Transformers and inductors introduce ringing, after sharp edge , there appears additional signal with falling amplitude .I talk about this .And it may have effect even after output filter . I don't mean core saturations or hysteresis effects on signal ,i talk just about parasitic capacitance caused artifacts .They are very good visible in smps supplies .
Member
Joined 2009
Paid Member
130khz cut is with many dB fall on this frequency .But in simulation you can zoom in lets say 1khz or 10khz ,effect of filter begins . Filter was simple - 1k ,1100pf , 33k load ( that amplifier had this ) .I changed cap to 470pf and tried to add another 470pf and a difference was heard at high frequencies. Filter was not sharp ,just typical input filter .If it would be parasitic distortion or some kinda of artifact ,it would appear first at high volume ,but i have not played loud at that test .And also simulation should show flat line even before 20khz .But as i said ,difference was small ,and in D class filters are always present ,just their cut frequency may be different ,depending on amplifier .
OK, we talk about the same. I agree, whenever you pass something through a coil with ferrite, or much worse iron, you are victim. When I built class AB audio amps with 'ultralinear' gate 2 taps at the output transformer, I was happy - in that time surprisingly good sound quality. No comparison to AB without gate 2 feedback.
In that aera, allmost all measurement instruments have been based on sinus signals. It was just a matter of broadly available instruments. That was the origin of a lot of unusable performance charts.
Today, we have lots more, any scope includes FFT analysis in the meantime.
I think, I might make a crazy project at all, but my personal challenge is, to get the best possible audio quality and high home audio power with tubes. And what do I mean with best?
Best impulse audio response possible. A flat frequency response and low distortion, too - of course. Why do I use a switched transformer? Because all hysteresis effects happen in a time domain, which you can't hear at all and disappear over lots of cycles and a feedback stabilizes audible frequencies or phase shifts.
Another factor: Today, we have better speakers, then many years ago - at least, we could have. My speakers are big 30cm woofers from Visaton, combined with some smaller. Coils are without iron, except for the spreaker coil itself
Last edited:
Pulse audio response is very very related to input filter ,which reason is to slow down input slew rate to a rate ,which amplifier still able to reproduce . If we saying 20khz ,its sine wave just in test equipment. But even 1khz sqarewave behaves different , after filter ... And who knows , how tracks are mixed in studio ,how sharp can change their amplitude . Natural sounds may not have that ,but electronic music has that ,and properly amplify that is challenge... So when someone says you can't hear 20khz or more ,thats almost true ,but cutting that will show you some loss on some material . After filter 1khz squarewave no longer looks like it was . But it just 1khz !
I want to remind on the well known theory behind Fourier transforms. A filter needs to be designed very carefully. They always have a purpose, right at their architecture. In my case, I would have to get rid of my base frequencies at first. Triangle and magnetic discharge have fixed, but different frequencies.
I return to my target: Best pulse response is a good starting point!
And by the way: I own some organ concert recordings with organ pipes near to 16 Hz - I want to hear them! (Simon Preston, Haendel, dome of Luebeck)
I return to my target: Best pulse response is a good starting point!
And by the way: I own some organ concert recordings with organ pipes near to 16 Hz - I want to hear them! (Simon Preston, Haendel, dome of Luebeck)
This is a switched, symmetric class d amp. The transformer is run with some ~hundred kHz and the ferroxcube core I use is designed for that.
Other then usual PWM driving a core which will saturate, I let the transformer run through magnetic discharge at another frequency.
Do not interchange with any analog, linear amplifier!
Yes , switched is switched ,but... What happens when amplifier will longer supply one polarity of output signal , lets say positive halfwave of 16hz ? Pwm duty cycle will be more for one tube ,and depending on how close it will be to maximum , this is like assymetric magnetisation current in push pull ,which magnetises core.This will cause similar effect like in flyback power transformer , you need air gap ,to prevent core saturation .Maybe if you will limit duty cycle to lower values like you said earlier 66percent that will not happen ,but then you can't utilise entire posible amplifiers power output at given power supply voltage .
