Friday, 30 September 2016

Substituting the output stage load

So far I've built the gain, splitter and driver stages on my test rig. While I am well pleased with the results, this has been with a higher B+ voltage than intended. As a consequence, I've had to use higher resistors than planned to get my HT+ voltages right for the valves.

Part of the reason for my high voltage is that I don't as yet have the output stages - each of the output valves will be biased to a quiescent current of around 50mA... representing a 200mA load on the B+ which I would expect to drop the voltage by some amount. 

Question is.... will the extent of that voltage drop require me to change the values of my dropping resistors?

Without the output stage present, my thoughts turned to constructing a dummy load, something that I could sink around 200mA at 550V DC into.

A few calculations later and I concluded that three 60W lightbulbs wired in Series would do the trick. So construction of the world's ugliest dummy load was commenced... this is the result:

We have tamed the B+ to 565V which is right where I want it.


In place and power applied. Some nice voltages flying around on my bench today

Just to make sure my calculations weren't completely wrong, I wired the bulbs in series with a 10ohm resistor. Which is dropping 2.25V across it, meaning that the lights are drawing 225mA

So about 25mA more than I'd planned... I can live with that. Valve electronics is not always an exact science.

So what did this do to my voltage? It took around 15-20V out of the supply. Perfect.

I changed my supply resistor to the 12AX7 from 270K to 220K and I will change the 12AU7 supply resistor to 27K from 30K as soon as I have a 27K 5W resistor... yep... back to RS for more parts!

Saturday, 24 September 2016

Fixing a serious error

I never said I was perfect at this stuff....

Reviewing the circuit plan and also others' circuit plans, I determined the design for my fixed bias circuit for the KT88 output valves had rather a dangerous flaw. The nature of these valves is that they require a constant negative voltage on the grid or else they'll go into full conduction, followed very shortly by the anodes getting red hot and then shortly thereafter by pyrotechnics.

The negative current is supplied by the bias supply, which in my design runs at around -80V and is adjusted by potentiometers until you're getting 0.5V across the cathode resistors (in other words, 50mA quiescent current, regulated by the bias voltage).

The potentiometers used to make this adjustment are a vital part of the circuit and if one of them should happen to suffer a failure – such as a wiper lifting – then in the current design, the bias voltage completely disappears. Followed closely thereafter by the output valve and potentially other parts as well.

Fortunately this is quite easy to correct. I've re-drawn the circuit to a fail-SAFE mode, rather than fail UNsafe as it was previously. If a pot wiper should lift now, it will send the bias voltage more negative, rather than cutting it off.

This is the relevant section of the circuit now:


The drawback with this design is that it requires a little more current. Now, Each side of the bias circuit will be passing between 2.6 and 4 mA - so total potential around 8mA ... the bias winding on my transformer is rated at 200mA so I think we're safe.

The values of the resistors may need a little tweaking by experimentation to get the exact bias currents needed through the valves. I'll find that out when it comes time to build the output stages.

But for now, I think this is a worthwhile improvement to the design.

Wednesday, 21 September 2016

More measurements

I'm reasonably happy with the initial gain and driver stage Proof of Concept build on the testbench, after having played around with resistor values to tweak bias voltages and measuring outputs. The small problem I had was that my oscilloscope was going into clipping on the input before the signal went into clipping, so I couldn't measure the gain accurately.

I ordered a X100 probe to remedy this problem and with its arrival today I was able to make some accurate measurements on the 'scope.

So I was interested in two things: Overall Gain, and Frequency Response

For testing the gain, I sent in a 200mV RMS signal and advanced my gain control all the way up (no attenuation) and measured the voltage at the outputs of the driver stage, loaded with 100K resistors to simulate the output valves.

At the outputs I am getting 148v RMS from a 200mv RMS input signal. This equates to around 742 X gain for the initial gain and driver stages, or put another way, or 57dB

The next measure is frequency response. This is typically measured up to 100kHz, but my (borrowed) function generator has a maximum frequency of 20kHz. Better than nothing.

