Hi Zigzagflux,
Have a great vacation.. Awaiting further reports upon your return.
Given that all of my amplifier designs except one in recent years have been fixed bias designs I have to strongly agree with 45's suggestion to try fixed bias - if nothing else you will eliminate the odd interactions between the cathode bypass cap and the opt. Depending on the cap you are using the HF performance of that cap may be nowhere as good as you think as well. In fact in some instances a no cap approach is better than any cap, regardless of how good, and IMO this is one such instance..
I had forgotten that built my ST-70 variant, thanks for the pleasant reminder!
Have a great vacation.. Awaiting further reports upon your return.
Given that all of my amplifier designs except one in recent years have been fixed bias designs I have to strongly agree with 45's suggestion to try fixed bias - if nothing else you will eliminate the odd interactions between the cathode bypass cap and the opt. Depending on the cap you are using the HF performance of that cap may be nowhere as good as you think as well. In fact in some instances a no cap approach is better than any cap, regardless of how good, and IMO this is one such instance..
I had forgotten that built my ST-70 variant, thanks for the pleasant reminder!
Hello,
I've been running a PP 300B IT coupled amp with DHT input and driver with Buds excellent OPTs (level 1) for a few years now.
This last weekend I substituted the output stage common bias resistor and associated bypass C with the CCS found at Gary Pimm's web-site - heatsink Q2.
Probably the single most positive change that I've heard in a while, very obvious and for the better... you might try it.
Shane
I've been running a PP 300B IT coupled amp with DHT input and driver with Buds excellent OPTs (level 1) for a few years now.
This last weekend I substituted the output stage common bias resistor and associated bypass C with the CCS found at Gary Pimm's web-site - heatsink Q2.
Probably the single most positive change that I've heard in a while, very obvious and for the better... you might try it.
Shane
Your experiences underscore the need for me to breadboard my amp before committing to chassis layout.
Another thing I'd like to try is a comparison between the 5687 > 45 & triode strapped D3A (PP still into the IT). The D3a nets you roughly the same gain, similar rp and 2/3rds the standing current. The miller penalty is mild. D3A is said to be quite linear.
The truth is in the testing. Would the D3A be able to compete with the two stage input, both sonically and objectively? It has the advantage of simplicity and lack of interactions of two stages and additionally is one less interstage.
None of that is related to your problems, so I apologize for the tangent, but it is related to this design.
I'd also be really curious as to how the fixed biased compares to the original.
Another thing I'd like to try is a comparison between the 5687 > 45 & triode strapped D3A (PP still into the IT). The D3a nets you roughly the same gain, similar rp and 2/3rds the standing current. The miller penalty is mild. D3A is said to be quite linear.
The truth is in the testing. Would the D3A be able to compete with the two stage input, both sonically and objectively? It has the advantage of simplicity and lack of interactions of two stages and additionally is one less interstage.
None of that is related to your problems, so I apologize for the tangent, but it is related to this design.
I'd also be really curious as to how the fixed biased compares to the original.
Back from Mexico. Wished I could stay. Water was pleasantly warm. We spent a day at both Grand Cayman and also Cozumel. Very easy for a old whimp like me to go all the way into the Gulf, instead of the typical up-to-the-knees. Weather perfect every day, upper 70's and plenty of sun. Someone remind me why I live in Wisconsin again ?
Wired up the LL1620 and repeated the measurements. Definitely found some improvement, but as others had suggested, this behavior at high frequencies may be somewhat normal. I do confess I have not messed around with winding configurations this time; I simply followed the manufacturer's recommended wiring being overly anxious to see how performance looked. Maybe I can improve on this further with some experimentation, not sure.
Shot of 300b plates at 4W output (original ONetics xfmr):
Shot of same with LL1620:
Now it does look like the output is less than 4W with this capture, but due to the lower distortion, less drive was needed to obtain 4W output.
