Scratch Build Tube Amplifier

[Kei]

Scratch build valve amplifier

...or Valve amp if you're British. Those from OCUK may recognise this. It will be the abridged version rather than the entire saga. That said, even OCUK got a more brief version compared to the poor souls on DIYaudio who were a godsend helping me debug and get this beast working properly.

Something a little different from the usual pc builds. I've been trying to get both the time and effort to scratch build my own power amplifier for nearly a decade. Originally, I had planned on building a clone of a Quad 606, but owning and tinkering with 12 different solid state amplifiers, it just didn't really "spark" my enthusiasm enough so it got put on the back burner in 2012. Late last year a friend in work asked to borrow a scope probe as he was repairing his vintage valve amplifiers. (Leak TL12's) Which brought up the idea I'd never considered, why not scratch build a valve amp instead?

The plus points from my perspective were:

• They look and sound great according to the hifi world (although I'll admit I've never heard one, which inspires my curiosity)
• Point to point wiring with no printed circuit boards means very easy to modify, if a little more challenging to build
• Fewer parts, so less to go wrong, should make fault finder much easier
• Valves are more resilient to problems in the circuit and are less likely to catastrophically fail unlike transistors and mosfets
• Valves are still in production so no issues buying parts (unlike certain older mosfets/transistors)

Disadvantages:

• Transformers and chokes used can cost a small fortune
• Valves aren't dirt cheap like transistors
  
• Valves wear out with use
• Valves can be susceptible to external noise and can be microphonic
• The high voltages used require considerable care to avoid an early departure from this world

Having spent some time looking around for something to work with, I came across a book by the British valve manufacturer Mullard called, Circuits for audio amplifiers. In said book were a few different designs one of which piqued my interest, the 5-20, so named because it uses 5 valves and would happily produce 20W. It might not sound like much to some, but 20W is more than sufficient to shake the windows, rattle the furniture and severely annoy my neighbours. The topology was used in many successful commercial designs that are very well regarded. (Dynaco ST70, Harmon Kardon Citation V, Eico HF60 & Beam Echo Avantic DL7-35 to name a few)

Schematic for said build.


I decided to approach the idea by designing and building a prototype to test the concept before committing to an exact design and paying for materials that may not work out. The first prototype was built from scraps of MDF as I had a load stored. By this point I had bought all the components including the rather excessively expensive transformers. Since the output transformers can make or break the sound of the whole amplifier, I opted for Sowter as they are renown for transformers that have vast bandwidth covering the audible range and beyond. The only off the shelf choke I could find that fits the design was from Variable Voltage Technology on the Isle of Wight. The HT transformer I chose a Primary Windings part as it was by spec the same as VVT but slightly cheaper with a decent reputation. So 2x Sowter UA-21 output transformers, 2x VVT VTL12158-1440 Chokes and 2x Primary Windings HT transformers were bought. (at considerable expense)

I tweaked schematic with altered fusing to improve safety and solid state diodes added to help protect the valve rectifier. (all provided by the helpful folk at diyaudio)


The standard Mullard tag board design.

My tweaked version to make for shorter and tidier cable routes between the board and the other components as I intended to build a different layout that is similar to their 5-10 amplifier.

The board placed in with the other components in their approximate positions (where possible)


I then drew up a rough cad design to show the approximate finished design layout. The frame will be made from wood and the main base plate will be aluminium ~3mm thick. I'm thinking of using Shaeffer AG for machining a pair of panels for the final design.

I went for new old stock input valves in the form of GE 6267 for the EF86 (american equivalent) and a GE JAN 5751 in place of the ECC83/12AX7 dual triode phase splitter as it has a little bit less gain. Not sure who made the 6267 for GE as they are west German, but the 5751's are USA made. Hopefully the older valves will perform well.

For the output valves I went with new russian made Tung-Sol EL34B and Sovtek 5AR4 for the rectification.

Roughly laid out prior to the bulk of the components arriving



When the mountain of parts arrived, I quickly discovered the main filter capacitors were stud mount which wasn't mentioned in the datasheet for this particular size. I was very glad I was only building a prototype at that point.

