I am in the process of designing an amplifier with DHT tubes in push-pull configuration. The "cathode" (i.e., one side of the filament) will be connected directly to GND, without a cathode resistor. Bias will be applied as fixed bias to the grid. The filaments should be supplied with DC by a Coleman-like DHT filament regulator, and it would be nice if both regulators could run off the same raw DC supply. See attached layout.
There are a few schematics of the Colement DHT filament regulators floating around the net, and they all show a CCS on one side the filament and a gyrator on the other side. So, with one end of the filament of each tube connected to GND, the corresponding regulator output will also be connected to GND. In other words, the regulator outputs are directly connected to each other. If both regulators would run off the same raw DC supply, the regulator parts between GND and the common DC input would be connected in parallel, which is a bad idea.
I discussed with @Rod Coleman about this, and he hinted that in his latest version of the regulator (V9), he changed the sequence of the gyrator-filament-CCS series connection to gyrator-CCS-filament (at least that's how I understood things). With the new arrangement, the filament is not sandwiched between the gyrator and CCS anymore. For both tubes, one end of each filament is connected to a CCS, while the other ends connected to GND. This means the raw DC can also be referenced to GND without causing the issue I described above.
Rod also indicated that he added a temperature compensation to the CCS to avoid drift of the filament current. The tube always sees the correct current (and hence the correct voltage).
Rod prefers to not share his V9 schematics, which I perfectly understand. I could simply use his boards/kits and get everything up and running. However, I'd prefer to implement a self-contained PCB for the entire amp, with the filament supplies included on the board. I therefore made an attempt at designing an incarnation of a DHT filament regulator that would suit my requirements:
There are a few schematics of the Colement DHT filament regulators floating around the net, and they all show a CCS on one side the filament and a gyrator on the other side. So, with one end of the filament of each tube connected to GND, the corresponding regulator output will also be connected to GND. In other words, the regulator outputs are directly connected to each other. If both regulators would run off the same raw DC supply, the regulator parts between GND and the common DC input would be connected in parallel, which is a bad idea.
I discussed with @Rod Coleman about this, and he hinted that in his latest version of the regulator (V9), he changed the sequence of the gyrator-filament-CCS series connection to gyrator-CCS-filament (at least that's how I understood things). With the new arrangement, the filament is not sandwiched between the gyrator and CCS anymore. For both tubes, one end of each filament is connected to a CCS, while the other ends connected to GND. This means the raw DC can also be referenced to GND without causing the issue I described above.
Rod also indicated that he added a temperature compensation to the CCS to avoid drift of the filament current. The tube always sees the correct current (and hence the correct voltage).
Rod prefers to not share his V9 schematics, which I perfectly understand. I could simply use his boards/kits and get everything up and running. However, I'd prefer to implement a self-contained PCB for the entire amp, with the filament supplies included on the board. I therefore made an attempt at designing an incarnation of a DHT filament regulator that would suit my requirements:
- One end of each filament can connect to GND, and so can the raw DC supply.
- I used a temperature-compensated voltage reference in the CCS.
(The LM4041 is said to be rather noisy; so I went for the TL431, as used in the Sony VFET 2017 amp by Nelson Pass, so it can't be too bad.)
Attachments
The Q9 section is not a Gyrator. That's actally a ripple voltage filter.
The V9 version from RodColeman can be used directly in this situation.
Its Raw DC GND is connected to one side of the filament. I havn't seen the Gyrator section in the circuit. Series connected Gyrator and CCS is meaningless because both of them intend to provide high AC impedance. There's only one power transistor, so it's impossible to be Gyrator and CCS at the same time.
A Gyrator should be a low impendance voltage source to DC and a high impendance current source to AC. The MOSFET in RC V9 is confiured as a CCS.
Like this.
I think the previous versions are mainly VIN-Voltage Filter/Stabilizer-Filament-CCS-GND. I remember the top side was a voltage stabilizer instead of a Gyrator. And only the earlier version use gyrator on the top side.
There's no "Correct Current" for a certain kind of tube. (Except those IDHTs whose heater is designed to be connected and powered in series directly from AC Input without transformer.) The current need to be adjust carefully for each tube. I don't agree with using potentiometers in the feedback loop. And the quiescent current of TL431 will go through the filament with set current, which will cause extra error and drift.
I'd like to build it like this, with a shared positive end of filament.
The shared Raw DC goes into a voltage stabilizer, providing higher PSRR. The output of the voltage stabilizer is connected to the Global GND next. A low side CCS set the filament current for the tube. The TL431 works as a single voltage reference instead of an error amplifier with voltage reference. Potentiometer is used to set the voltage which control the current. There's no potentiometer in the feedback loop. Temperature drift compensation is not needed in this circuit, but the stability compensation of the error amplifier is needed to provide high impedance across audio frequency while maintaining stable. With high performance OPAMPs and inadequate compensation, even a little bit longer wire with little inductance will result in oscillation.
