Standards conversion - the Third Way!

[David_Robinson]

It's possible to use a video scaler, of the type used to scale up standard TV to HD format, as a standards converter. I have a DVDO iScan HD+, and this will allow all output timing parameters to be altered and saved to a user setup. Unfortunately it will not quite go as far down as 405/50/2:1 (or 378i to use the modern designation), but it will certainly do 819/50/2:1 (786i). Absolutely no hardware modifications are required to get a mono output. I managed to display the result on a Bush TV135R, set to 625 with the line hold suitably tweaked. The only snag I found is that the Bush (and I suspect most 625 or 405 sets) would not accept the 819 standard's solitary field sync pulse, so I had to add more field syncs, similar to 405.

The quality of the scaling and interpolation is as you would expect, extremely high, partly because the scaler converts the input to progressive scan before scaling it.

The other standard I tried, which works very well indeed, is 405 progressive (405/50/1:1 or 378p). This will also display on the TV135R. I guess it would be possible to use this as the input to a very stripped down standards converter that would just convert it to interlaced, while retaining the high quality scaling.

Of course these scalers are expensive - I would not suggest buying one just for this. But for those of us that already have one, an interesting experiment!

[Kat Manton]

Hi,

This is, in effect, what I'm doing - the Linux software my system is based on was developed as a free open-source software-based alternative to expensive video scalers. Modern PCs are powerful enough to do real-time video scaling rather well, these days; with the right software and a decent operating system. So you can watch DVD or off-air TV etc at any resolution you can get out of a graphics card; and with Linux you've always been able to "get your hands dirty" and play around with the video timings, resolution, etc. to a very fine degree.

So, while it was developed by and for those using large high-resolution displays for home theatre systems; I worked out that with an nVidia GeForce4 card (possibly other cards, too) it's possible to get near spot-on 405-line timings, interlaced as well, out of it. (As the lowest dot-clock for the card is a touch high it comes out as 968 x 377 interlaced - the software doesn't mind this odd resolution and scales perfectly anyway.) And when I get around to running through the maths again 441, 819 and quite a few other standards ought to be possible; I just haven't worked it out yet

Results should be good; just as with the video scaler this system de-interlaces any incoming interlaced signal you're feeding it (composite video or analogue terrestrial from a 'framegrabber' card) - and with streamed video, DVD and digital TV it's not acting as a standard converter as such. The source being inside the PC, it takes the digital video data directly, scales it, then outputs this directly at 405-lines; there's no intermediate stage where analogue 625-line video needs to be processed.

Regards, Kat

[Kat Manton]

Originally Posted by David_Robinson

The other standard I tried, which works very well indeed, is 405 progressive (405/50/1:1 or 378p).

Just a thought, shouldn't that be "189p"? It's what I was running during my early experiments with an nVidia GeForce2, which doesn't support interlaced modes. Or does your scaler do something clever where it still manages to output the odd 189 lines followed by the even 189 lines of what it has scaled internally to a 378-line picture, just without interlace - such that following this with a circuit to fudge the syncs about does ultimately give a 378-line picture not a line-doubled 189-line one?

Or am I getting confused and you're producing what would be "810-line" if it was interlaced?

Regards, Kat

[David_Robinson]

Originally Posted by Kat Manton

Just a thought, shouldn't that be "189p"?

No, it's the same number of active lines as normal 405, but with all 378 lines on every field, non interlaced. The line frequency is 20250Hz, which is why I used a 625-line TV which was the nearest I had. Fortunately the old technology with wide tolerances everywhere, means plenty of adjustment range on the line hold!

I'm sure the performance of PC-based scaling is now very good, but I suspect dedicated hardware can still beat it. Obviously the difference won't matter much when the output is 405 - the more you are scaling down, the easier it gets, which is why my existing converter using 4 lines from 1 field looks subjectively very good. When converting up to HD it's not so easy - the DVDO for example uses a 4-field aperture (they don't say how many lines per field) for de-interlacing. That's an awful lot of Mips...

[evingar]

It's possible to use a video scaler, of the type used to scale up standard TV to HD format,

Yep, I would say so without a doubt. We have been developing a product to convert between HD and SD (and vice versa).

http://www.rezzer.tv/

The engine is a Gennum part: -

http://www.gennum.com/ip/gf9320.html

As you say though, it's pricey at around £60 - then you need RAM around it and a processor to load the registers (a small PIC type would do the job for a simple application - the 9320 is serially controlled.

