Phase modulation feedback averaging,yiehaa

gentleclockdivider

https://community.native-instruments.com/discussion/comment/199547#Comment_199547

Yes , you can really clearly see it on a scope

Simple 1;1 ratio mod-carrier structure .

All it takes is an feedback enabled operator (feedback to max value of 7 ) and a carrier , both set to ratio 1;1

Dexed and ops7 , should have a modulator value at 79(and velo sclaing to full) , everything beyond that and the noise will creep in ,

Dexed most def. uses averaging .

colB

https://community.native-instruments.com/discussion/comment/199813#Comment_199813

To me the most obvious demonstration is an operator modulating itself, getting close to a saw waveform. The averaging filter in the feedback (and removing the 1 sample delay) enables it to get closer to a saw sound, with more edge, before it starts collapsing into noise.

colB

https://community.native-instruments.com/discussion/comment/199584#Comment_199584

Thinking Some more about this, the original DX7 mapped it's envelopes to some log curve, likely from a small lookup table, so the feedback amplitude would have been steppy, this filter would have acted as a linear interpolator (that's what it's called in the core macro).

It wouldn't have fixed the problem entirely, but would make a big difference.

gentleclockdivider

https://community.native-instruments.com/discussion/comment/199947#Comment_199947

Aha you're right about a single (feedback ) operator being the best method to get a saw shape
But , on the dx7 , the only structure where a feedback operator is direcly heard (so it's a carrier ) is structure 32 , and the dx7 uses some kind of method where the more carriers there are , the lower the output sound of the carrier , and the output is directly proportional to the feedback amount .
Try it , algo 32 , last operator set to full output and feedback to 7 , it will give you a saw but will not turn into noise ,simply becasue of the operator output does not provide enough gain (becasue of the 6 carriers ) .
btw .

Dexed got this wrong , ops7 got it right ( wel it's a bit accurate replication )

gentleclockdivider

I 've got incredibly close with the numeric scaling of the operator output

On the dx7 it's a value between 0-100 , for the reaktor ensemble I've used a value between 0-1 to the 5th power , this gets really close .
I also added velocity scaling for the operator

Cabledx7.ens

gentleclockdivider

Nah forget about the visual scaling readout , still some work to do
Gainwise , it's o.k.

gentleclockdivider

Our own cute op-2 , with tx waveforms :)

2op.ens