Dynamically changing delay in gr_delay (or history in any gr_block)

GNU Radio
1

I made a simple example with a cosine and a delayed version of that
going through
a multiplier, and observing the output together with a slider that
changes the delay dynamically.
However i do not see any change (eg, elimination of the dc component
when delay ~ pi/2).

Looking at the code of the block gr_delay it shows that the delay can
dynamically change
and this results essentially in a call to set_history() in gr_block.
Looking further into the set_history() method, it sets the private
variable d_history.

The question is: does the dynamic change of history have any effect in
gr_block?
What happens in the guts of the block when d_history changes?
and why don’t i see any effect in the example above?

thanks
Achilleas

2

On 15/11/2011 3:00 PM, Achilleas A. wrote:

and this results essentially in a call to set_history() in gr_block.
Looking further into the set_history() method, it sets the private
variable d_history.

The question is: does the dynamic change of history have any effect in gr_block?
What happens in the guts of the block when d_history changes?
and why don’t i see any effect in the example above?

thanks
Achilleas
I ran into the exact same thing a couple of weeks ago. It has to do
with the deep-structure of the schedule, and Jonathan Corgan had
indicated he was going to look into it.

The weird thing is that FIR filters do the same thing when you change
the number of tapes (muck with d_history), but they actually
work dynamically after you change them.

3

I actually tried using filters to implement the delay and they do NOT
work as expected:

I used “Interpolating FIR filter” with taps equal to (0,)*delay+(1,)
and i didn’t see any difference as i was changing the delay parameter
dynamically…

Achilleas

On Tue, Nov 15, 2011 at 3:05 PM, Marcus D. Leech [email protected]
wrote:

Looking at the code of the block gr_delay it shows that the delay can
thanks

–

Achilleas A.
Associate Professor
EECS Department Voice : (734)615-4024
UNIVERSITY OF MICHIGAN Fax : (734)763-8041
Ann Arbor, MI 48109-2122 E-mail: [email protected]
URL: http://www.eecs.umich.edu/~anastas/

4

On 11/15/2011 05:01 PM, Achilleas A. wrote:

The filtering still seems to work, but the delay (based on d_history)
appears not to.

–
Marcus L.
Principal Investigator
Shirleys Bay Radio Astronomy Consortium

5

Take a look at the attached grc file:

As implemented, it does not work (changing the delay does not have an
effect).

If I introduce the “fictitious” filter (1,0,0,0,0,…)
it works as expected.

AM I doing something wrong in the first case?

Achilleas

On Tue, Nov 15, 2011 at 5:26 PM, Marcus D. Leech [email protected]
wrote:

Achilleas

–
Marcus L.
Principal Investigator
Shirleys Bay Radio Astronomy Consortium
http://www.sbrac.org

–

Achilleas A.
Associate Professor
EECS Department Voice : (734)615-4024
UNIVERSITY OF MICHIGAN Fax : (734)763-8041
Ann Arbor, MI 48109-2122 E-mail: [email protected]
URL: http://www.eecs.umich.edu/~anastas/

6

Marcus,

The grc file that i sent does not support your theory:
I have 2 filters:

one with taps: (0,)*delay+(1,)
and the other with taps (1,)+(0,)*delay

then I can change the delay dynamically, which also means that the
history is also changed dynamically (NOT ONLY AT INIT) and there is no
problem
whatsoever.
Clearly the second filter is not needed but if i get rid of it then the
whole
thing does not work…

I am really confused…
Achilleas

On Tue, Nov 15, 2011 at 6:21 PM, Marcus D. Leech [email protected]
wrote:

AM I doing something wrong in the first case?

Shirleys Bay Radio Astronomy Consortium
http://www.sbrac.org

–

Achilleas A.
Associate Professor
EECS Department Voice : (734)615-4024
UNIVERSITY OF MICHIGAN Fax : (734)763-8041
Ann Arbor, MI 48109-2122 E-mail: [email protected]
URL: http://www.eecs.umich.edu/~anastas/

7

On 11/15/2011 06:16 PM, Achilleas A. wrote:

It seems that the runtime machinery pays attention to d_history only
on block init, and at no other time, which leads to unexpected
results. But, surely, this must have worked at some point?

