Nick's little project

[NickK]

A bit slow in the ordering - I thought it would be good to create a BOM project in mouser then give it a day to recheck. The work got in the way, 15h day yesterday etc.

I've been looking at a number of areas:

* use of bi-polar caps (50V) for 330uF output caps (the cathode follower shouldn't get above 20V.. shouldn't!) but these are a matter of pence. vs a 450V Polypropylene 330uF cap being £45 each. If I blow up one it's not going to matter during the initial testing. Then I can order the PP caps.
Edit: more info regarding capacitors: https://www.analog.com/en/analog-dialogue/articles/capacitance-and-capacitors.html
So in reality the higher the frequency it's likely to result in the bi-polar electrolytics causing some roll of but perhaps not in audio frequencies.

* power inrush - most vacuum tubes are a matter of low capacitance ~70uF 450V etc due to vaccum tube rectifiers only do mA. OTL.. well that's 450V 6000uF to get a nice 1V peak-to-peak ripple (£45) and the 12V heater line gets a res cap of 10,000uF 63V in there to kill any noise.
So that's a butt load of capacitance. Plan on 450J of energy in a second is 440W. Inrush is a good 50A for the first cycle in modelling. So I'm trying different ideas to reduce the current flow. A set of switched 50W resistors in series that are switched out sequentially is the lowest tech approach. A better way would be to use a switched PWM supply to limit inrush - however that requires a lot more tech. All the other ways result in massive generation of heat (100W+) and voltage drop during normal operation.

* negative grid bias - this is easy todo by grounding the grid to 0V then supplying +24V through a 1Meg resistor to the cathode. The result is a -24V grid so the valve powering up doesn't attempt to randomly switch on at full current - not good for your headphones with a massive current/voltage spike.

* mute - muting with switch causes the output cap to hold a DC charge.. which when you plug in or come off mute goes POP with a big chunk of DC through the headphones.. so to combat this the mute button has a high wattage 32ohm resistor that provides a drain during mute. Coming off mute then doesn't have a massive cap build up.

 

[NickK]

I made a minor oops.

Both the bh7 and ecc99 have a max of 3.5w per triode (ie two per tube), which is fine for the bh7 at 200V but the ecc99 due to the higher current quickly ends up going beyond the 3.5W.
At 150V the ecc99 running at 23-25mA is basically 3.5W but as soon as the anode voltage is 200V that drops to 17mA to keep within the 3.5W.
So it means I need to drop the 200V to 150V. Issue here is I need the current at 150V and less current at 200V which is a little backwards and makes a simple dropper resistor not particularly efficient.

Another option is to run two rectification stages from the same transformer secondaries. That way it can be build to supply current at 150 and then 200V with less current.

 

[NickK]

So with no signal, this is a small run from 5-55 seconds using just passive components for one channel:


So this is telling me that I should have a reasonable noise floor as a basis in the design with a 6000uF 450V cap on the mains rectification. This is setup with a resistor before to give a 1Hz low pass filter.

Then there are two 150uF and one 200uF 450V caps 'smoothing' ie filtering for each channel to make up a bi-directional low pass filters.

Total capacitance for the amp will be 7000uF, or about 350 joules. To put that in context 1BTU is 1000 joules. So that's going to take at least 20 seconds for the system to have charged up. Normally valve amps are low current, using high voltage instead so caps etc are smaller. As this low impedance, output transformerless, class A it has both high voltage and larger current use - hence needing voltage and current = larger power = more capacitance for passive filters etc. It also means a massive inrush if it's not controlled, so I have a crude setup of power resistors (capable of 50W each) to limit and smooth the 50A+ inrush over a longer time to stop the transformer from melting or destroying the power transformer arching over internally. The net effect is a longer start time.

The secondary thing that large capacitance brings is more current is needed to keep the caps charged, so I may be loosing some mA of power through the caps, causing the caps to heat up a little but it means you need that extra current in the power supply to start with.

The important point about this modelling test is the noise floor that the power supply adds - in this case it's looking low (although modelling tends to be return more ideal results than real life, so getting it as low as possible is good.

Happy with that so far.

 

[NickK]

Bill of materials - as basic as I could make the amp.

Recorded as number channels x number required per channel x thing.

