Nick's little project

[NickK]

I think the initial voltage regulators will simply be a "maida" implementation. You loose at least 20V in the drop but basically it's a cheap starting point AND being a series regulator I can current limit at the same time.

A ON FQPK3N60C MOSFET (80p) will do 600V 3A (or 1.8A at 100degC) peak. Looking at the specs this will do DC (ie continuous) 200V (400mA) and 400 (100mA) without breaking sweat. With an LM317 behind it, it will give good -80dB noise voltage regulation without getting too fancy (at a later date).

That solves the voltage regulators for the front and backends but still leaves the high current low-voltage for heaters. These only need to cope with 12.6V but will need to cope with with a couple of amps each. The LM317 has a 2A limit and ideally you'd 1/2 that for comfort and low noise. I just need to build a current limiter for the heaters with a soft start option - this stops something that euro tubes and new tubes seem to suffer from - hot spots that exist in the heater filaments of the new tubes at startup (ie no resistance so the element gets a lot of current and it's uneven in it's resistances until fully working). Carlson has an example of this over on his YouTube channel.

Turns out someone has created a 500V 500mA capable HV regulator - using the Maida design but with a LT3080 and a 900V 10A capable part. I'll probably end up using that (he's had enough experience with it - it's been in design/use since 2012!).

 

[NickK]

Just put two voltage regs into the simulator - this includes AC and a power transformer, rectification too.



Nice soft start by the B+ voltage regs.... and see what it does.
Blue = current through ONE of the ecc99s
Green = voltage on the cathode rail
Blue = the voltage to your lovely 32ohm headphones..

Yes that's a nice starting voltage of 30V to your headphones (oh an 1 AMP) transient and an initial current of 600% through each ecc99 (they have 60mA max!). Typically this would be where the mute relays would be switched to disconnect the headphones during startup.

This is what your headphones see..


So muting the headphones during startup would probably be a good thing!

I've yet to tune the transformer or voltage regulators for the right voltage but you can see the voltage drop:


Looking at the input (yes I'm aware it's 350V!) this is the output from the transformer (blue) and the output from the full wave rectification that goes into the regulator (blue). You can see the effect that the 10,000uF cap has lol, in reality I will not be able to use that sized cap as 10KuF at 400V would be the size of a laserprinter. Instead I'll use smaller caps, but I'll have to put them in series to reduce the voltage each sees. I can then parallel them if needed to increase current and capacity. Given this is miliWatts of power.. I suspect I won't need that much.



Those eagle-eyed will note the DC offset.. due to the weird transformer model setup I currently have.

 

[NickK]

Seems the LT3081 has a programmable current limiting on too. So I could combing the voltage regulator and current limit too! New to look at it a little more.
It can limit from 3mA upwards, so it would allow me it limit the current to something sensible.

 

[NickK]

So looking at the power... a simplified model showing each valve as a current source simulates a valve at maximum throughput by limiting current to 100mA through each output tube and 20mA through the smaller. It occurred to me just now that the current max for an ecc99 is actually 60mA.. but I'll run that in a bit:


The first run looked awful - no voltage regulator.. and initially only 100uF smoothing caps with a single secondary running. I then upped the game.. making a far larger capacitance pool and had two transformer secondaries feeding current - this meant that the current draw through each cap is about 30 Amps rather than a single 60A draw previously.. Naturally - those spikes are ideally generated hence they're short spikes rather than broader smaller cycles.

Hmm started at 20V ripple.. and with caps got this down to about 1-2V..


Then ... I got busy with putting a voltage regulator in.. and the ripple dropped (given this is full power running at 1A at 270V - or 270W):


Not bad for initial attempt. 0.2V ripple. However this voltage regulator can actually do uA (micro-amp) ripple.. so I need to tweak this and get the right LTSPice models.

So now with the 60mA update, the model looks like this:



This model represents the maximum the tubes are all rated for operating (including the non-connected triodes in the second ecc99 per channel - for both channels. Total current would be 640mA but the regulator would be happy up to 1.5A.


You can see the yellow voltage regulator. This is the initial run.. If I can work out why it's messed up from a voltage perspective, then I'm laughing:



What ripple


The LT3080 LDO has a max voltage of 30V in-out but the way that the real live version of this works is to float the LDO so that it sits in a small voltage dip under the maximum voltage you set. It then works to smooth that small dip out. Hence this will work with 500V.. as long as not more than 30V is across the chip itself.