This is described earlier in the thread. The first frequency is triangle, creating PWM. The second frequency will be high enough to avoid saturation and the discharge time is shorter then the passing time. This is the reason for keeping the discharge winding smaller winding count and discharging back into the supply of 500V, while the GU50 anode will raise to almost 1800 to 2000 Volts - without danger or anode current.
Basically, the 3 triodes are a differential amp cathode coupled. Which is overridden by a third triode, which will force the both GU50 g1 to -55V. At least, this is the purpose of the third triode (pair). I noticed a design error there, which I will fix soon. During this phase, the core will discharge it's magnetic energy. Whatever came in during PWM operation, it will be discharged in a third of the working cycle.
Edit: The core is a EC120, not a EC102. That is another typo in the circuit.
I spent some hour to play with variants of the driver circuit.
What is good, is the cascode circuit, which is driven by the PWM. It will be a fast inverter for the opposite GU50 and pull 30mA on one side, or the other.
If I make this a bit careful, it will drive the g1 precisely, fast.
The problem is, to mix this with the magnetic discharge, where both triodes need to pull 30mA from the gate g1 GU50 until discharged and the tubes cut current, the transformer goes into discharge out of one of the EY500.
I play with the idea, to use another triode pair, anodes at the 2 10k resistors, pull both down and reach -50V at one, -100V at the other.
The only drawback is, that the charge in g1 is even higher. The resistor will have a current of 35mA (distributed by PCC89s).
After ending the magnetic discharge, one PWM triode set will keep the GU50 off. The other triode anode will raise and it's GU50 will start normal operation.
I also thought of using 2 small tube diodes (EAB(C) type) to keep swing limits fixed between my optimal values. Those diodes would clamp at regulated voltages, hardened by large caps, which sustain anything, which is low audio. They are very fast and easily carry 30 or 60mA.
This mixer is the most complicated subcircuit.
What is good, is the cascode circuit, which is driven by the PWM. It will be a fast inverter for the opposite GU50 and pull 30mA on one side, or the other.
If I make this a bit careful, it will drive the g1 precisely, fast.
The problem is, to mix this with the magnetic discharge, where both triodes need to pull 30mA from the gate g1 GU50 until discharged and the tubes cut current, the transformer goes into discharge out of one of the EY500.
I play with the idea, to use another triode pair, anodes at the 2 10k resistors, pull both down and reach -50V at one, -100V at the other.
The only drawback is, that the charge in g1 is even higher. The resistor will have a current of 35mA (distributed by PCC89s).
After ending the magnetic discharge, one PWM triode set will keep the GU50 off. The other triode anode will raise and it's GU50 will start normal operation.
I also thought of using 2 small tube diodes (EAB(C) type) to keep swing limits fixed between my optimal values. Those diodes would clamp at regulated voltages, hardened by large caps, which sustain anything, which is low audio. They are very fast and easily carry 30 or 60mA.
This mixer is the most complicated subcircuit.
Now I made certain calculations, datasheet inspections and now a fast sketch again.
Omitted is the part of square wave generators and the feedback, which was already sketched.
The focus here is, how I want to guarantee a fast and reliable switching between -1V and -55V at the GU50 g1 pins.
U1 and U2 are a usual complementary amp with common cathode resistor feed of 30mA. Each triode will run at 15mA when on.
Result is a 30mA charge/discharge drive at g1 of the GU50.
U3 and U4 are switched simultaneously, on or off. They have independent cathode resistors and common gate.
During on, the PWM from U1, U2 appears at the GU50s. U3 and U4 are off.
During on of U3 and U4, all 4 GU50 are switched off and the transformer will rise all voltages in the last magnetization direction and one of the EY500 diodes will pump energy back into supply of 500V.
New is the clamping of the g1 swing by a pair of EABC80 tubes. Their d2 and d3 each are capable of iA 80mA, which is way enough. Important is the real independence of the diodes and the handy clamp voltages.