300b plates at 4W output (original ONetics):
Shot of same with Lundahl:
Then an FFT of the plate voltage (ONetics):
Same with Lundahl:
Finally a shot of 8W output (ONetics):
And Lundahl:
So it appears there is a degree of distortion with both output transformers, but the Lundahl is clearly less. While some of these measurements were being made, I would check the voltage across the cathode bypass cap, to see if there was any signal voltage being dropped across it, which would indicate to me some interaction or rising impedance with frequency. I could not measure any increased voltage, so it would appear the affect of the cap is minimal in this case?
There are a number of suggestions I have heard and would certainly like to consider (fixed bias, differential, etc) but for various reasons I am somewhat limited to the design at hand. Foremost, my power supply for the 300b is fixed at 500V, and I would have to make undesirable changes to the power supply chassis to address this. So for now, I will have to remain in cathode bias, sorry.
One other observation I found was that the hum nulling circuit causes some odd wiggle of the grid voltage to the 46 tube. But only with the Lundahl IT; no obvious wiggle with the ONetics. So for now, I have the hum nulling out of the circuit; hum is 15mV at the speaker terminals without it, which is tolerable for me; my speakers don't put out a ton of bass anyway.
I will wait to hear some comments before I finish up with some of my thoughts and conclusions, and possibly some direction for me in the future.
What do y'all think? Soundwise, again in mono, it sounds good, but I wouldn't say it blows me away. Just a good amp, that's how I would summarize it. There's got to be something I'm missing that Lynn's amp can do well.
Thanks !!
Wired up the LL1620 and repeated the measurements. Definitely found some improvement, but as others had suggested, this behavior at high frequencies may be somewhat normal. I do confess I have not messed around with winding configurations this time; I simply followed the manufacturer's recommended wiring being overly anxious to see how performance looked. Maybe I can improve on this further with some experimentation, not sure.
Shot of 300b plates at 4W output (original ONetics xfmr):
An externally hosted image should be here but it was not working when we last tested it.
Shot of same with LL1620:
An externally hosted image should be here but it was not working when we last tested it.
Now it does look like the output is less than 4W with this capture, but due to the lower distortion, less drive was needed to obtain 4W output.
300b plates at 4W output (original ONetics):
An externally hosted image should be here but it was not working when we last tested it.
Shot of same with Lundahl:
An externally hosted image should be here but it was not working when we last tested it.
Then an FFT of the plate voltage (ONetics):
An externally hosted image should be here but it was not working when we last tested it.
Same with Lundahl:
An externally hosted image should be here but it was not working when we last tested it.
Finally a shot of 8W output (ONetics):
An externally hosted image should be here but it was not working when we last tested it.
And Lundahl:
An externally hosted image should be here but it was not working when we last tested it.
So it appears there is a degree of distortion with both output transformers, but the Lundahl is clearly less. While some of these measurements were being made, I would check the voltage across the cathode bypass cap, to see if there was any signal voltage being dropped across it, which would indicate to me some interaction or rising impedance with frequency. I could not measure any increased voltage, so it would appear the affect of the cap is minimal in this case?
There are a number of suggestions I have heard and would certainly like to consider (fixed bias, differential, etc) but for various reasons I am somewhat limited to the design at hand. Foremost, my power supply for the 300b is fixed at 500V, and I would have to make undesirable changes to the power supply chassis to address this. So for now, I will have to remain in cathode bias, sorry.
One other observation I found was that the hum nulling circuit causes some odd wiggle of the grid voltage to the 46 tube. But only with the Lundahl IT; no obvious wiggle with the ONetics. So for now, I have the hum nulling out of the circuit; hum is 15mV at the speaker terminals without it, which is tolerable for me; my speakers don't put out a ton of bass anyway.
I will wait to hear some comments before I finish up with some of my thoughts and conclusions, and possibly some direction for me in the future.
What do y'all think? Soundwise, again in mono, it sounds good, but I wouldn't say it blows me away. Just a good amp, that's how I would summarize it. There's got to be something I'm missing that Lynn's amp can do well.
Thanks !!
Glad you're back! Here's a quick thought - what is the quiescent current going through each 300B? I found that I was still getting some Class AB behavior at 60 mA per device - but that 80 mA got rid of it. Yes, I know that's a lot of current, but the distortion data was clear. The 3rd and higher harmonics dropped really fast in the 60-85 mA region, to my surprise. It changed much less in a SET amplifier with the same output stage and tubes.