A fixed conversion from 625 to 405 should present no problems. Seems a bit of a sledgehammer to crack a nut with though.

[tubesrule]

Hi,
I think you need to be careful with deinterlacing of the image. Interlacing was originally done as a way to achieve a 2:1 bandwidth reduction while still retaining an 'effective' 50 frame/second image. By deinterlacing you present a 25 frame per second image to the television, not a 50 field per second image, so motion will look very choppy and unnatural. Usually when deinterlacing is performed, it is done to a progressive monitor that can handle the full frame rate, 50 frames/second in this case. For interlaced televisions, the signal should remain interlaced for the best motion reproduction.

[David_Robinson]

Originally Posted by tubesrule

By deinterlacing you present a 25 frame per second image to the television, not a 50 field per second image, so motion will look very choppy

No no! Modern de-interlacers are much smarter than that! Motion adaptive interpolation is used to create a good quality progressive scan signal with good motion portrayal. If this is then used to feed a standards converter, the interpolation will be much better because the incoming lines are closer to the required output line, compared to using an interlaced input.

It is true that most de-interlacers have a mode where the output motion is at 25Hz, but this is only used where the source is known to be film rather than video.

[ppppenguin]

Originally Posted by David_Robinson

No no! Modern de-interlacers are much smarter than that!

You sometimes have doubts when looking at LCD tellies. I suspect that's a matter of cost. Good de-interlacing is non-trivial. The Gennum chip is reputed to be a good one though it's expensivefor use in domestic equipment.

[oldeurope]

Hi, I agree with Darryl.
I watched a HDTV picture from a satellite HDTV set top box on a Sony
and a Panasonic high quality flat screen TV. And the picture is still choppy.
I think in the future all sources are choppy. This is a problem.

Kind regards
Darius

[David_Robinson]

Originally Posted by ppppenguin

You sometimes have doubts when looking at LCD tellies. I suspect that's a matter of cost.

Yep, that's why I bought the DVDO scaler! On my Sony LCD, using its own scaler, there is a problem with fine detail that moves. If there is a lot of detail, for example a hedgerow, or someone with a finely patterned tie, and the camera is slightly unsteady, the Sony will continuously switch between interpolation modes. This results in a shimmering effect on the area in question. The DVDO is very much better, probably because it uses 4-field interpolation. This means that it requires (and provides) a compensating audio delay. The Sony doesn't, so it can't use that many fields.

The de-interlacing chip in the DVDO is made by Silicon Image. The combination of the DVDO with the Sony gives the best SD pictures I have ever seen - and I include Grade 1 studio monitors in that comparison. No line structure, no flicker, no motion artifacts, and subjectively higher resolution due to a reduction in the Kell factor!

[David_Robinson]

Originally Posted by Darius

I think in the future all sources are choppy.

Yes, it's already been happening! For some reason I don't understand, producers of dramas think it's clever to have 'filmic effect' added to the programme to give it a 'film look', even when it's been shot on video. This involves converting the motion so it is strobed at 25Hz. It appears from recent BBC shows I have seen, that were made in HD, that this will still happen with HD. So the resolution will be higher, but only in 2 out of 3 dimensions (not the temporal)!

[oldeurope]

Conclusion for me: If there is an analogue way I'll take it.

Kind regards
Darius

[channel405]

Originally Posted by Darius

Hi, I agree with Darryl.
I watched a HDTV picture from a satellite HDTV set top box on a Sony
and a Panasonic high quality flat screen TV. And the picture is still choppy.
I think in the future all sources are choppy. This is a problem.

The original source could well have been 720p then. Most HD displays are plasma or LCD so are native progressive and you would think it wouldn't be so apparent, though I think they are actually at 50Hz rather than 25. This should mean then you might get the "choppy" or strobbed affect on moving objects above a certain "speed", but there would be less motion blurring than on interlace so the moving object will be more clearly defined.

It seems as if there is a gradual shift away from the old interlace technique, PCs and Mac don't use it now, TV is starting to move away from it. I seem to remember a mechanical 3:1 interlace system mentioned on this forum not that long ago, must have been a nightmare televising football with it!!

Going back to scalers as converters, good idea but the cheapest I have found new I think is from Keene Electrical at about £160. Or risk fleabay.

Cheers
Tim

[Kat Manton]

Originally Posted by channel405

Going back to scalers as converters, good idea but the cheapest I have found new I think is from Keene Electrical at about £160. Or risk fleabay.

