What is a good weave-style deinterlacer (lag-free)?

[awe444]

With 480i content, the XRGB-3 emulates a CRT and thus renders each field with black lines every other line, with the black lines alternating between even and odd each frame (i.e., each 1/60th of a second). With this method, stationary images are half the brightness they would be at 480p, and vertical edges exhibit flickering.

By contrast, weave-style deinterlacing renders the previous field and the current field together (no black lines), which can cause a combing effect on non-stationary content but makes the overall signal 2x brighter and stationary content is completely non-flickery. No line-doubling or interpolation occurs with weave deinterlacing, so vertical resolution is preserved, and it is also lag-free (actually the total content you're perceiving at a given moment is 1/2 frame behind, but that's because half of it is current and the other half is one-frame old).

So my question is what is a good weave-style deinterlacer for 480i signals, accepting input via 15-pin DSUB or otherwise?

[Xyga]

Extron Andora ? I think...

(though it's not actually deinterlacing, is it ? and the input is 9-pin Dsub RGBs, check Fudoh's website)

[awe444]

Extron Andora ? I think...

I think you're right---quoting Fudoh's site's review of the Andora:

The single fields are just weaved together to get the proper 31khz output. This results in slight combing artefacts in high movement scenes

But just before that there's this statement:

it's completely lag free as there is no deinterlacing applied to the picture

This may merely be a terminology issue, since weaving is technically a form of deinterlacing. Really he meant to say there is no deinterlacing that introduces lag.

But is this really the only device that performs weave-based deinterlacing? Nothing more modern?

[telemetry]

My DVDO VP30 has a great weave-style "feature", but I call it "bad cadence lock" with a two-frame delay.

For my curiosity, wouldn't a good weave-deinterlacer *not* have one-field lag? Seems like it would always be showing you the latest field, interwoven with F-1.

Maybe this is what the Andora does. I mean, half the frame is unavoidably a previous field, but it should always be showing the latest field as soon as possible.

[awe444]

My DVDO VP30 has a great weave-style "feature", but I call it "bad cadence lock" with a two-frame delay.

Will take a look at that one, thanks

For my curiosity, wouldn't a good weave-deinterlacer *not* have one-field lag? Seems like it would always be showing you the latest field, interwoven with F-1.

Yes so what you're perceiving at any given moment is (on average) one-half frame old. I would consider anything less than one frame behind as lag-free. So the DVDO you cite probably isn't what I'm after if it has two frames delay...

half the frame is unavoidably a previous field, but it should always be showing the latest field as soon as possible.

Correct this is indeed what I mean by weave-style deinterlacing (though it's not clear to me that the Andora does this, can anyone confirm?)

I think the terminology is ambiguous because there are deinterlacing methods where one "weaves" together two fields wherein one or both of those fields is temporally-interpolated and hence at least one entire frame behind. "Weaving" to me describes a less-complicated alternative to spatial (vertical-dimension) interpolation, but it could still involve temporal interpolation and hence lag.

[Xyga]

The VP30 with ABT102 deinterlacing card has two modes for games, the first is fast with only 6m lag but very blurry, the second is full deinterlacing looking awesome (only beat by the Mini afaik) but with a 2 frames delay indeed.

Are you talking about one of the other available modes, telemetry ? I've only ever used gamemode 2 honestly...

@awe444: something like the Andora is probably better for what you want yes, but the article suggests there are visible combing artifacts if an SLG is not used in the chain.
Maybe you should do a thorough forum search, there ough to be at least a few posts or even a couple of dedicated threads about it.

[Fudoh]

For my curiosity, wouldn't a good weave-deinterlacer *not* have one-field lag?

you can design it in a way, that one of the fields doesn't lag. You still need to buffer the field before, but your output can be locked to "younger" field. As a result the even lines you see would be up to date, while the odd ones are lagging a field behind (or the other way around).

I still don't see the point though. Why would you want a weave deinterlacer?