To measure frequency response, I monitored input voltage (to ensure the signal generator is honest!) and voltage at the outputs of the 12AU7 into the 100K load.

A spreadsheet exercise resolves this into a graph, which wants to be as flat as possible:

Click for larger size


I am satisfied enough with this response. 

Up next: I'm going to re-design the adjustable bias circuit so that it's fail-safe, rather than fail-unsafe as currently designed. If one of those pot wipers should lift, the bias voltage will disappear from the KT88, followed shortly thereafter by pyrotechnics as the valve red-plates. Instead I'll use a design with fixed resistors and trim pots where if the pot fails it gets more negative bias, not less.

Also I've ordered my volume control and input remote switcher today.

I have also decided these cheap Chinese valves I've been using in my testing will not do for the final project, they are too variable. I'll get some better ones, along with my output valves. Fortunately I have a local source for those.

The Chassis is going to be the next challenge...

Monday, 19 September 2016

Listening tests

Having got the initial stages of the amplifier working to my satisfaction on the test bench, I felt rather impatient to actually put something through it other than test signals to look at on the oscilloscope. With the output from the 12AU7 gain stage being able to reach around 110v peak-to-peak before clipping sets in, I got to thinking if there was some way I could reduce this signal down somehow to a line-level amplitude, such that I could feed it into an Aux-in somewhere.

And so... with an appearance befitting the haste with which it was constructed, I introduce the leading contender for the title of world's ugliest potential divider:


That's a 1M resistor reducing the signal and a 33K load resistor that I'm taking the signal off.

I did say it was ugly. But it served the purpose perfectly and I was able to put a signal through it and listen through a portable speaker with an Aux-in.

What music did I choose for this auspicious occasion? That honour went to Bob Seger and the Silver Bullet band: "Still the same"

It sounded clean and clear and gave me great satisfaction to hear some signal coming through this unholy contraption.

The rest of the session was spent tuning up the bias to get the 12AU7 operating within its most linear range, as tested by increasing the drive and watching the output go into clipping on the oscilloscope. Once the top and bottom of the output sinewave start clipping at the same point, I'm happy.

In other news: All the way from Canada, my output transformers arrived today. They look a lot less ugly than my power transformer



Now "all" I need is:
  • Chassis
  • KT88 output valves
  • Some better preamp valves than the $10-a-pop Chinese ones I'm using for testing
  • Remote control motor-drive volume control and input switcher

My next step is going to be assembling a circuit board containing the delayed turn-on circuit for the HT. The intention is to wait until the valves are hot and the bias supply is stable before switching on the B+ power – this will protect the output valves (expensive) and the supply capacitors in the gain stages which are rated at 400V but at initial turn-on are shooting up to full B+ potential (currently 570V) before the valves warm up and start drawing current. 

This will be a small circuit board the size of a credit card containing a 5V regulator, 555 IC and a couple of capacitors and resistors to set the IC's parameters at a 20sec delay, plus a couple of SPST relays (one for each arm of the HT) as we'll be switching the AC going into the diodes.

Sunday, 18 September 2016

PoC Test Rig Continued

Continuing from the previous build, this phase of the test build is focussed on V2 - the driver valve. This is a 12AU7 twin-triode, and each half of it will be amplifying one phase from the previous inverter stage to the approximately 50v needed to drive the KT88 output valves.

This is a simple grounded-cathode topology, the same in fact as the single-ended initial gain stage. These are identical and to encourage the valves to balance, I have tied the cathodes and they share a common cathode resistor.

This is the schematic for today's build – the previous build is grayed out:

Click to see larger size

The yellow highlights are measured voltages .... well almost. The vagaries of valve electronics means that the current draw from each half of the triode is not exactly the same, which results in differing voltages at the anodes. In this case I had two 12AU7s to hand, and on the first, the measured voltages at the anodes were 180V and 189V, the other 190V and 180V

My original design called for a 10K pot to provide an adjustment in dividing the HT+ voltage between the two halves, to ensure identical amplitude output, this was omitted in this test build (no 10K pots handy!) but will be included in the final build.