Because of the capacitor mess up, I redesigned the layout.

One finished turret board. (minus a capacitor I forgot to order) The astute may notice that I mounted everything upside down so the turret board connections were backwards. I had to pull everything off and flip it over.

Basically built and just needed wiring up.

What with the compact design, it was a bit of a wiring nightmare. (the disadvantage of point to point designs)

Fitted 4mm banana sockets and a neutrik RCA

The wiring nightmare complete

First test showed 39Hz square wave at 47V p-p with no input signal. I'd built a 35W oscillator. Turns out, I'd wired the anodes of the phase splitter to the wrong output valves so everything was 180 degrees out of phase making the negative feedback positive, hence the oscillation.


Disconnecting the feedback loop is a quick test to prove it all works as it should. I got a nice clean result. The gain was extreme. 0.05V input, 8V output. (that's 8W into 8 ohms)

I swapped the anode/UL connections between the two output valves to fix the phase issue. Re-tested it and no more oscillator. Voltages as measured at that point.


1KHz square wave input showed some ringing.


10KHz square wave input also showed more ringing.


The ringing was the sign that there were problems with stability. The design utilised a huge quantity of "global" negative feedback which includes the output transformer. Inevitable phase shifts resulting from the transformer reduce the stability margin. Too much phase shift combined with a lot of negative feedback results in oscillation. At the time, I wasn't as well versed in the whole affair as I am now.

The other problem I encountered was excessive dissipation, evident in a mild red glow to the anodes. (called red plating) This massively shortens valve lifespan so also needed fixing.

Since I was seeing around 20V more than I should have after the rectifier, my first place to look as was fitting some dropping resistors in the 410Vac feed to the rectifier. I spent some time going over the datasheet for GZ34/5AR4 and realised that there is a minimum resistance per plate that I'd not known about as it simply wasn't mentioned in the book.

The valve wizard site also happened to have a handy formula for calculating it. (Rlim = Rsec + Rpri × (Vsec/Vpri)^2 + any extra resistance)
V Pri = 250
V Sec = 398
R Pri = 5.7
R Sec = 88.5
Result = 102.9464448

This put me a tad short of the 125 ohm minimum at 2x400V. (although they only list it for 60uF filter) I ended up fitting a pair of 5W 60R vintage RS branded resistors which I found in my stash of old components.

This reduced the issue but not quite eliminated, just a hint of a glow to the anodes. (dark grey metal structure inside the tube)

During further testing, I'd come up against a lot of oscillation problems. Most triggered by overload conditions or where the output transformer would start to approach saturation. You would expect to see distortion rather than severe ringing.

I thought a likely candidate causing the issue was my alteration to the layout and turret board, so set about building a Mk2 version which followed the standard design as closely as possible.

 

[Kei]

Cue the Mk 2 Version...When all else fails, do it by the book. This is pretty much as close to the book layout as I could get. I had to fiddle about with the two capacitors that live between the noval valves as it was supposed to be a multi-element electrolytic which I didn't have.


The connections to the output transformer from the output valves in the standard design are significantly shorter and the B+ runs in the opposite direction too which is a big part of why I thought the design was playing a part in my stability issues. (it wasn't but I didn't know that at the time)


I decided to cut the turret board down to a more sensible size this time which would shorten component legs, further reducing the likelihood for stray capacitance to have any significant effect.

On first testing I observed some familiar sights. I had to increase the dropper resistors from the 60R I had installed as it clearly was not enough here.

Still ringing at higher power levels



This led to me looking at the Claus Byrith modifications (a MK2.5 if you will) which converts the front end from a pentode input to triode strapped input. The rest of the modifications are far more difficult to achieve as it would require a negative supply in order to move to a fixed bias arrangement.

The alterations to the turret board and wiring were remarkably simple

Schematic for this design changes.

This was the first big step in the right direction as it tested significantly better albeit with some issues were still lingering. Much closer to stable, a combination of reduced gain and less negative feedback.

The only failing in the "Mk2" prototype was that there was significant mains hum that hadn't been there in the Mk1 and the HF performance regarding distortion was poor.