The V9 version from RodColeman can be used directly in this situation.
Its Raw DC GND is connected to one side of the filament. I havn't seen the Gyrator section in the circuit. Series connected Gyrator and CCS is meaningless because both of them intend to provide high AC impedance. There's only one power transistor, so it's impossible to be Gyrator and CCS at the same time.
A Gyrator should be a low impendance voltage source to DC and a high impendance current source to AC. The MOSFET in RC V9 is confiured as a CCS.
Like this.
I think the previous versions are mainly VIN-Voltage Filter/Stabilizer-Filament-CCS-GND. I remember the top side was a voltage stabilizer instead of a Gyrator. And only the earlier version use gyrator on the top side.
There's no "Correct Current" for a certain kind of tube. (Except those IDHTs whose heater is designed to be connected and powered in series directly from AC Input without transformer.) The current need to be adjust carefully for each tube. I don't agree with using potentiometers in the feedback loop. And the quiescent current of TL431 will go through the filament with set current, which will cause extra error and drift.
I'd like to build it like this, with a shared positive end of filament.
The shared Raw DC goes into a voltage stabilizer, providing higher PSRR. The output of the voltage stabilizer is connected to the Global GND next. A low side CCS set the filament current for the tube. The TL431 works as a single voltage reference instead of an error amplifier with voltage reference. Potentiometer is used to set the voltage which control the current. There's no potentiometer in the feedback loop. Temperature drift compensation is not needed in this circuit, but the stability compensation of the error amplifier is needed to provide high impedance across audio frequency while maintaining stable. With high performance OPAMPs and inadequate compensation, even a little bit longer wire with little inductance will result in oscillation.
Last edited:
Thanks for these insights, very useful!
Sorry for my poor use of the "gyrator" term. It seems the term gets used for just about anything, and I got used to it.
Your circuit makes sense, but I suspect it is a bit more complex than what Rod Coleman actually uses. I believe there are no opamps on the photos I found on the net.
Sorry for my poor use of the "gyrator" term. It seems the term gets used for just about anything, and I got used to it.
Your circuit makes sense, but I suspect it is a bit more complex than what Rod Coleman actually uses. I believe there are no opamps on the photos I found on the net.
What is the issue with the potentiometers?
I guess you are referring to my circuit draft. Can you explain where/how the TL431 quiescent current passes the filament? As I see it, R32 (R33) feeds the quiescent current to the TL431, which is connected to GND. The filament current passes thorough Q11 (Q12) and R36 (R37) -> filament -> GND. This loop is separate from the R32/TL431 loop.
This is a typical TL431 CCS circuit. The only difference is BJT vs MOSFET and there's no gate resistor.
If common GND is shared, then you have to float the TL431 to create a 2-terminal CCS. In your circuit, you just connect the FIL1_OUT with GND. I think that's wrong and the filament should be placed as my drawing. Then it'll be the same as the TL431 datasheet from TI. The TL431 has no directly connection with GND, so that its quiescent current will go together with filament current.
Building a CCS with adjustable shunt regulator, we need to keep Vref with Rs.
So that the quiescent current flows out from Anode, and Anode must be connected to the resistor. It's impossible to build a top side CCS while getting rid of quiescent current of TL431.
LM4041 is much different from TL431. Its Vref is the voltage between Cathode and ADJ. It's not the same as TL431.
Due to such difference, the LM4041 is capable building a top side CCS with no quiescent current together with output current.
There's actually no OPAMP in Rod Coleman's design. And there's also no TL431, LM4041 or similar device in his design.
Since TL431 and LM4041 have integrated OPAMP. I think that's the same as other OPAMPs. And separate OPAMP can show higher performance.
He's using BJT as error amplifier and Vbe of BJT as voltage reference. That's similar to a TL431.
In his V9 version, he change the NPN BJT into PNP. The transforming is like changing from TL431 to LM4041.
I guess a BJT acting as error amplifier and voltage reference has higher flat bandwidth comparing with OPAMPs?
Potentiometers have higher noise, lower reliabilities and higher inductance. It will influence the performance of the feedback loop.
Oups! Yes, you're right. I was blind. Thanks for pointing this out!
Makes sense. Rod Coleman told me that he found the LM4041 to be too noisy, and he wasn't successful when he tried this part.
Rod Coleman indicated that he achieved good thermal compensation in his last design. I am not sure how he did that, but there might be temperature-dependent resistor involved (NTC, PTC).
How much does the filament current drift with temperature in reality, and how important is thermal compensation?
He's using NTCs. But according to my test, I found that adjust current will unbalance the thermal compensation, and it's really hard to find the balance point if you'd like to build it by yourself. OPAMPs have extremely low temperature drift, thus no extra temperature compensation is needed anymore.