Now, I don't know how pentodes vs triodes handle the AB transition, and what bias currents they like to see, but typical pentode Class AB amplifiers are run around 35-40 mA per tube. I was expecting the generic 300B SET figure of 60 mA would be adequate to remove Class AB artifacts, but in fact it wasn't enough. In practice, this meant derating the B+ voltage, or using super-300B type tubes with high-wattage plate ratings. I got good results from beating up the Chinese tubes, though - they seemed to take the abuse just fine - something I'd never do to a NOS tube.
As for sonic preferences, what can I say. I've gotten to a point where listening to Class AB pentode amplifiers with 6~20 dB of feedback isn't something I enjoy that much. I prefer either all-triode DHT SET or Class A DHT PP. But that's just me. Other people like other topologies - OTL with huge amounts of local feedback, FET/MOSFET/tube hybrids, and the classic (Class AB pentodes with feedback) amplifiers from the Fifties.
There are certainly some speakers that seem to demand high-powered Class AB pentode with feedback or 60~200 watt Class AB transistor amplifiers - they just don't work very well with SETs or low-powered Class A PP amplifiers. So it's not entirely a listener thing - speakers play a role as well.
Now, I don't know how pentodes vs triodes handle the AB transition, and what bias currents they like to see, but typical pentode Class AB amplifiers are run around 35-40 mA per tube. I was expecting the generic 300B SET figure of 60 mA would be adequate to remove Class AB artifacts, but in fact it wasn't enough. In practice, this meant derating the B+ voltage, or using super-300B type tubes with high-wattage plate ratings. I got good results from beating up the Chinese tubes, though - they seemed to take the abuse just fine - something I'd never do to a NOS tube.
As for sonic preferences, what can I say. I've gotten to a point where listening to Class AB pentode amplifiers with 6~20 dB of feedback isn't something I enjoy that much. I prefer either all-triode DHT SET or Class A DHT PP. But that's just me. Other people like other topologies - OTL with huge amounts of local feedback, FET/MOSFET/tube hybrids, and the classic (Class AB pentodes with feedback) amplifiers from the Fifties.
There are certainly some speakers that seem to demand high-powered Class AB pentode with feedback or 60~200 watt Class AB transistor amplifiers - they just don't work very well with SETs or low-powered Class A PP amplifiers. So it's not entirely a listener thing - speakers play a role as well.
zigzagflux said:
As is, your amp looks unbalanced to me. Too high 2nd harmonic for a class A push-pull at a fraction of its rated power. It is true you are monitoring a quite demanding point (a rather high frequency). To be sure about that you should check for the same power level (namely, for the same input signal) at 1KHz and 100Hz.
The distortion of your amp does not surprise me as you did not consider any possibility for dynamic balancing. The only possible calibration I can see in your design is the cathode of the input stage. Remove the current source and use a classical cathode resistor. The total resistance will be a separate (bypassed) resistor for each side and a common pot (approx. 15-20% of the total value) with central pin grounded. I know it is local Fb but I cannot see another way and you will have to find a compromise anyway.
Choosing tubes with very high degree of matching is very difficult and expensive and above all is no guarantee of permanent balance because each tube will grow old its own way....
45
Tweeker said:
Higher standing current (in a Class A PP amplifier) also has a really nice payoff - much less sensitivity to load. As a loudspeaker guy, this is a win-win-win for me. Who cares if it puts out 20~30% less power (thanks to lower B+) if you can use a broader range of speakers?
That's the real reason I like transformer primaries on the highish side (6~8 K) and high operating currents (70~85 mA) - the amp as a whole is much less sensitive to the crazy gyrations of speaker impedance. And the best speakers have a bad habit of the most zany impedance curves.
The opposite would be a SET running at moderate currents and lowish primary values - with the right speaker, magic, with the wrong speaker, not so nice.
Feedback squashes out all of these differences, but the distortion spectra changes in character, so the amp as a whole sounds different. But if the starting point is pentodes, well, feedback makes a lot of sense for many reasons. Horses for courses, as the Brits say.
How true it is.