Signal path
2x 1x 1M ohm 1/4WW
2x 6x 300 ohm 1W
2x 1x 10M ohm 1W
2x 1x 0.1uF 450V PP (Aur OX)
2x 1x 10uF 450V PP
2x 1x 330uF 450V PP / 50V bi-polar electrolytic)
2x 1x cap clamp (if required)
2x 3x B9A valve sockets

Input Stage Power
2x 1x 700 ohm
2x 1x 8K ohm 10W
2x 2x 1K 5W
2x 2x 150uF 450V Electrolytic (Nichicon)

Output Stage Power
2x 1x 250K 1W
2x 1x 100 ohm 1W
2x 1x 5 ohm 1W
2x 1x 15K ohm 1W
2x 1x 55K ohm 1W
2x 1x 1K ohm 10W (1W if slow charge)
2x 1x 27K ohm 1W
2x 1x 80 ohm 5W
2x 4x 10 ohm 1-2W
2x 2x 2.1K ohm 20W
2x 2x 2K ohm 5W
2x 1x 10K ohm 1/2-1W
2x 1x 47uF 450V Electrolytic (Nichicon)
2x 1x 200uF 450V Electrolytic (Nichicon)

Output Jack
1x 1x Headphone socket
2x 1x 32ohm small speaker

Output Mute
2x 1x 23 ohm 1W

Output Startup Bias Control
2x 1x 1M ohm 1/4-1W
2x 1x 750K ohm 1/4-1W
1x 1x 10 ohm 1W
2x 1x 150uF 450V Electrolytic (Nichicon)
1x 4x 450V 10A switch (double throw - one throw for each channel)

Heater Power
1x 1x 10,000uF 63V Electrolytic (Nichicon)
1x 2x 10 ohm 50W
1x 1x 2K ohm 50W
1x 2x 450V 10A switch (bypass and heater on/off)
1x 1x 600V 10A Bridge Diode
1x 1x 240V - 12V 30VA transformer - 240V connects into primary mains-side power.

Secondary Main Power
1x 2x 10 ohm 50W
1x 2x 100 ohm 50W
1x 2x 500 ohm 50W
1x 2x 1K ohm 50W
1x 2x 2K 0hm 50W
1x 1x 6000uF 450V Electrolytic (Kemet)
1x 1x capacitor clamp
1x 1x 600V 10A Bridge Diode
1x 5x 450V 10A switch (bypass and main B+ on/off)
1x 1x 240V - 220V 1.3A 225VA Transformer
2x 1x fuse holder 2A.

Primary Main Power
1x 1x 450V 10A switch (bypass)
IEC socket + EMI filter + power switch + fuse
1x 1x Antherm NTC
1x 2x Chassis safety earthing lug

Development components
1x 4x 1K variable pot 5W 500V
1x 1x 10K variable pot 5W 500V

Wiring
Shielded signal wire
20AWG solid wire 600V power wiring
2x 2x solder tag board.
Janzen solder

Final bit of discussion with the folks at diyaudio and there was a couple of points about the low frequency roll off due to the 330uF cap, a larger cap means a better, lower rolloff. 2200uF replaces the 330uF output cap. Also dropping the output cap terminator resistor to 1K so that the DC doesn't build up.

Just modelling to see.

 

[NickK]

This is why I'm taking it slow and steady with this project. This is after changing the 330uF to 2200uF for better bass response.. it means additional capacity and thus inrush..



This is the startup over the first 20 seconds. Green is the secondary 240V mains current. Then each of the resistors that are acting as inrush limiters.

The inrush limiting startup is manual, and made of (in this model) the following in parallel: 1K (50W) -> 1K (50W) -> 500R (50W) -> 150R (50W) -> 10R (50W). If you take 1.3Arms you're ok for around 2.4A peak, with a little more for single cycle or two.

Each step is doing well until the last ~ 8 second switch at which point we're seeing 5.5A which will certainly make the transformer hum and possibly cause it to fail over time. Also at 270W peak the 50W resistors aren't going to like that repeatedly on power ups. Remember these figures are for a single resistor - there are two.. in parallel. so that's 560W peak. So we need to adjust the resistors and lengthen the timing to reduce the spikes....