From a cost prospective, I may look at a transformer with a dual outputs - use that singularly with a a single rectifier (the 450V 4500uF electrolytic cap is £40!). Later enhancements I can update - add regulators, rectifier and cap and add a second secondary to separate the audio channels into, effectively, mono blocks

 

[NickK]

I did an overnight simulation, ripple wise:


Then zooming in on the stable area:


So 263.95-263.25 = 0.7V ripple.

I R Disappoint. So there's obviously something messed up in the model for the diodes etc in use (I don't have the precise diode models etc) given the circuit has been physically built and measured at 20 microvolt ripple at 200mA 500V load (0.7V = 700mV = 700000uV vs 20uV). Also that's not all the story, looking at the FFT.. Dang that's noisy:


I can see a 100Hz and then images (ie replications) that go up from there - nyquist.. this looks like the switching noise (probably the noise from the 0.7V ripple itself).

As this is a single ended amp (the topology doesn't cancel/reject power supply noise) and it's a headphone amp (the output is 1V), the power supply needs to be whisper quiet.

The reason for using a regulator, other than noise, is that it allows me to reduce the smoothing capacitor size and reduce the number of secondaries in use. This voltage regulator can cope with a 20V peak-to-peak ripple and output 20uV ripple. The cost of the cheapest 4500uF 400V cap is in the region of £40-50 each. So reducing the cap size required means the regulator cost is offset by the reduction in cost..

The guy that designed this regulator back in 2012 still sells the SMD-populated board for $89CDN. It's been tried and tested by lots of audio valve guys. Add caps, fet etc to that cost. I trust his design so it's definitely my model. He's also modelled it in LTspice so..

Looking at the input from smoothing.. this looks like the model isn't even running (I get 0.7V ripple out of the smoothing caps).

Dropping down to one secondary, and only 470uF of smoothing caps gives me a output smoothing ripple of about 12V peak-to-peak. Current draw is about 0.6A -1.7A spikes. Mouser has 470uF 400V caps at £5 each (KEMET).

 

[NickK]

This is the problem - blue from rectification, green output from the voltage rectifier:

 

[NickK]

Lastly.... You sure you are going to be happy with all that hard work and then fat electrolytic cap as the last filter....thought about parafeed or even just singlend (sister?)

I’d feel safer with a a cap. The 330uF is needed to support 32ohm low frequency, the bypass 10uF shows a good frequency response.

The Thermistor is my first step and that will do for the short term but I’d like a soft start. The issue of the LT3080 is down to the junction mosfet and bad model but I’m open for any regulator initially as long as it can cope with the current demands of OTL low impedance - this has been the problem with other regulators.

There’s an interesting differential parafeed 1/2 way down the page: https://www.tubecad.com/2014/09/blog0308.htm

The first step is getting the amp up and running. I nearly have the BOM - OTL calls A means I’ll be looking at - quite a large transformer:l:
8*60mA*150+ 8*20mA*200 = 72W + 32W = 104VA just for the signal.
4*400mA*12.6 + 4*300mA*12.6 = 36VA heater

 

[NickK]

Jesus... That's some requirement for a headamp!

My parafeed pulls 80ma @ 350v. Heaters... 1.2a if using the Russians, or d3a Germans only eat 700ma.

I may run the heaters off a 15V SMPS given some filtering/12.6V regulator it could easily take some load off.

I can reduce the power requirements as the model but I want some “tinker room” plus a few more VA is needed for headroom sees something approaching 180VA. Which is bananas

 

[NickK]

I have re read the whole thread. Love the design, it is wonderfully over the top.

The bit I can't get past is the choice of the Otl final stage. All that glass, all the design, then a fat lytic cap right at the output....sure I can't tempt you to go transformer coupled?

The minimum would be 5 valves (2 front end and three AI output tubes) should cover stereo. The pain is the wiring is the reason to add a 4th output but only using 1/2 the tube). I have plans.. towards differential and the. push-pull but best to get this built first and I’d like to get a DSD512 integrated too.

Also with the ecc99 that will drive a 2a3-based power stage if I want to add a pair of speakers

The reason for going transformerless (OTL) is to experience without a transformer, the downside is the power inefficiency - OTL is typically quoted as 1% or less, then with class A that's 40% of 1%.. hence the stereo output stage is, per channel, idling at 75mA with ±75mA swing. The headphones are 104dBV so that's probably plenty loud enough, if I used the remaining triode/channel that increases to 100mA with 100mA swing (ie 100mA*1V = 100mW = 0.1W).

The power requirements are basically the same between running a 110mA capable 6AS7 and two 60mA capable ecc99s per channel. The ecc99 is more linear and gives me more options for tinkering without the large heater current required by the 6SN7 & 6AS7s. However the main reason for multiple triodes (two per class tube) is the parallel structure reduces output impedance. Add negative feedback and you get a good damping factor.