So after completing this project, I should probably close out the thread and update everyone on the end result. Nice to know, after so many posts and reads, that there was resolution. Indeed, in this case there was definitive resolution.
A lot of very good input by all, which is much appreciated. There are not many avenues to which one can go to get solid advice on a challenging project such as this. DIYAudio has been a life-saver for me more than once! I would like to summarize by reflecting on some pertinent posts, and provide confirmation of the advice. There were, at the end of the day, two fundamental issues that had produced the poor performance of this amplifier, and one misconception that had to be purged from my mind.
First issue was the interstage transformers. I had tried many different things to attempt to both eliminate some rising response, as well as correct some very serious phase shift issues at higher frequencies. To little avail:
I had tried tubes with Rp’s ranging from 1K to 2K, single / parallel, various winding configurations, with really very little difference in performance across the IT. Maybe it's just me, but I could not for the life of me get improvement in the IT's, especially regarding high frequency phase shift. I basically gave up.
Per K&K Audio’s suggestion, I resorted to the LL1692A, designed specifically for Rp’s in the range I was dealing with. Measured performance was extremely encouraging, but since I was purchasing these 4 units one at a time (cautious upgrades and monetary considerations), I never got a good chance to hear a completed version in stereo until only last night. Alas, these are not amorphous core !
Second issue is the measured distortion on the output stage at lower power levels. Grid drive to the output stage looked very good, plate voltages did not. Purchased a LL1620 output transformer to compare to the existing O-Netics. While I was able to get more power out of the Lundahl, the basic behavior was similar, just less pronounced.
Decreased the cathode resistor in the self-bias output stage, and that significantly improved performance at the higher power levels. Running about 71mA per 300B at 410V plate. A very welcome improvement, thanks !
Last item was a misconception on my part. Was seeing decreasing performance over 10k when running in excess of 4-5 watts.
This was right on the money. It is not practical to expect a 16W amp to produce 5W at 20 kHz. I lessened my expectations in this area, and put down the scope.
With both amps upgraded to Lundahl IT’s (kept the O-Netics OPT’s) and no zobels anywhere, I had a listening session last night. If I may quote one more time:
Yep. There we go. I didn’t need to swap in the ST70 to do A/B testing anymore. There was no question I was getting what I had expected. Very glorious sound, amazing detail and soundstage (insert audiophile superlatives here). Couldn’t stop listening; I have never quite heard cymbals sound so real before !
To provide two tangible examples, there is a divergence in the Beatles’ Octopus’ Garden where they are simulating being underwater. You can hear a man’s voice gurgling; he moves from stage left to stage right. Very distinct, but I never heard it before.
Another example is Fleetwood Mac, Dreams. When Stevie sings her own background vocal, she is in a different location than her lead vocal. Listened to this song hundreds of times, never caught that.
Far as I’m concerned, this project is done. Maybe in a few years I might toss in some amorphous core, but for now, it’s time to sit back and enjoy. Been 3-4 years in the making. Thank you everyone for your help and advice. What a wealth of information here ! If anyone is interested in measurements or futher info, feel free to let me know.
With no feedback there is nowhere for any subjective colorations to hide - even very small colorations are obvious and gross. In addition, with Class A PP throughout, there isn't the wonderful warm bath of 2nd-harmonic distortion of a SET amplifier. You can have a fair amount of 3rd-harmonic distortion, and if there's enough 2nd present, the 3rd just gives a pleasing "bite" to the sound. 5th, 6th, 7th, and higher just sound harsh, though, so I try and find tubes that have very low upper-harmonic content.
The combination of the two characteristics makes for an extremely transparent but also brutally unforgiving topology - fortunately, it doesn't have that effect on source material, it's actually quite forgiving that way - but small errors in circuit or power-supply design are shown in sharp relief.
It's useful to keep in mind this amplifier, with no local or global feedback, has less distortion at 1 watt than a Golden Age Class AB pentode amplifier with 10-20 dB of feedback. The analogy I use is: When the tide goes out, you see all kinds of creatures clinging to the rocks that you didn't know were there before.