 

[NickK]

I decided to go back to the maths for the inrush control.

A 200 ohm resistor with 7000uF limiting to about 1.6A per cycle initially is reasonable. A capacitor pull max current that then slopes down current as it charges. So RC time constant gets about 36% charged but to get to 99% takes 5x. So calculating 5RC is the task.

If I calculate the max current I want to pull of 1.6A, then calculate a resistance that gives me that - 200ohms. This, on paper gives me just over 10 seconds for 5RC.

The issue here is that 200 ohm resistor is running at 500W/cycle.. so 1/200 = 10/2K means I can use ten 2K 50W rated resistors in parallel and they were' in business. This sounds nuts, chassis mounted resistors 50W are about £3-4 each, but then having additional switches also end up at £3-4 each.

The current slopes off sopes off so the 50W is only a few milliseconds.



Top is the secondary winding
Middle is one of the ten 50W resistors
Bottom is the voltage for the capacitor and B+ rail.

A smooth soft start. No surprises so far - just waiting for the model to hit 11 seconds when the resistors are bypassed. See if there's a spike or not (there is likely to be spike) but the cap will be virtually completely charged to 220V.. so there's likely to be a few amp spike as it runs up a bit higher.

Yep, 90V jump and that meant a few cycles of 14A inrush, so I think a 56ohm step may be good here - there should be less current so less power needed.

 

[NickK]

Almost there..

 

[NickK]

Better. Just doing the final model for this. However this is showing - blue as the capacitor inrush at the secondary, the green showing the B+ voltage rise.

There's a final switching I'm modelling where the 16 ohm resistor is switched out after 20 seconds. I'm hoping this only has a small jump for about 10V so the current and power shouldn't cause too much of a spike.

 

[NickK]

Yay!


That's good - I think the transformer wouldn't die with that.

 

[NickK]

Hmm looking at this closer there’s quite a spiky current meaning although it would work, the power supply would have a low power factor. It would produce harmonics and take more power.. commercially anything over 80W typically has a power factor correction.

People done make 300V output switched power supplies but they can be made to work - a switched psu rectifies then shapes the power factor so the current is shaped like a sine wave close to the voltage sine wave. The output high voltage ~400V is then rectified and then it’s stepped down by rechopping the current.

So in theory it should be possible step down by a minor amount plus the transformer would be smaller.

 

[NickK]

The startup idle current on the output tubes - spot which has negative start up biasing with zero signal:



Would take the top trace every day!

 

[Kei]

This makes me appreciate the simplicity of an indirectly heated valve rectifier. Built in 15 second soft start. The only real disadvantages are the very limited filter capacitance (56uF for 5AR4), voltage drop & sag and the power consumption of the filaments. (5AR4=1.9A @ 5V= 9.5W)

Fingers crossed for your project, reality behaves like the simulations.

 

[NickK]

This makes me appreciate the simplicity of an indirectly heated valve rectifier. Built in 15 second soft start. The only real disadvantages are the very limited filter capacitance (56uF for 5AR4), voltage drop & sag and the power consumption of the filaments. (5AR4=1.9A @ 5V= 9.5W)

Fingers crossed for your project, reality behaves like the simulations.

That’s what I’d noticed with valve power supplies - they work well when used as intended (voltage rather than current). The same with the inrush surge - only a 20J for a 200VA transformer. As soon as you need current reserves that’s additional 10x the surge caused by the caps.

I’m not expecting real life to be that close, the modelling forces some analysis be thought before things go bang.

The day job is 24/7 at the moment so this has taken a back seat during the week.

EDIT:

So there's an option to replace the 10 50W 2K resistors with one 500W 2.2K resistor. So that way the size required for resistors is more manageable. The 500W resistor is, wait for it, ... 30cm x 8cm x 1cm and needs a heatsink for full duty use as it would get to 250degC...

Now the capacitors are 350J capacity. So in 1 second DC the inrush would be 350W. However given this is AC and we only get power in 100Hz (50Hz rectified) cycles in DC, that initial spike is higher hence 500W. Annoyingly they don't do 2K but 2.2K. Not a big amount but needs to check but this, for the £58 (figure that against 10x£4-5), is a nice manageable option. Max voltage on this is 1500V so it's not going to arc.
It can also be used to discharge the capacitors if need be.

