Recommended windowing FFT

[ChiefOfRocketSurgery]

Had a question on some windowing recommendations for FFT VI. The context is as follows but the post is specifically for any windowing recommendations, not necessarily troubleshooting our signal noise source (if it's a source) at this point.

Context: This came up during some periodic signal noise investigations my group is examining during data collection for test of a unit we are working on. Unit context: 4 channel 12 bit DAC w/ 10 kHz output. +/- 10 V.  We are using compact C chassis with NI-9205 32 channel analog inputs setup for floating differential measurement in a RSE configuration.

We were sampling 10kHz and noticed periodic signal noise raising the floor on our FFT plots. this is shown in the attached video. We haven't had an opportunity to independently scope the hardware with our Tektronix to see if our unit is producing the noise but will be doing so with the NI equipment disconnected so as to see if this is just the nature of our production unit. Normally we sample at 1.6 kHz. The attached video shows the output of 4 independent DAC's and returns the display in DB as well as pk values.

Question: We have left the windowing of the FFT VI defaulted to Hanning. Ultimately we will be using this NI hardware and VI combination to acquire balancing data (1x) (f<60 hz) and wont particularly care about higher frequency signals (f>1000 Hz). Is a different windowing technique better than hanning or is the default hanning the most recommended?

[algarcia]

I think scoping the DUT directly with a scope directly is a great step to verify if the unit is producing noise.

However, the behavior of an instantaneous rise in noise floor happening at various times on all of the channels looks suspiciously like the averaging is resetting on the FFT. Looking at your code snippet it doesn't look like any averaging is set currently though. 

A Hanning window is generally a good starting point for most types of measurements. However, the problem with the noise floor randomly rising isn't indicative of a bad window function. This white paper gives an overview of windowing.

[Henrik_Volkers]

My (crystal ball) guess: A missing bandwidth limitation. If you sample with 10kHz and look for signals up to 1 kHz, a simple RC filter migth do the job. (depending on your source a simple capacitor addet migth work)

According to the spec the analog input bandwidth is 370 kHz, so every signal in that range will fold into your measurements.