SO YOU WANT TO BUILD A STANDARDS CONVERTER REVISITEDÖÖÖÖ
Some further thoughts and reflections on generating 405 line signals.
When I wrote my original article in 2002 I couldnít have anticipated the flowering of converters that has happened in the last four years. Although itís a lot easier now than ever before I still stand by my initial comment:
No-one who ever tried it said that making a 625 to 405 converter was easy. That privilege is reserved for armchair designers.
I have also had the pleasure of having had some of my claims proven wrong. This is in line with the belief that if an expert says something is possible, then he is usually right but if he says that something is impossible then heís probably wrong. I will point out the errors of my ways as they arise.
HALL OF FAME
The hall of fame has grown considerably since I wrote in 2002. Obviously there are no new professional designs since there is no current professional requirement. However some of the new designs have been done by electronics design professionals.
While there has been a lot of work on converters themselves we mustnít neglect modulators. Simple modulators have always been within the scope of the reasonably keen enthusiast and David Looserís design was published way back in 1984. Most modulators since then have followed his basic design but there have been two new approaches, one quite conventional, the other very different.
I donít want to repeat my 2002 article but itís worth recapping a few of the most important converter milestones.
In the beginning was optical conversion, a camera pointed at a monitor. The first electronic converter, dating from 1963, was the BBC CO6/501. The signal path was entirely analogue and there was hardly a chip in sight. It was superseded by the digital CO6/509 which continues to be the reference point for all future designs.
David Boynes developed the first amateur built unit for his own use sometime around 1990. The two commercial designs of the early 1990s were the Dinosaur and Pineapple. Both have been extinct for some time and have been known to change hands at quite high prices. The Domino converter of 2002 included a modulator. It was an excellent design and did much to enhance interest in 405 line TV. With the availability of other converters, some at much lower prices, I doubt that Malcolm Everiss will make any more of them.
There are four newcomers to the hall of fame. Two have followed conventional digital approaches while the other two are strikingly different.
Darius clearly didnít read my 2002 article or if he did read it, he was wise enough to ignore this comment:
Analogue storage was used in the first BBC converter and it might be possible to build a modern equivalent with charge coupled delay lines if they are still available. Not a practical option.
He has used CCD (Charge Coupled Device) analogue delays, normally used for dropout compensation in VCRs, to make a relatively simple design. The design uses a simplified 2 line interpolator which can either pass a line or take the average of 2 adjacent lines. The results are proof that even the simplest interpolation is vastly better than none at all. Dariusís skills are clearly in analogue and particularly in RF design. He has designed and built beautifully engineered modulators and VSB filters. Unlike most designs, his modulator doesnít use the 1496 balanced modulator chip. The VSB filter is a nice refinement since all but the very earliest TV transmissions suppressed most of one sideband in order to conserve bandwidth. It is not important for closed circuit use since all receivers can handle the full double sideband signal without difficulty. The earliest version of the converter rather lacked bandwidth due to a low clock frequency on the CCD delays. Later versions improved this and it is now entirely adequate. Darius has always been happy to publish his designs and at least one enthusiast has built one in the UK. Until recently Dariusís converter was built on pinboard and was subject to many modifications. Darius has produced PCBs for the design which should now be available. This will make it much easier to build. The total build cost is likely to be between £100 and £130, depending heavily on the cost of the CCD delays.
Darryl has designed and marketed not one but two converters, both under the name Aurora. Darryl is a design engineer working in the video industry and has much knowledge of modern design techniques. In particular he is a skilled and inventive user of FPGAs. (Field Programmable Logic Array) Both his designs use Xilinx Spartan FPGAs to provide all the logic. They are both fully professional designs with multilayer PCBs and fine pitch surface mount devices. Both converters use 3 line interpolation and give excellent results. According to BBC research, confirmed by Darrylís own simulations, extra lines beyond 3 contribute very little to picture quality.
The first Aurora converter, dating from 2004, could take PAL or NTSC input and convert to any output standard from French 819 all the way down to 30 line mechanical taking in just about every possible vintage standard in between. Darryl also did a special version of the programmable logic to give CBS sequential colour. Conversion between frame rates was possible because this design used a framestore.