I will, today, re-check my capacitor sizes. It may be the at I can get away with a smaller capacitor given I have two sets of low pass 1st order caps now in the system.

If that's all good, then I'll put together the initial power bill of materials.. and order. Then I can check the power is all good outside with a plastic bucket over the top to stop explosions (I'll put a current limiter in).

 

[NickK]

Ok, mrs said a 'quiet' weekend (code for you've been working too hard).

Output from a simulation given the current model including start up - this is 10KHz and 3.16V big power wave. I've also increased the coupling cap between the stages to 0.2uF instead of 0.1uF. The 12BH7s have way more current than the original designed 12AT7s.. so this will help with driving the ecc99s. Lastly the output caps are now not 330uF but 2200uF on advice.. so:



Looking at the input B+ it's quite quiet but that mains noise:



The output from the smoothing cap is a little noisy still:



I've also spotted a 'oops' - the 2.2K 500W only needs to be 200 ohm (as 10x in parallel needs 2.2K to be 200 ohm). So I'm re-checking and I'll model a few changes. No difference other than the resistor is about £32 not 58 I'm also seeing if I can reduce the capacitor size a little.. 6mF is big and if need be I'll get it but I'm thinking there's better options.

Looking forward getting a final commitment in the order box..

 

[NickK]

So I've been struggling with the power:
1. Power, simple maths, is relatively simple.
2. Power + passive low pass filters drop the voltage considerably or if you want small resistance then you must have massive caps (all hail the inrush god!).
3. Power factor is a biatch - without it your little need becomes a pointy spike and you end up frying torridal.
4. OTL + Class A + SE = power monster.

I've been playing with LTspice and the 300VA (300W) power torroidal but that is not enough (point 3).

Note: A point here is that I'm looking at a level of power that would blast my current headphones.. and this is at +20dB transient power sustained - essentially 120dB+. However the emphasis here is not volume but will my power supply blow if it gets a transient (short answer - no given nothing gets to this level).

Now a 300VA or even a 225VA is probably as happy as Larry if it's using active power factor correction. Switch mode power supply often have this built in by legal regulation requirements (certainly anything over 75W).

So I could easily do 300VA IF I didn't have any power filtering.. endured the lovely 100Hz hum in the background.

So what other options?

Well you can get more current by paralleling the secondaries - but then you have only 117V. So if I want 150V on the output and 150-200V on the input stage, then this could work - we only need a max of 80mA on the input (the higher voltage) and 480mA (ok 500mA) on low voltage 150V..
We could then build a parallel full bridge rectification that would give me 117*1.414 = 168V but is less current efficient, but the we could add into that bridge rectifier a voltage doubler that gives me 330V (from 117V!) the down side is it doubles current if you want the same..



It almost works to solve my problem. I get an acceptable 180V 80mA and 150V 500mA (top) but.. the bottom still shows the problem. In green is the current for one of the secondaries - a peak of 1.64A per secondary or 3.4A required. Now the 300VA supports a 1.36A per secondary or 2.72A max. short.

Now.. the 500VA can push 234V 2.14A or 117V at 4.28A. So crazy? a 500W headphone amp.

I'm now currently trying a model with a lower average mA on the output tubes, as if it was playing at normal volume.

Running at 'normal' volume and given the odd transient wouldn't harm the power supply, the peaks per secondary at 1.16A. Which is good enough to fit with the 300VA transformer. However I need to run my slow model first.. and see.

EDIT: Tempted to make the move to PP OTL right now - change the configuration and design so front and back are PP and differential. No need to worry about PSU noise as it cancels out.. Unlike an SE which I'm finding to have an SE, needs lots of voltage head room to drop over regulation/ripple reduction with OTL and class A that makes it a power nightmare.

So an alternative is to switch back.. to something a little more normal - a PP, run + and - rails. I think with a PP and 300VA you'd be laughing.. let's see what I can design..

EDIT2: There is one possible way - tune the amp's use of the power supply noise to cancel it out. This requires feeding the noise into the amp stages to create a inverse wave to cancel out the power supply noise.

 

[NickK]

So I've taken the plunge in looking at alternative power options.. after exhausting the linear power supply option...

