Hi All,
I was on this forum years ago and have been building tube audio amplifiers for a while now but am still learning. So I want to share a current project and ask for some suggestions on how I can improve its high-frequency performance.
The backstory:
After a jig of working mainly with small signal tube amplifiers (preamps, phono stages, etc.), I decided to take a stab at a power amplifier with toroidal power and output transformers.
The transformers I purchased were part number VDV-4070-CFB-PPS from Bureau Vanderveen. According to their published specifications, they looked like very good units. Their wide bandwidth is primarily what sold me.
https://www.mennovanderveen.nl/index.php/nl/producten/specialist/vdv-4070-cfb-pps-detail
Having no experience with this kind of transformer, I made the following circuit:
This is a markup of a partial schematic. I only am showing the audio circuit and not the microprocessor control circuitry (as it is not relevant). The markings are what I did to improve the performance already. Purple markings show DC voltages. Red markings are circuit and component value tweaks. Cross-outs are in blue.
Measurements:
All measurements were taken with a an 8-ohm resistive dummy load hooked to the speaker output. Measurements were taken with an H8903B audio analyzer, a Siglent SDS2204X oscilloscope, and a Fluke 287 multimeter. Square wave plots were generated by a BK Precision 4045B waveform generator. VRMS measurements were taken just before the amplifier clipped. Nominal power output is 50 watts into 8-ohms.
1. Open loop frequency vs. gain plot.
2. Closed loop frequency vs. gain plot along with THD:
Questions and Observations:
The current driver I am using rolls off open loop starting around 10kHz or so. I observed the tubes are the main contributor to the rolloff and not the transformer. So naturally the GFB loop, when introduced, increased the gain more at higher frequencies with respect to the lower frequencies. Distortion increased due to the higher gain. What do you suggest I do to improve the performance of the open-loop bandwidth? Are there better driver circuits I could look at using?
I have more questions but will save them for another post.
Thanks very much.
I was on this forum years ago and have been building tube audio amplifiers for a while now but am still learning. So I want to share a current project and ask for some suggestions on how I can improve its high-frequency performance.
The backstory:
After a jig of working mainly with small signal tube amplifiers (preamps, phono stages, etc.), I decided to take a stab at a power amplifier with toroidal power and output transformers.
The transformers I purchased were part number VDV-4070-CFB-PPS from Bureau Vanderveen. According to their published specifications, they looked like very good units. Their wide bandwidth is primarily what sold me.
https://www.mennovanderveen.nl/index.php/nl/producten/specialist/vdv-4070-cfb-pps-detail
Having no experience with this kind of transformer, I made the following circuit:
This is a markup of a partial schematic. I only am showing the audio circuit and not the microprocessor control circuitry (as it is not relevant). The markings are what I did to improve the performance already. Purple markings show DC voltages. Red markings are circuit and component value tweaks. Cross-outs are in blue.
Measurements:
All measurements were taken with a an 8-ohm resistive dummy load hooked to the speaker output. Measurements were taken with an H8903B audio analyzer, a Siglent SDS2204X oscilloscope, and a Fluke 287 multimeter. Square wave plots were generated by a BK Precision 4045B waveform generator. VRMS measurements were taken just before the amplifier clipped. Nominal power output is 50 watts into 8-ohms.
1. Open loop frequency vs. gain plot.
| Frequency | VRMS | Gain | Gain (dB) |
20 | 16.5462987 | 165.462987 | 44.3740172 |
50 | 19.0918831 | 190.918831 | 45.6169753 |
75 | 20.0111219 | 200.111219 | 46.0254288 |
100 | 19.9404112 | 199.404112 | 45.9946822 |
500 | 20.0818326 | 200.818326 | 46.0560668 |
1000 | 20.0818326 | 200.818326 | 46.0560668 |
5000 | 19.9404112 | 199.404112 | 45.9946822 |
10000 | 19.6575685 | 196.575685 | 45.870596 |
15000 | 19.0918831 | 190.918831 | 45.6169753 |
20000 | 18.243355 | 182.43355 | 45.2220942 |
2. Closed loop frequency vs. gain plot along with THD:
| Frequency | Distortion | Amplitude | Amplitude (dB) |
20 | 3.15 | 21.16 | 26.51031327 |
25 | 0.32 | 20.89 | 26.3987688 |
30 | 0.25 | 20.91 | 26.40708066 |
50 | 0.23 | 20.73 | 26.33198604 |
75 | 0.25 | 20.7 | 26.31940691 |
100 | 0.23 | 20.67 | 26.30680953 |
250 | 0.24 | 20.66 | 26.30260634 |
500 | 0.25 | 20.67 | 26.30680953 |
750 | 0.25 | 20.66 | 26.30260634 |
1000 | 0.26 | 20.66 | 26.30260634 |
5000 | 0.57 | 20.52 | 26.24354713 |
7000 | 0.78 | 20.42 | 26.20111476 |
10000 | 1.04 | 20.23 | 26.11991766 |
15000 | 1.7 | 19.7 | 25.88932452 |
20000 | 2.3 | 19.2 | 25.66602457 |
30000 | 3.67 | 17.81 | 25.01327839 |
50000 | 5.87 | 14.37 | 23.14913536 |
Questions and Observations:
The current driver I am using rolls off open loop starting around 10kHz or so. I observed the tubes are the main contributor to the rolloff and not the transformer. So naturally the GFB loop, when introduced, increased the gain more at higher frequencies with respect to the lower frequencies. Distortion increased due to the higher gain. What do you suggest I do to improve the performance of the open-loop bandwidth? Are there better driver circuits I could look at using?