The swing at g1 of all GU50 is the same for all and always in required, allowed values only.
U1 to U4 degradation or drift are without danger.
Here the sketch:
Depending on a secure dimensioning of the resistors, the clamping is not excessive but will not drift. The clamping voltages are regulated.
Omitted is the part of square wave generators and the feedback, which was already sketched.
The focus here is, how I want to guarantee a fast and reliable switching between -1V and -55V at the GU50 g1 pins.
U1 and U2 are a usual complementary amp with common cathode resistor feed of 30mA. Each triode will run at 15mA when on.
Result is a 30mA charge/discharge drive at g1 of the GU50.
U3 and U4 are switched simultaneously, on or off. They have independent cathode resistors and common gate.
During on, the PWM from U1, U2 appears at the GU50s. U3 and U4 are off.
During on of U3 and U4, all 4 GU50 are switched off and the transformer will rise all voltages in the last magnetization direction and one of the EY500 diodes will pump energy back into supply of 500V.
New is the clamping of the g1 swing by a pair of EABC80 tubes. Their d2 and d3 each are capable of iA 80mA, which is way enough. Important is the real independence of the diodes and the handy clamp voltages.
The swing at g1 of all GU50 is the same for all and always in required, allowed values only.
U1 to U4 degradation or drift are without danger.
Here the sketch:
Depending on a secure dimensioning of the resistors, the clamping is not excessive but will not drift. The clamping voltages are regulated.
Last edited:
Depending on difference of reality and dimensioning. But they start to conduct at 0V and go down to 40 Ohms.
Nearly precise setting is possible.
I can adjust operating points of the PCC89 and the clamping voltages - easy and secure.
Funny - this very old EABC80 is ideal.
The aim is, to securely switch off all GU50 at once and to keep the differential amp tubes in operating point conditions.
When the U3, U4 return to off, current will immediatly charge g1 of the correct GU50 pair.
Last edited:
Lots of. You may replace the anode resistors against tubes, constant current. But the performance win is low, if any.
But doing it with tubes exclusively, is my personal challenge. I want even very, very low lower frequency limit and this is the way - dropping all bad behavior of iron.
And a reliable gate driver (with tubes) is essential for sound quality.
This was the first concept, which convinced me, that I will not outburn the GU50s by drifting drive circuits.
Today, you can buy ready made class D amps with good quality, using modern semiconductors. I own one, was surprised about capability and pulse response.
But I never owned a full set of tubes of high quality, now I do.
If you use simulator, repeat that with - lets say - 2x470pF with accuracy and then 1100pF with a 10% (+10%) value -> 1210pF.
Of course, you will not even see it different at square wave 1 kHz, it is already significant in measurement.
Remind the Fourier decomposition of a signal in frequency, phase, amplitude values. You will hear the difference, I beleive that immediatly.
But we can safely omit anything above 20kHz.
I remind on FM transmission quality, which does a clear cut at 15 kHz - and sounds good, still.
If I return to alternate driving circuits, I expect a lot of high frequency ringing everywhere, but I want to avoid such at the GU50 gate1 as much, as possible. When I install a constant current source in the anode rails of the U1, U2 amp, I expect anything, but no clear signal.
Just because these constant current is not constant. It is an active circuit with dynamics.
I made calculations, some pages up, for charge and discharge times. I already reach slew rates, which create the GU50 run in some tenths of MHz, which will already in the domain, where the transformer is ringing like a bell (parasitic inductances, capacitances) and I will use ferrite core armed cables to the anodes, mabe even small resistor values at the cathodes - just to reduce RF emission.
But overall, the new driver circuit could be quite precise (timing) and therefore will not introduce distortion.
Here is the starting point for the signal generators. It is a standard opamp circuit. Both generators are essentially the same:
Too opamps form a loop between schmitt-trigger and integrator.
One is buffered and level-adjusted at the sqarewave output, the other at the integrator output. Level adjustment is required for the PCC89 gates of the driver circuit. Their swing is below 0V only, at around -2V the triode is in low current condition already.
