[mythtv-users] Enlighten me about 1080 "interlace"

[Doug Larrick]

Joe Barnhart wrote:
> I thought I understood this interlace stuff but Myth
> is challenging my concept of interlace vs.
> non-interlace video.
> 
> I have a TV capable of displaying 1920x1080
> interlaced.  It's a CRT based RPTV from Pioneer, if
> you care.  I have watched terrestrial broadcasts on
> HDTV and it looks breathtaking.  The image consists of
> two alternating 540-line frames, each 1/60 second,
> displayed at even and odd scan lines for 1080 unique
> lines of information.
> 
> Now I get my Myth box running and it looks very, very
> good.  But not quite as good as my TV with a direct
> OTA receiver.  Why not?  I've set everything up as
> carefully as possible.  I'm on stock 0.16 Myth, I have
> an Intel 3GHz processor running an nVidia 5500 card
> with Xv.  The modeline is correct and the card is
> doing interlaced out, as evidenced by the flickering
> on 1-pixel lines with an xterm on screen.
> 
> Some people on the list whose advice I trust very much
> have suggested that 540p and 1080i output are
> _identical_ as played by Myth.  This certainly seems
> to be true.  I have played with the "deinterlace"
> setting and the "bob and weave" deinterlace (which I
> understand duplicates one of the frames and tosses the
> other) looks the same as interlaced on my set.

There is no "bob and weave" deinterlace in MythTV.  The relevant 
algorithms are "onefield," which behaves as you say -- tosses half the 
fields; and "bob," which shows each field sequentially (one field each 
1/60 sec.).

The other two algorithms are "linearblend," which retrieves resolution 
from low-motion scenes at the expense of ghosting during horizontal 
motion; and "kerneldeint," which does better but is too 
resource-intensive for HDTV on today's hardware.

It's worth noting once again that if you're using XvMC, none of these 
algorithms are used.  If you choose any of them, it will use the only 
method XvMC is capable of: bob implemented in hardware.  This is not a 
software limitation.

There are more-advanced algorithms out there, notably from the 
open-source (but for Windows <boggle>) DScaler, and from TVTime.  The 
software architecture of how Myth displays video is not amenable to 
easily adapting these filters, which is a shame, because the TVTime ones 
seem to be developing into somewhat of a standard.

When displayed in a 540p frame, bob deinterlacing should look identical 
to native 1080i, at least on a CRT with relatively long-persistence 
phosphor.  Make sure you're using a 1920x540 video mode (not 960x540), 
or you *are* tossing out half your horizontal resolution.

I believe that on some sets, native 1080i output would look better than 
the fake 540p that we use.  However, there appears to be a bug in 
nVidia's drivers, where *video* is not displayed at the full resolution, 
but instead undergoes some sort of deinterlacing.  Pause some 
horizontally-panning scene when in 1080i and you should see flicker, but 
you don't.  Normal GUI stuff is fine; it's the Xv overlay that's screwed up.

In conclusion, I think that for a 1080i-native set, 1920x540 with bob 
deinterlacing is the best you'll get out of Myth right now.  If your set 
does 1080*p*, it should be even better, though some of those 
more-advanced algorithms would help.

-Doug
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 256 bytes
Desc: OpenPGP digital signature
Url : http://mythtv.org/pipermail/mythtv-users/attachments/20041221/ccc57d06/signature.pgp