I’m using a flow graph with no GUI to demodulate a signal. I have a
selector block that allows me to select the source of the signal between
the audio card and a wav file. I can change the source using an xmlrpc
server/client.
If I don’t use a throttle block with the wav file, the flow graph takes
all
the CPU and the server/client is very slow.
If I put a throttle, I get audio underruns when using the audio card.
Same
problem for a constant source.
Can anyone please recommend a better way to do this?
I’m using a flow graph with no GUI to demodulate a signal. I have a selector
block that allows me to select the source of the signal between the audio card and
a wav file. I can change the source using an xmlrpc server/client.
If I don’t use a throttle block with the wav file, the flow graph takes all the
CPU and the server/client is very slow.
If I put a throttle, I get audio underruns when using the audio card. Same
problem for a constant source.
You can’t get audio underruns from an audio source, so I assume you
have an audio sink that you didn’t mention in your description that
you’re using in both modes for the output of the demodulator.
It sounds like you have the audio sink set not to block. In its
properties, enable the OK to Block checkbox. Do not use a throttle.
This will allow the audio sink to limit the data rate from the wav
source (by backpressure). When an audio source is being used, if the
source and sink are from the same sound card and have the same sample
rate requested then they should be happy too.
Sorry, it was a mistake. For the list, just clarify that
I have an audio sink using the same audio card @ 96 KHz with the
settings
set to OK to block YES.
Here are some better solutions, from simplest to best:
else that acts like it.)
(the demodulator and sink) into a GRC block which is used by the Python
program. Either way works.
Thanks, I’ll try these solutions tomorrow. I assume that they also apply
to add a constant block into a selector block.
usage in any other way.
that its documentation lies. It does not leave its unused inputs
Using some method to force the wav source and audio source to match
sample rates. Specifically, you could use a “Multiply by Matrix” block to
replace the function of the Selector entirely: give it a matrix value of
either ((1, 0),) or ((0, 1),) to select just one input.
Can I make a multiply to matrix block have several inputs and one only
output? If not, I was thinking on multiplying one of them by a constant
block zero and then adding both outputs. Would this be a valid
workaround?
Using some method to force the wav source and audio source to match sample
rates. Specifically, you could use a Multiply by Matrix block to replace the
function of the Selector entirely: give it a matrix value of either ((1, 0),) or
((0, 1),) to select just one input.
Can I make a multiply to matrix block have several inputs and one only output?
If not, I was thinking on multiplying one of them by a constant block zero and
then adding both outputs. Would this be a valid workaround?
The number of inputs and outputs are determined by the dimensions of the
matrix value entered. If you put in ((0, 1,),) with the comma (actually,
I just remembered you can write [[0, 1]] as well and you don’t have to
worry about the extra comma (blame Python for that)) then it will have 2
inputs and 1 output. And, for example, if you wrote [[0], [1]], then it
would have 1 input and 2 outputs.
By the way, you can of course enter any numeric values to scale the
inputs as you like, so for example you could use [[0, 0]] as a mute
mode, or enter an arbitrary number as a gain control, if you wanted one.
(If you did have an extra output that you couldn’t get rid of, a null
sink would be simplest there.)
Even selecting source and sink from the audio card, if I have a wav file playing
in another input of the selector without a throttle, the CPU goes to 100%. To
avoid it I put the wav file, then the throttle and the the selector. I’ve read
that is bad practice but I couldn’t reduce the CPU usage in any other way.
(Please make sure to CC the mailing list in your replies.)
Ah. I just looked at the implementation of the selector block, and found
that its documentation lies. It does not leave its unused inputs
disconnected (which is actually good, because that would cause errors in
some cases), but connects them to null sinks (which consume all the
input they’re given).
So, using a throttle is a workaround for this. (The cost is that since
the throttle’s timing is based on the CPU clock, not the audio card
clock, you’ll occasionally get either underrun or overrun.)
Here are some better solutions, from simplest to best:
Using some method to force the wav source and audio source to match
sample rates. Specifically, you could use a Multiply by Matrix block to
replace the function of the Selector entirely: give it a matrix value of
either ((1, 0),) or ((0, 1),) to select just one input.
Modifying, or creating a replacement for, the selector block which
does not connect to a null sink but rather to a block of some sort which
never accepts any data and so stops that branch of the flowgraph. (I
don’t think such a block actually exists, but it could, and there might
be something else that acts like it.)
Actually remove from the flowgraph whichever of the wav source and
audio source are not currently in use. This is the most general,
efficient, and straightforward solution, but unfortunately it cannot be
built within GNU Radio Companion – you must write some C++ or Python
code. If you don’t want to stop using GRC entirely, you could isolate
that part by putting the connection changing logic inside a hierarchical
block; this would be similar to the selector block itself, but it would
be a source, which has the wav and audio sources “built into” it.
Alternatively, you could write your main()/top block as a Python program
and put the rest of your logic (the demodulator and sink) into a GRC
block which is used by the Python program. Either way works.