It seems so. The noise spectrum density of LM4041 is about 200nV/√Hz @1KHz with Vref = 1.2V, while the noise of TL431 is about 125nV/√Hz @1KHz with Vref = 2.5V.
I've done a brief test over a 300B.
1.3A 4.395V,
1.31A 4.480V,
1.32A 4.560V,
0.77% variation in current results in about 1.5% variation in voltage.
The temperature coefficiency of filament conductor will amplify the error in current into a much larger variation in voltage.
In order to maintain 5% filament voltage variation, the current variation need to be limited within 2%.
The amplifier can meet the heat balance finally, so we only need to consider the ambient temperature change. But the final heat balance takes a long time, so it also takes a really long time to adjust the current carefully. Every single adjustment will be followed by a long time of monitoring the voltage changing gradually. That's painful. A slight rotation to the potentiometer will cause the filament voltage keep changing in the following 3-5 minutes. If the filament current change along with the chasis temperature obviously, the adjustment will be far more complex. That's the reason why I'm building the filament CCS module with automatic adjustment. I'd like to build an open source version with reduced cost later, but I have to finish my recent project first.
Last edited:
Thanks for all these insights!
In the meantime I found this article, which also has some useful information:
https://www.mvaudiolabs.com/tubes/the-taming-of-the-dht/
Looking at the last schematic on that page, I guess I could just flip the polarity of the ripple filter and move it to the lower side of the filament, between the CCS and the raw DC supply. (Or, the opposite: move the CCS to the upper side of the filament.)
Or maybe a single ripple filter would be enough to provide the DC to multiple CCS filement supplies.
In the meantime I found this article, which also has some useful information:
https://www.mvaudiolabs.com/tubes/the-taming-of-the-dht/
Looking at the last schematic on that page, I guess I could just flip the polarity of the ripple filter and move it to the lower side of the filament, between the CCS and the raw DC supply. (Or, the opposite: move the CCS to the upper side of the filament.)
Or maybe a single ripple filter would be enough to provide the DC to multiple CCS filement supplies.
Problems of this kind are avoided in my V9 regulator. This is in part due to the thermal design and layout of the PCB itself - a single small board.
Setting the V9 with a real filament does not involve such fussy adjustment. Just power on, adjust to 1 or 2% of nominal over the first minute, then wait for the amp to warm up. Then make the final adjustment. This is how I set up my 300Bs, using stock V9s, and anyone can do it, following the V9 manual.
On subsequent power cycles, The typical filament voltage keeps to within 1% of nominal, after a few minutes (at most) warm up, so long as the installation follows the instructions given. It requires a little care, but not fuss. The kits have been available for years, and I do not get complaints about adjustment difficulties.
Yes, this is one of the major challenges of purpose-designed regulators.
You can't expect to drive power transistors with Ampere level current without carefully tailored compensation. This is true of fully discrete circuits, or 431s or opamps.
The stray inductance of PCB components is only one factor; the power interface, and load inductance (cable plus filament inductance) must also be accommodated.
The V9 compensation is comprehensively tested in practice, as well as designed for 90⁰ plus of phase margin, irrespective of gain variations in the error amp. There is a version of the V9 that can drive loudspeaker field coils all the way up to multiple Henry level inductance.
It is very satisfying to complete this work successfully, but design and verification takes a very long time.
This schematic gives an example of forward path and compensation that makes for too little audio bandwidth before the impedance drops. The difference is easily audible!
In practice, it depends on the manufacturer and the actual sample.
I have seen 431s measuring a 0.5mV of RMS audio noise. Some are microphonic in addition.
Adding a large electrolytic capacitor reduces noise, but kills bandwidth.
Q3 should be a N-ch MOSFET for better current accuracy, R6 could be reduced to about 100Ohms, and a emitter follower stage could be added between Q4 and Q3.
These modifications will allow the power transistor to act faster, so that the bandwidth of the feedback loop could be increased.
C3 is for loop stability compensation. In my own test, I found that about 200pF for C3 is enough. Larger C3 value will lead to lower CCS impendance. With different models of power MOSFET, the minimum value for C3 will be different. Generally, power MOSFET with larger capacitance will need a larger C3 value to keep stable.
P1 is for adjustment, and R7 is actually a NTC tied with Q4 for temperature compensation.
In order to calculate the temperature compensation value, you need to consider the Vbe temperature coefficient vs collector current, the resistance changing with temperature of your selected NTC. The temperature compensation can be done only within a small range of working conditions.
In this circuit, Q4 has the same function as a TL431.
It's using Vbe of a BJT as the voltage reference, and higher voltage on Vbe (or voltage between ADJ and Anode of a TL431) results in a higher collector current (or current from Cathode to Anode in a TL431).