I would hazard a guess that most (if not all?) are designed to produce anything from 525i/60 and/or 625i/50 upwards only, so may be fine for 819i/50 with some tweaking but not exactly a lot of use with vintage single-standard 405-line 50Hz field-rate, 10.125kHz line-rate interlaced television receivers - unless it can be persuaded to produce and work with an odd 'rectangular pixel' resolution with the correct line and field rates as I'm doing with PC graphics cards.

Originally Posted by David_Robinson

I'm sure the performance of PC-based scaling is now very good, but I suspect dedicated hardware can still beat it.[...]

As someone primarily a software engineer these days, I'd have to doubt that - any signal proccessing which can be done by dedicated hardware can be done with software; though it may take quite a bit of processing power

This is the direction a lot of industries are heading; processing power is relatively cheap and getting cheaper; and software is more flexible than dedicated hardware... Once you've designed a device and committed it to silicon you're stuck with it; unless you stick a CPU core on the die along with some reprogrammable memory. (In which case the 'software' becomes 'firmware' but that's just semantics, it's still a processor running software and it can be reprogrammed.)

Originally Posted by David_Robinson

[...]That's an awful lot of Mips...

...which a quad-processor 3.66GHz Xeon 64-bit machine would undoubtedly have - that may be a high-end machine now, but in a few years it may be what a few of us are using as a general-purpose desktop machine... now repurpose that with dedicated software rather than a general-purpose desktop operating system and you have a very powerful box of digital signal processing...

Regards, Kat

[tubesrule]

The problem with running a progressive 25 frame/sec image to an interlaced monitor is your eye is only seeing 25 images per second, not 50 images per second as with true interlaced video so motion will never be as smooth. It really doesn't matter how good the deinterlacer is in this situation. The deinterlacer is critical when converting an interlaced image for display on a progressive monitor. Then how much information the deinterlacer takes into account from adjacent frames becomes critical.

An easy way to see this is to view filmed material that was simply sped up to 25fps (not half fields inserted as some systems do), and then compare to true interlaced video, like that from a live sports feed, and the difference is fairly extreme. Progressive video at twice the frame rate of interlaced video is far superior to interlaced, and has none of the flicker effects on stationary images, with all the same motion information, which is why computers and digital television have gone this way. Interlacing was an extremely cleaver way to reduce bandwidth while keeping the apparent frame rate up at 50 in the early analog days.

David, you mentioned how producers seem to like the "film effect" when shooting video. I've had the unfortunate experience of dealing with Hollywood for the last 6 years for both episodic television and theatrical film, and there is a definite tendency to believe "film just looks better". You'll get all sorts of reasons why they believe this is true, but I think it is mostly due to familiarity. If television had come along first, I really think these guys would be trying to achieve that "electronic effect" for their films because it just looks better ;-)

Darryl

[evingar]

..which a quad-processor 3.66GHz Xeon 64-bit machine would undoubtedly have - that may be a high-end machine now, but in a few years it may be what a few of us are using as a general-purpose desktop machine... now repurpose that with dedicated software rather than a general-purpose desktop operating system and you have a very powerful box of digital signal processing... [IMG]file:///C:/DOCUME%7E1/ADMINI%7E1/LOCALS%7E1/Temp/msoclip1/01/clip_image001.gif[/IMG]

Hi Kat.
As someone from a hardware background, I still back hardware solutions over software solutions running on a standard platform for simple single use applications.

Cheap desktop PC's are "messy" with their monitors, keyboards and mice, “lappys” are neater but more expensive - WRT either type, I personally can’t be fagged waiting for them to boot up, "fart about", do a disk check etc etc. I understand what you say about dedicated software, but you are always going to have to run an operating system kernel, and this gives fundamental problems irrespective of the power of the machine.

We provide service for 2 kind of machine at work, a Doremi server based on bespoke hardware and a simple time delay server based on standard PC hardware running a Linux kernel and dedicated software. You would not ***believe*** the problems we have had with the latter in areas of dirty power etc – it’s just not as robust as the dedicated hardware solution!

This is in no way a reflection of what you have achieved with your PC standards converter - which IMHO is nothing short of miraculous, but a neat tiny standards converter takes no time to boot up, takes minimal power, does not have a noisy fan and does not "crash”, take aeons to boot, corrupt its hard drive, do disk check etc etc every time the power dips.

A dedicated hardware solution is preferable in my eyes for my particular situation - even if ultimately it does cost more (I have a Pineapple standards converter which I am very pleased with).

