I was asked by a friend if I could sort out a problem that had struck his highly-specialised professional record-transcription equipment. On quiet notes the Right channel was slightly low in output but louder notes produced violent blasts of distortion; the left channel was unaffected.
The problem seemed to be originating in a rack with a Prism A/D converter, a couple of Cedar de-noisers and a Prism D/A converter. By-passing the Cedar produced no improvement, neither did a substitute A/D converter. The fault was definitely coming from the D/A converter; swapping analogue leads around showed that the fault stayed with Right analogue output.
The Prism is a complex piece of kit and, in working order, is probably worth as much as a small secondhand car, so I approached it with some trepidation. There was no service data available anywhere, not even a circuit diagram, so I had to deduce what I could from the board layout. The tracks were almost invisible and there may have been multiple layers to the board - it wasn't obvious.
Working from the chip pinouts, I established that there was a large DC offset on the Right balanced XLR output connector. It was symmetrical about earth, so it probably originated before the balanced output stage. There were four stereo D/A converters on the board, they appeared to be arranged with two chips (four channels) to the Right and two chips (another four channels) to the Left. One of the Right channel outputs was clipping hard against the rails.
Looking at the data sheet for the Philips TDA1574 D/A converter chip, I discovered that there is an op-amp built-in for each analogue output, with its gain set by an external resistor. Philips suggest a value of
13k for this resistor as the op-amp forms part of a filter. On the board I found the relevant resistor ...and it was open-circuit!
Changing the resistor was a bit fiddly but the equipment is now working properly again. A failed resistor is a rarity these days, so this was quite a surprise (and a big relief in view of the value of the equipment).
The question arises: "Why go to all the trouble of putting the outputs of four D/A converters in parallel?" The answer appears to be that Prism achieved an incredibly low noise figure by this method. Every time you parallel a pair of signals the coherent signal increases by 6dB but the noise (which is not coherent) only increases by 3dB. Every doubling produces a 3dB improvement in the signal-to-noise ratio, so Prism gained a 6dB advantage.
I don't know whether they also staggered the clock pulses to increase the frequency of the clock ripple on the outputs (to give better filtering), but it wouldn't surprise me if they did.