Power Supply Considerations

Overall Design Criterea

I think it's relatively rare to start designing anything at all with a completely free hand, able to simply select what will best meet the main objective. "Design a widget to accomplish function X within budget Y (great so far . . .), but then there's a "Z", a rider, a qualifier, and there's anything BUT a free design hand! Such "riders" end up governing the topology of the widget. As with all my projects, that was certainly the case here.

I started with a list of requirements for the power supply for this amplifier, and this of itself reduced the topology options considerably. Design criteria for my power supply were;

That was pretty well it. But this was enough; as a direct consequence of this wish-list the structure of the power supply was pretty well self evident (at least to me anyway(!);

The Power Supply used for the project; a full wave voltage doubler, off a non-centre tapped secondary, with bleeder resistors and capacitor filtering.  Almost the simplest of all power supply arrangements.

Mains transformer and the two main filter electro's of the doubler circuit.  The base of the left electro has a cardboard coller just doubly ensure the isolation of its metal outer case, which is not at chassis-ground potential.

Measuring Load Regulation

So at this point the power supply topology basically suggested itself. However when I'd done something like this before, working with such a power supply while fiddling with the design of the actual amplifier stages, I'd had all sorts of trouble. Poor power supply regulation meant that every time I altered circuit conditions in one stage of the amplifier, the load on the power supply altered, and the B+ level altered - usually sufficiently to alter the expected effect of the circuit change made in the first place! I'd ended up changing a cathode bias resistor three or four times, rather than once to achieve a desired change, on several occasions! Add this to other variables that change the B+ - a volt or two change in my 240Vac mains from one day to the next, returns a three or four volt B+ change, and the amount of hum on the B+ rail changes things too, and on-the-fly designing can be frustrating to say the least.

Normally, and sensibly, one might use active regulation to circumvent all such issues. The B+ level remains set in stone for most practical purposes. I took another tack, and decided to try and chart the B+ regulation closely, and then make all changes include the V(B+) and I(B+) characteristic accordingly. By doing this, and by implementing the major current consumer, the output stages for both channels, first, I didn't experience much of a problem with variable B+ levels this time.

A large dummy load was something "interesting" though. I calculated that my full quiescent current load would be about 30~40mA each for four output EL84's, 5mA each for two phase splitter triodes, and say another 10mA for the input valves, and some bleeder resistors for the main filter electro's, and anything else I might have forgotten! Worst case, say about 180mA all up. Hard to imagine the B+ wouldn't have been dragged down to +300V by then, so I need a dummy load of 300V/180mA = 1.7kOhms able to briefly dissipate 300V x 180mA = 54W or so. Hmmmm.

Actually, that wasn't so very difficult. I had a bag of 10K 10W resistors somewhere - obviously a "gem" purchase at one junk shop or another . . . . I just kept adding them in parallel across my test set-up power supply. Six in parallel hit the design max load current (10K/6 = 1.7kOhms) and sequentially lesser numbers gave a string of values to plot graphically. In a sprit of true scientific endeavour I decided to hook up the oscilloscope too, for good measure, and record the amount of ripple observed. It came out as follows. I've plotted what the final curve actually looks like too - it's a bit lower in level, I imagine because my original set-up didn't load the filament supplies and this probably had some effect too.

Measured Power Supply Regulation

The major surprise of course was that the off-load voltage was considerably higher than I expected, as was the level of the whole curve actually. The B+ at a VA of around 39, (i.e. the nominal 130V x .3A of the secondary winding), was about 395V, not the 367V (or likely lower) that I'd rule-of-thumbed earlier. A tad under 10% higher, now what could it be . . . .?

D'oh! Jeez I HATE it when I make stupid, obvious errors and oversights! I had my "normal" 240Vac mains supply strung across what, according to the same data written on the mains transformer that I was using everywhere else, was intended as a 230Vac primary winding. Not good for the transformer, at least in theory, but I know from experience that they usually survive such ill-treatment well enough. It would mean too that the filament voltages would also be "high" too, and this was quickly confirmed by direct measurement; 6.9Vac off load instead of 6.3Vac. This was definitely "not good" as such a 10% over-voltage here results in much greater than 10% reduction in valve life-time. This issue would need addressing with some filament supply dropper resistors.

Using again my 39VA rule-of-thumb, I calc'd that the primary current would be around 39VA/240V = 160mA at a point representing slightly more than half max design secondary loading. To drop 10V (from actual mains 240V to the transformer spec of 115+115=230V) would then need a 10V/160mA = 62 (say 68) Ohm resistor, able to continuously dissipate 10V x 160mA = 1.6W (so at least a 5W resistor would be advisable). Trialing a 68 Ohm, 5W resistor confirmed things well enough and I knew I'd found the source of the over-voltage. In the event I actually ditched this resistor, and stuck with 240Vac across the 230Vac winding. In practice, in the completed amplifier project, I ended up drawing about 180mA @ 370V (370V x 180mA = 67VA) from a winding designed to supply a nominal 39VA. While anything BUT best practice, and although the transformer audibly protests just a little, it's thus far so good!

Filament Supplies

Filament supply for the 6AN8's at left, the EL84/6BQ5's at right.

Each of the two filament windings has a 47 Ohm W resistor to ground, a simple, conventional practice to ground reference the filament supplies, and do so symmetrically.

For the 6AN8 filament supply, the measured transformer secondary voltage, (higher than normal since I was supplying 240Vac to the primary not the rated 230Vac), was 6.76V. This high value was actually still a surprise since I was apparently loading the winding more heavily than it's rating, and expected to find it pulled well down. Still, it was a "nice" surprise! Now I'd read that longevity of valves is decreased significantly by excessive heater voltage, so I needed to look at adding some dropper resisters. The rated heater current for the 6AN8 is 450mA; 900mA for two of them. So, I needed to add in series one (6.76V-6.3V)/0.9A = 0.51 Ohm resistor (suitable for continuous dissipation of (6.76V-6.3V) x 0.9A = 0.4W), or, to preserve the symmetry of this AC supply, two resistors of half these resistance and power values. I used one 0.27 Ohm 0.5W resistor in each "leg" as shown in the diagram, and got plenty near enough to 6.3V actually measured across the filaments.

Underneath one end of the chassis, showing major power supply components and filament supplies.

In the case of the output stage(s) filament supply I decided to leave things alone. The loading on this second 6.3V secondary winding was lower than the nominal rating, and with the slight over-voltage on the primary I expected to see something significantly higher than the 6.43Vac actually measured in practice. That's mean a (6.43V-6.3V)/3.04A = 0.04 Ohm (or two 0.02 Ohm) resistor. This value was too low to be of much point, so I decided to leave things just as they were.