[PD] high frequency control data

Andrew W. Schmeder andyschm at eh2o.ws
Wed Nov 7 04:55:29 CET 2001


> I am attempting to use a phasor~ linked to a threshold~ unit to generate
> high frequency bangs to then trigger a reasonably complex sample playback
> mechanism, attempting to mimic granular synthesis.  The problem is that
> when i bring the phasor frequency up into the range of audible pitch, say
> above 100hz, i can hear a low frequency cyclical shift in the sound
> texture, which theoretically should be completely even.

Seems like just using metro for triggering would accomplish the same thing.

I believe that control messages can only be processed on vector endpoints - 
i.e. the finest possible resolution is about 1.4 milliseconds with a 
64-sample dsp vector...  someone please correct me if I am wrong.

This means that the theoretical maximum rate of reliable triggering is ~689 
events per second -- although jitter will probably arise at significantly 
slower speeds... in fact you are guarenteed to have jitter if the frequency 
of triggering is not an integer submultiple of the vector-processing speed.

However this is not to be confused with the behavior of the clock when using 
delays, qlists, metros, etc - where you can schedule events to the 
millisecond at 64-bit floating point accuracy--- see the pd documentation for 
the details on this.

For granular synthesis try using the output of phasor~ directly to control 
the playback operation.. i.e. use phasor~ to drive tabread~.  Then convert 
all parameters to signals and use samphold~ to achieve sample-accurate 
control.... all in the signal domain.  I believe that this method will be 
less computationally expensive also.  

Since phasor~ cannot take signal input for phase I've considered using 
sample-delays to implement accurate phase shifts for very small grain 
windowns (untested theory).  The other options are to do tricky stuff with 
the block size and overlap or use some completely different method like 
convolution (also untested ideas for me).


andy




More information about the Pd-list mailing list