Switched mode power This gets fun

So the market for high power SMPSs is small, expensive and well there is another option

Take a 240Vac SMPS that output 12V at 500W... means a shed load of amps, it also must by law.. have a PFC built in (hehe) and also can be 'isolated' so that all the mains power goes through magnetic (and the feedback goes optically).. thus if anything happens mains side.. transients of KV etc.. then the supply keeps the person with their headphone safe.. did I say they also solve their own inrush problem..

So now we have a whole lotta 12V goodness.. but SMPS hate inductive/capacitve loads.. give a capacitor for nice ripple rejection and they will emergency stop as the cap looks like a dead short.

So what if you find a LT3750 or a LT3751 which is DC-DC capacitor charge controller. That can take ANY sized cap and charge it, controlling inrush and charge as an output up to 500V then regulate power (LT3751).

We now have a 240VAC (inrush limited, power factor corrected, switched mode) -> 12V 60+Amp -> LT3751 Charge controller (variable voltage, variable current limiting) -> 200-500V B+ DC -> 5.1ohm resistor -> 6,000uF (6mF) -> B+ (the 5.1ohm+6mF gives a hard 0.5Hz low pass filter..)

So is the output any good? Well the LTspice model seems to show this (note this is just the DC-DC charger and 6mF without any additional smoothing caps) - I found the a thread with the model and updated it to closer to the example applications in the LT3751 data sheet, also updated it to use the 50A capable coil craft transformer:



Err.. what ripple and the switching noise up into the MHz level..



If I need to change the power.. I can simply alter the voltage or current limit..

I think I may have found the power supply..

 

[NickK]

The LT3751 is an interesting chip - it is designed to cope with capacitors.. in fact it's a capacitor charger chip. Unlike normal SMPS that break with capacitive loads this knows what todo.

The chip has a number of modes (charge, regulate and no-load), it also has the ability to current limit and voltage regulate. With a coil craft transformer, it also copes with up to 500V output and 24V 50A input! It can charge at 24V 42A!



Green is the 6mF 450V capacitor, the current was set to a maximum of 24A on the 24V side. The blue is the softitart of the 3080-based voltage regulator which takes 30 seconds to startup but will remove the final few mV of ripple. It's set to charge to 200V and regulate there.. so the system now goes for maximum charging to charge the cap, then when the feedback indicates it's at the right voltage is switches to regulate which uses a quieter switching and can mix/match charge/regulate/no-load as the system needs. Lastly it has a charge pin - essentially a start pin, and a done pin, so startup and sync with other power can be automated.

So this means it can be adapted to the main 648W pfc 240ac->24dc 27A capable power supply.. after charging at 600W it's only drawing about 20-40W whllst the soft start continues.. I assume that the power will then grow to about 100-120W for the mA required by the load (set to 280ohm to give about 500mA load).

Using a different LT3780 chip, another flyback chip, it can convert 24V to 5-30V and regulate it.. so I can then use my 24V supply to run the heaters with a conversion to 12.6V. That chip has a soft start too and current limiter which means I can current limit the heater startup (prevents hot spots in the heater wire due to inrush) without any additional circuitry. The example design does 12V 2A so with a little tweaking that should give me more current (change the mosfets etc). It also has a on-off pin, so it can be synced with the other power supplies.

Lastly when I add the digital DSD512 integration, the 3.3/5V power can also be supplied from the same 24V power supply.

I can then stagger the power up as the 648W PSU is actually a medical PSU I have in mind has a 5V 300mA output - which would be enough to run a microcontroller.. so a push to start microcontroller sequence would work nicely. It also means (a) it has the power factor correction present already, (b) it has EMI filtering, (c) it's an off the shelf unit leaving just the 3751 board to be made and lastly (d) has safety features like over/under etc.

So now the power will go 240Vac mains -> 24Vdc 27A -> a number of PSUs such as 24V -> 200V, 24V->12V and 24V-> 3.3/5V.

Simples.

Oh.. noise? This is from a partial previous run..

 

[NickK]

So I've been spending a lot of my spare time on this, the approach seems very very flexible and works nicely.