[cfx]

.

[awe444]

you can design it in a way, that one of the fields doesn't lag. You still need to buffer the field before, but your output can be locked to "younger" field. As a result the even lines you see would be up to date, while the odd ones are lagging a field behind (or the other way around).

You've exactly described the method of processing I'm looking for. Can you confirm that the Extron Andora or any other device provides this type of 480i lag-free weave processing (in which each frame consists of the current field woven with a buffered, one-frame old field)?

I still don't see the point though. Why would you want a weave deinterlacer?

I'm wondering the same thing. From the OP, it sounds like it's to make the picture "birghter" beasue of not having scanlines (black lines). But isn't the only time this is even possibly an issue is on a CRT? Even there, the darkening effect I've only ever noticed on 240p or 480p with scanlines, not an interlaced picture, or a "fake interlaced" like what an older XRGB does.

In other words, isn't the OP based on a faulty premise?

The "fake interlaced output of an older XRGB" that you describe is precisely what I'm dealing with in my current setup, so yes, my premise as stated in the OP is indeed to achieve a brighter and less flickery image compared what my XRGB-3 does. What I'm not willing to give up is the lag-free nature of the signal, though I am willing to live with comb artifacts.

[Fudoh]

though I am willing to live with comb artifacts.

this is what I can't imagine since it really renders the output mostly unusable.

Yes, as far as I remember the Andora works this way.

[SGGG2]

The "fake interlaced output of an older XRGB" that you describe is precisely what I'm dealing with in my current setup, so yes, my premise as stated in the OP is indeed to achieve a brighter and less flickery image compared what my XRGB-3 does. What I'm not willing to give up is the lag-free nature of the signal, though I am willing to live with comb artifacts.

What kind of games are you looking to use this with? It sounds like fast action games; most 480i games can be forced into 480p (with varying results). Setting the scanlines to 0 should take care of any brightness issues.

[philexile]

I have an Andora available for sale if you are interested.

[awe444]

though I am willing to live with comb artifacts.

this is what I can't imagine since it really renders the output mostly unusable.

Yes, as far as I remember the Andora works this way.

OK---one thing I haven't considered (and probably what you guys are thinking when you say its a bad idea) is how the Andora handles 240p content, as I've considered only 480i till now in this discussion. I'm guessing it line doubles the 240p, with one of the doubled lines being current and the other a frame behind. This obviously will make combing artifacts, which can be masked with an SLG, however the masked image will have half the frame rate (30fps as opposed to the original 60fps) and will be 1 frame lagged and will lack transparency effects.

So I'd want to manually switch to the Andora for 480i content only, and XRGB for 240p. So its probably no longer a desirable option. Though as most PS2 and many N64 titles are 480i always, might not be so bad...

What kind of games are you looking to use this with? It sounds like fast action games; most 480i games can be forced into 480p (with varying results). Setting the scanlines to 0 should take care of any brightness issues.

I guess I'm considering all game types---really I'm trying to optimize the look of 480i still images (e.g. menus and HUD elements) by eliminating their flicker without compromising responsiveness during action.

[SGGG2]

The combing artifacts from weave-deinterlacing with 480i are going to look HORRIBLE. If you want to get a better idea of what I mean -- boot up a 60fps fast action game on PS2 via B1 on your XRGB-3 and turn the scanlines way up.

Given your requirements for lagless 480i and no flickering, you're not going to get any better image quality than feeding B1 into an SLG (for 240p style scanlines) and setting scanlines off on the XRGB-3 itself.

edit: Just checked Fudoh's review of the Andora, that plus an SLG should beat out the XRGB-3 combo for your needs. It *might* look identical, though.

[Exidna]

The combing artifacts from weave-deinterlacing with 480i are going to look HORRIBLE.

You would only use weave deinterlacing on 30 FPS games.
For 60i games you should either use bob deinterlacing which line-doubles alternate fields, or have something convert it to 240p by shifting alternate fields. The latter would be preferable.