I have more questions but will save them for another post.
Thanks very much.
Why the 150pF cap on the cathodyne cathode ?
The cathodyne is quite loaded 100k with an ac load of 150k. Maybe another tube ( ECC82 ?) with a much smaller
plate/cathode would do better.
The cathodyne is quite loaded 100k with an ac load of 150k. Maybe another tube ( ECC82 ?) with a much smaller
plate/cathode would do better.
Hi Peter,
My apologies but that 150pF does not connect to the cathodyne. I had marked that schematic incorrectly. Below is the corrected schematic.
The power supply was tweaked as well.
Also, the open loop testing was performed without the step network (36 ohm and 150pF capacitor).
My apologies but that 150pF does not connect to the cathodyne. I had marked that schematic incorrectly. Below is the corrected schematic.
The power supply was tweaked as well.
Also, the open loop testing was performed without the step network (36 ohm and 150pF capacitor).
I would simplify the circuit. I'd use the LTP as the phase splitter, and 1/2 12AX7 to drive it, dc-coupled, similar to EICO HF-89. This reduces the number of tubes for a stereo and the number of capacitors in the signal path. You will need to modify the LTP CCS, to work at much higher voltage.
I am using KT90 tubes right now, even though I have all the common types of power tubes listed.
The resistors in the anode are for measuring current through the tube.
The resistors in the anode are for measuring current through the tube.
The source of high frequency roll-off is Miller capacitance of the input stage 12AX7. As this roll-off is relatively minor in the audible band (about 1 dB), it is acceptable by common standards. If you want to improve HF response, you can use a tube with lower mu in the first stage, for example 6N1P. The tradeoff will be lower open loop gain, but your circuit has more than enough of it. If you want, you can compensate for lower gain by choke-loading or CCS-loading the first stage.
The resistors in the anode are for measuring current through the tube.
Not my way: Dangerous exposure on this. I would scrap the anode resistors and opt for the cathode monitoring as much safer to use 10 ohm wirewound resistors in each o/p tube cathode. They will create some regeneration that reduces distortion.
BTW; I notice the values of interstage coupling caps C96/C65 of 0.1uF and 100K cathodyne resistors, and you are using a toroid output transformer that has no LF cutoff, unlike inbuilt lamination gaps of E&I styles. On my slide rule quickie, the excessive loop gain pole at the lower LF might make the circuit unstable into LF oscillation when the global NFB is connected. For starters I would use much lower interstage capacitor values for C96/C65. This is the old Williamson problem.
BB
Not my way: Dangerous exposure on this. I would scrap the anode resistors and opt for the cathode monitoring as much safer to use 10 ohm wirewound resistors in each o/p tube cathode. They will create some regeneration that reduces distortion.
BTW; I notice the values of interstage coupling caps C96/C65 of 0.1uF and 100K cathodyne resistors, and you are using a toroid output transformer that has no LF cutoff, unlike inbuilt lamination gaps of E&I styles. On my slide rule quickie, the excessive loop gain pole at the lower LF might make the circuit unstable into LF oscillation when the global NFB is connected. For starters I would use much lower interstage capacitor values for C96/C65. This is the old Williamson problem.
BB
That is indeed much safer, just keep in mind that cathode monitoring will include both plate and screen grid current.
While I agree that measuring the voltage drop on anode resistors bring certain level of danger, the fact is that using proper equipment (battery powered multimeter, appropiate leads for the voltage, insulating gloves etc) reduces the possibility of an accident.
