both to ground, either by shorting or through a resistor?
I am aware that the data I posted contains no desired signal, since my source was not connected at the time I acquired the data.
This signal comes from the open leads acting as a pure receiver of an AM modulated RF signal that is being broadcasted from nearby.
To answer your question, I just short AI0+ to AI0-. I havent yet connected them to ground through a resistor. Maybe I should do that.
Looking at the manual, it seems that by changing the input range, the resolution of the acquired signal changes.
By playing with my acquisition VI, I changed the input range from 20 Vpp to 1 Vpp and somehow the noise level was reduced. Does that make sense?
Thanks for your time.
With a smaller range you increase both the resolution and the accuracy of the device because you decrease the step size. You'll notice at 10 Vpp (maximun voltage range) the accuracy is listed as only 1.74 mV, but at .1 Vpp (minimum voltage range) the accuracy is 38 µV. This is because at 10 Vpp there is a stepsize of 20/(2^16) and at .1 Vpp it is 0.2/(2^16), considerably smaller. It's likely that the smaller step size reads the noise lower than the larger steps you were reading at 10 Vpp effectively giving you a smaller noise reading.
As for the resistors, it always a good idea to soldier components for final deployment as a way to insure a good connection, but for testing purposes you should get a good enough connection just splicing them in. Make sure you ground them to the probe like described in the Field Noise and Wiring Considerations article I posted for you. Below is a good summary of the different wiring configurations and when/how they should be used.