Alpair 12P MLTLs - Part 4

Day 4 - 4 hours

Goodness, it's over 7 months since my last blog, and over 10 months since I last had chance to play with the speakers. It's fair to say audio has been on a back burner for much of this year, but today I made a little progress with the MLTL speakers :-). But first let's have a look at temperature-watch.

Parts of the country are experiencing terrible flooding, no doubt made worse by the unseasonably mild weather which is probably retaining a lot of moisture. 12 degrees C, mild for the end of December.

So first job was to have a quick tidy, then remember where I'd got to. I'd made a start on routing the first hole for the 12P at the end of the last session, but I was rushing and hadn't noticed that the template had slipped slightly and was no longer centred. Fortunately I noticed in time before serious damage was done. So I marked out the circles with a compass to be sure, then carefully reset the template and clamped well.

More haste less speed. Lesson learnt - measure twice cut once.

The first hole needed three phases. The first was to rout the inner hole for the driver to pass through, 170mm diameter. The plunge cutter only reaches part of the way through the baffle though, so needs cutting from the other side too - the third phase.

The second phase is to rebate the driver, 203mm diameter and 11mm deep. This was a bit of a laborious process as there was a lot of wood to cut away. Here's the finished hole. Note the misalignment from cutting from the front and then the back. This was largely due to the initial cut that had slipped. It shouldn't be a problem though, as long as the driver passes through and the driver makes an air tight seal.

Next was the hole for the Monacor pseudo ribbon I'm going to use to fill the treble out over 12k Hz. I started to cut away the all the wood so the unit sits within a cavity...

...but then decided that I was probably going over the top so then cut full depth before turning over and finishing the cut. Ideally I guess I would have preferred to isolate the driver from the main chamber and the air pressures within, but hopefully it shouldn't make too much difference. I might even add a plate to the back just to seal it.

And here's the two holes viewed from the rear.

Next the 3" port, which requires a 79mm diameter hole. This is perhaps just a little slack, but probably better than being too tight and struggling to insert the port. I can always use a bit of glue or silicon to secure them.

From the rear

And here's the first baffle with the initial holes made. It needs a good clean up and a rear 45 degrees chamfer cutting on the rear of the hole for the 12P. And a check to make sure everything fits of course.

For the second baffle I fully marked out all three holes both front and back first, now I had a little more confidence with what I was doing.

And started with the largest hole for the 12P. Here's the first 170mm diameter hole cut and the start of the wider 203mm rebate.

The finished hole from the rear.

And from the front.

And that was the end of the today's audio fun.

Design Of A Single Ended 2A3 Valve Amplifier - Part 1

This is a simple single ended valve amp I "threw together" a couple of years ago. It was meant for a colleague to try valves at home, but never found it's way out of the house. It's been in the project room ever since (as attested by the obligatory dust to be found in all the best male domains) and makes a particularly good match for the Quasars.

There's nothing fancy about the design - a triode strapped D3a pentode capacitor coupled to a 2A3, both in self bias with AC heating. The HT power supply is a simple LCLC passive filter with a GZ37 rectifier. Here's the audio schematic without part values, all straight forward stuff.

But it has decent parts and sounds very good. Surprisingly good really. Not the best amp I've built (that's a copper GM70 SE amp, but that's another story), but it fits in a single chassis which makes a change from most of my more recent builds... And I haven't felt the need to replace it which speaks volumes. And in the spirit of openness here's a photo of the inside just to prove how thrown together it was. I certainly wouldn't encourage anyone to study my wiring!

The chassis was actually recycled from my very first amp build - a Bluebell Audio 2A3 Loftin White built 10 years ago. Parts were supplied by Philip Ramsey using Shishido san's circuit. I just needed to drill a few more holes to accommodate different output transformers, chokes and a volume pot.

Preamble over, let's talk about the design. Bear in mind this is intended to be a simplified explanation of simple valve amp design - if you want the in-depth theory there are many good sources, try Valve Wizard's website or Morgan Jones Valve Amplifiers. I would also highly recommend reading Gordon Rankin's write up of his Bugle 45 amp - a great amp and really good design primer. First though, recognition must go to Nick Gorham as it was a circuit he posted that inspired me to build this.

The Output Stage

The usual place to start when designing a valve amplifier is the output valve, in this case a 2A3. By looking at the datasheet we can see whether it's likely to be suitable for our system, or what we might need in our system to make it "work". Assuming we have some reasonably efficient speakers a 2A3 makes a lot of sense.

