jmturn@RINGWLD.UUCP (03/14/86)
> There are several production companies >in New York which specialize in converting video from-to various formats. >The gear to do this is VERY expensive. The networks do this in certain >instances, but even they contract out to these companies. As far as doing >it on a consumer level, I seriously doubt you are going to be able to get >it done for a reasonable cost. >David P. Harpe >University of Louisville > >BellNet: (502) 588-6303 >Bitnet: DPHARP01@ULKYVX > How to do format conversion... 1) Get a s**tload of fast RAM. 2) Digitize the incoming signal. 3) Do line averaging 4) Do timebase correction 5) Spit it out the other side. The reason that this has been expensive historically is item 1. Remember 5 years ago, when 16Kx1 200ns RAMs cost $1 a piece? Today you can get 64Kx1 for 89c, and that places the cost of a megabyte of RAM at about $60. Now, even if you need 8 megabytes of RAM to digitize the signal (which you don't...), your total memory cost is still only $320 for the memory. All the rest of the items (2-5) could be performed by a few custom VLSI chips. Why are format converters expensive? 'cause there's not enough demand to produce in consumer quantities. Same reason a vectorscope or waveform analyzer costs 2X a 100Mhz dual trace scope. >NTSC - [N]ever [T]he [S]ame [C]olor SECAM - [S]omething [E]ntirely [C]ontrary to the [A]merican [M]ethod Save Your Vertical Blanking Intervals _ for Big Cash Prizes! / \ | * | James Turner \ _ / (The Ringworld Engineer) --------------------------------------------------- | ARPA | ringwld!jmturn@CCA-UNIX.ARPA | --------------------------------------------------- | | decvax \ | | | sri-unix \ | | UUCP | !cca!ringwld!jmturn | | | ima / | | | linus / | --------------------------------------------------- | MAIL | 329 Ward Street; Newton, MA 02159 | --------------------------------------------------- (The views expressed in this message have been computer enhanced for your pleasure.)
dsi@unccvax.UUCP (Dataspan Inc) (03/27/86)
> > How to do format conversion... > 1) Get a s**tload of fast RAM. > 2) Digitize the incoming signal. > 3) Do line averaging > 4) Do timebase correction > 5) Spit it out the other side. > > The reason that this has been expensive historically is item 1. Remember > 5 years ago, when 16Kx1 200ns RAMs cost $1 a piece? Today you can get > 64Kx1 for 89c, and that places the cost of a megabyte of RAM at about > $60. Now, even if you need 8 megabytes of RAM to digitize the signal > (which you don't...), your total memory cost is still only $320 for the > memory. All the rest of the items (2-5) could be performed by a few custom > VLSI chips. > There are at least two more requirements for M<->B conversion, which you are totally neglecting. Suppose you are going from system B to system M. SOMEWHERE, you have to "make up" 10 more frames per second while preserving motion continuity. This task is nontrivial. It is not as simple as repeating the frames at appropriate intervals due to the motion discontinuity involved. Actually, virtually all B to M framestore converters are very poor in terms of preserving temporal continuity of the original signal. There are also equally difficult issues in going from M to B, but the temporal "jitter" artifacts are much less noticeable because there is more temporal data going from M to B. Second, you need precision composite - to - RGB (or YIQ, which seems to be "in" and "chic" these days) converters before digitisation. There are also very expensive antialiasing filters which need to be implemented as analog filters prior to A/D conversion (ever price a Matthey (UK) antialiasing linear phase filter -- ONE costs more than a cheap home VCR); sin(x)/x correction of the output signal, local generation of SC-H phased sync and burst, genlock circuitry for the input to recover the write clock, etc. Digital filtering is sometimes used for the aforementioned filtering tasks to lessen the requirements on the output postaliasing filter. Third, there is this matter of colour correction. EBU PRIMARIES ARE NOT THE SAME AS NTSC PRIMARIES. The basis matrix for EBU must be multiplied by a suitable function matrix (including gamma correction) to arrive at the correct matrix for NTSC. I agree, standards conversion and time base correction for home VTR's is trivial, because there is not enough video information there to corrupt any worse than the combed, cored, edge enhanced, full-of-differential-and- group-delay-distorted video already is. What disgusts me is that all this effort goes into standards conversion framestore design, but the end product ends up on a monitor which is barely suitable for cue purposes in a dressing room. (Yeah, there's video there....) David Anthony DataSpan, Inc