I was already planning on building the MK3 at this point anyway as I wanted to build an aluminium top plate to try and reduce the noise/interference pickup. I took the opportunity to try and make some improvements to the original design based on the suggestions made on DIYaudio. (at this point it was late May, I'd started designing in January, building in early March)

 

[Kei]

Made a start on the metal prototype build by re-purposing a 6U rack blanking cover to make the top panel. After plenty of drilling and punching the larger holes out I was left with this. Layout remains standard albeit a lot larger as I lack the tools to cut aluminium accurately in straight lines

Since the standard design has some known drawbacks with the choice of valve for the phase splitter and I needed to make a few tweaks to the feedback anyway, I took the opportunity to integrate some bigger changes. With the help of some kind fellows on DIYaudio I got the following schematic allowing the change from 12AX7/ECC83 to 6SN7/6CG7 for the phase splitter. A CCS (constant current source) was been integrated into the "tail" in order to force AC balance on the anodes and improve linearity. Zeners were introduced into the cathode bias on the output valves in order to reduce the standing bias but also reduce the losses across the larger resistors that would normally be used to reduce the bias. The advantage would be lower idle power dissipation without sacrificing power output. The disadvantage would be a lower tolerance to mismatched valves

New turret board layout. The CCS part was to be made om veroboard that would tie into the turret board.

The living room table was been turned into organised chaos.

One thing I did discover during my teardown was, what I'd thought was a 220nF capacitor off the link between the phase splitter grids turned out to be a 22nF cap. (which has been that way right from the start as somehow I ordered the wrong value)

I decided that due to the chaos I'd been having with feedback that I needed to know more about my specific transformers and their frequency & phase characteristics. I ended up modelling the transformer using just the signal generator through a 4.7K load across the anode connections linking the HT centre taps together and an 8R load on the secondary. Connected one channel of the scope to the primary and the other channel to the secondary and measured the time difference between the primary and secondary.

The constant current sink built on veroboard and fitted to the main turret board.

 

[Kei]

Decided to have a rejig this time and move the capacitors about and also have a partition between the power supply and the amplifier.


This time I also planned on making a permanent infrasonic filter on the input which would reduce the input signal level below 15Hz to minimise the problems it could cause. I had tried it on the Mk2 but it seemed to introduce noise due to the length of unshielded cable and components between the socket and the grid of the first valve. This time I decided to use some more veroboard and shielded cable.

The design

Built. I decided to mount it using some M3 threaded inserts that I had from something I'd done in work. I had no <1W resistors on hand in the right values so ended up having to put them in stood up at one end to fit in the space.

Mounted in the amp. I made a foil backed piece of card to act as a shield which I grounded to the input socket plate. I also rotated the input valve by 180 degrees to make the control grid pin closest to the input to shorten the wiring.

The Mk3 prototype finished.


I ended up having to rotate the CCS transistor in order to fit the heat sink. Thought it best to have one as it would need to dissipate ~400mW


Decked out with valves, it definitely does look a lot better in aluminium. It would probably look even better if the finish on the aluminium was clean and even. (It will be in the final build)

 

[Tolemac]

Loving this mate it's very well put together

 

[Avet]

That is some proper work. I am amazed. Well done.

 

[Kei]

Thanks. It's proven to be a much bigger challenge than I originally expected.

First testing wan't entirely to plan as I quickly discovered I'd got the pinout of the transistor in the CCS backwards. (ECB rather than BCE if that makes sense) The operating point of the EF86 is slightly out which can be tweaked by adjusting the resistors.

It's performance doesn't look too bad considering that is without feedback. (which corrects distortion, improves frequency response and reduces output impedance) It's certainly better than the 12AX7 phase splitter of the original design. It still gets quite out of shape around 11Vrms output, particularly on the negative cycle which is down to the V1/V2 operating point. There's also the obvious signs of crossover distortion too.

Even a 10K square wave doesn't look bad. (tube amps rarely look great here due to the output transformer) The frequency response is remarkably flat at lower power outputs. Towards that 11Vac output, the response starts to fall slowly from 14KHz.