PS coloration is a deep subject all by itself. With a SET amplifier, the audio-frequency current modulations appear directly on the supply, which means you hear both the noise and parts colorations unmodified. With Class A PP, the audio-frequency current modulations are multiplied twice in frequency, and essentially full-wave-rectified. This isn't such a nice-sounding modulation as the SET, which is the same as the music itself.
So despite the measured noise figures being lower - which is due to PP cancellation - the sound of the noise, and audio-frequency modulations, are much worse and much less desirable as a residual coloration. That's the real reason for the shunt regulation and putting the bypass caps on physically very short current-loop paths between cathode and plate circuits. Much of the subjective tuning of the amplifier comes down to little more than details of the power supply and its physical wiring and layout. Attention to RF-style noise rejection techniques (small loop areas) and instrumentation-style grounding provides major benefits here.
The current direction for further improvement (whenever I conclude the Beyond the Ariel project) is a better filament supply that is immune to AC line noise, while also taking advantage of the inherent 20~30 dB noise rejection of a balanced DHT tube, and also having low subjective coloration. The attempt to use an off-the-shelf 3-pin DC regulator was a complete failure, with gross colorations compared to conventional AC heating. Pre-filtering the AC going into the primaries of filament transformer, though, does give cleaner and sweeter sound.
The combination of the two characteristics makes for an extremely transparent but also brutally unforgiving topology - fortunately, it doesn't have that effect on source material, it's actually quite forgiving that way - but small errors in circuit or power-supply design are shown in sharp relief.
It's useful to keep in mind this amplifier, with no local or global feedback, has less distortion at 1 watt than a Golden Age Class AB pentode amplifier with 10-20 dB of feedback. The analogy I use is: When the tide goes out, you see all kinds of creatures clinging to the rocks that you didn't know were there before.
PS coloration is a deep subject all by itself. With a SET amplifier, the audio-frequency current modulations appear directly on the supply, which means you hear both the noise and parts colorations unmodified. With Class A PP, the audio-frequency current modulations are multiplied twice in frequency, and essentially full-wave-rectified. This isn't such a nice-sounding modulation as the SET, which is the same as the music itself.
So despite the measured noise figures being lower - which is due to PP cancellation - the sound of the noise, and audio-frequency modulations, are much worse and much less desirable as a residual coloration. That's the real reason for the shunt regulation and putting the bypass caps on physically very short current-loop paths between cathode and plate circuits. Much of the subjective tuning of the amplifier comes down to little more than details of the power supply and its physical wiring and layout. Attention to RF-style noise rejection techniques (small loop areas) and instrumentation-style grounding provides major benefits here.
The current direction for further improvement (whenever I conclude the Beyond the Ariel project) is a better filament supply that is immune to AC line noise, while also taking advantage of the inherent 20~30 dB noise rejection of a balanced DHT tube, and also having low subjective coloration. The attempt to use an off-the-shelf 3-pin DC regulator was a complete failure, with gross colorations compared to conventional AC heating. Pre-filtering the AC going into the primaries of filament transformer, though, does give cleaner and sweeter sound.
Lynn:
I filter the AC feeding the filament transformer with a Sola ferroresonant transformer. While it does show some 5th, 7th, and 11th harmonics in its spectrum, there is essentially nothing higher that I can measure. The noise rejection on these units is extremely good; 120dB common mode and 60dB transverse. The regulation is of course another benefit, which I then step down to about 108V. With this method, I can get my 120Hz residual down to about 1 mV rms, which is very quiet and unobtrusive on my system. The tweeter is almost completely silent, and the woofer just a gentle rolling rumble. I can't say I yearn for a quieter system, to be honest.
I've even considered using a clean sine wave driving a class D amplifier to produce my needed 108V, to see how a 'perfect' sine wave would improve the hum. I have the sine source and output tranny, but haven't gone ahead with the amp. I suspect the bigger problem with that idea is noise, rather than harmonic spectrum. Might be a fun experiment, at least.
A very fine design you came up with. I wouldn't expect such transparency out of sooo much iron in the signal path, but it's there. Just have to choose wisely; as you say, this IS a very unforgiving design.
