[PD] 0 length delay in delwrite~

[Matt Barber]

There's one trick that works for zero delay and [delwrite~ 0], based on the
code of [delread~]: provide a negative delay time (with abs(delay time) > 1
sample).

Here's what's going on in Alexandre's patch.

[delwrite~ 0] sets up a buffer that is 68 samples long, plus four extra for
interpolation in [vd~], which we needn't worry about here.

delwrite's float method is this:

/*---------------------------------------------------------------------------*/
static void sigdelread_float(t_sigdelread *x, t_float f)
{
    int samps;
    t_sigdelwrite *delwriter =
        (t_sigdelwrite *)pd_findbyclass(x->x_sym, sigdelwrite_class);
    x->x_deltime = f;
    if (delwriter)
    {
        int delsize = delwriter->x_cspace.c_n;
        x->x_delsamps = (int)(0.5 + x->x_sr * x->x_deltime)
            + x->x_n - x->x_zerodel;
        if (x->x_delsamps < x->x_n) x->x_delsamps = x->x_n;
        else if (x->x_delsamps > delwriter->x_cspace.c_n - DEFDELVS)
            x->x_delsamps = delwriter->x_cspace.c_n - DEFDELVS;
    }
}
/*---------------------------------------------------------------------------*/

Here, I believe that because the write and read are sorted, x->x_zerodel ==
0. Also DEFDELVS == 64, and delwriter->x_cspace.c_n == 68;
So with [delread~ 0], x->x_delsamps is initialized to 0 + 64 - 0. The first
if statement fails, but the second one passes, and x_delsamps is set to c_n
- DEFDELVS = 4.

Let's pick it up after [delwrite~]'s second block calculation. It's just
gotten the range 64:127 in its input vector, and written indices 64:67 with
64:67, and indices 0:59 with 68:127. Its phase is now at index 60. With
x_delsamps set to 4, [delread~ 0] begins reading at index 60-4 = 56, which
contains the value 124. For four samples, the difference between input and
output (the [-~] in the patch) is -60. Indices 60:67 contain values 60:67,
and 0:59 contain 68:127; the difference between input and output is
therefore 8 for the 60 remaining samples of the block. Hence the
oscillation between 8 and 60 in the number box, with 8 appearing most of
the time.

If you set a negative delay, the first if statement passes, and x_delsamps
is set to 64, as it should be to get zero delay (that is, it's reading the
last 64 samples written to the delay buffer, so there's zero difference
between input and output).

Matt




On Sat, Dec 12, 2015 at 12:00 PM, Alexandre Torres Porres <porres at gmail.com>
wrote:

> > Order forcing works well for me. Just set the [delwrite~] to 10 (
>
> but weirdness arises from 0 length delay.
>
> > If a delay of zero WAS actually permitted, these would form infinite
> loops.
>
> Don't think so, depends on what a 0 delay is. If it is "no delay", and by
> that I mean "No Buffer", then it only outputs 0 values, right?. In other
> words, nothing happens, no infinite loop, nothing, just zeros...
>
> but if instead of zeros it does have some buffer length in it, then a
> feedback delay will always de delayed at least one block size, so no
> worries about feedback loop.
>
> Actually, you need to set delread to "0" for a minimum block size delay.
> Perhaps you meant you couldn't or shouldn't put a "0" delay time in the
> delread~ object for feedback, but actually you NEED to do that.
>
> Thing is that I just use delwrite~ and delread~ with 0 length arguments
> for both and a block size of 1 to allow single sample feedback. I do it
> cause I wanted the minimum delay buffer size as possible and I didn't want
> to write in tiny and long and boring numbers according to one sample size
> depending on sample rate.
>
> Since it was working, I had just always assumed it would create a buffer
> of one block.
>
> This is not what's really happening as I see it.
>
> I don't really care that much on what happens, doesn't seem like a big
> deal, but it was nice to understand this behaviour. It doesn't seem very
> consistent, that's all I can say...
>
> Now, what it actually does is really just a matter of design choices. It
> could very much just create no delay buffer at all, where you'd get 0
> values perhaps, like I imagined. That's silly anyway...
>
> Or... it could be only one sample... or one block... I had assumed out of
> nowhere that it could be a block size, but it could much be just a single
> sample, which seems to make sense and it'd be cool I guess.
>
> What's really bad is that you need to always put a value that is at least
> one block size. It's a bug considering the documentation clearly stated
> that the design was really supposed to be a delay between 0 and max delay
> size, but one way or another, it's really annoying doing all this math as
> a workaround, when it's just a matter of coding it properly to allow any
> size greater than 0 and smaller than a block size (in orther words, to fix
> it).
>
> > If you, however, want a simple block delay in a feedback loop,
> > just use a pair of [send~] and [receive~].
>
> don't work for block size < 64
>
> > specifying the buffer size makes much more
> > sense then giving a maximum delay time
>
> Those two things means the same to me, where maximum delay time = buffer
> size. I don't get this.
>
> > [delwrite~] object would need to keep track of this
>
> sure, whatever, why not?
>
> by the way, that's the one that defines the max delay length (or buffer
> size), (and there can be only one, by the way) - so it only needs to keep
> track of its block size to work out the proper buffer size.
>
> I might see an issue if delread~ is in a subpatch that has a longer block
> size, but I don't wee why anyone would need that, and perhaps just say you
> shouldn't do it.
>
> I think we've discussed this before, perhaps just make sure both are in
> the same block size. I for one, never needed them to be in different block
> sizes, makes no useful sense.
>
> But anyway, I guess Miller is the one that should hop in and share his
> thoughts.
>
> and lets not forget the "clear" method, also important :)
>
> cheers
>
>
> 2015-12-12 12:33 GMT-02:00 Roman Haefeli <reduzent at gmail.com>:
>
>>
>> On Fri, 2015-12-11 at 14:26 -0200, Alexandre Torres Porres wrote:
>> > hi, I'm checking that if you put a 0 length delay in delwrite~ you
>> > still have some buffer
>> >
>> >
>> > what's up with that? and how does it work? how big is it when you
>> > don't define it?
>>
>> Don't know. I confirm that weirdness happens when setting [delwrite~] to
>> 0. In other words: The only value that does not justify using a delay at
>> all shows unexpected behavior. I can live with that.
>>
>> > I always get a minimum dealy even with order forcing (and different
>> > values depending on extended orvanilla), it seems the delay needs to
>> > be at least the block size to work properly
>> >
>> >
>> > moreover, I can't have an order forcing
>>
>> Order forcing works well for me. Just set the [delwrite~] to 10 (and
>> [delread~] to 0) in your patch and the number box shows 0 (this means
>> zero delay, right?).
>>
>> Roman
>>
>>
>> _______________________________________________
>> Pd-list at lists.iem.at mailing list
>> UNSUBSCRIBE and account-management ->
>> http://lists.puredata.info/listinfo/pd-list
>>
>>
>
> _______________________________________________
> Pd-list at lists.iem.at mailing list
> UNSUBSCRIBE and account-management ->
> http://lists.puredata.info/listinfo/pd-list
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.puredata.info/pipermail/pd-list/attachments/20151212/824edf3a/attachment-0001.html>