In my 2002 article I thought about methods of generating NTSC 405:
A clever approach would be to do the whole NTSC encoder digitally in programmable logic. I reckon this will need rather more programmable logic resource than the whole of the rest of the converter.
Darryl and I discussed this possibility in some detail and we think it feasible but he has not explored it in practice. For once I was right and we were sure that the coder would consume a lot of FPGA resources. There would be no hardware changes, purely new code for the FPGA. This would be a demonstration of the power and flexibility of using programmable logic.
The new Aurora was released in April 2006. Unlike its predecessor it is not inherently multistandard and does not contain a framestore. It is available in several different versions including 405, 819 and older French, German and US standards. There is almost no hardware difference between these versions, another demonstration of the flexibility of FPGAs.
In my 2002 article I said:
Converting digits to analogue is much easier than the other direction. A discrete DAC is nothing like as scary as a discrete ADC but Iím still glad we donít have to do it that way now.
Darryl was unaware of my wisdom and designed a novel discrete DAC for the new Aurora. This 10 bit DAC combines a conventional R/2R ladder with pulse width modulation for the lower order bits. It saved a few dollars and contributed to the very low price of that unit. Darryl has achieved what I previously believed to be impossible and is selling the new Aurora at $260 (around £150) which is truly a bargain. Affordable high quality conversion is now a reality. I have reviewed all the commercially available converters in 405 Alive and later in the BVWS Bulletin. Until now I have always felt the need to justify the relatively high prices but now there is no need.
The new Aurora also has a multichannel modulator using a pair of MC44BS373CA devices. This approach to System A modulation was pioneered by David Robinson and adapted by Darryl to make a fully flexible modulator for all vintage standards from 343 up to 819 lines.
Darrylís converters include other useful facilities, in particular they can capture still images to use as test patterns. Aurora user manuals contain a lot of technical detail and are a good introduction to the technology. They can be freely downloaded from his web site.
David has made several important contributions to the art of standards conversion. His design has four line interpolation based on the BBC CO6/509. Most importantly he has made the only converter capable of producing NTSC colour on 405. While the original Aurora might be able to be reprogrammed to do this it has not been.
In my 2002 article I said:
Annoyingly, standard colour encoder chips could easily do the strange subcarrier frequency needed but not the 405 syncs.
In my own conversion experiments I attempted to force an Analog Devices ADV7171 encoder to produce 405 line monochrome. I never managed to solve the problems but, much to his credit, David has made it produce 405 NTSC. In fact the colour is the easy bit since the ADV7171 contains a fully programmable digital oscillator to generate subcarrier. The hard part is sorting out the sync timings and avoiding spurious blanking where you donít want it. He still has to make 405 syncs separately and add them to the rest of the signal.
David also proposed the Freescale (was Motorola) MC44BS373CA as a System A modulator. This is a programmable modulator that covers all the VHF and UHF bands. It is capable of both AM and FM sound but cannot do the System A specification of 3.5MHz sound to vision spacing with the sound carrier 6dB below vision. This meant that two devices had to be used, one each for sound and vision.
The digital technology is thoroughly modern, using an Altera Cyclone FPGA with SDRAM for a framestore. He took the decision to scale the picture by 23:15 since this correctly converts the number of active lines. If you multiply the 576 active lines of PAL by 15/23 you get 376 which is within the theoretical range of 374 to 379 lines of the 405 line system. Most other designs use 125:81 which converts the total number of lines or even 3:2 which is within 2% of correct. The difference is of no practical importance as it only affects the aspect ratio by a tiny amount but indicates the amount of thought that has gone into the design.
David made his converter out of interest and has no plans to publish the design or make it commercially available.
In 2006 Kat Manton achieved conversion using standard PC hardware running Linux. There was a minimal amount of external hardware needed to complete the design. She ignored all the people who said that you canít do it with a PC. These people included me:
Some have suggested using computers or digital signal processors (DSP). Conceptually there is no problem. Iím a pretty lousy programmer and I could write you the conversion algorithms in a few lines of a high level language such as BASIC. These would take a 625 line image, already in the computer, and convert it to a 405 line image, also within the computer. And thereís the snag. You still need to get the picture in and out and you have to ensure that the computer can keep up with the data. Video comes at you continuously so real time means exactly that Ė you cannot put your hand up, take time out and catch up later. So when you say you have built a converter with a cast off 486 PC I wonít believe you.