Anode (Plate) Characteristics

Let's look at the anode characteristics graph on the datasheet

This graph shows a series of curves representing the relationship between voltage across the valve and the current through the valve for different grid voltages. Let's consider the classic 2A3 operating point, i.e. how the valve is "set up": 250V, 60mA, -45V on the grid. If we look back at the first page of the datasheet we're even told what might be a good operating point ;-). So looking at the graph above, 250 plate volts and 60 plate milliamps just happens to intersect with where a line representing -43.5 grid volts would be. (Okay, that's not quite -45V but it represents the lower limit of the AC heating, 2.5V.)

If for instance you maintained 250V across the valve but went with -35 grid volts you'd draw around 115mA. And burn out your 2A3 pretty quickly. So we need to check that our proposed operating point is within the valve's rating, and for a class A1 amplifier the datasheet tells us that we shouldn't go above 300V across the valve, and we can plot this line on the graph.

The datasheet also tells us we shouldn't let the anode (plate) dissipate more than 15W. So we can add this to the graph too as a curve with a series of points where voltage x current = 15W.

To draw the 15W curve we could plot a point at 100V and 150mA (which is 15W) say, and another at 200V and 75mA (= 15W), and another at 300V and 50mA (also = 15W), and so on for as many points as we feel necessary. For safe operation of the valve we need to operate in the area of the graph below the curve.

As we're designing a class A1 amplifier the grid should always remain negative relative to the cathode too, so we also need to be to the right of the 0V grid line too.

And if we plot all three on the same graph we find the acceptable area of the curves that we want to work within, the green shaded area below.

 And 250V anode voltage and 60mA is right at the limit of this region, thus maximising the power obtained from the valve. In general it's usual to maximise the power obtained from a valve, if you're lucky enough to have rare old monoplates you might choose to be a little kinder to them and maybe run them at 50mA instead...

Loadline

The next thing to consider is the loadline, an example of which is also shown on the datasheet, rather conveniently. It's the sloping straight line labelled "LOAD RESISTANCE = 2500 OHMS", centered about the -43.5V grid line, ranging between 0 grid volts and -87 grid volts. This loadline represents the load resistance of the output transformer, which in the case of the datasheet is 2500 ohms. But why 2500 ohms? You could just leave it to the RCA engineers and accept they knew what they were doing, which they most certainly did, but the longer answer is it's a compromise.

What do we want from our amplification stage? Usually we want to maximise power and we want to minimise distortion. The relative importance of the two depends on what we're trying to achieve, but as this is an output stage we want a good balance of the two.

If we consider distortion we can see how linear the valve is (or isn't) at our operating point by looking at how evenly spaced the grid lines are along the load line. Between 0V and -60V the spacings look fairly even, but beyond -60V the grid lines tighten a touch. So when the music signal is larger than + or - 15V (so lower than -30V and higher than -60V) the distortion will increase.

In general, if we wanted to reduce distortion we could increase the load resistance which would flatten the loadline i.e. make it closer to horizontal. But if we do this we will lose some power. And that's the compromise. We're considering the classic 2A3 operating point, and there's good reason for doing so as it's a good balance.

The rule of thumb for an output valve's load resistance is 3 x the anode (plate) resistance. So if we look at the datasheet again, the plate resistance is given as 800 ohms, which multiplied by 3 gives us 2400 ohms, or our 2k5 output transformers. 3k5 output transformers are often seen used in 2A3 schematics and trade a little output power for slightly lower distortion.

If we look back at our graph of anode characteristics with our limiting conditions marked on we can see the loadline is right at the very top of the 15W line. In fact part of the loadline actually crosses the 15W line but it's transient and overall the valve dissipates 15W.

Cathode Resistor

Okay, that was a bit heavy, sizing the cathode resistor is much easier now we've decided how we want to operate our valve. As we're designing a simple valve amplifier we're going to use self bias. We know that there will be 250V across the valve and we need to bias the grid at -45V relative to the cathode. Ohm's law is our friend

voltage (volts) = current (amps) x resistance (ohms)

or

V = I x R

Rearranging Ohm's law gives us R = 45 / 0.060 = 750 ohms. The datasheet suggests 750 ohms, there's a surprise!