This difference is clearly visible in the oscilloscope traces, one trace per anode, from test points A and B in the circuit schematic. See below:

Note the differing peak-to-peak voltages of the signals.

I've changed the colours of the traces on my oscilloscope to match the identifying colours on the probes. Makes for less confusion with a powered-up circuit on the bench.

These different voltages will need to be balanced and if I switch to the different valve, the difference is reversed. The outputs from the phase inverter are much closer together. 

This is how the test rig looks with the second valve wired in. It's not meant to be pretty it's just meant to work, please be gentle with your comments on my construction technique, the final build will be a thousand times better

All wired up with voltmeter and oscilloscope probes:

Astute readers will spot the series resistors needed to tame the B+ to usable voltages for the 12AU7

According to the oscilloscope I'm getting the signal amplitide I need (after a worrying half hour where the signal was clipping, only to find I'd left the probes on X1 not X10 – D'oh!)

Also according to the spectrum analyzer the THD in this section is clean, at around 0.08% at 50v drive. The readings are taken through 0.68µF capacitors loaded to earth thru 1M resistors.

Once again only informal measurements of freq response have been made from 20Hz to 20kHz, the amplitudes on the 'scope screen did not change. However more measurements will be made to be certain of the results.

Next steps - get a 10K Pot  to try to balance the outputs from the 12AU7.

Final thought for the day


This shit is real. I'm also really hoping that'll come down a bit when the OPTs are connected and the output valves biased up. I think I might see if I can contrive a 200mA load to simulate the 4 KT88s at quiescent, and see what this does to my voltage. Perhaps a couple of 100W lightbulbs connected in series might do the job.


Friday, 16 September 2016

Proof-of-concept test rig

So after a few trips to Jaycar, and after the components that I'd ordered from RS had arrived, I had what I needed to start building the test rig:D

Watch as I light up the initial gain stage, then read on for some measurements and discussion


The point of the test rig is just to get the initial gain, inverter and driver stages modelled and to check the bias parameters for the valves with resistor substitution, ahead of the main build. That way, I'll need to do as little component swapping as possible when doing the final build.

So this is the circuit I built today. This uses a 12AX7 (ECC83) as an initial gain and DC-coupled Cathodyne/Concertina Phase Inverter.


The highlighted voltages are actual measured values. 

The letters are signal test points for the oscilloscope traces (to follow).

Calculating the bias, we have a voltage drop of 195V across R3, which for 390K suggests a bias current of 0.5mA which is right where I wanted it to be.

On the inverter we have a voltage drop of 115V across R8 which at 100K suggests a bias current of 1.15mA, a little higher than I was aiming for but a potentially a consequence of DC coupling.

This is what the build looks like. Those of a sensitive disposition my wish to look away at this point, it is not pretty (hey, it is only a proof-of-concept/test build, it only has to work, not look good as well!)

Red and Black wires for power, yellow wires for heaters, white wires for signal

100K volume control to allow easy adjustment and duplicate the setup in the final build

And yeah, I know that 150K resistor feeding the supply capacitor is too low-power. Tell me something I don't know! But it works and hasn't gone up in smoke yet. Obviously the final build will have a resistor with a more appropriate power rating for durability.

Below is the power supply. Those of a sensitive disposition PLEASE look away now. This is rude but it works

Unloaded supply voltage is 580V DC... extreme caution needed! 

YES the chassis is very carefully earthed!

Yes I do work with one hand in my pocket when the circuit is energized!


The transformer is 400-0-400V centre-tapped. the diodes I'm using (1N5408) have a peak inverse repetitive voltage of 1000v which according to PSUD2 this design will exceed, admittedly only by a couple of volts, so I've doubled the diodes and paralleled them with 220nF ceramic capacitors.

Also note the electrolytics - these are 470 µF 450V capacitors, but the power supply's open-circuit voltage is running at around 580V so I've connected them in SERIES so that each cap has around 260V across it. It does mean that the body of the second capacitor is at around 260V potential to the chassis, so this design will need special considerations in the final build. Note also the 100K bleed resistors across the caps, after power is removed it takes around one minute for the voltage to come down to around 30 volts.