I made some adjustments to the cathode on EF86 and ran some tests. I swapped out the 1.2K resistor for a 1K. I also swapped out the JJ GZ34/KT77's for EL34B/5AR4. The JJ GZ34 clearly has more sag than the sovtek 5AR4 as the voltages out of it were ~10V higher. The behaviour of both V1 & V2 seemed to be far better with the voltages being much closer to expected. I also noticed the asymmetric distortion on the output was now gone, only showing slight signs of crossover distortion as the output passes 13Vrms. (21W into 8 ohms)

This is how things were with just the cathode of EF86 changed.


Changed the R6 screen grid resistor to 470K as I wanted to actually see the effect on the voltages across both V1 & V2. I was surprised at how big a difference it actually made everywhere else, yet the screen voltage change was pretty small.

V1 Anode increased by 26V, grid 2 decreased by 8V and the cathode decreased by 0.15V. V2 anodes remained the same but the cathode increased by 11V.

Anyway, I put the 330K back in as that seemed to produce voltages in line with expectations considering the higher supply voltages. I can't remember where I read it but it said about the screen being a 1:4 ratio with the anode, so 82K*4=328K.

Here is a short video of it under a music test run with speaker attached. (not hifi sound here as the D810 mic isn't great)


I actually left it run for nearly 6 hours as it sounded great. Not a hint of red plating (anode hot spots) this time, all looked healthy in the glow department.

 

[Kei]

Closing the feedback loop didn't entirely go to plan as the usual feedback issues returned. However, things were not as bad as they had been with the previous builds.

The usual 58ish KHz ringing observed as signal size rises. (shutter speed caught the screen mid refresh hence the dim readout text)

Zoomed in on the oscillation

Made some tweaks to some component values which made a decent improvement but didn't fix it.
Comparison at 100Hz 500mV input. Left is C11/C12 @ 100nF & R4 @ 10K, right is C11/C12 @ 470nF & R4 @ 22K. C2 was 270pF for both.

I looked at reducing the amount of feedback as it had been suggested that I may have a greater level of feedback than originally planned. This meant increasing R27 from 8.2K. After a quick search through the components on hand, I put in 10K and rechecked. A small improvement but not cured yet.

It made sense at this point to look at how much gain there was in the system at each level in order to figure out how much feedback was currently being applied.

Open loop gain = 49.54dB
Closed loop gain with R27 @ 8.2K = 27.31dB (22.23dB feedback)
Closed loop gain with R27 @ 10K = 28.75dB (20.79dB feedback)

That explained a lot. The design was supposed to have 17dB of feedback. Even with a 10K resistor in there, I've still got a fair bit more than I need. Increasing R27 to 18K should bring things in line with the numbers I'm after.

I dug out an 18K resistor and did another test run. It now appeared to be stable. With R27 @ 18K I'm seeing 33.26dB gain so about 16.3dB of feedback.

This is how things look now.

100Hz @ circa 11.5Vrms out (matches the previous testing) 11.5V translates to 16.5W into 8 ohms

100Hz @ the limit before it distorts heavily. 15.64Vrms converts to 30.6W into 8 ohms which is not bad going for a 20W amp. There are some very slight signs of crossover distortion where the amp is moving into class B operation and the zero crossover points between the positive half and negative half of the push-pull roll over slightly leading to a nasty S shaped distortion in the waveform.

20Hz @ the limit. 27.3W and no sign of transformer core saturation says the transformers have decent LF performance. The original mullard design could only manage 20W @ 30Hz.

100Hz square wave testing shows the LF bandwidth. The slope of the of the flat section shows that the LF bandwidth is limited. (artefact of the transformer and my infrasonic input filtration) The steeper the slope, the worse the bandwidth.


1KHz square is pretty much perfect with a tiny ripple which is the merest sniff of instability


10KHz square. The rounding over at the rise and fall appear to be in the input signal if taken after the input veroboard suggesting stray capacitance. The curving over shows limited HF bandwidth as it requires bandwidth out to 100KHz to achieve a perfect square. Again, this is normal for a valve amp due to the transformers, high impedance and miller capacitance effects on triodes.