We also need to calculate the power rating of the resistor.

power (watts) = current (amps) ^2 x resistance (ohms)

or

P = I^2 x R

(Blogger doesn't seem to allow superscript for the squared term)

Which gives us P = 0.060^2 x 750 = 2.7W

But we must derate this as a 2.7W resistor (even if we could find such a thing) would burn up very quickly. Typically we derate by 3 to 5 times, so our 2.7W becomes 8.1W to 13.5W. Even at 3 x derating the resistor will get very hot and I prefer to go to 5 x if I can find something suitable. In this case I would go for at least 12W, maybe a nice Mills if you're feeling flush.

Our output stage is starting to take shape now. Under the cathode we have a 750 ohms 12W resistor which raises the potential at the cathode to 45V. As there's 250V across the valve (i.e. between the cathode and anode) there will be 250 + 45 = 295V at the anode.

Cathode Resistor Bypass Capacitor

If we don't use a capacitor to bypass the cathode resistor the stage will have lower distortion and higher headroom, but more importantly for an output stage it will have a small fraction of the gain. This is why common cathode stages usually have a cathode resistor bypass capacitor. Calculating the size is a little involved.

capacitance (farads) = 1 / (2 x pi x f-3 (hertz) x R (ohms))

or

C = 1 / (2 x pi x f-3 x R)

f-3 is the frequency at which bass will have rolled off by 3dB. 3dB represents a halving of the bass output in this case so we need to set f-3 somewhere below the frequency at which we want full bass output. If we want full bass output down to 40Hz, not unreasonable for an output stage, let's set f-3 at 20Hz.

R, unfortunately, isn't simply the value of the cathode resistor. It's actually the cathode resistor in parallel with the cathode resistance, rk.

cathode resistance = (anode resistance + load resistance) / (amplification factor +1)

or

rk = (ra + Rl) / (mu + 1)

So rk = (800 + 2500) / (4.2 +1) = 3300 / 5.2 = 634.6 ohms

And therefore R = 1 / (1 / 634.6 + 1 / 750) = 344 ohms

Finally we can calculate the value of the cathode resistor's bypass capacitor

C = 1 / (2 x pi x f-3 x R) = 1 / (2 x 3.142 x 20 x 344) = 23.1uF. So we'd choose 22uF as it's a commonly available size. If you had a 47uf capacitor to hand then f-3 would be 10Hz. And 10uF would give a f-3 of 46Hz. I had a couple of 100uF capacitors handy so used those.

As you can see, if we had simply used the value of the cathode resistor, Rk, instead of R then we would have made the cathode bypass capacitor more than twice the size it actually needs to be.

As there's 45V at the cathode the capacitor needs to be rated higher than this. 50V is perhaps a little too close for my liking and I would choose at least a 63V rated capacitor and possibly 100V.

Grid Leak Resistor

We're nearly there now, just the grid leak resistor to specify. It's purpose is to tie the grid to ground and provide the cathode bias via the cathode resistor. The datasheet usually specifies a maximum value, in our case 500k ohms. Higher isn't necessarily better, but it can't be too low otherwise it would draw significant current and we want to keep the current draw under a milliamp. 100k is a reasonable value for the 2A3, and 1W should be more than enough as it should see very little current.

Final Output Stage

And here it is, our finished output stage. If you're wondering why there are two resistors between the cathode resistor and the cathode, this is in lieu of a humpot. A humpot allows any hum caused by AC heating of the valve to be minimised. But instead of a hum pot two resistors will put the cathode resistor at the centre of the potential difference between the two ends of the filament. In practice 2A3s are virtually hum free with decent construction of the amp and I've never felt the need for a hum pot.

In a later blog I'll look at design of the driver stage.

Alpair 12P MLTLs - Part 3

Day 3 - 4 hours

Today was all about learning how to use the router guide and some large circle cutters. It was the first use for the router too, a £12 special from either Aldi or Lidl, I don't remember which. It's not worth thinking about how they managed to manufacture and ship it half way across the world for 12 quid, even when reduced to half price.

So, nearly 8 degrees C today.

First job was to finish off the remaining three braces including a light sand. I might just rout an arris around the internal edges later.

Rather than bodge some rather rough holes for the various bits and pieces that are to be mounted in the cabinets I wanted to make a proper job so used the template guide that came with the router. This of course means I need some hole templates, so I bought an adjustable hole cutter from Toolstation for 8 quid.

Decent ones look to be four times the price so I was a little wary about its quality. As I haven't used any others I can't compare, but it's a bit fiddly to set up (so might others) and though it reckons to cut up to a depth of 30mm I struggled to do much more than 3mm in mdf. The cutting tips are pointed, and once most of the point has entered the mdf the mdf started to burn.

