1080i playback blur

Assuming you have an nVidia card, I'd guess it's just the same issues everyone else has with Linux nVidia drivers and 1080i content output at 1080i.

I have a Hitachi RP CRT TV, so 1080i is of course the native resolution. It kills me that I have to enable deinterlacing in MythTV to make 1080i content watchable because nVidia just can't get it together with 1080i content with their Linux drivers (especially via DVI). I agree, even with the 1080i deinterlaced, 720p content looks better.

For me, this has been the only notable flaw in an otherwise awesome MythTV system. Looking at how long this has apparently been an issue, combined with the fact that's it's only an issue under linux, and only a real issue for people with 1080i TVs (CRT), I wonder if it will ever get fixed.

Tom

That's where I am too. Kernel de-int on 1080i just about pegs my 2.6 GHz P4, which is OK as long as I don't have a commercial flagging job running simultaneously.

From what I've read on the mythtv-users list, there seems to be a timing problem in the presentation of interlaced fields from mythTV to the nVidia driver. Quite often then the TV ends up pairing the bottom field from frame n with the top field from frame n+1, which produces a blurring effect - especially noticeable when there is motion or panning. Pausing and restarting playback is a roll of the dice - it gives myth and nvidia another chance to get in synch (or not).

1080i looks "ok" on my RPCRT with kernel de-int. It's not as good as de-int off when the driver is in sync, but it is better than the blurry image from de-int off when the driver is not in sync.

Very good question. I've seen mixed reports on this. Some say it worked ok until after the 8776 drivers. I've seen some indicate it's even older. I'm using DVI output. Some claim to have had luck with component out, but most seem to say they had just as much trouble with that.

To be specific, here's the problem I have if I try to disable deinterlacing and watch 1080i content: It may look awesome for like 15 seconds and then seems to drift out of sync and showing awful trails and motion blur...it seems to go back and forth like that.

I have no way to test anything I've read about older drivers due to enough catch-22's to make you run screaming into the night. First of all, the 8776 drivers won't work correctly with my 7100gs card...others have found this as well...I get a black screen trying to watch TV. Also, with the 7xxx card, I can't try any really old drivers, as the chipset isn't supported. I toyed with the idea of trying older drivers and getting a 5100 card until I realized that it would mean a new frontend...my frontend is a small form factor Dell with only PCI Express. As far as I can find, no older chipsets are available in PCI Express cards.

I think we need to start posting more about this on the nVidia forum. This brings me to another issue however. They naturally want you to attach a bug report for these sorts of things. My understanding (correct me if I'm wrong) is that they don't support custom modelines. In spite of the fact that the newer nVidia drivers are supposed to support 1080i without custom modelines, I only get the top half of the screen (stretched to fit the screen) if I use them. As a result, I have to use a custom modeline for 1080i. This problem is described here (apparently nVidia has a bug assigned to it):

http://www.nvnews.net/vbulletin/showthread.php?t=90084

I attached a bug report on one of the threads about 1080i tearing issues...not sure if anyone ever looked at it. I've never seen a sign of nVidia acknowledging this 1080i tearing issue at all.

Tom

Well, I can vouch for the fact that this problem is not due to custom modelines.

I am just using the 9755 drivers on my setup, with this nvidia-specific option in my xorg.conf
Option "TVStandard" "HD1080i"
I do end up with a lot of overscan, but nothing like the stretching symptoms you describe.

My blurring symptoms are exactly like yours, however. I have a test clip from the Stanley Cup playoffs which will be blurry one time I play it back, and not the next. If I pause and then restart it, it often switches between blurry (out of sync) and clear (in sync), or vice versa.

I'm using that TVStandard as well...I don't think that has anything to do with it. If you're not using a custom ModeLine, what do you have for the 'Modes' in the Display subsection of the 'Screen' section? If you aren't using a custom ModeLine, you must be specifying one of the built-in mode names, like "1920x1080" for example. If I try and use that I (like others) get only the top half of the screen...almost as though it's treating it like 1920x540 or something.

In any case, I don't think any of that is related to the blur issues.

Tom

Right now I probably have the best overall picture I've had by using bob deinterlacing with my 1080i setup. Doing so required a patch like the one here:

http://svn.mythtv.org/trac/ticket/2903

In order to use bob, the refresh rate must be at least twice the frame rate. For some reason I don't understand (and you can see this if you start the backend with -v playback) when outputting at 1080i@60Hz, the frame interval is reported as 16666 (1000000/16666 = 30FPS), but the refresh rate is reported as 33333 (1000000/33333 = 30Hz) rather than reporting the refresh rate as 60Hz. The patch changes this for interlaced output thus allowing you to enable bob.

I don't know why the frame rate is reported that way. Also, I've read in several places that you shouldn't use bob with an interlaced display, but have never read a specific reason. The reality of course is that we shouldn't need to use any de-interlacing with interlaced displays.

In any case, I've seen no ill affects from the patch, it looks better than it ever did with kernel de-interlacing, and is using less CPU.

Tom

Check this out...I can't help wonderring if the interlacing issue here isn't in some way related the hopeless interlacing we have at 1080i with newer drivers:

http://www.nvnews.net/vbulletin/showthread.php?t=82141

Note from the last post that they're addressing this in the legacy drivers. Do they actually think that interlaced output works correctly in the newer drivers?

Tom

Ouch! Thanks for pointing that out. I never noticed he wasn't using a binary operator there. I guess the other bit flags aren't getting used (at least in my case anyway), as it only did that when it was supposed to. Actually I don't think you need the '== 0x010' though, as the bit operator will only be non-zero if the 0x010 bit is set.

The bob deinterlacing does make a difference though doesn't it? In my case it's noticeably smoother, uses less processor, and looks much better than any other resolution I've tried.

I've been meaning make to post to the nVidia forum with links to all the complaints about 1080i interlace issues and ask if they plan to address it. They've shown no sign of acknowledging it yet.

Tom