[awe444]

boot up a 60fps fast action game on PS2 via B1 on your XRGB-3 and turn the scanlines way up

At first this statement confused me, then I realized I was misunderstanding how the XRGB-3 actually processes 480i signals. Previously I had thought the XRGB-3's scanline darkness setting was irrelevant for 480i because (I had mistakenly assumed) the XRGB-3 always just weaves pure black lines into each field to form the frames (i.e., CRT emulation of 480i).

Now I realize (and have verified through testing) that the XRGB-3 does in fact spatially double each field-line in 480i, specifically it places a copy of each field line above each original line to form the frame. Setting scanlines to full (lowest setting = 7) masks the copied lines to give the CRT emulated look, while setting scanlines to off (highest setting = 255) gives a brighter image albeit with the same amount of edge flickering and 50% less vertical resolution overall. Passing the latter (non-XRGB-scanlined) version through a miniSLG does gives a near-perfect simulation of 240p (which I consider luck because if the miniSLG had the opposite of its unadjustable even/odd line layout, it wouldn't work).

This realization is actually a nice solution to the longstanding problem I've had of getting MegaMan 9 and MegaMan 10 on the Wii to play with a real 240p look without lag (since these games have 240p graphics but the games output in 480i only).

Nonetheless this is still different from the technique the Andora uses, which I may indeed want to test out at some point.

You would only use weave deinterlacing on 30 FPS games.
For 60i games you should either use bob deinterlacing which line-doubles alternate fields, or have something convert it to 240p by shifting alternate fields. The latter would be preferable.

With bob deinterlacing (what the XRGB-3 does when its scanlines are turned off), you loose vertical resolution, gain brightness, retain flickering, retain 60fps, and don't perceive any combing.

With weaving (as the Andora does), you retain vertical resolution, gain brightness, loose flickering, drop to 30fps, and introduce combing.

[SGGG2]

For 60i games you should either use bob deinterlacing which line-doubles alternate fields, or have something convert it to 240p by shifting alternate fields. The latter would be preferable.

I agree! But OP had something else in mind originally.

Passing the latter (non-XRGB-scanlined) version through a miniSLG does gives a near-perfect simulation of 240p (which I consider luck because if the miniSLG had the opposite of its unadjustable even/odd line layout, it wouldn't work).

FYI, the SLG3000 can mask odd or even lines.

[awe444]

Passing the latter (non-XRGB-scanlined) version through a miniSLG does gives a near-perfect simulation of 240p (which I consider luck because if the miniSLG had the opposite of its unadjustable even/odd line layout, it wouldn't work).

FYI, the SLG3000 can mask odd or even lines.

Yes, but the miniSLG can't change this setting---which is why I guess its luck that the mini seems to have the correct layout.

[Fudoh]

With weaving (as the Andora does), you retain vertical resolution, gain brightness, loose flickering, drop to 30fps, and introduce combing.

no. You still get 60fps, since every field is used two times. Consider A, B, C, D, E... the fields, then you get AB, BC, CD, DE, EF... on the output.

[awe444]

With weaving (as the Andora does), you retain vertical resolution, loose flickering, drop to 30fps, and introduce combing.

no. You still get 60fps, since every field is used two times. Consider A, B, C, D, E... the fields, then you get AB, BC, CD, DE, EF... on the output.

You're right---corrected statements:

With bob deinterlacing (what the XRGB-3 does when its scanlines are turned off), you loose vertical resolution, retain flickering, retain 60fps when using a fixed scanline mask, and don't perceive any combing.

With weaving (as the Andora does), you retain vertical resolution, eliminate flickering, drop to 30fps when using a fixed scanline mask, and introduce combing.

And both bobbing and weaving are lag-free techniques and both result in a brightened image.

[Exidna]

With weaving (as the Andora does), you retain vertical resolution, gain brightness, loose flickering, drop to 30fps, and introduce combing.

no. You still get 60fps, since every field is used two times. Consider A, B, C, D, E... the fields, then you get AB, BC, CD, DE, EF... on the output.

