Thanks again for your reply! Interesting comments.
RFICDUDE wrote on Dec 15th, 2010, 4:36am:People have developed semi-analytical relationships between OIP3 and ACPR/EVM for signals that have similar statistical characteristics (PDF, CCDF and moments) as Gaussian or complex Gaussian signals. Practical signals that fall into this category included WLAN/WiMAX/LTE OFDM and basestation CDMA/WCDMA. Signals that do not typically fall into this category are handset CDMA/WCDMA, EDGE, GSM and simple modulation such as QPSK.
I've found semi-analytical expressions (using empirical correction factors) to relate two tone 3rd order non-linearity (IP3) to modulated waveforms such as WCDMA. The reference you provide below is more tractable that the ones I've read, so, thanks
The trouble is the GSM/EDGE waveforms. For a simple amplifier without memory effects and a polynomial gain function (mclaurin series), none of the simulations match my calculations! Although the behaviour is closer to sinusoidal than the CDMA or OFDM family, the analysis or finding an expression has been much harder! If you have any papers or references relevant to multi-carrier GSM/EDGE non-linearity for base stations, that would be very interesting indeed!
RFICDUDE wrote on Dec 15th, 2010, 4:36am:A paper you may be interested in by Wu et. al. that relates OIP3 and OIP5 to ACPR for complex Gaussian-like signals is:
Q. Wu, H. Xiao and F. Li, "Linear RF power amplifier design for CDMA signals: A Spectrum Analysis Approach," Microwave Journal, Dec., 1998.
Search around and I think you maybe able to find it on the web.
I managed to find this paper. Thanks for the reference.
RFICDUDE wrote on Dec 15th, 2010, 4:36am:If you want more general nonlinear models (not signal models) to run different signals through then I can provide references for that, but those models do not provide analytical signal models (you have to do that work yourself).
I have already a set of time domain multi-standard signals that I can use. What sort of models do you refer to? Behavioural models of non-linear devices?
RFICDUDE wrote on Dec 15th, 2010, 4:36am:In designing transceiver transmitters, my experience always been that the signal which has the largest peak to average ratio tends to dominate the design since it dictates the worst case for achieving some average output power with a linearity constraint. This would be true if the peak of the peaky signal is at least as much in power as all your other signals. If max/peak powers are not similar then optimizing the design could be trickier since you would need to scale the design a bit to maintain optimum performance for different peak output power requirements.
Sounds like interesting and challenging work.
Challenging indeed!
I think I see what you mean. My assumptions are equal carrier power distribution for all types of carriers in my composite signal, so what you say holds for my case. I hope that makes sense.
I can imagine that for a device that is adequately backed-off, a signal with higher PAR will result in higher non-linearity on the output spectrum. However, I'm not able to connect the two, because the typical PAR on the CCDF curve considered for design is 10E-2% which is a rather low probability and the Spectral Mask requirements for all standards are 'ensemble and time' AVERAGE measurements (not PEAK-HOLD measurements), made over sevaral thousands of sample periods. I mean to say that the 'peaking' will occur so rarely within the observed time period, that, logically, it should not be a factor at all in the resulting non-linear distortion products (IM3). Do you have any comments here?
-sandman.