Maybe more coarse materials would cut better but I simply flipped the mdf over and cut from the other side. The battery drill struggled a little to drive the cutters but it's a lot easier to hold. When I cut the larger holes for the 12Ps I think I'll probably need a mains drill.

Now to check to see if the template is the right size. I couldn't find much scrap wood, apart from a rather narrow piece, but it worked just fine. The template guide is shown on the underside of the router below.

The 6mm router cutter...

First pass of the router, about 6mm deep

After a couple of passes, 12mm ish

Three passes and done. There's a slight lip at the bottom around about a third of the circumference. I tried cleaning it up but the router didn't touch it so I must have managed to get a little bit of twist in the router. But it's not an issue because it will be covered. If I was going to use it anyway.

Pretty much perfect size, about 1mm larger than the speaker binding tray.

I cut two more templates, one for the ports and one for the "supertweeter" L pad attenuators. The port template produces a snug fit, maybe just too snug, another half millimeter clearance would be better I think. On the other hand the L pad attenuator was completely wrong and I had to cut another. Just shows how easy it is to get it wrong, and how important it is to check before assuming it's correct and ruining a precious piece of birch ply!

Alpair 12P MLTLs - Day 2

Cabinet build - Part 2 (2 hours)

Just a short day today, only a couple of hours spare in the morning to play.

The next job is to cut the internal braces. The braces add a little stiffening, but perhaps more importantly they should help assembly by keeping the sides square and of equal width. One per cabinet would probably be enough but I had plenty of spare mdf so cut two per cabinet.

Then a quick trial fit with the front and back and a single side. The braces were a pretty good fit, only one needing just a little easing.

Meanwhile on temperature watch we have a heady 7 degrees C today :-).

Once cut, the braces need the middles removing. Four holes cut with a 19mm wood drill bit in the corners first, and then the rest removed with a jigsaw.

And that's all there was time for.

Alpair 12P MLTLs Part 1

I've had a pair of Alpair 12Ps for well over a year now which have been "running in" in the Quasars. But the intention has always been to build a pair of MLTLs in an attempt at domestic compromise. It's a design by Scott Lindgren which he very kindly allowed me to use. Thanks also to Colin Topps for his invaluable advice, particularly on routing techniques.

The 12Ps are a nice driver with a decent if not stellar 92dB sensitivity and a well balanced sound. For a wide band driver they have a well judged frequency compromise and reports of this MLTL cabinet suggest there is good bass. But the 12Ps do roll off about 12k Hz which I find audible, missing that little bit of sparkle, so I'm going to use a little bit of support above this frequency.

Cabinet build - Day 1 (4 hours)

Starting the build on the last day in January is perhaps not the most sensible time as it was a little chilly, a heady 3.5 degrees C. Fortunately I was out of the wind at least. (The thermometer reads 5 degrees C too high. My better half never really liked it, which is why it's relegated to the garage, but whenever I look at it I think of beautiful Seville, from whence it came.)

So here's the starting pile of wood, a single sheet of 2440mm x 1220mm x 18mm solid birch ply.

I took the easy road and got the timber yard to convert the sheet for me to my cutting plan. The upside is it saved me a whole lot of effort for a tenner, the downside is I didn't get to control the cutting. 4 no. fronts and backs at the rear, 4 no. sides in front, 4 no. tops and bottoms to the left, and some bits left over to the right which I might use to thicken the base up.

Examining the fronts and backs I wasn't particularly happy as there were the odd marks on the faces, a couple of well filled but unsightly knots, and quite a lot of the grain had been ripped out by the saw cutting across the grain of the exterior plies. Still, no going back now, just pick the best compromise and get on with it!

The front baffles need thickening up with a second piece - this improves the rigidity of the baffle but is also required because the depth of the rebate to suit the 12Ps would leave precious little of the thickness left to attach the driver to. In an ideal world I would probably have used more birch ply but I had some 18mm thick mdf offcuts and wasn't about to spend another £55 on a second sheet of birch ply. It probably makes little difference to the sound anyway.

I drilled and countersunk the secondary baffle then brushed on some plenty of pva, perhaps too much...

And then clamped the first edge together and fixed some screws to hold the two pieces together, then turned them around to clamp and screw the opposite side. Perhaps just a little too much pva. But better than not enough I guess. I took quite a bit of care to ensure the screws didn't damage the fronts, but even so I was a little nervous until I was able to examine them later to put my mind at rest.

Both baffles completed and left to dry after quite a bit of repeated mopping up of squeezed out glue. Really the glue needs to dry overnight, and preferably 24 hours. And be a bit warmer.

