[Discuss-gnuradio] Audio Source and Sink, mostly in GRC

[Art Botterell]

Hi All -

Noob here, but I've spent a number of hours googling and fiddling this to no 
avail, so I'll risk bringing it to the list:

I just did a fresh install of Karmic on an i5 Intel box and installed 3.2.2.  
Starting out at baseband I have an on-board Intel sound plus two USB audio 
devices (one of them built into an HF rig), and they all work fine... except 
with GNURadio, where I'm having some... um... issues.

Please excuse the length and my basic grasp of things, but here's what I'm 
seeing:

1) Initially dial_tone.py works fine from the command line (although the first 
time I launched it it said "gr_vmcircbuf_createfilemapping: createfilemapping 
not available")

2) Using Ubuntu's Sound Preferences pane I set the USB device in the Icom as 
default input, then open GRC and send audio from an audio source associated 
with "default" to the fft_sink, all at 24K samples.  That works fine, although 
CPU utilization goes to 100% on one of my cores and every 5-30 seconds the 
thread appears to switch from one core to another.  

3) Adding a throttle in the middle brings the CPU down off the ceiling a bit, 
varying between about 70-100 percent, and stops the core-swapping... but after 
a few minutes it's back on the roof and switching cores again, albeit not quite 
as frequently.  With all three components at 48K it takes about 15 seconds for 
the busy thread to get up to 100%, as opposed to about 5 seconds at 24k.

4) But I substitute an audio_sink for the fft_sink, and set the chain to 24K 
samples, I get a few seconds of stuttering audio and about thirty "aU" 
underflow flags, and then it appears to freeze with CPU utilization stuck at 
100%.

4) dial_tone.py still works fine to the same output, and I can swap back to the 
fft_sink with similar results as before.

6) OK, so I did a simple mod of dial_tone.py, substituting an audio.source for 
one of the tones and disabling the other, and adjusting the connections 
accordingly.  Now I get fairly stable audio from the USB source at 48K, with a 
few underflows at the start and 100% on one of the cores... but the latency is 
about two seconds!

So I'm wondering:  Is it possible there might be something going awry between 
audio_source / audio_sink and my ALSA (1.0.20) driver that's making them drive 
the CPU so hard?  Or am I just missing something basic about how to wire things 
in GRC and/or GNURadio?

Also I'm wondering what's the difference between connections under GRC and 
directly from Python?

(And also I'm a little vague on what values to use for the "PCM Device Name" in 
setting up audio sources and sinks using various devices.)

Yep... all told I'm pretty much a lost ball in high weeds.  Please have mercy!  
Any insights or pointers will be much appreciated.

- Art