Looks good to me. You got a HF dominate pole so have thought about stability. The LF cutoff is tricky as it depends on a few things. Sweep at 1Hz or so the check the gain does not raise too much on say the grids on the EL34's. Sometimes the LF caps need to get smaller it depends on whether the transformer or the coupling is the dominate pole.
My first suggestion is the big dumb hammer approach: remove the secondary feedback and measure the interstage performance ahead of the output tubes.
jCalvarez
I think the insulated approach is also risky ; I tend to work with several bench meters and been caught off guard having one set on amps and getting stung. Reckless/careless one can say, but it does happen even in company development labs.
There is the other "not to be ignored" issue on a new design that touching a probe (meter lead in this case) on an o/p tube /primary with a circuit that is only marginally stable, fester HF instability as give misreadings. This HF feedback could result in a orange anode. Any oscillation depending on proximity of input tube/layout circuits. In the old days this is where the screening can came in.
Often with such newling amps I run with a lower B+just to check various performance issues before going the full B+ hog.
BB
I think the insulated approach is also risky ; I tend to work with several bench meters and been caught off guard having one set on amps and getting stung. Reckless/careless one can say, but it does happen even in company development labs.
There is the other "not to be ignored" issue on a new design that touching a probe (meter lead in this case) on an o/p tube /primary with a circuit that is only marginally stable, fester HF instability as give misreadings. This HF feedback could result in a orange anode. Any oscillation depending on proximity of input tube/layout circuits. In the old days this is where the screening can came in.
Often with such newling amps I run with a lower B+just to check various performance issues before going the full B+ hog.
BB
I would be tempted to remove the LM334 and use a resistor instead. A simple resistor will ensure nicer overload recovery and more headroom, and balance should still be pretty good despite variation in the 12BH7. There is a lot of RC inside your feedback loop, so it pays to keep everything simple and compliant.
I would move R95 to the other side of the relay, so the tube has a permanent grid leak path (or add resistors on both sides of the relay).
Be prepared to add a small capacitor between grid and cathode of the input tube, in case you discover you need it for feedback stability. You'll only discover this through testing.
I would move R95 to the other side of the relay, so the tube has a permanent grid leak path (or add resistors on both sides of the relay).
Be prepared to add a small capacitor between grid and cathode of the input tube, in case you discover you need it for feedback stability. You'll only discover this through testing.
If you see peaking in the LF response then perhaps lower the 0.47uF coupling to output stage to at most 0.1uF, but increase the coupling cap values from PI stage a lot, say to 1uF, as a way to push the 1uF influence to well below the transformer resonance to give better stability.
Perhaps do some squarewave testing to determine how stable the HF response is, including seeing if any capacitance loading or no load causes an unstable output. If you have damped HF oscillatory response for a squarewave, then you may be able to lessen that by adjustment of a comp cap across the feedback resistor (I use a small modern radio tuning cap to set that up).
Perhaps do some squarewave testing to determine how stable the HF response is, including seeing if any capacitance loading or no load causes an unstable output. If you have damped HF oscillatory response for a squarewave, then you may be able to lessen that by adjustment of a comp cap across the feedback resistor (I use a small modern radio tuning cap to set that up).
This type of 4 stage amp without global NFB will be quite sensitive perhaps requiring perhaps around 50mV input for full o/p.
I notice in the circuit diag 20K grid leaks plus residual resistance of the trimmer (assuming mid travel of 25K = 45K) will punish the 68K driver anodes with considerable signal loss.
My view on interstage tube drivers is nasty and blame them alot for poor tube amp performances. Maybe it´s the type of music I listen to, dynamic organ concertos which test both tube amps and speakers.
Trobbins....I see conflicts here at the LF end with a Toroid design as increasing interstage coupling cap values to lower the LF pole can also result in "blocking" when fixed bias is used. Compromises to get a win-win on this one. Morgan Jones Valve amps 4th ed page 208.
Probably best wait until this design is actually running.
BB
I notice in the circuit diag 20K grid leaks plus residual resistance of the trimmer (assuming mid travel of 25K = 45K) will punish the 68K driver anodes with considerable signal loss.
My view on interstage tube drivers is nasty and blame them alot for poor tube amp performances. Maybe it´s the type of music I listen to, dynamic organ concertos which test both tube amps and speakers.
Trobbins....I see conflicts here at the LF end with a Toroid design as increasing interstage coupling cap values to lower the LF pole can also result in "blocking" when fixed bias is used. Compromises to get a win-win on this one. Morgan Jones Valve amps 4th ed page 208.
Probably best wait until this design is actually running.