Schematic as things stand. Also note the voltage at C2 and L1 are quite a bit higher than the model. The high anode/screen voltages may explain the additional power before clipping starts.

Measured frequency response for each version of the amp. The current config is shown in yellow with feedback and green without feedback. The Mk2 is shown in blue with feedback and red without feedback. Whilst some may point out that the blue plot has a wider frequency response, it was also highly unstable and this graph doesn't show distortion. The Mk2 showed quite pronounced second harmonic distortion from 10KHz and up once the power output rose beyond 10-12W. (the slight kink in the upper part of the sine wave where it rolls over)


Gave it a major trial by fire by running it all day on the other set of valves. It passed with flying colours and sounded great.


These are not well suited to hot summer usage. Looking at that, I will avoid coming into contact with the glass bulbs.

 

[Kei]

Fitted the Electro-Harmonix 6CA7's today and am giving it a test. I think it may be my imagination but it definitely seems to sound a bit better. I think there may be more gain or something, but either way, they sound really good.

They also look good too as they have a slightly broader bottle and a bigger plate structure.




Power valves side by side. Even though all 3 are rated virtually identically, I'd be surprised if the 6CA7 couldn't dissipate more power simply due to the size and fining of the plates.

 

[Kei]

I've been using it regularly since the start of August and it's been running great. It sounds superb. It's surprising just how loud 20-30W is capable of. Having resampled all of the valves, at best, the differences between them are subtle. I'm currently sticking with the JJ KT77's as they are a matched pair rather than a quartet and they are also the cheapest so I can wear these down whilst I work on building the second amp. Swapping valves around all the time isn't going to bode well for the sockets or the valves and I wouldn't want to wear half a quartet. I'll use the quartets once I have both amps finished. Currently I've got the left channel powered by my solid state amp and the right channel via the valve amp and the channel levels set to match. They sound remarkably similar, though I think the valve amp has a tubbier bass which I reckon is down to the low damping factor due to the higher output impedance.

I moved it from behind the door to where the turntable was which has helped with cooling. The turntable needed to move anyway as the top of the "flexi-rack" is not stable enough for the turntable. I had to tip toe around whenever it was on. That will get it's own wall shelf at some point.


Did another video with better audio this time, although it still isn't great. The mic preamp in my D810 is a bit naff as the sound I get out of the soundcraft mixer I use better than this. I also have an old Sony HVR-Z1 which also sounds great with the ME-66 mic I use but it's HDV and I don't have firewire anymore.


I'm still working on the "how to build an all metal chassis" issue. I'm torn between the simpler 3mm aluminium plate on a wooden frame like I've already build against an all metal box like a hammond 1444 folded aluminium chassis. An american company called Landfall Systems seems to make an extremely good looking item but shipping and taxes are likely to be painful.

 

[Kei]

After much testing, I came to the conclusion that although the amp sounded great, it had one issue; too much gain. The open loop gain was at 49.54dB and closed loop was 33.26dB. The problem this causes is that minute levels of noise can be amplified sufficiently to be come audible and you have less range on the pre amp volume control before it gets excessively loud.

The big issue I was seeing was noise caused by an HDMI ground loop in my marantz AV7701 pre amp. The noise is at a very low level where it only starts become apparent on 90dB efficient speakers with gain above about 25dB. (a not so pleasant sounding 50Hz hum) The best solution would have been to break the ground loop but unfortunately I only found one ground loop isolator for HDMI and it was nearly £800. (Basically an HDMI to LC fibre converter set) The easier solution was to find some way of simply dropping the gain level. The simplest but also most bodgtastic answer would have been to simply attenuate the input signal. This does work as my other solid state power amps all have gain control pots on them which actually don't adjust gain but simply attenuate the input signal. Turning these down makes the hum quieter and can make it disappear altogether.

Another simply approach to reduce the gain is to increase the level of negative feedback which subtracts from the open loop gain. More feedback results in lower closed loop gain. Unfortunately, I couldn't resort to this as I was pretty much at the limit with feedback as my stability margin was well tight.