I still believe that you canít do it with an old 486 but Kat has used the low cost and great processing power of modern PC hardware to devise a largely software solution. If you had to buy all new hardware the result would not be financially viable but if you can use a PC thatís normally doing other things, or possibly an old PC, then it can make economic sense.†
Katís original concept was not actually a standards converter but a means of putting pictures on the screen of a vintage TV. Moving pictures are now widely handled by PCs. DVD replay, digital TV and streamed internet video are all digital sources of pictures. The main problem that Kat has solved is using a PC to provide a 405 line output from these pictures.
Her starting point was the MythTV Personal Video Recorder (PVR). This is open source software for use with Linux. The scaling needed for 405 line output (or any other number of lines) is an inherent part of MythTV so her main task was making a PC graphics card give a 405 line output.
Linux has a lot of flexibility when it comes to setting up a PC graphics card to give a custom resolution. There is no fundamental reason why it shouldnít do 405, 441 or 819. The important thing is to use the ordinary VGA RGB outputs. The CVBS and YC outputs that are often available from graphics cards are not useful since they cannot be programmed to do other than PAL 625 or NTSC 525. Graphics cards are not really designed for such low resolutions as 405. Kat used a dot clock of 12.07MHz which is rather higher than you would expect for 405 and gives a resolution of 968x378 with near perfect 405 timings. The sync combiner is very simple. The separate horizontal and vertical syncs from the VGA output are combined with an exclusive OR gate to make a simplified mixed sync waveform. This does not have the proper half line information during the field sync so itís possible that some receivers will suffer from poor interlace or horizontal pulling at the top of the screen. Kat has proposed using a simple phase locked loop and additional logic to generate correct field syncs. The RGB signals are added with suitably weighted resistors to make monochrome and the sync is added too.
If you want the PC to do the traditional work of a standards converter you will need an analogue capture card or TV tuner in the PC. This can impose a much greater processing load which might result in freeze frames and other artefacts. The remedy is a more powerful PC.
Kat had to face the problem that you canít really see what you are doing on the PC. The VGA output is 405 line and standard computer monitors will not display this. If you have a reliable 405 line monitor you can see the PCís output but Katís solution during development has been to control the PC remotely from another one. It may be possible to use a dual output graphics card with one output as 405 and the other as VGA.
The great power of a software based solution is that it is easy to change after you have done the initial design. Kat is currently looking for alternative solutions to MythTV. As a PVR, this software is actually overkill for the application and uses more PC resources than are strictly necessary. She hopes to make her software solution freely available in a form that is very simple to install on a PC, even if you donít have much knowledge of Linux.
At the time of writing, at least one other enthusiast has successfully implemented Kat Mantonís design.
In my 2002 article I discussed a design for a hypothetical standards converter. It wasnít quite hypothetical as I actually started to implement the design using some old hardware that was available from a project I had done for a client. Like so many projects it worked up to a point and delivered 405 line pictures after a fashion but I got distracted and never returned to fully debug the design. The methods I described were the conventional digital techniques that a professional designer might use to provide a commercial solution. Kat Manton and Darius have taken very different paths which bear no resemblance to my ideas and I have covered these in the hall of fame above. The Aurorae and David Robinsonís design all use conventional digital approaches so letís take a look at how they compare with each other and with my hypothetical design.
Great minds must think alike for surely we are not all fools. Thereís definite unanimity here. I proposed several methods but my favourite was an integrated decoder such as the SAA7113 by Philips. This approach fixes the input sample rate at the industry standard frequency of 13.5MHz. The first Aurora used the SAA7113 though I can assure you that Darrylís decision was independent of mine. He programmed it using code executing within the Xilinx thus avoiding the need for a microcontroller. The new Aurora uses a newer and better chip, the Texas Instruments TVP5150A. This incorporates a multiline comb filter which can extract all the available luminance bandwidth from a signal. David Robinson used the Analog Devices ADV7183A which also has a comb filter.