This design is not anything you can listen to; without output valves and an output transformer there is no chance of driving a speaker with this setup, but that's not the idea. The point of this is simply to inject test tones from a function generator and look at the outputs on the oscilloscope and spectrum analyzer.

So to the oscilloscope traces: Firstly, the gain stage, measurements at point A and B in the circuit. (A is the blue trace, B is the yellow trace)


The initial gain stage is inverting the phase as expected, and will give up to 10v peak-to-peak before harmonic distortion starts being visible on the spectrum analyzer.

THD at this amplification is low at 0.05% and the spectrum analyzer looks pleasingly clean.


The next step is to use the other half of the 12AX7 to perform the phase inversion necessary to drive a push-pull output stage. The topology here is to use the other triode as a cathodyne/concertina/split-load with equal resistors on the anode and the cathode. This ensures an equal amplitude output at both ends, and an overall gain of around 0.95.

The potential difficulties here are getting the bias level correct; the grid needs to be at a lower potential than the cathode, but DC coupling from the previous stage means that the grid is at the same potential as the anode of the preceding stage.

This can be resolved by using a high value resistor for the anode - it will be noticed this design relies on a 390K resistor.

Measurements taken while the circuit is running indicate the valve to be self-biasing, to an extent. The voltages on the diagram above are taken from actual measurements while the circuit was running. This seems to indicate the idea of a DC-coupled cathodyne with a 12AX7 is technically feasible, at least on paper. This is the same idea as the Williamson design, except that uses a 6SN7 in this topology.

My idea in this test rig was to see if it was even possible to use a 12AX7 in this topology. The oscilloscope measurements seem to indicate it will work.

Below is the output trace from the Anode (Point C in the circuit, Yellow trace) and Cathode (Point D in the circuit, Blue trace):


This looks the way it should - I tried the test with frequencies from 20Hz to 20kHz and the results were consistent. 

The spectrum Analyzer shows no significant presence of harmonics:


Based on these initial measurements, I am optimistic this will be a viable initial stage for my amplifier. I have yet to make measurements of frequency response, so far the level of attenuation at higher frequencies appears to be minimal, but I'd rather have some numbers rather than just "minimal" so these will be my next measurements.

The next stage of building will be to add the second valve, the 12AU7 driver, to boost each phase of the signal from 10v p-p to approximately 50v p-p.

More measurements to follow...

Sunday, 11 September 2016

Back on again

No pictures with this entry...

So an overseas work trip (I tend to have a lot of those in my line of work) put a temporary halt to progress, however it was not completely wasted time as I did manage to get some online shopping done... most of the internals of the amplifier (resistors, capacitors, rectifier diodes, valve sockets etc) were ordered online and are on their way.

The peer review of the circuit design was in equal parts useful and discouraging; a thin veneer of useful pointers and advice, wrapped around a solid core of "you don't know what you're doing, go away and leave us alone". Yeah, definitely outstayed my welcome on that one :(

I've also decided that I'm going to build up an experimental/test version of the gain, splitter and driver stages of the amp in a small test chassis with wide spacing between tagboards, to facilitate experimentation and component substitution. I want to DC-couple the anode of the initial gain stage to the grid of the cathodyne/concertina phase inverter, Williamson-style. Using a 12AX7 this should be possible, but the calculations are marginal.

So, the plan is to build up the input stages on a test chassis, using a function generator for input, and monitoring the voltages and waveforms at critical points in the circuit using the voltmeter and oscilloscope. Of course with no output stage this won't be anything that can be connected to a loudspeaker or headphone and listened to, but the voltmeter and oscilloscope will tell the story.

If successful, I shall post conspicuously on this blog the operating parameters necessary to use a 12AX7 in this configuration, with oscilloscope traces etc – in the hopes it may help someone else.

And if it is not possible to get an effective result, I'll switch to using the 6SN7, but I want to try with the 12AX7 / 12AU7 valves first.

When the components start arriving, I'll start building...