I'm trying to perform an IP3 measurement on an RF front end I've put together, its a SAW/LNA/SAW/LNA chain with the first LNA being low gain, high output/high linearity to keep out of band interference from pushing it non-linear, while overcoming the loss of the 2nd SAW, and to set the noise floor. While the second SAW/LNA give the rest of the out of band rejection and provide the rest of the gain needed. So of course "well, just how linear is it?" is going to come up, so IP3 measurements are being performed.
2nd order intermodulation products (IM2) are supposed to rise at 2dB out per dB in, and 3rd order products (IM3) are supposed to rise at 3db out per dB in, but in my measurements during a two-tone test, the IM3 products are only rising at 2-2.2 dB/dB instead of 3, and I'm trying to reconcile where the problem is creeping in.
I'm using a vector signal generator with internal baseband to generate the two tones, at centerfreq+250 KHz and centerfreq-250KHz, though a circulator and a 5db attenuator to ensure the sig gen and amplifier input both presented with a nice match and output goes into a spectrum analyzer for doing the measurements manually with the peak table. The tones are all well within the passband of the filters. Simple setup, should be easy right?
The signal generators IM3 products are way below that of what I'm getting when measuring the amplifier (~30dB better IM3 ratio on the sig gen than the amp-under-test), and they stay at a nice 1dB/dB rise, which makes sense given its just adjusting its output attenuation most the time as I change power output. There is some carrier bleed though at ~-60dBc, but I'm not seeing any mixing products pop up from that.
If I exclude the measurements made near amplifier compression, when I fit lines to the measurement points it makes for a very nice fit (R=0.996) so its not that screwing me up, and all of the measurements were made with the IM3 products at least 15db above the noise floor of the spec an.
The only things that I'm questioning now is that I choked down the VBW/RBW of the SA down quite far (10 Hz) to drive down the noise floor and put it into FFT mode instead of swept SA mode, and perhaps I have it too tight and frequency error (despite clocking them both from the same reference) managed to put the IM3 tones shifted just far enough to get extra attenuation, but I dont see how that would make for a 2dB/dB slope instead of 3dB/dB.
I'm planning on re-doing the test with wider VBW/RBW on Monday with an off the shelf single stage unfiltered LNA, and beating on it till I can figure out what is going on, but the physics defying results managed to stay on my mind for the weekend, and I cant figure out what I'm doing wrong or missing.
Any sage advice about how I have no idea what I'm doing?