How much HD2 suppression is needed for the LO synth going into a quadrature demodulator or modulator? How does LO signal even-order distortion affect demod image rejection and modulator side-band suppression?
Myself, Ian and Eamon Nash have been looking at this - still on-going. From prelim measurements it does look necessary to filter the
LO signal, if the PLL contains an integrated VCO. For example the ADF4350 has a 2H level of -19dBc and a 3H of -13dBc.
In terms of how much suppression is required, it looks like you need to get the harmonic levels at or below -40dBc. This is achievable with a
5th order Chebyshev filter.In-band ripple is typically not a concern.
We have not figured out which harmonic 2H or 3H is having the larger effect.
One theory as to why the harmonic level is effecting the modulator specs. like sideband suppression and image rejection is due to the way the
quadrature is being generated on-chip by means of a polyphase filter, which requires low harmonic content to generate accurate quadrature.
Ian has done the most work on this and could comment further....
It is my understanding that the mod/demod performance degradation is due to the even order distortion content. All of the LO paths on our mods/demods employ significant gain or limiting to ensure the LO waveform that is driven into the mixer cores is squarewave. A squarewave with perfect 50% duty cycle will have no even order distortion and rich odd-order distortion. If you review fourier coefficients for a 50% duty cycle square wave you should find that the 3rd harmonic is ~9.5dB below the fundamental. That's crummy HD3, but we know from past work with our mods/demods/mixers that the 3rd harmonic is not an issue. If it was we would have problems with our waveform limiting in the LO paths of our existing mods/demods. I suspect the even order distortion performance is critical in mod/demods that use RC polyphase networks, and that components that utilize a quadrature divide by 2 LO path that the even order distortion is more tolerable, at least for waveforms where the fundamental and odd order harmonics dominate. The tough part about measuring the sensitivity to even order harmonics lies in the waveform generation. We may need to consider a few test setups in order to have a means to control the level of even order distortion we inject into our devices under test.
I honestly don't expect any of us to have the perfect answer to this complex question just yet. I just wanted to use this issue as a test pilot to help me learn to navigate the forum (and to elevate the importance of even order distortion on wideband synth solutions).
Thx for your reply,Eric
I'll toss in my 2 cents worth...
I agree with Eric, the mixer core is fundamentally a switch to flip the phase of the carrier +-180 degrees (or multiple by +-1 or 0,1 as does the ADL5350), so you want to switch the core with a square wave with minimum rise and fall times (minimizes noise in the transition zone), which means a square wave with ideally an infinite number odd-order harmonics (I saw a cool lecture demo of this once upon a time).
But. based on what Eric said, you need to drive the polyphase filters with a well-filtered sine wave (which makes sense, they're designed around a reasonably narrow bandwidth vs. octaves) THEN limit the quadrature outputs to drive the core. The exception is a /2 or /4 quadrature generator.
Sounds like fodder for an app note!
I guess another consideration is the half-IF issue. By my calculation, if the HD2 of the LO has energy at (RF - IF/2), then you will create co-channel. So we could work the numbers backwards from there for particular SNR and co-channel rejection requirements...
Hi Guys. Attached are some simulation results that seem to prove that the sideband-suppression/image-rejection degradation is dominantly due to 3rd LO harmonic quadrature amplitude mismatch more than anything else. We should be able to support this claim with some 1xLO mod/demod measurements using a notch filter to tailor the levels of HD2 and HD3 applied to the mod/demod. I suspect if one goes through the rigorous quadrature math including LO harmonics and frequency dependent quadrature errors the responsible mixing term will be more apparent. I'll save that pen-and-paper exercise for another day.
Thx for everyone's thoughts on this subject. I'll close out the question for now, but more lab work and engagement with our design teams is still warranted.
Retrieving data ...