Here is the circuit:
You might be surprised, I brake my rule for tube circuits only. The reason is simple:
It would take me too much time, to design matching circuits. Too many options, with drawbacks of unknown extent, for both generators.
Another reason: These generators shall run with full specs, before any heater will allow significant currents at the anodes. It is important to note, the PCC89 is heating up quicker then the GU50.
Each channel needs 12 tubes for the amp.
I estimate, if I continue with tubes only, I would reach target, but the innovative part is already in sketch. The remainder is just time consuming.
The opamp is fast. Settling time at g of 2 is 40nsec to 0.1% accuracy. Gain/Bandwith is about 130 MHz - way enough!
It can drive up to 50mA and uses ± 5V supply - ideal for the PCC89 gates.
The circuits are basically the same. One schmitt trigger is symmetric, the other is not. The symmetric is for triangle, the other for magnetic discharge control.
I will need a comparator, after the triangle. Comparing audio to triangle. Because of simplicity, I omit that. Circuit is trivial.
Not trivial would be a usable limiter for audio. If audio is exceeding triangle, ugly , sharp clipping would be result. I'm not sure, if I take measures.
No option is a tube stage for that.
It is not clear, but my speakers will probably burn after short time of full throttle, anyhow.
They are specified 100 W sine - for a few seconds.
Another point for the use of opamps is the feedback processing. I already made the sketch for a Philbrick opamp. Maybe I will try a prototype with the PCC89 and evaluate it. For the requirements of feedback and comparator it might be good enough. But as first shot, I will use those easy to handle ADA4851 devices.
Too opamps form a loop between schmitt-trigger and integrator.
One is buffered and level-adjusted at the sqarewave output, the other at the integrator output. Level adjustment is required for the PCC89 gates of the driver circuit. Their swing is below 0V only, at around -2V the triode is in low current condition already.
Here is the circuit:
You might be surprised, I brake my rule for tube circuits only. The reason is simple:
It would take me too much time, to design matching circuits. Too many options, with drawbacks of unknown extent, for both generators.
Another reason: These generators shall run with full specs, before any heater will allow significant currents at the anodes. It is important to note, the PCC89 is heating up quicker then the GU50.
Each channel needs 12 tubes for the amp.
I estimate, if I continue with tubes only, I would reach target, but the innovative part is already in sketch. The remainder is just time consuming.
The opamp is fast. Settling time at g of 2 is 40nsec to 0.1% accuracy. Gain/Bandwith is about 130 MHz - way enough!
It can drive up to 50mA and uses ± 5V supply - ideal for the PCC89 gates.
The circuits are basically the same. One schmitt trigger is symmetric, the other is not. The symmetric is for triangle, the other for magnetic discharge control.
I will need a comparator, after the triangle. Comparing audio to triangle. Because of simplicity, I omit that. Circuit is trivial.
Not trivial would be a usable limiter for audio. If audio is exceeding triangle, ugly , sharp clipping would be result. I'm not sure, if I take measures.
No option is a tube stage for that.
It is not clear, but my speakers will probably burn after short time of full throttle, anyhow.
They are specified 100 W sine - for a few seconds.
Another point for the use of opamps is the feedback processing. I already made the sketch for a Philbrick opamp. Maybe I will try a prototype with the PCC89 and evaluate it. For the requirements of feedback and comparator it might be good enough. But as first shot, I will use those easy to handle ADA4851 devices.
Last edited:
Hybrid technology is very good sometimes ,not everything needs to be in clear rules ,when you are DIY something. You may get best of each type of technology then and combine that .Simplicity of opamps circuits , their speed , and combine with retro tubes style and high voltages .For me tubes are associated with something retro , like b/w tv we had in 80's or very old radios with vinyl on top ...
I think it would be easier to use non-tube based driver for GU's ,but thats are your choices and challenges .
I think it would be easier to use non-tube based driver for GU's ,but thats are your choices and challenges .