The Adventure of the Ghostly Images

Engineers wrestle with a cranky CRT-based imager and a prospective customer is getting antsy

By Charles Glorioso, Contributing Writer

A Los Angeles-area company I worked for was using an innovative and patented technique to “burn” images on 35mm slide film one pixel at a time when the original IBM PC was announced, and the owners saw an opportunity to greatly reduce the cost of building a film recorder to create presentation slides.

The engineering team took on the challenge and developed a CRT-based imager that wrote images one pixel at a time, 4000×2000 by 24 bit color, directly onto 35mm slide film.

The digital engineer designed an ISA board for the PC that collected 4000 bytes of data for one image line and provided that to the film recorder. Looking for a place in the PC ISA address space to safely place his 4k register, he decided to put it into ROM space. His logic was that the ROM addresses are by definition read only, and he could put a write-only register into that space without causing conflict.

We built the prototypes and it all worked beautifully. PC software created images and wrote them one line at a time into this 4k register. The film recorder serial shifted the data out and imaged it pixel by pixel onto the film. Once developed, the resultant 35mm slides were very sharp and very bright. In short, spectacular! And the total solution cost, including the PC, was less than half that of existing, competitive products.

While we were finishing the design the PC AT (Advanced Technology!!) was released. This was a 16-bit machine and promised faster throughput. We borrowed a machine, made a few test images, and developed the film. We were elated that it not only worked, but it also made images more quickly.

We built multiple pilot versions of the product, shipped them to prospective customers, and eagerly awaited the orders.

Then we got a call from a prospective customer on the East Coast. The images coming out of the unit were dark and totally unacceptable. Oh well, we guessed, the prototype failed during shipment. We sent him a second, 50-pound unit. In a few days he reported the same result.

After a week of trying things unsuccessfully by phone, we bought an air ticket for the design engineer and sent him and his oscilloscope to the East Coast. He hooked his scope up to various places without understanding the failure. However, the developed images were all very dark.

Finally, he hooked his scope up to the ISA bus, and the answer became clear. While the AT hardware did 16-bit writes and reads from most addresses on the ISA bus, reads and writes to ROM space were constrained to 8 bit. Why the IBM engineers had done this was never clear to us, especially since no data writes were expected in ROM space, but it was an unexpected behavior that was likely screwing us up. But why did it work in our office, but not on the East Coast?

Figuring this out took the engineer several more days. In the end he determined that in ROM space, the lower 8 bits was presented on the ISA bus and latched, and then the upper 8 bits was presented on the same 8 wires as the lower 8 bits and latched.

Our application placed a 16-bit word on the ISA bus, the AT latched in the lower 8 bits, then expected 8 new bits and latched those. However, our application, thinking it had done a 16-bit write, didn’t provide any data for the second latch operation.

In LA where it was summer and dry, a ghost image of the data remained on the bus and was captured by the second latch. On the East Coast where it was summer and the humidity was high, the ghost image bled off the bus before latching, and all upper bits were always latched as zero. So, while the upper 8 bits were wrong in LA, they had values, and our quick testing had missed that the image was not quite right. On the East Coast, the images were always 1/2 exposed or less, since the upper data bits were always zero.

The fix was to create an AT only version of our board that grabbed all 16 bits when the ISA bus thought it was only latching the lower 8 bits. The product worked as advertised and was a technical success. The company, however, was not successful.

Contributing Writer Charles Glorioso has a BSEE from Purdue and an MSEE from Illinois Institute of Technology. He has over 40 years experience in electronics design and management for industrial and consumer products.Employers have included Teletype Corporation, Cadence, The Exploratorium, at least 6 companies which no longer exist, and most recently Davis Instruments. Charles retired earlier this year after six years as Director of Engineering at Davis Instruments, and is now working there part time on special projects.