demers

I've been trying to create a one-oscillator feedback FM patch based on the Roads book, but obviously I'm missing something. I scoured the Internet, and haven't found any clearer explanation either.

Any suggestions? A patch is below, it produces something, but I'm sure it's not what's intended! Thanks, Inouk

feedback-fm

On Nov 11, 2007, at 4:20 PM, Inouk Demers wrote:

> I've been trying to create a one-oscillator feedback FM patch based 

> on the Roads book, but obviously I'm missing something. I scoured 

> the Internet, and haven't found any clearer explanation either.

One thing to bear in mind is that, in the digital domain, there is no 

feedback without at least some latency, and this latency is going to 

put strange effects into the phase relationship of the feedback FM 

that are totally different to the analogue domain.

In your patch, there are two points of latency - one is the send~/ 

receive~ that will in this case introduce one vector size of latency, 

and the other is the fft~ which will incur your fft size of latency, 

in samples. If your vector size is e.g. 128 then there's 1152 

samples total, which corresponds to 26ms. That means a 360 phase 

delay at 38Hz, double that delay at each octave above.

Not sure what sonic effect this will have though.

An option to get around this (or minimize the delay at least) is to 

run the feedback fm inside a poly~ with vectorsize set to 2, and 

reading your phase directly from a driving phasor~ rather than using 

>
> Hello all,
>
> I've been trying to create a one-oscillator feedback FM patch based  
> on the Roads book, but obviously I'm missing something. I scoured  
> the Internet, and haven't found any clearer explanation either.

One thing to bear in mind is that, in the digital domain, there is no  
feedback without at least some latency, and this latency is going to  
put strange effects into the phase relationship of the feedback FM  
that are totally different to the analogue domain.

In your patch, there are two points of latency - one is the send~/ 
receive~ that will in this case introduce one vector size of latency,  
and the other is the fft~ which will incur your fft size of latency,  
in samples.  If your vector size is e.g. 128 then there's 1152  
samples total, which corresponds to 26ms. That means a 360 phase  
delay at 38Hz, double that delay at each octave above.

An option to get around this (or minimize the delay at least) is to  
run the feedback fm inside a poly~ with vectorsize set to 2, and  
reading your phase directly from a driving phasor~ rather than using  
fft.


> I've been trying to create a one-oscillator feedback FM patch based on the Roads book, but obviously I'm missing something. I scoured the Internet, and haven't found any clearer explanation either.

Does roads actually suggest to phase information of an fft and use that in the signal domain? Since you name yamaha, I think that model below comes close to what the dx7 does. Also I think that vector delay is not a problem at all for this application.

Here's another effort. I don't have the Roads book in front of me, so I can't be sure this is what he specifies. I would bet $$ this is a simplified version of a patch previously posted to the list.

Try running your patch with pitch at, say, 20, and offset at say 0.2, 

Now change the vector size and hear the differences.

They sound kind of cool... but since the vector delay has such a 

strong impact on the sound, I think it should be incorporated as a 

parameter itself; e.g. place a comb~ or tapin~ tapout~ in the 

Hey, why not also make the delay time variable by feedback too to 

On Nov 12, 2007, at 12:37 AM, jvkr wrote:

> Also I think that vector delay is not a problem at all for this 

Try running your patch with pitch at, say, 20, and offset at say 0.2,  
feedback at 0.25.

They sound kind of cool... but since the vector delay has such a  
strong impact on the sound, I think it should be incorporated as a  
parameter itself; e.g. place a comb~ or tapin~ tapout~ in the  
feedback loop.

Hey, why not also make the delay time variable by feedback too to  
make it doubly nonlinear...

>  Also I think that vector delay is not a problem at all for this  
> application.

Johan, you seem to have two things being fed in to the phase input of cycle~ (thanks for that: I was feeding back into the primary input), yet each produces different things.

The Roads diagram only shows the feedback being multiplied by a modulation index before being added to the 'offset' frequency. That seems to match the right hand portion of your patch, which works on its own. The middle portion seems like a version of simple fm using the phase input. Do I have that right?

Johan, you seem to have two things being fed in to the phase input of cycle~ (thanks for that: I was feeding back into the primary input), yet each produces different things. 

The Roads diagram only shows the feedback being multiplied by a modulation index before being added to the 'offset' frequency. That seems to match the right hand portion of your patch, which works on its own. The middle portion seems like a version of simple fm using the phase input. Do I have that right?
