I'm working on an interesting project for a biotech customer, and I could use some wisdom from the assembled multitude.
They're building a scanning fluorescence microscope that scans several lines at once with a single photomultiplier, using a clever scheme to make each line come out at a different RF frequency.
The frequencies are inherently evenly spaced, so third-order IM products cause image artifacts. So they asked me to come up with a way to reduce the IMD.
I've been doing some measurements, and have found the usual cubic dependence of the IM3 products up above about 50 uA of anode current. That's generally a combination of voltage sag in the dynode bias string, plus some space-charge effects. It can be reduced by changing the dynode voltage distribution so that the last few stages have more bias.
Interestingly, though, below ~50 uA there's a broad range of anode currents (20 dB or so) where the IM3 products go up linearly with the signal, so that the SFDR is a nearly constant 60 dB, plus or minus a few. This is without changing ranges on the spectrum analyzer or anything like that.
Ahah, you say, the IM3 is in the illumination source, which runs at a constant drive level. So I thought, but it ain't so.
My illumination source is a pair of blue LEDs, each driven from a separate amp with separate power supplies. To get the spurs big enough to see, I'm using two synthesizers locked together--one is just a frequency doubler hung off my 10 MHz rubidium reference, and the other is a PTS1000 locked to the same reference, running at 20.001 MHz. The analyzer is an HP 8566B, locked to the same reference. That way I can use a 10-Hz resolution bandwidth to see the spurs.
Using a photodiode, I've verified that the spurs in the light sources are below -80 dBc.
There's mention in the literature of a space-charge effect at the photocathode, which would vary less with gain than in the last stages. If that's it, the -60 dB would be set at the very beginning, so it should depend mostly on the actual light intensity at the photocathode.
However, the linear behaviour occurs regardless of whether I'm using a fixed light level and adjusting the anode current with the bias voltage pot or leaving the voltage alone and using a variable optical attenuator.
It also doesn't depend on the attenuation of the signal in the RX chain--adding a 10 dB pad doesn't change the SFDR in that anode current range.
This is, in other words, weird as hell.
Any ideas?
Thanks
Phil Hobbs