This reverts commit 8e0a1cbdc8.
c.f. #772
The underlying issue is still the same. You will always find something that does not match when trying to compare the DDS with the parallelized DUC. They are just different. I could correct it for phase but then it will fail for amplitude. Or you'll compare the offset channel to phase1 or amplitude1. Let's state that equal things are well synchronized but unequal things may have a deterministic latency difference of strictly less than one coarse RTIO cycle.
Don't try to tweak out the last bit of latency by feeding the HBF input
early. Instead feed it late so the interpolated samples are early and
the latency is an even multiple of the super-sample cycle.
* unify I and Q data limiters. there is no conceivable way why they
would be different.
* reorder clr bits to be like consistent
* move the sat add limiter to before the hbf again
* give 1 bit headroom to interpolator to handle overshoot
* fix Config limiter widths (NFC)
* move clipper to behind the HBF to correctly shield DUC
This leaves a factor of two headroom for the sum of the following
effects:
* HBF overshoot (~15 % of the step)
* A1/A2 DDS sum
While this is technically not sufficient and can still lead to
overflows, it is unlikely that one would trigger those. It would require
doing large amplitude A1, large amplitude A2 and additionally doing
amplitude/phase jumps that would overshoot the HBF. No sane person would
try that, right?
closes#743