I may be tempted by that differential parafeed but getting a good match transformer is an art/design all to itself - more than just windings, it also resistances and capacitances. I can then replace the transformer and reuse the transformer for a discrete power stage. The power transformer I had in mind would be a£32-50 230V:230V 1.7A 180VA. However good output trannies (two required) means £60-120 per transformer.

I'm also intent in exploring different topologies. This basic amp provides a basis for exploring where each stage is separate. Just replacing a stage is easy by desoldering a few wires at a point board. That way I can "top-roll" drop back in a set of valve bases with associated caps & resistors or even just hook them up with a set of alligator clips.

 

[NickK]

Another good point is that transformers rarely if ever go wrong. Valves on the other hand wear out over time and they can go wrong, sometimes spectacularly. (shorts and thermal runaway) I've not had any so far that have gone wrong like that but I've had a few dodgy valves with operating points that were well outside of spec. OTL does have the benefit of not having to deal with the limited frequency response and phase shifts introduced by the transformer.

Safety is at the front of my mind.

Transformers - act as an isolation transformer as long as the windings don't short un tube/amp failure modes.

OTLs need careful design in this and then some testing too..
* startup is the same with most transformers - the surge is there and handled by mute relays normally, or tube rectifiers
* cap shorts. With a cap short - single ended and any other design typically have a circuit from B+ to ground through the headphones.
* tube shorts. Result in high voltage swings (AC) caused by high current, so regardless of output cap integrity, will output to the headphones. If the cap fails then you have a high current and therefore high voltage DC path from ground through the headphones.
* CCS failure to short. large current, AC rise causes headphone voltage spike and unlimited current to all output valves, the bias will rise and negative bias the grid but will cause tubes to fail.
* shutdown as the shutdown occurs, the grids need to be kept negative to the cathode or you will have a big current & voltage spike..
* cross section shorts (resistors/caps/wires) all should cause a short and thus blow something rather than output higher voltage.
* headphone socket short - simply discharge to ground from the output cap
* headphone disconnect/replace the caps should handle this, and there is an inbuilt resistor to ground.

An upper current limiter on the B+ as well as the CCS on the cathode, voltage crowbar on the output headphone side so anything over 2-3V would cause a blown fuse. Also B+ circuit collapse as the B+ to the valves will have relays so if the power goes, the relays drop the circuit connections - restart is via soft start again. As they say - the electronics are there to save the fuse... so the B+ fuse and mains fuses are slower than designing and having some safety features.

I'm being ultra careful, simply because a shock to the head is unlikely to have any form of response from the shocked individual.

There's an awful lot of OTL style headphone amps that don't have any protective circuits and simply rely on the mains fuse..

 

[NickK]

This is what I will want todo for the negative bias - this shows the input valve

Green B+ (no voltage regulator and only 470uF)
Red - cathode bias
Blue - headphone output
Cyan - current through the first

You can see the start up - the B+ (green) is off, and the system clamps the cathode bias to make a -30V grid bias (red). Then the B+ kicks in and then only when the grid bias is released do you see the current flow through the valve. Afterwheich you can see the cathode sine wave and the current sine wave.

You can see that the headphone has a small -2V spike but the output mute relay (not modelled) would stop that.

 

[NickK]

Looking at the output stage:
Blue - headphone voltage
Green - headphone current through 32ohms
Red - is the cathode feedback into the initial stage
Pink - cathode line at output stage (before output cap)
Cyan - B+
Grey - Current across one valve. See about 0.1 seconds.. 16A spike then 2-4A for a period before dropping off.. that's not healthy!

Note that only one ecc99 is in the output stage just to make things faster.

This still needs more work around the signal and getting the correct grid for the e99. It also shows *why* a voltage regulator is needed or large smoothing caps!

 

[NickK]

I did note after posting that the second has a problem - the timing on the output is wrong (easily solved) but if you look at the grey trace it (tube current) - it spikes to 2A. This would occur if not all tubes conducted at the same time (which they won’t).
So a minor correction on the timing is needed. One nice thing is putting a 1ohm resistor in series allows the current to be tested using a the voltage across it. It can then be shown on a voltage meter for example.
A voltage could then be used to set the lower CCS current limit.. something for a later date perhaps.

Today’s goal, once the chores are out of the way, is to finalise the signal path and design ready for the first ordering.

 

[NickK]

Inrush current limiter set up on the mains AC side of the transformer.