The issue comes down to MHz switching, selecting the right mosfet and diode. There are some gotchas such as capacitance and power reflected when the diode stops conducting plus the diode selected needs to minimise reverse current.This reflected power at switch off then jumps back primary side, increasing the primary side voltage and current. So essentially you see higher voltage after the transformer and into the mosfet. This spike can be very high, so the advise was to forego the 200V or 600V mosfets and go with 900V+. The next is that the gate capacitance and timing needs to be careful reviewed vs the MOSFET gate Q values. Otherwise there will be a delay in starting to switch on or off. (this switching is 78,000 times a second). Get that wrong and the system will allow full current for longer and result in a higher minimum amount of power being delivered which makes regulation harder.
So in short, from my calculations, at full load the reflected power is equivilent to 24W.

I've also estimated that the supply will give me about 625mA max at 200V from napkin calculations. Given the efficiency is more like 80% then I'd figure on about 500mA. Enough for the amp, an alternative option is to simply provide two - one for each channel.

I've been looking at SMPS for the 24V supply - the Meanwell MPS 600-24 is a medical grade SMPS. It has all the power factor management, EMI filtering, safety features, high efficiency and 5Y+ warrantee.. it also has a 150mVpp ripple over 20Mhz bandwidth. So in short, 0.1V ripple. Consider the rectified linear PSU needs a 6mF capacitor to get to 1V.. we're already onto a winner.
I may not even need a full 600W.. however the beauty is that as long as I stay above 10% (60W) (relatively easy with a tube class A with heaters being 30-40W) then it will simply only draw that needed power from the mains.. ie if the amp draws 100W it only draws 105W (for example if it's 95% efficient) not 600W continuous.

The cost.. well the MPS is £136. Then look at the cost of a 600VA+ transformer, plus required rectifier, PFC, high wattage soft start resistors, etc.. it's cheaper and you still need more regulation. The cost of the LT3751 will be about about £50 with a prototype PCB. The cost of components £4 for the LT3751, £4 for the mosfet, then varied surface mount components.

The only different I may make is that I may switch from a single 6mF to 6x1mF in parallel to reduce the heating effect, this also reduces the ESR that is higher in the smaller caps. That way I still keep the 0.5Hz filter/reserve cap to really help reduce ripple, remove noise and provide support for transients. If you look at the ripple and current, you can calculate actually with the SMPS I don't actually need 6mF - I can use much smaller caps.

C = I / (2fVripple) = 0.500 / 2*70,000*0.15 = 0.0002380 F or 238uF.

So the question is do we need 6mF? The short answer depends on two things - transient support and noise filtration.

Transients - well given that the tube transient is supported by the B+ value and the current each tube is able to transfer, these are then at the limit of the tube itself in terms of power dissipation and the load line. In watts that's per second, so as long as we're not blasting away and the current x internal resistance doesn't cause a voltage spike over the arching point - we can support a higher voltage as long as the split second transient is, when smoothed over, less than the max plate dissipation. In short I'd plan for 60mA as I don't know the length of a transient in the music..
Therefore a 300uF close to the output tubes would help in transient handling, coupled with two 500ohm resistors we could drop from 200V to 150V and provide bi-directional filtering, lowering noise from the channel onto B+ and from the B+ into the output.
The front end could also do the same - although their power handling is 20mA each, which makes me a little concerned putting a 300uF over the input triodes.

In the end what I may do is start with a couple of capacitors and add more as I need them in the locations. Just doing a 6mF run but I may run a 1mF reservoir and then see if there's much difference. In the end the simulations won't make much of a difference until the kit is together.

 

[NickK]

So I have a model running.. this is the last sub-second sample as the amp is just about stabilised.. this is the full amp model - one channel running 30Hz and the other 10Khz to check for cross talk.. there's no cross talk and no noise. The real 24V PSU would have some noise but that is likely to be attenuated by the low pass filters.

The obvious thing is that the signal is quiet - 447mV input but low output.. need to look at the scaling but the main thing is the power supply.. that's decent enough I think.

No if only I can get this into a proper power supply.

There's an evaluation LT3751 but it's unlikely to be able to provide the current required for the entire amp and the eval board is £100, it's pretty steep.

 

[NickK]

So I’ve began creating a bill of materials for the 24V-200V power supply