There was a lot more scope for invention here. I suggested a rather outdated 8 bit DAC which in retrospect was a mediocre choice. Integrated coders such as the ADV7171 are readily available and have the potential to provide 405 line NTSC. David Robinson succeeded where I had failed and made an ADV7171 deliver 405 line NTSC. Darryl used a much newer 10 bit DAC chip in the first Aurora and a novel low cost discrete design in the new Aurora. Both these 10 bit DACs have sufficient resolution to allow the sync to be generated as part of the video data coming out of the FPGA. This saves adding the syncs in the analogue domain.
I barely touched on modulators in 2002. I suppose I thought that the problem was largely solved with several designs all using the conventional approach of 1496 balanced modulator chips. Most needed custom made crystals for their oscillators though you could use LC tuned circuits. All had separate sound and vision modulators. This approach was fine for channel 1 and usable on channel 4 but the 1496 is not suitable for Band 3 and switching the channel was difficult or impossible.
There have been two distinct developments since then. Darius has devised and published a simple modulator using only discrete devices and this is very suitable for home construction. David Robinson pioneered the MC44BS373CA for System A. It can synthesise any VHF or UHF output frequency from a standard crystal but the penalty is the need for external control, typically from a microcontroller. Also its surface mount package makes it much harder for the enthusiast to use. Darryl showed that with some ingenuity it is possible for a single device to come close to the System A specification but a pair of devices, one each for vision and sound, provides a much more accurate and flexible solution. Darryl used them on the new Aurora with control from code in the FPGA. The result is an exceptionally flexible modulator which can be made to work on any standard and channel, both historic and modern.
The digital bits
Again we are unanimous about the logic. There is now no sensible alternative to an FPGA. Xilinx and Altera are the leading manufacturers but there are others and the choice really boils down to cost, availability and prejudice. You normally get most performance for your money with the newest parts and Darryl has used the Spartan 3 and 3E series. The latter is very new and very low cost. Darryl has incorporated all the control system in the FPGA while David has used a separate microcontroller.
All standards converters need memory. At least a few lines for any design and at least a whole frame if you want a framestore based design. There is no overriding reason to use a framestore unless you also want to do a frame rate conversion such as 60Hz NTSC to 50Hz 405. Other less important reasons for using a framestore are the absolute guarantee of rock solid output syncs at all times and the ease of including full frame test patterns. There were several different approaches to memory. I proposed specialised frame memory such as the Averlogic AL422. This is very simple to control and was successfully used in the Domino. David Robinson went for standard SDRAM, an approach that I would now use, while Darryl used large SRAMs for the framestore in the first Aurora. Large fast SRAM is relatively costly but much easier to manage than SDRAM.† I can speak from personal experience, sometimes bitter, of the problems of SDRAM for video framestores in professional applications. The new Aurora does not have a framestore and holds up to 11 lines of video in the internal RAM blocks that are a feature of most modern FPGAs. This is value engineering at its best, using the latest technology to minimise the parts count and cost while not compromising performance.
Many of the BBC Research Department reports are available here including several from the 1960s and 1970s that are relevant to standards conversion.
There are many useful articles on this site as well as a very active discussion forum. All the people in the Hall of Fame are regular contributors to the discussion forum and have posted a lot of information about their standards converters.
The site for the Aurora converters. You can also download the manuals here.
Lots of information on the MythTV Personal Video Recorder
Kat Mantonís site for her PC based 405 systems. Under development at the time of writing.
Data on the various FPGAs and other chips can readily be found on their manufacturers web sites. A useful tip is to enter the part number into a search engine such as Google. This will often take you directly to the relevant part of the manufacturers web site.
I have made frequent references to my 2002 article. If you donít have the relevant issue of the BVWS Bulletin I have posted the text (without pictures) in the UK Vintage Radio discussion forum mentioned above. Look for the thread PC as a standard converter in the Standards Converters, modulators etc section.
Designing converters is still only for the knowledgeable or the brave but itís no longer the route to madness. There is no true commercial market Ė you couldnít make a living out of converters so all are labours of love to a greater or lesser extent.
I continue to salute all those who have tried and succeeded in the wonderful world of standards conversion.