From: John Costella Subject: TECH: Galilean Antialiasing project Date: Mon, 20 Nov 1995 14:49:11 +1100 (EETDT) A few weeks ago I mentioned here that we are revitalising a project on spatio-temporal antialiasing of VR graphical hardware ("Galilean Antialiasing") at the University of Melbourne, beginning in 1996. I got a number of responses to the post (and a jump in accesses to my WWW page), so I thought that for the benefit of those who weren't reading this newsgroup in 1992 I'd review briefly what on earth the project is all about :-) and let you know what we have in the pipeline. The basic idea is that the frame buffer (the chunk of RAM that contains the pixels) at the end of your graphics card, which has been essentially unchanged in form since Sutherlandian times, is ill-suited to realtime 3-d graphics: when you draw into the frame buffer, the image remains *static* until the next update. This is great for things that *should* stand still (e.g. the window drawn around your Netscape application!), but is terrible for objects that are moving, stretching, rotating, and so on: it causes your virtual world to "clunk, clunk, clunk" along, jumping every time a new update is supplied. So I wrote a somewhat rambling paper back in '92 in which I showed expicitly that it is feasible to build a "smarter" hardware frame buffer, which "knows" how to move the pixels *between updates*, hence keeping the world flowing smoothly along, while the graphical engine is generating another update. The "blueprints" for several "generations" of such frame buffers were described, with various levels of trade- off between graphical acceleration, and design complexity and cost. It should be noted that this approach can be applied to the end of ALL graphical architectures; for example, a 3-D polygon-rendering accelerator board can be FURTHER accelerated by having a "smart" frame buffer between it and the display device. So where does sci.virtual-worlds come into this? Well, back in '92, there was lots of brouhaha about Eyephone and Dataglove patents, and lots of lawyers flying about with law suits, which I admit didn't sit well with my own academic background in Engineering and Physics research. So I decided to put my paper directly into the public domain, and render the concepts unpatentable (at least under U.S. patent law). I also had this great idea that people might be able to sort of "publish" things on the Internet, bypassing paper-based publishers altogether. Of course, peer review is an important part of the publication process, so I needed to "e-publish" the paper in such a way that I could have it "reviewed online". So I asked our honourable moderators of sci-vw at the time, Bob and Mark, whether it could be "e-published" on sci.virtual-worlds, complete. They agreed. And so some time in November 1992 the paper (broken into four parts, because it was too damn long!) was posted, and also desposited (in one piece!) in the HITL sci-vw archives. Of course, since then the World Wide Web has come along, and now every second computer article is espousing the new idea of "publishing on the Net". But at the time many people thought it was all a bit too radical. ;-) Anyhow, there were some fine critical reviews of the paper in this group (probably still there in the archives somewhere), which helped smooth out a few rough edges. Most helpful were those from employees of Silicon Graphics, who could see potential in the algorithm for a *software* implementation, but who preferred the "buy more polygons per second" approach to hardware. I disagreed (and still do); but the reviews are a good read. So where has it gone since then? Well firstly I posted a somewhat more polished, concise paper to sci.v-w in early 1993. (Both papers are now also on my WWW page.) The idea was applied to the realm of *non*-real-time 3-d graphics rendering (again, as a software algorithm rather than in hardware) by Chen and Williams (see SIGGRAPH '93). But, essentially, no-one has yet built an accelerator card based on the idea, unless they have kept it very quiet. So here we are in 1995, and I am again in a position to pursue some good VR research. We will have an Elec Eng postgrad using some of this new FPGA technology (Field Programmable Gate Arrays---they're quite amazing :-) to build some research prototypes in 1996, and I would hope that we would have proof-of-concept prototypes, perhaps with something in VLSI, by the end of the year. We are pursuing this as a research project, again with the hope that it will be taken up by a manufacturer once we show them how to build it. :-) Of course it is harder to give something away than to sell it. 8-) So why am I telling you all this? Well we would like to continue to provide sci-virtual.worlds readers with the same updates on progress as when the project was seeded three years ago. We will be working first on expanding on and optimising the theory of the device, using software simulation, and then begin to implement the simplest "generation" in hardware. Of course, the Net now has the WWW, which means that papers and reports can be deposited on a web page; I'll simply post a URL to the group, and a brief comment. Assuming, of course, that our honourable moderator Toni does not object to such updates? :-) John Costella ==================================================================== Dr John P. Costella School of Physics, The University of Melbourne jpc@physics.unimelb.edu.au http://www.ph.unimelb.edu.au/~jpc Tel: +61 3 9344-5435 AH: +61 3 9768-9268 Fax: +61 3 9347-4783 ==================================================================== -----BEGIN PGP SIGNATURE----- Version: 2.6.2i iQCVAgUBMKyNR2rlpFIGn9BdAQFUMwP/ZaopThDTfYvJ8Oy11ZZ+zkf11GCoqz4R FGs3kzwEJilWF1Q5W3ZB7ceieAgSkaekzMr/XYuJh1jDjKuav7I4z1lEE0RyY3lx n9zMGTYIq+A6Q6ktOaT+lPxMrmvB5/XWRUYjVSSTPCEr5kOa4VyZvcIPXW5gjGAL 6i9KwV9d84A= =PB3F -----END PGP SIGNATURE-----