BB
Blocking is far more likely in your output stage, where signal level is a lot higher, which is one reason to aim to reduce that 0.47uF. The PI coupling is likely to be at a much lower signal level, and if so then its coupling cap can be pushed to a much higher value. In the old days they used a large cap but as a step filter to lower phase shift from that stage to improve unstable operation.
For those to illustrate the issue, the early KT88-50 amplifier of 1957 shows it between the first and second stages. For enthusiasts the text explains the problem well.
http://elektronikjk.pl/Elektroakust..._88-50-a_Low-Distortion_50-Watt_Amplifier.pdf
BB
http://elektronikjk.pl/Elektroakust..._88-50-a_Low-Distortion_50-Watt_Amplifier.pdf
BB
That article by GEC/MOV shows a circuit with multiple low-frequency CR phase shift contributors. Typical amps try and only deploy two CR networks along with the output transformer resonance, and even then motorboating/peaking behavior can show up without refinement of the phase shift contributions. In that amp there are four networks that can contribute low-frequency phase shift: the input to PI stage CR, the PI to driver CR, the driver to output stage CR, and the output stage cathode decoupling caps.
The PI to driver stage CR network is pushed below concern due to the parallel R's (R14, R15) although that shifts the dc bias of the driver stage which can cause problems and is why amps that use this technique normally add an additional series cap to make it a step network.
The input to PI stage CR is pushed below 1Hz by means of the high resistances used in its self-balancing paraphase splitter, which would also likely suppress its phase shift contribution.
The other two networks are in the 1-10Hz range, which is where the output transformer resonance would also sit, so the amp may well not be an exemplar of how to manage low frequency stability.
What I did find interesting in the article was the commentary on a need for close-coupling of the UL winding in the output transformer to achieve improved HF response, and the likely benefit that may accrue from using a higher turns ratio tapping (as compared to the typical aim of a UL tapping derived just from distortion minimisation). I would surmise that they ran into some instances of HF instability with certain UL output transformers of that era, given they went to the effort to identify three commercial offerings that must have been tested as acceptable.
There was also a related glib comment about using a 12AU7 driver due to its low plate resistance to keep high frequency RC phase shift above 50kHz, but no mention was made of the elephant in the room of the choice to use a 10k grid-stopper for the KT88.
The PI to driver stage CR network is pushed below concern due to the parallel R's (R14, R15) although that shifts the dc bias of the driver stage which can cause problems and is why amps that use this technique normally add an additional series cap to make it a step network.
The input to PI stage CR is pushed below 1Hz by means of the high resistances used in its self-balancing paraphase splitter, which would also likely suppress its phase shift contribution.
The other two networks are in the 1-10Hz range, which is where the output transformer resonance would also sit, so the amp may well not be an exemplar of how to manage low frequency stability.
What I did find interesting in the article was the commentary on a need for close-coupling of the UL winding in the output transformer to achieve improved HF response, and the likely benefit that may accrue from using a higher turns ratio tapping (as compared to the typical aim of a UL tapping derived just from distortion minimisation). I would surmise that they ran into some instances of HF instability with certain UL output transformers of that era, given they went to the effort to identify three commercial offerings that must have been tested as acceptable.
There was also a related glib comment about using a 12AU7 driver due to its low plate resistance to keep high frequency RC phase shift above 50kHz, but no mention was made of the elephant in the room of the choice to use a 10k grid-stopper for the KT88.
As an FIY, I did not want to delve into the high-side measurement facet of my design in this post, but just the audio part of it.
So for those concerned about safety, the plate resistors are not for manual measurement. It is part of a high-side measurement circuit that goes back to the microprocessor for measuring plate current (this circuit is currently working). In order to correctly measure anode current, the measuring circuit needs to be in the anode. Putting it in cathode is simpler, but the the sum of the anode and screen currents will be captured.
But yes, using a high-side shunt for measuring bias current is dangerous and would not recommend it for manual measurement. I validated the circuit by using my battery powered meter and oscilloscope. A mains powered meter's isolation transformer may not have robust enough insulation for tube plate duty.
So for those concerned about safety, the plate resistors are not for manual measurement. It is part of a high-side measurement circuit that goes back to the microprocessor for measuring plate current (this circuit is currently working). In order to correctly measure anode current, the measuring circuit needs to be in the anode. Putting it in cathode is simpler, but the the sum of the anode and screen currents will be captured.
But yes, using a high-side shunt for measuring bias current is dangerous and would not recommend it for manual measurement. I validated the circuit by using my battery powered meter and oscilloscope. A mains powered meter's isolation transformer may not have robust enough insulation for tube plate duty.