In 1982, I was a manager with a technical consultancy in the United Kingdom. At the time, there was a boom in simple home computers like the Radio Shack TRS-80 and Sinclair ZX. The U.K. government was keen to develop new employment opportunities in Wales and helped the company Dragon Data develop a low-cost, U.K.-made competitor, and our consumer products group designed the hardware.

The Dragon 32, as it was called (dragons are big in Wales), was highly limited and the display quality [using a radio frequency (RF) modulator generating a signal to feed into a TV antenna socket] widely criticized. Every vertical feature was accompanied by a colored “ghost” just to its right, looking like the effect of an echo on a broadcast TV signal. You could even estimate the echo delay time by applying a ruler to the screen; it was about a half microsecond. Sales were suffering, customers were returning products, and our client was getting very impatient.

The design team was focused on the problem but totally stumped—the phase alternating line (PAL) video signal going into the modulator was completely clean, with no sign of any overshoot or echo. The team ex­­plained that, to save money, they used a Motorola 525-line National Television System Committee (NTSC) integrated video encoder chip, with ancillary logic to add 100 lines to get the 625 needed for PAL. Using this chip, already made in large numbers for U.S. products, simplified the design and cut costs. A few more pence was shaved off by using the same crystal, with the by-product that the actual line period was the NTSC 63.5 rather than the PAL 64 ms. TV sets seemed to happily cope with this, sometimes with a little adjustment of the horizontal hold.

Suddenly, I recalled the difference between “simple PAL” and “delay-line PAL.” Early color TV sets used simple PALs, but once the component industry had worked out how to make a cheap, compact 64-ms ultrasonic delay line, sets quickly switched to

delay-line PALs. This delayed the color signal for each line and averaged it with the next—because successive scans are highly correlated, it cancelled out differential phase errors making the system much more robust. Of course, if you have a signal with a 63.5-ms line period, the delayed signal is “averaged” a half microsecond later. The video system in the Dragon was doing exactly what the designers intended, but there was a fundamental mismatch between how they assumed TV sets work and reality, hence the “ghost.”

A solution was far from easy, requiring a complete redesign of the circuit. A quick fix added a SCART video output, bypassing the RF system. As part of mollifying the client, we also developed a SECAM modulator for the French market—the Dragon was the only computer to have this. But it was all too late, and Dragon Data collapsed in 1984.

One lesson I hope the company learned is that if you are going to do something clever to save money, make sure you thoroughly understand every part of the system that might be affected. Also, if you see anything that doesn’t make sense, make sure you properly understand why it’s happening before releasing a product.