It was about 6pm when I finished by which time it had reached a balmy 5 degrees C. Hopefully it'll be a bit warmer the next time!

Coleman Regulators Part 2 - GM70

Rather a long time ago now I posted about Rod Coleman's filament regulators which I use for heating GM70s (and other DHTs). They're an excellent solution but they do require a decent raw DC supply which is worth writing about. I had intended to do this some time ago but audio has very much been on a back burner whilst other priorities (like work and the garden) have occupied me. Above is my implementation of Rod's suggested choke input supply.

At the time of writing Rod has a "preliminary" website for his filament modules, though it's due an update soon apparently. I guess I ought to say that I have no connection to Rod and have never even met or spoken with him. I'm very happy with his modules though, and his support and advice have been super. Indeed much of my scribblings here are a distilling of Rod's advice over the months and years.

So where to start? Fortunately Rod supplies suggestions for raw DC supplies for his modules, both his usual capacitor input power supply and a choke input power supply. They both do a similar job, but each has its pros and cons.

Here's Rod's basic circuit for the capacitor input supply, with thanks to Rod for permitting me to use it.

Note the trafo only needs to be rated for 21V but a whopping 7.1A because it's capacitor input. Here's a handy page on Sowter's website on rectification. (And here's another on Hammond's website.) Rod recommends we should be shooting for a nominal 25V raw DC to feed the modules (24.3V minimum, 28V maximum).

Looking at the second circuit on the Sowter page, in simplistic terms for 25V on the secondary we need a trafo rated at 25 x 0.71 = 17.75V. Great! That's less volts than we need. Well, sadly, the laws of the universe (and physics) dictate that we don't get something for nothing. We need a maximum DC current rating of 3.3A, so the current rating of the trafo needs to be at least 3.3 x 1.61 = 5.31A. Ouch. So in ball park terms the GM70 will consume 17.75 x 5.31 = 94W (or 94VA). A general recommendation is to derate by at least a factor of 2, so the trafo for each GM70 should be in the region of 188VA. Nobody said big SET amps were cheap!

Here's Rod's basic circuit for the choke input supply.

The trafo now needs a higher voltage rating, but lower current rating. Using the fourth circuit on the Sowter page, voltage = 25 x 1.11 = 27.75V, current = 3.3 x 1.06 = 3.50A, power = 2 x 27.75 x 3.50 = 194VA.

So perhaps unsurprisingly the trafo does the same amount of work whether capacitor or choke input is used. So which to use? Well, it's personal choice, but there are some other factors which are worth considering.

Following the trafo are the diode rectifiers. As the trafos have different secondary voltage ratings depending on whether capacitor or choke input the diodes will be at different potentials. Here are the outputs from Rod's PSUD files with voltage across the diodes (yellow) and final raw DC supply to the filament reg modules (red), first for capacitor input

and second for choke input

Unsurprisingly the choke input supply will need diodes with a higher voltage rating. Fortunately diodes aren't hugely expensive in comparison to the other items. Schottky types are preferred as they don't have the switching noise that other types have meaning there's less noise on the supply, which is a good thing.

Next in the circuit is a capacitor or resistor, depending on the supply. Let's concentrate on the capacitor input supply first. Here's Rod's power supply, this time showing current.

There's a LOT of ripple in the first capacitor, between -4A to 12A, so 16A. So the capacitor needs a big current ripple rating, and might even need to be shared by two or three parallelled capacitors.

Rod pointed these 15000uF 40V 9.5A ripple BHC monsters out to me - they look a good option to simplify the number of components in the power supply if the ripple is high. And they might even be a visual match for any Cerafines, Mundorfs, JJs or WKZs you might be using for your HT power supply.

Here's the choke input version

Okay, the curves showing current through the diodes and first cap aren't as pretty but look how much smaller the peak current is: 6A rather than 16A. That's a huge difference when it comes to specifying capacitors.

The smaller current in the first capacitor of a choke input supply is because the choke has had a beneficial effect on smoothing the ripple (it actually stores charge). Technically, the conduction-angle is longer as the choke stores energy during the lower amplitude portions of the mains sinusoid waveform.

Rod has shown a resistor before the first capacitor, and the choke before the second capacitor in the second filter stage and I presume this is because the choke would need to have a rather large current rating. Chokes with (relatively speaking) large inductance, high current rating, and low DC resistance are very large, very expensive, and very difficult to find. By using a resistor first a sensibly sized choke can be used, in this case a Hammond 159ZJ. The choke also has the benefit of filtering some of the spikes that can be present on a supply.