A field is half a frame. You would never weave fields like that because it's mixing fields from different frames together.
You either transmit 24/30 FPS at full resolution, or 60 FPS at half resolution.

With 30 FPS, each field is half the vertical resolution and they are weaved together.
Fields: a1 a2 b1 b2 c1 c2 d1 d2 e1 e2 are weaved together to produce Frames: A B C D E at 30 FPS

With 24 FPS, fields are stored using a 3:2 cadence.
Fields: a1 a2 a2 b1 b2 c1 c2 c2 d1 d2 have the duplicate field discarded, and produce Frames: A B C D at 24 FPS

With 60 FPS, each field is a unique frame. You either alternate between odd/even scanlines (480i60) or tell the display to show all fields on odd scanlines only. (240p60)
480i60 will flicker without any deinterlacing. Bob deinterlacing will eliminate the flicker and scanlines.
Bob deinterlacing doesn't halve your resolution because each field is unique, so the image only has 240 lines of resolution.

I suppose if you had a 480px tall source that updated at 60 FPS and were only sampling the odd/even lines for your 480i60 output, then you could theoretically have a higher resolution image at 60 FPS after deinterlacing, but you wouldn't use bob or weave for that. You'd need fancy motion compensated deinterlacing for that, and that won't be low latency.
I don't think that's really applicable to games, though I could be mistaken.

Bob deinterlacing could theoretically be done with practically no latency, but weave deinterlacing must introduce at least one field of latency. (16.67ms)

[awe444]

With weaving (as the Andora does), you retain vertical resolution, gain brightness, loose flickering, drop to 30fps, and introduce combing.

no. You still get 60fps, since every field is used two times. Consider A, B, C, D, E... the fields, then you get AB, BC, CD, DE, EF... on the output.

A field is half a frame. You would never weave fields like that because it's mixing fields from different frames together.
You either transmit 24/30 FPS at full resolution, or 60 FPS at half resolution.

With 30 FPS, each field is half the vertical resolution and they are weaved together.
Fields: a1 a2 b1 b2 c1 c2 d1 d2 e1 e2 are weaved together to produce Frames: A B C D E at 30 FPS

With 24 FPS, fields are stored using a 3:2 cadence.
Fields: a1 a2 a2 b1 b2 c1 c2 c2 d1 d2 have the duplicate field discarded, and produce Frames: A B C D at 24 FPS

With 60 FPS, each field is a unique frame. You either alternate between odd/even scanlines (480i60) or tell the display to show all fields on odd scanlines only. (240p60)
480i60 will flicker without any deinterlacing. Bob deinterlacing will eliminate the flicker and scanlines.
Bob deinterlacing doesn't halve your resolution because each field is unique, so the image only has 240 lines of resolution.

I suppose if you had a 480px tall source that updated at 60 FPS and were only sampling the odd/even lines for your 480i60 output, then you could theoretically have a higher resolution image at 60 FPS after deinterlacing, but you wouldn't use bob or weave for that. You'd need fancy motion compensated deinterlacing for that, and that won't be low latency.
I don't think that's really applicable to games, though I could be mistaken.

Bob deinterlacing could theoretically be done with practically no latency, but weave deinterlacing must introduce at least one field of latency. (16.67ms)

Visual description of how I understand weave and bob are defined:

[Exidna]

Yes, that's how it would work if you had a 480p60 source transmitted as 480i60 by alternating lines from the 480p source when creating the interlaced output instead of scaling it first. (though a weave deinterlacer would not display anything for the first field)
You would use motion-adaptive/compensated deinterlacing for that, not bob or weave.

For a 480p30 source transmitted as 480i, weave deinterlacing gets you the source frame back whether it's moving or stationary.
For 240p60 source transmitted as 480i (alternating scanlines) bob simply line-doubles it to 480p, eliminating any flicker. You aren't losing resolution.

