Hi Ferrograph, at the bottom left on the QA401 display you will see a status messaging toggling from streaming to processing while it’s running. On my machine, at 192K sample rate at 256K FFT it shows streaming about 3X longer than it shows processing.
While “streaming,” the data is being acquired. If you are sampling at a 192K sample rate, then in one second you’ll acquire 192K samples. But a 256K FFT means 256K samples, so it will take 256/192 = about 1.33 seconds minimum–there’s no way to shorten that. Now, there are things you can do such as continuously streaming and updating every 1024 samples. That can give the appearance of a very high sample rate once you’ve streamed a full buffer across (the first 1.33 seconds). But practically unless you are looking at transient events in a waterfall that doesn’t really buy you much.
When “processing,” the data that was collected during the streaming session is being analyzed, and here you are bound by the performance of your machine. But generally the streaming portion takes more time than the processing portion.
As to the question what FFT size is adequate, you can run a sweep for the measurement you are interested in.
For example, below I set the QA401 to run at 192K, and then ran a sweep for THD at different FFT sizes. This gives you a good idea of the measurement differences you might see. You could make most measurements with a 64K FFT and know you are just a few dB from the final value. And when you are ready to make your final measurement, for that one measurement go to the big FFT.
Also, keep in mind that you’ll get 4X more resolution for a given FFT size at 48K versus 192. If you don’t need so see beyond 20K for a measurement, stick with 48K. A 16K FFT at 48K has as much resolution as a 64K measurement at 192K.
Below is another plot, which is shows a THD sweep with a 16K FFT at 48K and a 64K FFT at 192K. Note this is the same resolution and same update rate. The measurements are similar, but the 48K sample rate wins by a dB or so in some places. It is normal for converters to have slightly different performance at different sample rates. And as the data here shows, the slower sample rate delivers a bit better performance in a few places.