When I first saw the 0.001 ohm resistor I was a little surprised as I thought it was so small that the supply might act as a capacitor input anyway, but it didn't. The only 0.001R resistors available at my usual suppliers were crazy money so I parallelled two 0.01R to give 0.005R. Sounds like quite a difference, it is a factor of 5 after all. The 0.005R resistor drops 0.0272V more than the 0.001R resistor. There are bigger issues to concentrate on.

Having written all the above, there is another way to provide a raw DC supply and some of my friends have used SMPS power supplies to good effect. If you can find the right voltage and current rating, and they're reliable, they can be a cost effective solution. Indeed some of my friends just use SMPS supplies as a DC supply, and to be fair sound better that I thought they should. But SMPS are known to have a lot of (radiated) noise which can affect the sound unless they are an appropriate distance from the signal part of the amp (>2m).

There is a further issue Rod warned me of. Rather than risk diluting his thoughts by paraphrasing these are Rod's words:

"The "off-line" (i.e. mains input) SMPS have loads of capacitance bridging primary-to-secondary, in order to meet statutory EMI regulations. This leads to leakage current from primary (mains) to secondary. Regardless of any regulator, the leakage current, which has LF and broadband noise components, will pass to safety earth through the filament, through the cathode resistor (if present) and into the B+ supply circuit and onward to earth.

This leakage current varies is size wildly according to the quality of the supply. Only in Medical-grade SMPS is it really small (1uA level). The current should be compared with the Anode Current not the filament current, since it will mix directly with the returning current in the cathode. So a noisy 100uA can wreak havoc if you consider that the music signal might only be a few mA typically. Also, the HF noise-current will return to earth through a low impedance path, and this risks re-radiation into signal wiring etc.

Next, the SMPS is usually built to a cost, and the stress on parts is high. Filament supplies run 100% load, at all times, so de-rating is needed. But even then, I expect the lifetime is not great, and the degradation is the electrolytics may worsen noise before they expire."

So you can try them if you feel lucky but for my money Rod's filament modules give our DHTs the best chance. And mine sure sound good - they're fit and forget.

So, taking all the above into account I went with a choke input supply. Using the PSUD circuit below, these were my component choices.

The trafo uses a Hammond 185F28 which is 14V or 28V depending on whether the secondaries are connected in parallel or series. I got mine from everyone's favourite Hammond supplier Philip Ramsey at Bluebell Audio.

Next are the diodes and I used 100V 10A rated Schottky. Don't be tempted to parallel diodes with smaller current ratings as they will switch on and off at fractionally different times which could lead to premature failure. The heat dissipation of the diodes is marginal without any heatsinking so I mounted them on a small piece of aluminium angle, but the amp chassis would probably be more than adequate. Don't forget to isolate each diode if they're conductive. A benefit of choke input, with its lower RMS current, is lower stress on the rectifiers so they should run cooler too.

Then comes the 0.001R resistor, or in my case two 0.01R resistors parallelled. Looking at PSUD again, the resistor passes nearly 7A so each resistor could pass 3.5A. Therefore the power in each is 0.01 x 3.5 = 0.035W. Even derating by between three to five times is not a lot, around 1/8W, so I used whatever I could find that was cheap! Which turned out to be a massive 3W!

And now the first capacitor, C1. I didn't want to split an order across more than one supplier so the best compromise I could find was to parallel two 4.7uF 160V MKP 1839 polypropylene capacitors.

Next the choke, a Hammond 159ZJ, again from Bluebell Audio. A reassuringly heavy unit, though it is only rated at 10mH.

Finally the last capacitor. I parallelled two Panasonic TS-UP 22000uF 35V electrolytics.

And finally, finally. I added a snubbing network across the secondary of the trafo, a 100nF capacitor and 47R resistor in series. It doesn't matter which way round they're connected. The optimum values need to be evaluated on test with a scope and test equipment, but Rod reckons 100nF and 47R are good enough in most applications so that will do for me.

And it works well. The voltage range of the reg is only just above 20V so in an ideal world I would prefer a few more volts out of the raw DC supply, so maybe a 30V secondary rather than 28V would be better. But then you'd probably need to get a custom trafo wound.

And this is what GM70s look like when heated. On the left a graphite plate, and on the right a copper plate. The photo doesn't really do the beauty of the copper full justice...

And here's a close up of the copper plate GM70