I mean, I regularly use filters whose parameters I change dynamically,
and they apparently do what I want, although, perhaps at the moment
of changing parameters, the phasing isn’t “right”, but they seem to
work.

Someone with more exposure to the gr-runtime stuff should comment here.

–
Marcus L.
Principal Investigator
Shirleys Bay Radio Astronomy Consortium

8

On 11/15/2011 07:24 PM, Achilleas A. wrote:

whatsoever.
Clearly the second filter is not needed but if i get rid of it then the whole
thing does not work…

I am really confused…
Achilleas

Yes, there’s some deep weirdness.

The only time the runtime pays attention to history, in my second pass
through there, is in forecast() computations, via a little
inlined function called “history()”, which just returns “d_history”.
Not sure what this means.

But clearly, there’s some brokenness there, and it’s not clear to me
how to fix it, although I desperately want dynamically-settable
gr_delay() blocks to work as well, for some interferometry work.

–
Marcus L.
Principal Investigator
Shirleys Bay Radio Astronomy Consortium

9

On Tue, Nov 15, 2011 at 16:44, Marcus D. Leech [email protected]
wrote:

But clearly, there’s some brokenness there, and it’s not clear to me how
to fix it, although I desperately want dynamically-settable
gr_delay() blocks to work as well, for some interferometry work.

Sorry, I’ve been buried in some commercial projects that haven’t left me
with much time to look at the issue more in depth.

I can confirm that dynamically changing the delay in gr_delay does not
(and
never has) worked. The fix for this is not that complex but it would be
a
change in the runtime that would need a lot of regression testing.

In the FIR-based delay scenario, it would be useful to know if there is
a
difference between changing the taps from, say [0, 0, 0, 0, 1] to [0, 0,
1], so the taps end up changing length, or changing the position of the
1
but leaving trailing 0’s so the tap length would remain the same.

Johnathan

10

On 11/15/2011 11:19 PM, Johnathan C. wrote:

me with much time to look at the issue more in depth.
What, you want to feed your kids in preference to helping a buncha
feeloaders like us? :slight_smile: :slight_smile:

11

I can confirm that if my FIR taps are defined as:

(0,)delay+(1,)+(0,)(19-delay)

then i can change the delay dynamically for any values between 0 and 19
in the grc file that i sent earlier (attached here as well) EVEN
without the second (trivial)
FIR filter present!

So it is clearer now that the problem with delay/filters is in the
setting of history…

thanks for the hints!
Achilleas

PS: Johnathan, could you give us a simple explanation of what happens
in the internals when there is a history change so that at least we
understand the computational model.

On Tue, Nov 15, 2011 at 11:19 PM, Johnathan C.
[email protected] wrote:

never has) worked. The fix for this is not that complex but it would be a
change in the runtime that would need a lot of regression testing.

In the FIR-based delay scenario, it would be useful to know if there is a
difference between changing the taps from, say [0, 0, 0, 0, 1] to [0, 0, 1],
so the taps end up changing length, or changing the position of the 1 but
leaving trailing 0’s so the tap length would remain the same.

Johnathan

–

Achilleas A.
Associate Professor
EECS Department Voice : (734)615-4024
UNIVERSITY OF MICHIGAN Fax : (734)763-8041
Ann Arbor, MI 48109-2122 E-mail: [email protected]
URL: http://www.eecs.umich.edu/~anastas/

12

On 11/15/2011 12:00 PM, Achilleas A. wrote:

and this results essentially in a call to set_history() in gr_block.
Looking further into the set_history() method, it sets the private
variable d_history.

The question is: does the dynamic change of history have any effect in gr_block?
What happens in the guts of the block when d_history changes?
and why don’t i see any effect in the example above?

I dont think that set history works quite that way. I made a new delay
block in gr-basic. Demo in GRC attached. Here is the implementation:
http://gnuradio.org/cgit/jblum.git/tree/gr-basic/lib/gr_basic_delay.cc?h=gr_basic

-Josh