OK I'll try and clarify my view By the way I 'think' where we divert on this issue is that you interpret the IMD contribution as an unwanted thing and are seeking to quantify its contribution to PEP as an error. I'm just interested in accurately measuring PEP according to the classic definition and IMD contributions are valid even though they are unwanted. Is that what we disagree on? Anyway, here's the way I see things: First of all I'm assuming there is typically a big difference in terms of the RF frequency vs the highest modulation frequency. eg many MHz vs just a few kHz. Let's first assume the transmitter has a really good LPF and the harmonic terms have been removed to -60dBc or better. We know the period of the modulation waveform is going to be typically hundreds of microseconds or more and the period of the RF is going to be hundreds of nanoseconds or less. Now we can both agree that any unwanted IMD terms can cause summing and cancelling and so can the individual tones in our speech. We can also agree at some point they can all phase together and we get a crest in 'modulation' (and sometimes we get a null) Now this next bit is where I think we have a different view. Both are technically valid views but I am being a bit of a pedant and sticking to the classic definition of PEP. For the determination of PEP I think this RF cycle at the modulation crest is perfectly valid despite it having unwanted IMD contribution. This is because PEP is only interested in what the average power is at that top RF cycle (not how clean the modulation is) Deep inside that crest is the single RF cycle that is subtly bigger than it neighbours. It is the daddy RF cycle where PEP is ideally measured.Because the period of any IMD term and the wanted modulation is so large (thousands of times bigger?) it can't realistically distort this RF cycle away from being a sine wave. The IMD will have had a part to play in its amplitude but it won't realistically affect the purity of the sinewave. So if it were possible to extract this single sinewave and measure its Vpeak you could still calculate its Vrms with good accuracy because it is still a sine wave. So the readacross from Vpk to PEP in the meter would give a good result. If we now look at the case where the harmonics aren't as well filtered then there won't be 'just' a pure fundamental sine wave in that little timeslot of a few hundred nanoseconds. There could be two (or more) sine waves squeezed in there and they might not be fitting into the slot with the same starting phase as the fundamental sinewave. So you get a slight summing and cancelling effect depending on the phases of the fundamental and the harmonic. So the Vpk of the composite waveform in that little slot becomes uncertain wrt the true Vrms of the composite waveform. This is because we don't know the relative phase of these two sinewaves. When this happens then you can't accurately convert from Vpk to Vrms anymore. if you explore all possibilities for the phase angle then you get a surprisingly large error in the simple Vpk to Vrms conversion and this makes the meter suffer very large measurement uncertainty due to the (uncertain) phase angle of the harmonic(s). In reality the angle will be set 'somewhere' and should stay like that but there is no knowing if it is at an angle that makes the meter read slightly low or makes it optimistic or anything inbetween. If you arrange a test setup that lets you rotate the phase of the harmonic you can explore the size of the uncertainty window. You can do it all with a SPICE simulator as well. You can also predict it using some simple sums. All three should agree pretty closely.