So much for my role as part-time cable puller and system administrator. My real job at Bell Labs was to do research and publish papers. Around 1985, I changed my interest from database systems to computer graphics. I was attracted by the visual feedback of making synthetic images and the interesting underlying problems in signal processing, numerical methods, geometry, and the physics of light and materials.


Image with aliasing

Antialiased Image

I’ve particularly enjoyed working on ray tracing and antialiasing. Images are usually synthesized by calculating infinitesimal points, and if this is not done properly, you get a familiar pattern of “jaggies” or aliasing, as seen above left. An antialiased version is seen on the right, and that’s about all I’m really going to say about a problem that occupied years of work and thought.

Turner Whitted worked at Bell Labs in the early 1980s, one of the primary inventors of the ray tracing method of image synthesis. People had worked on limited forms of this rendering technique, but Turner published a classic paper describing ray tracing with reflecting surfaces, antialiasing, and hierarchical bounding volumes, three fundamental inventions in one paper! Rob Pike got interested in my first ray tracer, and the two of us optimized the daylights out of that program. Turner’s ray tracer took hours to make a picture, and ours took minutes. He was delighted, and gave a lot of help and encouragement.

Actually my first ray tracer, sort of, was at Caltech in 1979. Among other things, it did a Monte Carlo simulation of photons emitted and scattered in the Cherenkov detector. It did not result in a picture of anything, just a sensitivity map for the detector. But it did perform a lot of the basic geometrical intersection and scattering calculations that all ray tracers do.


Bell Labs Graphics (Subramanian, Amanatides, Naylor, Mitchell)

By the mid 1980s, we had a nice little graphics group in our lab. Seen above, KR Subramanian had just graduated and was working with Bruce Naylor on real-time geometry algorithms. John Amamatides and I were working on a variety of ray-tracing problems, and we had written a renderer called “FX”, which was fast enough to generation animations. FX was designed to run on the AT&T Pixel Machine, a parallel graphics computer designed at the Holmdel lab around 1987.


Scene from Megacycles

John and I generated a short ray traced animation with FX and the Pixel Machine, called “Megacycles”. I found it on YouTube (Megacycles). It’s crude by modern standards, but a huge computation by the standards of its day.

My department head then was a talented electrical engineer named Ed Szurkowski. At the annual SIGGRAPH convention, Ed and I checked out the equipment exhibition together, and we stopped at the booth of a company that had just made a new 3D graphics board. The man in the booth was an engineer, and an old friend. He was happy to see us, and when we asked about the board, he shut off the machine and pulled the board out. Ed looked at it for about 15 seconds. As we walked away, he said, “That’s interesting, I never realized you could strobe DRAM that way, they’re saving one extra cycle. I’ll have to try that someday!” I thought to myself, never show your circuit board to Ed.

The director of our lab was also an Electrical Engineer, named Arun Netravali. Arun was an expert on digital image processing, optimal control theory, and he had a powerful command of mathematics. We worked together on several projects, and he was the first to apply control theory to computer animation. Arun would one day become the head of Bell Labs, and I saw first hand the kind of talent and courage to act that it takes to earn that position:

Around 1988 or 1989, Zenith and AT&T formed a partnership to work on high definition television (HDTV). Zenith had an analog format and was hoping Bell Labs could design some of the microelectronics. Arun realized that the time was right for a digital standard, using modern image compression instead of television’s ad hoc interlacing scheme. He gathered a few top engineers, and in almost no time, they had a complete system proposal for a non-interlaced (progressive) video signal format and the processing electronics. It would later be called “720p”.

Getting vice presidential approval to proceed, Arun and Scott Knaur hired rooms full of engineers. Xilinx had recently invented the FPGA, a programmable chip designed for rapid prototyping of hardware, and this was used to build the first working implementation of the 720p video system, in almost no time. The first 720p system was a solid cube of electronics about 2 feet on a side.

The CEO of Zenith came to see it, still perhaps thinking that we might build their analog system, but he was overwhelmed by what Arun showed him. At one point during a demonstration, he said, “So Arun, are you telling me that our analog system is shit?” Arun said, “Mr. Perlman, I’m telling you the analog system is shit.”. A new partnership was immediately made, where Zenith would build a radio frequency modem to modulate the Bell Labs digital signal. It’s not often that big corporations turn on a dime, but in this case, there was no time to fool around.

I thought a lot about the theory of interlacing and what it meant in terms of aliasing and flickering artifacts. Amanatides and I modified the FX ray tracer to produce perfect digital video, by correctly synthesizing an interlaced signal. We presented the work at the SIGGRAPH 1990 conference. Rehearsing the talk at Bell Labs, I was somewhat more informal than in public, and I ended the talk by saying “thus proving that interlacing looks like shit”. Arun slapped his knees and said, “I’ll give you a hundred bucks if you say that at SIGGRAPH!”. Not in front of 5000 colleagues, but my final slide said, “Interlacing sucks”.

The broadcast industry fought progressive scan standards for years. The battle was till going on when I joined Microsoft years later. The computer industry supported progressive scan, because text and iconic graphics flicker terribly on an interlaced display. It is only in recent years that progressive scan has really gained support, now that the public has seen the superior quality of 480p video generated by new DVD players.

In the early 1990s, Arun met my future boss Bill Gates at a meeting of Microsoft and AT&T. AT&T had rented a private dining room at a very expensive New York restaurant. When the head waiter asked Bill Gates for his order, he said, “I don’t see anything I like here. But I noticed a Burger King across the street. Could you send someone over for a hamburger and fries?” The waiter was calm and polite, but the executives from AT&T were completely unnerved.

Advertisements