PA PAR vs. EVM?

I'm designing a 1W OFDM power amplifier for 500~700MHz. The signal PAR is about 12.5dB and ESG reading is about the same, which means the PA P1dB should be better than (30+12.5) = 42.5dBm. In my design P1dB rolls off to 38.5dBmat each end, which is 4dB compressed, however EVM measurement is better than -39dB.

Can someone help me understand this? Could a compressed PA provide doog EVM?

LNA

I don't have an answer to your question, I'm sorry.
But remember that P1dB is just an approximate way to model circuits nonlinearities and, the way nonlinearities impact you systems depends on circuit implementation.
I mean, if the circuit has a hard saturation or a very smooth one, the results can be very different (In WCDMA sometimes there is a 0.1dB compression point specification!).
If the main parameter you are interested in is EVM (what about spectral regrowth?) and your measurement results are in spec, well, it's OK.

I remember few years ago working on a WIFI receiver we had similar numbers to deal with (12 dB PAR), but we demonstrated that, clipping at only 8 dB PAR, there was no degradation (if I well remember CCDF showed that the probability to go in the upper 4 dB PAR was really low...).

I would like to know which application are you targetting and the exact frequency BW.

Mazz

Thanks Mazz for sharing your experience.

This is a Wimax radio with 6MHz BW. EVM is one of critical parameters in TX chain, others like ACPR, Gain, output power all meet the spec, except P1dB, which is 4dB lower than spec at freq low and high side. I know the probability of clipping for 12.5dB backoff is low--0.01%, but I'm not sure how each 1dB compression tranlate to probability or EVM.

Anyone else?

From what I found practically, different compressed PA’s (having the same P1dB), don’t have the same phase change of the signal (AM/PM) near the compression point.

I think this is related to the semiconductor process.

WiMAX EVM seems to be affected first of all by the AM/PM, and after that by the AM/AM.

So, this could be one reason that you can go with the PA more in compression before breaks the EVM.

Thanks for sharing your experience.

Hi, Vfone, that's a good point.

I confirmed the similar results on a class A PA-- similar EVM with 7~8dB backoff (4~5dB compressed).

Is there any artical to support this point?

Just out of curiostiy, what instrument(s) are you making the EVM measurement on?

Aligent E4440A + digitizer + 89600

EVM of PA at about P1DB level is 2%. After P1DB level it rises very quickly. In waveforms with high PAR the peaks does not contain's data - it result of transition from one symbol to another.

Peaks doesn't contains data?
So, why we don't remove all the peaks placing a limiter?
Why all the clipping methods actually affect the EVM with the same amount they improve efficiency?

Is this rule of thumb number for PAs? Any theory to back this number?

I have the same question of Vfone regarding this. ?

Peaks in such waveforms are result of transition from one state to other (especially when transition through zero at the constellation plot). transition itself didn't contain data.
EVM of 2% at P1DB i guess is rule of thumb.