[Nickthedentist]

Originally Posted by evingar

I personally can’t be fagged waiting for them to boot up, "fart about", do a disk check etc etc.... it’s just not as robust as the dedicated hardware solution!

I agree wth you, but it's swings and roundabouts really. Both Kat's and Darius' efforts are amazing, and I'm sure both will prove to be excellent in practicel

[channel405]

Originally Posted by tubesrule

The problem with running a progressive 25 frame/sec image to an interlaced monitor is your eye is only seeing 25 images per second, not 50 images per second as with true interlaced video so motion will never be as smooth. It really doesn't matter how good the deinterlacer is in this situation. The deinterlacer is critical when converting an interlaced image for display on a progressive monitor. Then how much information the deinterlacer takes into account from adjacent frames becomes critical.

Hi Darryl,

Interesting you saying this, so this would mean that modern new such as 720p at 25fps are not a smooth even on native progressive plasma/LCD displays as old fashioned 2:1 interlace video? So the source video format is the crucial factor rather than the processing afterwards. In that case why go 720p at 25fps, or does just look jerky on CRT displays which will soon be out of date anyway?

Cheers
Tim

[Kat Manton]

Hi Chris,

I fully understand; though I'll add a few points:

Yes, my PC system takes a while to boot. But:

(Once it's all sorted out, anyway)

• The 405-line TV is the monitor.
• No keyboard or mouse needed, just the IR handset that comes with the digital TV card.
• I've never had one of my Linux machines (somewhere between 5 and 8 running 24/7 since I got into it 7 years ago) corrupt its hard drive or crash.[1] (Silicon Graphics XFS filesystem is very robust and I use it on all my Linux systems.)
• Problems with crashes on a brown-outs would indicate a lousy PSU, I tend to use decent ones in the machines I build and haven't noticed a problem.
• If you felt like splashing out it would be possible to build a neat, small-footprint mini-ITX fanless system; though as it's basically a 405-line PVR you still need a hard drive in it for timeshifting and pausing live TV etc. Think of it as more a 405-line TiVo and less a standard converter - and the TiVo is Linux-based, too...

My target audience I guess is really those who have an older PC lying around, necessarily still half-decent spec (possibly around 1GHz; I need to investigate system requirements properly and finalise a minimum spec.) It's the sort of system that can be had cheaply enough; I've had three machines around the right spec given to me by people who've upgraded.

But, once it's polished and tidied up, you just turn it on, (yes, and wait for it to boot)[2], then sit back and play DVD, watch televison etc. all controlled from an IR remote with on-screen display on the 405-line TV. Ease-of-use is a priority for me with this project.

[1] I have had machines crash, though this unusual. In one case this turned out to be due to dying RAM (and I didn't suffer any disk corruption) and in other cases far too many to mention I've had applications I'm developing crash, though they never take out the whole machine and require reset/reboot. I like Linux, I find it to be stable. Very stable.

[2] Though as it's a hard disk video recorder as well you might feel inclined to just leave it turned on if you use these features; it's not going to record anything while it's off. It should be stable enough to be left on 24/7 for months on end if you like. And that saves waiting for it to boot.

Regards, Kat.

[ppppenguin]

Originally Posted by channel405

Interesting you saying this, so this would mean that modern new such as 720p at 25fps are not a smooth even on native progressive plasma/LCD displays as old fashioned 2:1 interlace video? So the source video format is the crucial factor rather than the processing afterwards. In that case why go 720p at 25fps, or does just look jerky on CRT displays which will soon be out of date anyway?

720p25 looks horrible on a CRT simply because of large area flicker. However it's no trouble at all to show each field twice (or more times) and this does not introduce any extra artefacts. This is analogous to the multibladed shutter in film projectors. You are stuck with the fact that there is no temporal information at 50Hz so fast action such as sports will look like film rather than TV. 720p25 would be fine on an LCD provided it can scan as slowly as that because each pixel is held until it is switched again. In practice it is converted to 50Hz etc as you would with a CRT.

Interlace was a great video compression system in its day. Unfortunately it's now a curse. It is not possible to achieve perfect de-interlace of all pictures simply because it is not possible to distinguish between some vertical detail and certain movement. Some de-interlacers are very good indeed, others are not.

The Pineapple standards converter had a frame mode which interpolated between the interlaced pair of fields. Superb on still pictures and revolting when anything moved. In practice you always used the field mode.