More transparency in active speakers

Those of you who read my earlier posts about improving my Fayou actives might have realised by now I'm a bit fanatical about grounding and its effect on sound quality. I had an idea - could a system be put together without any ground at all, and might this solve many of the sound quality issues I've found in my setup? I could only find out by trying it. So, the idea was born - 'Goundless Sound'.

First up, a groundless amplifier is just another name for a balanced, or bridged design, at least in respect of the output being balanced. A truly balanced amp would also need a balanced input - that's not really ground-breaking (pardon the pun, could not resist) either. Putting the two together is hardly rocket science. But I decided on a few twists - more about those in just a moment.

Convention is a useful thing. Ponder a world without convention in respect of computer keyboards - Asus and Toshiba having one layout of the keys on their notebooks, Dell and Lenovo having another. A nightmare, right? Yet convention also has its downside. Why do we have the convention that a power amp needs to have roughly 1V input sensitivity? To my way of thinking this is too low - sending a signal across a cable means its subject to interference, so the bigger the signal at the start the lower the interference will be as a proportion.

Opamps generally work on supplies up to +/- 18V (some go as far as 22V, not many) and can put out 30V peak-peak without breaking a sweat. With a balanced signal, this becomes 60V peak-peak. Therefore to me it makes sense to have input sensitivity somewhere in the region of the maximum level to preserve SNR across a pre to power cable. The professionals have been using higher levels on balanced cables for ages, but it hasn't penetrated down to the consumer market. Time for diyers to wake up - what's the point of going beyond 16bits if there are real bottlenecks at other points in the signal chain?

So I've decided not to call my 'Goundless Amp' an amp at all, because it barely amplifies. Actually it does in the prototype, it gives 6dB gain but I'm thinking about reducing this. So here's another name for it - 'Balanced Drive Unit Buffer' - BDUB for short. Calling it a buffer rather than an amp highlights something else important too - its a module, not a separate system component.

Running chip amps bridged is fraught with difficulties because they're really not designed for low impedances and bridging effectively halves the impedance the amp sees. So to bridge chipamps, its really necessary to parallel them too. Starting out with a TDA2030a (because its the cheapest chipamp for me right now) then a minimum of 4 would be needed, giving a parallel bridge configuration. However this is limited by thermal considerations to only 80W or so (and requires a minimum 4R load) and I was looking for something to make a worthwhile improvement over the old LM3886. So I have adopted an 8 chip design, with four paralleled ICs either side - this is very comfortable into a 4R bridged load as then each IC sees 8R, and can cope with lower.

Based on experiences on this forum, getting paralleled amps to work well is a bit of a headache. One of the reasons is misunderstanding about how closely the gains need to be matched across the whole frequency range (including bass and potentially infra-bass). National's AN-1192 has to bear some of the responsibility here as its misleading in various places. Sharing resistor tolerance does not make much of a difference but gain setting resistors do.

My BDUB (wrote amp first and had to delete it) operates without the need for sharing resistors because I've designed it as a system module, not a standalone amplifier component. The sharing resistors are provided by the speaker cable - it only works with Cat5 or Cat6 network cable, not your average twin-core stuff. Cat6 is slightly thicker (23AWG) so is slightly preferred, but here it has a minimum length requirement of 2m. Few people will want their speaker cables shorter than 2m so I didn't consider that to be too much of a limitation. A 2 - 3m length of Cat6 cable provides an optimum value for the sharing function (of the order of 0.1 - 0.2R). If you really need a shorter cable, I have found some really cheap Cat6 patch cables here in China which appear to be constructed out of what looks to be 28AWG, so that'll work nicely down to 1m.

... to be continued