A simple 10ohm resistor with a time delay bypass switch (blue is the trigger to bypass at 0.2 seconds).

You can see instead of a large 50+ amp inrush current (green) the inrush is smoothed over a short period but limited to about 5-8 Amps. The poor 10ohm resistor gets a 600W impulse on the first AC cycle. I can smooth further but this simulates a thermistor in a very basic way.

 

[NickK]

So.. from the same run.. let's look at current. Although the final signal currents and voltages are not configured this looks promising as the same concept is valid:

Internal currents across the valves.


You can see the switching lines for the different startup stages - there is a big 3.16V peak-to-peak sine wave on the input from the start.
* not shown is the 0.2 mains trigger to bypass shown above.
* 0.5s (cyan) = B+ start
* 0.7s (pink) = clamp switches off, the blue pre-section tube current and green output stage current raise to idle current (operating class A bias always has current flowing through the device - this is what makes the class A 'hot'). You can also see a slight movement in the blue as the negative feedback returning to the input section starts to cause it react.
* 1.0s (red) = headphone mute switches off what which point you can see the green output currents rise as it pulls & pushes current through the headphone. Also the input section current (blue) causes changes to occur as the output section starts driving the headphones.

Note that there's no 1A spikes..

Now let's see what the headphone current overlaid on to that (grey):

No big current spikes

And let's have a look at the voltages.

Not sure this is quite as rosy..
* switch signals are the small 5V signals at the bottom.
* 0.5s B+ start, the green B+ rectified voltage goes to 270V which is correct. The blue which is the valve anode voltage initially starts perfect but then takes a nose dive followed by spiking between the cathode and intended anode voltage.
* 0.7s the bias clamp switches off and as the currents start flowing the matt green (output cathodes) and grey lines (input cathodes) start up.
* 1.0s the headphone mute disables and the amp appears initially stable for about 100ms.. before the B+ issues starts to cause impact.

So what does the headphone see?
In gold overlaid is the voltage that the headphones now see.


Phew.. so although there's a problem - the headphones simply see no voltage spikes output normal working voltages.

Now.. what causes that problem - I suspect is the current limiter not being right which I've added at the top of the output valves to prevent a unhindered flow from the headphones to the B+ line in case of shorts or tube failure.

Looking at the output CCS, the current limiter and the output capacitor you can piece together what is happening:


* 0.5 B+ start - the B+ rail in red is steady.
* 0.7 the cathode clamp stops and the tubes are free,
1. you can see the lower CCS (pink) takes a hit and then provides 75mA sustained idle current.
2. you can see the blue anode voltage of the output tube then takes a complete dive.
Voltage = current * resistance.
So this looks like the current is now flowing through a different path and so incurs a different resistance. Given the cathode voltage (cyan) increases this is likely that it's pilling through

The current through the valve at this stage begins to climb, as it conducts. As the valve is less resistance we have our change in resistance.

The valve (actually this would be all three but I only have on in there) is now conducting all the 75mA idle current that is being limited by the lower CCS. This is normal (the normal idle current per valve would actually be 25mA).
Unlike an anode follower.. we're not following the voltage to to make the sound for the headphones.. we're using the cathode and it's current flow however we need a voltage across the tube for it to remain stable.

So what is causing it? Well V=IR the relationship sort of points to current.. or the controlling of the current. The lower CCS is fine, it's working and doing it's job. The upper CCS is limiting the current flow. However when the resistance falls then there is no extra current to compensate and the voltage drops. Although the output cap has energy to provide current, there's no outlet.. or resistance to cause it.

* 1.0s headphone mute disable
Now the system finds the resistance on the other side of the output cap changes from 10,000 ohms to 32.. although DC current doesn't flow, AC currents do.. thus you see a increase in the currents flowing through which causes an oscillation in the blue valve anode response.. that finally dies out at about 1.4 seconds.
The current through the valve anode (green) shows this..

I think part of this is down to not having the final signals voltages (ie grid voltages setup) thus larger currents are allowed but.. this also points to the fact, at the moment, the use of an upper CCS rather than perhaps a more traditional current limiter is causing an issue.

V=IR.

If less current flows through the valve we see less anode voltage.
If the valve conducts more (or paths present a lower resistance) then resistance drops. Then we see less anode voltage.

An upper CCS that is being triggered into limiting (and assumes the model is not magically creating current) would cause (a) higher resistance and (b) less current.. the tube sat behind it.. would then see a limit to the current, it's own resistance would drop and the voltage would drop.

