[mythtv] AVSync2 Refinements

[Tim Pletcher]

On Mon, Dec 9, 2019 at 10:11 AM Tim Pletcher <pletchtd at gmail.com> wrote:

>
>
> On Mon, Dec 9, 2019 at 9:06 AM Mark Kendall <mark.kendall at gmail.com>
> wrote:
>
>> On Mon, 9 Dec 2019 at 00:32, Tim Pletcher <pletchtd at gmail.com> wrote:
>>
>> > Thank you for your insights Peter.  I have been testing and debating
>> whether to actually include any filtering or not. I do think true
>> proportional control with gain settings less than 1 is better as this
>> minimizes the control oscillation once near setpoint while still giving
>> somewhat faster initial sync. I do notice a significant difference in the
>> amount of variability from my two intel apollo lake based front-ends using
>> VAAPI versus my combined i5 frontend/backend using an NVidia GT1030 with
>> NVDEC so I think the proposed change is more impactful on low end hardware.
>> >
>> > As David suggests, I will create a ticket and submit a patch.
>>
>> Tim
>>
>> Many thanks for this - very handy:)
>>
>> I recently got caught out by the fluctuating av sync on the raspberry
>> pi. I've taken the liberty of producing a more extensive patch
>> (attached). This adds your filtering as well as removing all of the
>> the old a/v sync code and the settings for AVSync2 - i.e. it moves
>> over completely to the new av sync code and removes the audio
>> correction adjustment. (N.B. the patch is against today's  latest
>> master - I pushed some cleanup of  MythDVDPlayer and the patch won't
>> apply cleanly before that).
>>
>> It appears to work very well. On two (linux) desktop dev machines and
>> a pi 4 I can't see any issues - playback is nice and smooth. As you
>> say, initial sync seems to be much better as well.
>>
>> Certainly gets my seal of approval!
>>
>> One question - last_fix - should that be carried over between calls to
>> AVSync()? At the moment it is a local that gets reset between each
>> call - which I'm not sure is the intended behaviour.
>>
>> Thanks again and regards
>> Mark
>>
>>
> Hi Mark,
>
> Thank you for looking at this in detail.
>
> You are exactly right in that last_fix doesn't get carried over between
> calls. The last_fix value is actually needed to correctly do the signal
> filtering to the control loop.  Over the weekend, I eventually figured out
> the filtering wasn't actually working after adding more debugging detail to
> the logging for better visibility of what was happening mathematically
> between calls.  This is also the reason I arrived at the conclusion that
> filtering is probably not really needed since performance is good anyway
> and a filter brings the drawback of adding lag to the control response.
>
> In reality, with the broken filter application, the code was effectively
> using a control gain of (0.4) * (1 - 0.9) = (0.04).  This setting seems to
> still provide rapid initial sync while preventing over-control in response
> to natural audio timestamp jitter.
>
> The patch looks great.  Based on discussion above, I think we should
> remove the filtering and set:
>
> s_av_control_gain = 04;
>
> The last_fix and s_sync_fc references should go away alltogther and the
> new control calculation becomes:
>
> float fix_amount = audio_adjustment * sign * s_av_control_gain;
>
> Should I still create a Trac ticket for this?
>

Correction:

It should be:
s_av_control_gain = 0.04;

Tim
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mythtv.org/pipermail/mythtv-dev/attachments/20191209/35b0f265/attachment.htm>