Why are there big power differences on IP3 sides of PA?

pa+harmonic+ip3

I am working on a 900MHz RF PA tuning now and was surprised by the OIP3 of this device. The lower side IP3 (899MHz) and higher side one(902MHz) could be as much as 10dB different, as the output power level changes. Here I use 2 tone input with 1MHz spacing as the input to the PA. Anybody knows the root cause for this difference?

ip3 contours

Hi The difference is due to bad matching networks. regards, Paul.

oip3 pa

The return loss actually looks good. About -15dB for both input and output return loss around 900MHz. It is actully too good for a PA...

Re: IP3 of PA

Hi, The RL does not tell you anything in this. You need to find the matching impedance points for which your amplifier (Transistor/FET) has the best IP3 performance and match for this point. What you maybe did was match for maximum power or gain and this is not the same. Matching is very complex in digital systems the best is to use a load-pull equipment but I can't pay for that nor do you I think ?.

If you want the best IP3 than make a simple matching network (narrow band) and match by tuning your inductors/capacitors until you get MAX ip3 max gain and anything else. Then measure your networks impedances witha VNA and do this for more frequencies. Use these values for you matching network you design.

regards,

Paul.

Re: IP3 of PA

Hi,
Thanks for the response. Certainly you are right there, we could match the amplifier for best IP3. We could do an loadpull for IP3 contours and likely we will find the sweet spot that will give me max IP3. However, It is questionable that equalization of H/L IP3 sideband amplitude will be achieved at the same time.
If we know the reson for this amplitude discrepancy in h/l IP3 tone, we probably will have a clue to solve this issue and hopefully improve overall IP3 performance. It seems the discrepany in sideband IP3 indicates margin for PA performance improvement.

BR,
Rick

Re: IP3 of PA

Hi, It could be that the transistors your using are pre-matched and that you use them out of band ?. Maybe they are not the good technology for your requirements ?. A lot of components today are really matched for special frequency bands unsing them at different frequencies does give problems and does not provide maximum performance. Are your numbers close to the man. specifications ??. The current trough the device is a very important factor in IP3 so maybe the problem. Gain is different at lower freq. But think you know this already.

Paul.

Re: IP3 of PA

Here are a couple of things to ponder.

1-You really do have a narrow band system, with 1 MHz separation between tones. Because of that, I don't think that the problem is phase changes between the two tones causing differences in the third order products.

2-How good is the bias circuitry bypassing? It's possible that there is an issue with the bypassing on the bias lines.

3-How is the low frequency bypassing (DC to 10 MHz) Without a good low frequency bypass, the difference frequency could be re-modulating the bias point

4-How about the 2nd harmonic termination? Third order products are dependent on 2nd harmonics.

5-Could you be having some 5th order term's coming up? As you increase the power in the input tones, are there 5th order terms coming up at 898 & 903 MHz? Fifth order interference products also have components at the third order frequencies.

Dave

Re: IP3 of PA

Predicting IP3 looks like to be a very challenging part of high linearity PA. I’ve been using various kind of nonlinearity models, none of them are able to predict IP3 level of reasonable accuracy. There is very often more than 5dB different between measurement and simulation. So model extraction is obviously the key for IP3 simulation. I would appreciate it if you could able to shed some light on the necessary modeling technique here.

Thanks

Re: IP3 of PA

Try to do a two-tone IP3 test on Low, Mid and High frequencies, sweeping on each of these frequencies the spacing between tones. Let’s say from 500kHz to 3MHz, in steps of 500kHz. If you get different IP3 for different tone spacing there is a “memory effect”. This can be fixed playing with the bias impedance that is strongly frequency dependent.
Also some deliberate mistuning on the high Q factor input match of RF power transistors might improve the linearity vs frequency.