So it's good that it's seen it has a positive safety effect.. but I need to design it so it only starts infringing when it needs to limit. The current input signal is a 3.16V peak signal (I said peak-to-peak earlier but that's incorrect) which is +20% dbV for consumer line-in. This is deliberate to ensure the amp has headroom to cope with big transients such as cymbals or electronica without compressing the signal. A normal signal on a consumer system should be 0.447V peak.

 

[NickK]

Going to the big full model
* both channels operating sharing a common set of B+ rails, a single power supply transformer, voltage regulation on both B+ rails.
* full compliment of valves (ie two triodes and three output triodes per channel).
* 0.2s soft start second before it's bypassed
* 1s output mute disabled
* No cathode biasing and no upper current limiter.



Starting from the bottom
* green = input triode current (this is the one that takes the signal)
* blue = current through input triode common cathode that provides the amplified output and the grid is where the negative feedback is presented.
* red = current through one of the output triodes (multiply by 3 to get total output current flowing)
* cyan = input cathode voltage
* grey = output cathode voltage (ie between output tubes and the output cap)
* pink B+ voltage

Not bad. The model is slow.. only 4 seconds of operation but it's enough to get a good understanding.

The common spike/dip at 1.0 second is the mute switching the headphones on.

What is clear is that the headphone mute needs longer before it's disengaged, the voltage regulator is still starting up (it has a soft start itself). I suspect the amp will need 10-30 seconds to be fully stable before disengaging the headphone mute override.

Next - sort the voltages for the B+, signal and sort out the damn voltage regulator model..

Just to show the headphone output - looking promising but that 6.3V 200mA spike on de-mute needs to go!

 

[NickK]

So.. On the signals.

It turns out that when muted, the output section behaved itself but soon as mute was lifted the current path - headphone > output amp > B+ caused it to go ape, as I suspected.

I've also been playing with the top CCS - it's very very sensitive. Too much and resistance and it switches off. However it appears I've found a good middle ground (9.6 ohm, yes the tenth of ohms is all important):



So you can see the mute disable. Then the CCS keeps the system in check from drawing too much current through the headphones.

An added benefit is that the CCS means I can drop the B+ directly from the 200V line rather than having a separate 150V line. So reducing complexity!

Still got a little more way to go on the signal configuration but I'm happy I have found a solution. Only concern is it's sensitivity vs resistor 5-10% tolerances that change over temperature too!

Just checked the wave form.. poop!



That pulsing is caused by the CCS. Back to the drawing board.

 

[NickK]

First rule.. strip back when you find a problem.
* removed CCS and regulators, switches and anything else. Replacing the CCS with good old resistors has solved the problem but has allowed the system to wildly oscillate - probably the CCS was the only thing keeping it in check.
* cut the negative feedback, thus allowing a bias to be setup. The system has stopped oscillating and allows me to properly set the system levels up.
* then once it's working nicely.. I'll start adding bits back in.

 

[NickK]

The original design had some pretty tight voltage swing 'tolerances'. So the output of the 12AU7 was a mere 2V peak but at few volts of negative bias. The 12BH7 is a stronger beast so it upsets the applecart.

The final teaching of a valve amp is like solving simultaneous equations - the instability was caused by the biases and the voltage swing causing the the output valves to cut and start, cut and start, the negative feedback then caused the front to oscillate in a cut and start in sympathy. It was bad that even setting a zero source input (ie no signal) caused it to oscillate. Positive swings on the grid didn't help.

However once you have a rough understanding of the voltage swing from the 12BH7s, the current allowing a much more forceful swing thus you see a good 20Vp-p. The ecc99 then needs to be rebiased, so that it's grids never go positive and the valves always remain in their output safety zones.

At the headphone:


Not bad output, just a quick run FFT but it's showing promise.


This is at full volume and pelting about 220mW. With 3.16V peak input 10KHz sine wave simulating a full headroom +20dbV transient. Note - this is using a voltage source not the rectified signal (hence it's quiet).

So in normal playing the output would be far quieter.

This is looking good for a bill of materials

Next is sorting out the voltage regulator - I may take you up on your salis regulator, but just need to look into it. From that it really defines the power requirements from the transformer.

The nice thing is the system is now running on 200V throughout without CCS or regulator. So I hope that they just add the icing on the cake.

 

[NickK]

447mV results are good.



The only issue is that perceived volume is a logarithmic scale, so this is probably way too quiet Valves are no logarithmic output, unless the volume plays with the feedback and caused it to increase the gain (amplification) as well as the volume (size of signal). This is current poking out about 4.5mW at the headphones.

The nice thing is the system has some leeway to adjust the bias to increase the output. So I'm not done just yet.