The heart of the amplifier is the custom-made 400VA torroidal transformer. This feeds a pair of bridge rectifiers made from MUR860s, and high quality 10,000uF BHC capacitors.
The transformer
Original prototyping was with a standard 500VA which I inherited from a friend, but this made far too much mechanical noise. It also had enough leakage current to affect the hum performance under certain conditions - both of these are common problems with large torroids.
With this in mind, I approached a number of companies with a view to getting a low-flux model made. Buying custom-made transformers in single quantities is an expensive business - a glance in the major catalogues will show you how (relatively) cheap standard transformers are, but remember that these are manufactured in huge quantities and the set-up costs per unit are small. If you're making just one, you pay for the design and one-off costs all at once. But as this amplifier was largely made from recycled components, a little extravagance here was easily justified.
This was an interesting experience - of the 4 companies I approached, only 2 of them bothered to get back to me! The first response was from Avel-Lindberg, who were rather surprised, indeed almost indignant, that I'd suggested that their standard models weren't quiet enough! To be fair, I haven't tried one of their normal models, but I wasn't about to buy one to find out! They did suggest a low-flux design which would have been physically larger than the dimensions I specified and wouldn't have fitted in the chassis.
A day or so later, Antrim sent me a quote. I was pleased to see that they appeared to have understood all of the requirements - having asked for low mechanical noise, and supplied maximum dimensions, they'd designed a model that used a selected core and de-rated the transformer to 400VA in order to have a lower flux density while still fitting my specified dimensions.
A couple of days later, they contacted me to follow up the quote. This is when I discovered that they are very "hobbyist friendly" - they are able to take credit-card orders other the phone, and are well-used to dealing in small quantities for people like me. This made my mind up, and I didn't bother chasing up the other companies that didn't respond.
As I said, this will be expensive. The initial quotes from both companies looked quite good at around £50, but you need to add delivery and VAT to that.
There was some confusion a bit later, because once I was ready to order the transformer, I provided a list of extras and asked for the prices of each (for example, black tape, an electrostatic screen, specifying the lead-out positions, etc). The response was an extra £20 on the price, but I couldn't get a clear idea of how much each item cost - clearly I wouldn't want to pay a large premium for black tape, but I would have been prepared to pay a reasonable amount for the electrostatic screen. After a conversation, the price came down to around £60, but I'm still confused. Things would be simpler if there was some sort of rate card, but the price varies according to quantity and the sort of customer you are, so I guess it must be very hard to put raw numbers on these sorts of things.
Anyway, once past that, the transformer arrived after about 12 or 13 working days. It was clearly a quality product, and I couldn't wait to try it. An initial test with a variac showed it to be hum-free up to 250V - the standard-grade model was OK up to around 235 volts, at 240V it got very noisy indeed. When bolted into the chassis, it's not absolutely silent all of the time, but much, much better than the standard model and inaudible for all practical purposes.
So would I go down the custom-made route again?
The only drawback is the price. The final cost, including VAT and delivery came to just under £90 - that sort of money would buy 2 or 3 standard 500VA transformers from Farnell. But this price would quickly come down as the quantity rises, so that's a justification for dual-mono supplies!
Against that, you have complete freedom. You can specify the ratings of as many windings as you like, and you can even specify where they emerge from the transformer. You can choose a potted centre, which does away with the need for the normal dished washer and saves a surprising amount of height. But most importantly for an audio power amplifier, you can specify low-flux operation, and an electrostatic screen. Both these features are essential for any torroid larger than around 200VA, and you won't get an off-the-shelf model with these features.
So yes, I'd definitely do it again. And I'd use Atrim - the service was excellent! (Disclaimer - no connection with the company other than as a satisfied customer).
Rectifiers
Following on from the transformer, the DC supplies are made using two bridge rectifiers. There wasn't room on the chassis to accommodate a pair of standard packaged rectifiers, so that was an excuse to build the rectifiers using some spare MUR860s that I bought for the gainclone monoblocks. As I mentioned in the gainclone experiments page, the audible differences were marginal-to-imaginary, but as they would fit nicely in the available space, in it was an easy decision.
As you can see, these have been made using aluminium angle-stock, and FR4 copper-clad PCB stock. The simple layout has been formed using a scalpel, and the excess removed using heat from the soldering. The wires from the transformer connect straight into the PCB.
The DC from these rectifiers passes to the smoothing capacitors, and then to the PSU board. The wires carrying these rails are tightly twisted to ensure the radiation caused by the charging currents is contained. The board forms a central connection point for the power supply rails and the earths. It also has a pair of LEDs to indicate the presence of DC and safely bleed the charge in the event of no load. Additionally, I provided space for snubber components - this is something I can experiment with when my reference system is out of storage, but for now I used 1Ω and 100nF components as a starting point. However, I can't help thinking that the inductance of the connecting wires is going to negate any possible effects of this - intuition suggests that impedance correction should be placed at the chip if anywhere. Assuming that "snubbing" actually has any effect at all - it's one of those DIY Audio crazes I've never been wholly convinced about.
You can see the two sections of aluminium angle that attach to the top and bottom front rails and support the board. The multiway Molex connector is the interface between the amplifier and the control circuits - the yellow wires form the Mute bus, the blue ones are the DC detect bus, and -UNREG is supplied to the control circuits via the black wire. This cable assembly passes under the false floor chassis and makes its way to the control board.
This view shows the rear of the board. It was recycled from an old piece of equipment many years ago, and is 3mm thick, so very rigid. It was a simple matter to make the cuts in the tin-plated copper to form the simple layout. The amplifier's ground connects to the chassis here via the 100Ω resistor and 100n parallel capacitor.
Not shown yet are the four connections to the negative speaker terminals or the connections from the smoothing capacitors, which meet at the centre of this board. This all worked out rather neatly in the end.