On 14.07.2016 00:18, Tanu Kaskinen wrote:
On Sun, 2016-06-05 at 21:05 +0200, Georg Chini wrote:
Hi,
finally I got around to split the loopback patches. The current version has
a lot of improvements over the last version and hopefully incorporates the
comments from Tanu's previous review. Main features of the new code are:
- Reduce latency jitter by a factor 100 from around 30 ms to 300 usec
- Implement modified P-controller and adaptive re-sampling
- Underrun protection
- Latency initialisation to avoid underruns
- Adjust to specified latency if possible, else use minimum possible latency
- Allow smooth switching between different sources and sinks
- 3 new parameters
The intermediate results are tested to compile but not for functionality.
More details can be found in the individual commit messages. The theoretical
background of the controller is covered in the document "rate_estimator.odt"
which I will send in a separate mail. The document also contains some results
achieved with the new code.
Georg Chini (25):
loopback: Save time stamps during latency snapshots
loopback: Improve latency estimation in snapshots
loopback: Improve latency estimation
loopback: Change memblockq length
loopback: Move timer restart and snapshots to time callback
loopback: Restart timer after 1/3 second
loopback: Adjust rate based on latency difference
loopback: Stop tracking max_request and min_memblockq_length
loopback: Refactor latency initialization
loopback: Do not skip audio at startup
loopback: Calculate and track minimum possible latency
loopback: Implement underrun protection
sample.c: Allow module-loopback to exceed PA_RATE_MAX
source.c, sink.c: Implement
pa_{source,sink}_get_raw_latency_within_thread() calls
alsa-{source,sink}.c: Handle PA_{SOURCE,SINK}_MESSAGE_GET_RAW_LATENCY
loopback: Reset sink input rate when source or sink changes
loopback: Track and use average adjust time
loopback: Limit controller step size to 2.01‰
loopback: Implement adaptive re-sampling
loopback: Add latency prediction and Kalman filter
loopback: Track prediction error; debug and cosmetic changes
loopback: Allow shorter adjust_time
loopback: Add low_device_latency parameter
loopback: Add adjust_threshold_usec parameter
loopback: Add log_interval parameter
src/modules/alsa/alsa-sink.c | 25 +-
src/modules/alsa/alsa-source.c | 20 +-
src/modules/module-loopback.c | 786 +++++++++++++++++++++++++++++++++--------
src/pulse/sample.c | 5 +-
src/pulsecore/sink.c | 30 ++
src/pulsecore/sink.h | 2 +
src/pulsecore/source.c | 30 ++
src/pulsecore/source.h | 2 +
8 files changed, 732 insertions(+), 168 deletions(-)
I pushed the first 8 patches now, since they're fine and I don't need
them to be resubmitted. (I complained about a variable name in patch 2,
but the variable can be renamed in a separate commit.) The 9th patch
had some issues, so patches from that onwards probably need to be
resubmitted. You can resubmit at any time you like, or you can wait
until I've commented on the rest of the patches.
Thanks for review and pushing the first patches. I'll wait until your
review is complete before I re-submit, otherwise I will probably have
to change things too often.
_______________________________________________
pulseaudio-discuss mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/pulseaudio-discuss