The last option available was to remove the cathode bypass capacitor on the first valve which would drop the gain and also potentially allow for a little bit more feedback to be injected. The guy who helped to design and model all the tweaks to the original mullard 5-20 gave me a hand in making the changes to the design and then modelling it in LTspice.


It all looked good so I took the amp apart and made the necessary tweaks to the turret board layout.



On testing, it behaved similarly although there were some residual issues in the LF resulting in some slight oscillation. (anything lower than 13Hz above 0.5V in) The voltages across EF86 seemed a bit out of kilter which will always have a big effect on everything downstream. I reduced the dropping resistor for it's supply to raise the anode voltage which seemed to be just the ticket. No more oscillation at very low frequencies. All I get with very large infrasonic signals now is transformer core saturation which is without a doubt ugly and not pleasant sounding but it is not potentially amplifier destroying like oscillation can be.

10Hz clean @ 12.6W is incredibly good going for an output transformer.


10Hz @ technically 22W going by p-p values. The ugly look of core saturation.


10KHz square wave performance is pretty similar. The gain difference is obvious when you compare the input signal levels against the RMS value on the scope. I'm thinking of testing the input infrasonic filter to see if I can reduce the HF losses by reducing the 22K resistor to 10K.


These tweaks successfully dropped the gain from 49.54dB open loop and 33.26dB closed loop to 45.77dB open loop & 25.87dB closed loop. That also took feedback from 16.5dB to 19.9dB.

Under test, working very well. (still a prototype build)

 

[Kei]

I've started construction of a Mk IV amp which shrinks the overall package down. The Mk III underneath is the same size aluminium sheet, 19"x10.5" down to 15"x9.5". Even though the overall chassis size has been reduced, I have been able to increase the separation of the output valves in order to improve thermals.


So far I only had time to get the piece cut down to size and mark the holes with a punch. Tomorrow I will be drilling the holes out.

 

[Kei]

I've now deconstructed the first amp I built so that I can rebuild it as the Mk4. It's surprising just how quickly I took the whole thing to bits. 1.5 hours and it was done. I swear it took me over a week to build it.

 

[Kei]

More progress today. All holes now drilled and countersunk. I quickly found that I had undersized the holes for the cable grommets by 2mm so need to sort that tomorrow.


I've cut the turret boards down so that the cathode bias circuit is now placed under the output transformer. I need to buy some shorter nylon standoffs for it.


Test fitted the transformers to get an idea of the size and distances. The output transformer does sit a bit closer to the valves than I'd like but it's not extremely close so shouldn't pose a problem.

 

[Warchild]

Jeez that thing is a mass monster

 

[Kei]

Jeez that thing is a mass monster

It is a tad large and heavy. To get a stereo pair, it'll weigh around 25Kg.

Sorted the turret boards today.


I also decided that since the finish on the aluminium was a bit marred, I would sand it smooth. I'm now trying to decide whether to go all out and polish it up or try to get a smooth brushed finish. Currently part way through as it's taken more effort than I expected.

 

[SPS]

I don't understand much of it but looks like an awesome effort!

 

[Kei]

Much sanding and polishing later...

I'm also about 90% of the way through building the boards. A few capacitors, the zeners and the veroboards are left to do.

 

[Kei]

I finished up the turret boards yesterday after building the veroboards for the CCS and the input filter.


I found some more inner tube which I've cut to the shape of the transformer base which acts as a vibration damper. I did the same on the Mk III build and it seems to work reasonably well. This time I had sufficient to cut one piece to do the whole base where I did the 4 mounting points previously. All the bits and pieces are now fitted to the plate.


The polished plate definitely looks better than the anodized version. I've noticed a few small niggles with the layout that I've corrected for the final version. The rectifier valve is a bit close to the choke and the position of the phase splitter valve is closer to the input valve than it needs to be. Neither is a big issue though and shouldn't pose any problem. The only thing I need to make now is the chassis framework to hold the plate in place.

 

[Kei]

I've now made up some side panels. The sides are actually recycled from the MK II wooden build as this happened to be the same width. Sadly they come with some holes from their previous use.

I just need to create a front and rear panel to complete this version. (plus obviously finishing off the wiring.)