Is reducing frequency span cheating when measuring noise floor of amplifiers

[tron9000]

So I'm working on a Pre-amp for a sensitive sensor. I'm currently measuring its noise floor using a spectrum analyser.

I've not done this test many times so bear with me, trying to get my head back into the game.

The amp is in a shielded metal box and power is supplied internally from 2 batteries through some linear regs providing a split rail supply. With the BNC inputs of the amp shorted to GND using 50 terminators and the output connected to the spec analyser, with the frequency span set to max (100kHz - its an old spec analyser), the noise floor looks greater than it does with the frequency span set to say 195Hz.

So does anyone with experience with this sort of test know: is the test invalid if you take a load of different measurements with the small frequency span up to the max bandwidth? (so: 0-195Hz, 195-380Hz, etc up to 100kHz)

I know it'll take a while to do it this way, lets say I have time.

cheers

[nfmax]

Yes, it is cheating!

It's not clear from your question if you are using an FFT analyser or a swept-frequency spectrum analyser, but the underlying measurement is much the same. The SA is displaying the input power as measured in its resolution bandwidth. In a swept-frequncy SA the resolution bandwidth is manually selectable; in an FFT analyser it depends on the FFT 'bin spacing' and the noise bandwidth of the window you are using. As you reduce the span, you (generally) reduce the resolution bandwidth as well. With a smaller bandwidth, less noise gets through.

What you should do is set the SA marker to 'noise' mode. This should show the display trace level at the marker location as a power per unit bandwidth, e.g. in units of dBm/Hz (or you may get a voltage display in dB/root Hz). This compensates for the change in resolution bandwidth as the span is changed, and should give you an almost constant value, within the measurement uncertainty of your SA.

[tron9000]

its an FFT analyser

[nfmax]

On a decent FFT analyser you may be able to set the y scale to display V/root Hz (or W/Hz) directly, so you won't need to use a marker. What is the make & model of your analyser?

[tron9000]

its an SR760.

I've solved this now. Big conceptual misunderstanding. I've managed to plot the PSD (V/root Hz) Vs. frequency:

Thanks to Dave's video on Op-amp noise (EEVBlog #528) I figured out that my SA has this function, and got the results I was looking for.

thanks nfmax

Cheers

PS: Ha! you and I have equal # of posts!

[nfmax]

PS: Ha! you and I have equal # of posts!

Not now we don't! 
That's a nice analyser, it should do what you want

[eb4fbz]

Really it's not, as you should allways normalize the measurement to 1Hz (dBm/Hz, V/Hz, uV/Hz...) or specify the used BW instead.