I filter the AC feeding the filament transformer with a Sola ferroresonant transformer. While it does show some 5th, 7th, and 11th harmonics in its spectrum, there is essentially nothing higher that I can measure. The noise rejection on these units is extremely good; 120dB common mode and 60dB transverse. The regulation is of course another benefit, which I then step down to about 108V. With this method, I can get my 120Hz residual down to about 1 mV rms, which is very quiet and unobtrusive on my system. The tweeter is almost completely silent, and the woofer just a gentle rolling rumble. I can't say I yearn for a quieter system, to be honest.
I've even considered using a clean sine wave driving a class D amplifier to produce my needed 108V, to see how a 'perfect' sine wave would improve the hum. I have the sine source and output tranny, but haven't gone ahead with the amp. I suspect the bigger problem with that idea is noise, rather than harmonic spectrum. Might be a fun experiment, at least.
A very fine design you came up with. I wouldn't expect such transparency out of sooo much iron in the signal path, but it's there. Just have to choose wisely; as you say, this IS a very unforgiving design.
Thanks!!
I will surely refer to the original schematics of the Karna, as they also demonstrates the grounding techniques, etc.
Your schematic rose another question, referring to the CCS. I saw you apparently went with the easier solution of employing depletion mode mosfets for the CCS (instead of the more complex enhancement units as used by GP). Which chip did you use, the 10M45S, the DN2540, or other? Have you used the cascoded version, or the version with only 1 chip? I have been playing around with cascoded 10M45S, but could not achieve very high current levels (as required in the Karna). Single chip CCS will do it, however...
Cheers, Erik
I will surely refer to the original schematics of the Karna, as they also demonstrates the grounding techniques, etc.
Your schematic rose another question, referring to the CCS. I saw you apparently went with the easier solution of employing depletion mode mosfets for the CCS (instead of the more complex enhancement units as used by GP). Which chip did you use, the 10M45S, the DN2540, or other? Have you used the cascoded version, or the version with only 1 chip? I have been playing around with cascoded 10M45S, but could not achieve very high current levels (as required in the Karna). Single chip CCS will do it, however...
Cheers, Erik
For those of you interested in using a Constant Voltage Ferroresonant transformer, use only a CV S rather than a CV T. The CV S puts out a sine wave and the CV T puts out a square wave. Keep in mind that both throw off an immense amount of stray field energy, very close to 5 %, rather than the typical 0.1% of an E/I transformer.
Bud
Bud
ErikdeBest said:
I have successfully used the DN2540 in cascode at currents up to about 80mA. I know some people (Tubelab, I think) really like the 10M45S. They do behave a little differently, but both seem to work well. How much current are you trying to push? I would have to do a search to find some recommendations as to a better device (lower voltage, lower Rds) for the bottom current setting device, which then drives the upper DN2540 for the voltage capability. This is supposed to give you a little more extension.
To be honest, I'm not convinced you NEED a cascode for the power supply CCS's anyway. Lynn originally used a simple resistive dropping network to feed the gas tubes, and said they worked just fine. So a "perfect" CCS probably isn't critical for the power supplies I would guess.
BudP said:
Hi Bud. They certainly do. I played around with my E/M field strength meter, and also looked at hum pickup on the IT and OPT's. I found I needed about 18 inches of clearance to completely eliminate pickup. Orientation didn't seem to matter, either. The field was just spilling all over the place. You also need to put the crazy things under a box, the radiated audible noise is so high. To be honest, they're a pain in the butt, but they do regulate and isolate very well.
Hi Zigzagflux
Thanks for the information! I have a couple of the DN2540 IC's as well, but have not played around with them, yet. It is good to know a cascoded version can handle 80mA (I don't even need that much anyway!)
I am certainly not in the position of even guessing about the need for a cascoded version in a power supply, and its possible advantages towards a simple version, so I can't comment...
Cheers, Erik
Thanks for the information! I have a couple of the DN2540 IC's as well, but have not played around with them, yet. It is good to know a cascoded version can handle 80mA (I don't even need that much anyway!)
I am certainly not in the position of even guessing about the need for a cascoded version in a power supply, and its possible advantages towards a simple version, so I can't comment...
Cheers, Erik
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.