Games like Mega Man 9 & 10 are 240px tall, not 480px.
At most you want to apply bob deinterlacing to get a 480p output, though personally I would prefer to convert that to 240p. (flip the field order for every second field)

[Fudoh]

For a 480p30 source transmitted as 480i, weave deinterlacing gets you the source frame back whether it's moving or stationary.

that's not right, sorry. Weave on a 480p30 source (well, 480p30 is such a confusing term, let's say 480i60 with matching fields) gets you combing artefacts on every second frame (AA, AB, BB, BC, CC).

You're actually thinking of a 2:2 film mode applied to a 480p30 source. The film mode would weave only the matching fields, then repeat the output and then move on to the next fields.

Not the same thing.

[Fudoh]

For 240p60 source transmitted as 480i (alternating scanlines) bob simply line-doubles it to 480p, eliminating any flicker. You aren't losing resolution.

that's not 100% right either. It depends on the vertical offset the deinterlacer adds between the fields. That's why - for example - the OSSC's output looks different from the XRGB-3's output to the XPC's output.

[Exidna]

For a 480p30 source transmitted as 480i, weave deinterlacing gets you the source frame back whether it's moving or stationary.

that's not right, sorry. Weave on a 480p30 source (well, 480p30 is such a confusing term, let's say 480i60 with matching fields) gets you combing artefacts on every second frame (AA, AB, BB, BC, CC).

You're actually thinking of a 2:2 film mode applied to a 480p30 source. The film mode would weave only the matching fields, then repeat the output and then move on to the next fields.

Not the same thing.

I see the distinction you're making, but I've never seen anything which does weave deinterlacing on all fields like that.
What would the application be? That just guarantees that you're going to have artifacts all the time.

For 240p60 source transmitted as 480i (alternating scanlines) bob simply line-doubles it to 480p, eliminating any flicker. You aren't losing resolution.

that's not 100% right either. It depends on the vertical offset the deinterlacer adds between the fields. That's why - for example - the OSSC's output looks different from the XRGB-3's output to the XPC's output.

Sure, you can apply an offset to bob deinterlacing to try and minimize the "bobbing" from it, but that's not going to affect your resolution. It's still 240p.


Still, you don't want any of this for 240p games being transmitted as 480i, as with the Mega Man 9 & 10 examples.

[awe444]

Still, you don't want any of this for 240p games being transmitted as 480i, as with the Mega Man 9 & 10 examples.

True, I agree---I was citing the Wii MegaMan games as a good use of the XRGB-3's bob deinterlacing (when combined with an SLG). Totally unrelated to the OP's topic of weave deinterlacing---sorry for the confusion. I very much agree with you that weave deinterlacing is best suited for deinteracling true 480px content that is primarily static as opposed to fast moving.

[Fudoh]

I see the distinction you're making, but I've never seen anything which does weave deinterlacing on all fields like that.
What would the application be? That just guarantees that you're going to have artifacts all the time.

I guess you're thinking of software deinterlacing solutions. That's actually the only case where you'll ever encounter 30p from a 480i stream. All hardware deinterlacers, line doublers or scalers will strictly render 60p from a 480i stream. That's - by the way - the reason why PAL 2:2 film modes are so much harder to get right than NTSC 3:2 film modes.

What would the application be?

you simply don't encounter 30p material out in the wild. Most material is mixed at best. Better processors do have a 2:2 option and I've used it before on video games (e.g. on the Metal Slugs on the PS2, 480i at 30fps), but without actual detection algorithms, it's not possible to ensure perfect results. A straight weaving mode (with half the target framerate) could easily miss the right cadence and ONLY produce "combed" frames (AB, BC, CD, DE). Again one of th reason why PAL film modes are difficult. There're lots of materials out there, where odd+even results in the "perfect" frame throughout one half of the movie, while even+odd is required on the other half (bad edits are the reason for that).