Just completed the following amp build, basically a Lynn Olson Karna amp with a few differences. This is my first non-GNFB build, so I have some questions about performance.
http://www.just4sheep.com/public/schematic.pdf
Things appear to operate well statically; hum is between 400uV and 1.5mV with AC heating throughout; pretty happy about that. All DC idle conditions are appropriate, and the amp puts out close to 30W at 1kHz at the onset of visual distortion.
Sound-wise, haven't been all that happy. Midrange is truly excellent, no complaints there. Hearing a lot of details I haven't noticed in the past. Midbass is poorer than my ST70, although I am still experimenting with the transformer coupled HPF at the input.
Biggest concern is in the high frequencies. Washed out, and audibly quieter than expected. Cymbals are just not there.
Did some measurements to see what was going on. For the purposes of eliminating a possible issue, the 33nF HPF cap is BYPASSED for these measurements. Following is a simplistic gain/phase plot:
http://www.just4sheep.com/public/response.pdf
So not only do I have a rising response with frequency, but the phase shift is waaaay out there (output lagging input). Did some further testing to see where the shift occurs.
20kHz input, loaded with 4 ohm noninductive resistor (all phase shifts relative to input, measurements are differential):
at grids of 12B4: 0 degrees
at plates of 12B4: could not obtain safely
at grids of 46: -39 degrees
at plates of 46: -73.4 degrees
at grids of 300B: -78.5 degrees
at plates of 300B: -90.7 degrees (measurements taken through decoupling 68nF orange drops)
at speaker terminals: -115 degrees
I am surmising most of the phase shift is occurring at the differential stage, and not across the interstage transformers. Output transformer does have around 25 degrees of lag.
One other observation: With a 20kHz input, the voltage at the grid of the 300B's is very clean. However, starting at about 5W output, the 300B stage begins to visibly distort (seen at the plates). Grid stays clean, tubes are distorting. With only 5W output? Clean output at 1kHz, distorted at 20kHz.
ideas ???
Power transformers are Electraprint, audio/filament transformers are O-Netics, input transformer is Tribute. 300B's are JJ.
http://www.just4sheep.com/public/schematic.pdf
Things appear to operate well statically; hum is between 400uV and 1.5mV with AC heating throughout; pretty happy about that. All DC idle conditions are appropriate, and the amp puts out close to 30W at 1kHz at the onset of visual distortion.
Sound-wise, haven't been all that happy. Midrange is truly excellent, no complaints there. Hearing a lot of details I haven't noticed in the past. Midbass is poorer than my ST70, although I am still experimenting with the transformer coupled HPF at the input.
Biggest concern is in the high frequencies. Washed out, and audibly quieter than expected. Cymbals are just not there.
Did some measurements to see what was going on. For the purposes of eliminating a possible issue, the 33nF HPF cap is BYPASSED for these measurements. Following is a simplistic gain/phase plot:
http://www.just4sheep.com/public/response.pdf
So not only do I have a rising response with frequency, but the phase shift is waaaay out there (output lagging input). Did some further testing to see where the shift occurs.
20kHz input, loaded with 4 ohm noninductive resistor (all phase shifts relative to input, measurements are differential):
at grids of 12B4: 0 degrees
at plates of 12B4: could not obtain safely
at grids of 46: -39 degrees
at plates of 46: -73.4 degrees
at grids of 300B: -78.5 degrees
at plates of 300B: -90.7 degrees (measurements taken through decoupling 68nF orange drops)
at speaker terminals: -115 degrees
I am surmising most of the phase shift is occurring at the differential stage, and not across the interstage transformers. Output transformer does have around 25 degrees of lag.
One other observation: With a 20kHz input, the voltage at the grid of the 300B's is very clean. However, starting at about 5W output, the 300B stage begins to visibly distort (seen at the plates). Grid stays clean, tubes are distorting. With only 5W output? Clean output at 1kHz, distorted at 20kHz.
ideas ???
Power transformers are Electraprint, audio/filament transformers are O-Netics, input transformer is Tribute. 300B's are JJ.
The leakage inductance (or stray capacitance) in each of the interstage transformers is what is causing the HF phase shift you are seeing in each of your amplifier stages, Fisher and others installed caps across the primary to secondary windings (plate to grid) to reduce the effect of leakage inductance on the HF phase and frequency response. (You could try something like 0.022uF or larger since you are using 1+1:1+1 interstages - this might work reasonably well.)
Sometimes a little resistive loading on the secondary of the IT will help with the HF response issues.
It is highly likely that capacitance in the output transformer is responsible for the bad waveforms you are seeing, practically speaking you do not need much power response at 20kHz, try checking it at 15kHz where hopefully it is at least 6 dB greater than at 20kHz.
The design I published in VTV 10 yrs ago had a usable bandwidth of 30kHz and was mostly limited by the effects of miller capacitance in the output stage. Leakage inductance in the output transformer also contributed.
Generally I think more than one interstage in the signal path is design overkill and now you probably know why...
Somewhere in there there is also some ultrasonic peaking going on - you should try to establish what stage(s) are responsible.
Sometimes a little resistive loading on the secondary of the IT will help with the HF response issues.
It is highly likely that capacitance in the output transformer is responsible for the bad waveforms you are seeing, practically speaking you do not need much power response at 20kHz, try checking it at 15kHz where hopefully it is at least 6 dB greater than at 20kHz.
The design I published in VTV 10 yrs ago had a usable bandwidth of 30kHz and was mostly limited by the effects of miller capacitance in the output stage. Leakage inductance in the output transformer also contributed.
Generally I think more than one interstage in the signal path is design overkill and now you probably know why...
Somewhere in there there is also some ultrasonic peaking going on - you should try to establish what stage(s) are responsible.
zigzagflux said:
I agree with your assessment.
What is the actual idle current in the output stage? I use JJ300B in both my 300B SE amp and the PP 300B mentioned in a previous post- they perform well. Take a look at the ac signal voltage appearing across the cathode bypass capacitor, shouldn't be much, but if there is try some additional bypassing.
Thanks Kevin:
Won't be the first time you come to the rescue.
For starters, I appreciate your feelings about too much iron; you're not the first one to give that advice. Maybe I'll a little thick headed, but I still believe this can work. I've heard the same number of people swear by xfmr coupling as those that avoid it. I must be somewhere in the middle
I'll do some measurements in the next few hours in accordance with your previous post. The statements about leakage inductance got me thinking, and I went back to Morgan Jones on transformers. He says to expect an 18dB/octave LPF based on the characteristics of a transformer, the frequency dependent on transformer design. Also has some suggestions regarding loading and/or zobels to improve performance.
I think the O-Netics are a good tranny, so I'll continue to burn in the amp (currently have about 10 hours on them) and see what happens in about 20 hours. I will swear by a break-in period with some Lundahl's I have (love them). Who knows, these may require the same.
Quickly checked the AC voltage across the cathode cap. It's essentially immeasurable from 1kHz to 20kHz; no change that I can see. The search continues.....
Won't be the first time you come to the rescue.
For starters, I appreciate your feelings about too much iron; you're not the first one to give that advice. Maybe I'll a little thick headed, but I still believe this can work. I've heard the same number of people swear by xfmr coupling as those that avoid it. I must be somewhere in the middle
I'll do some measurements in the next few hours in accordance with your previous post. The statements about leakage inductance got me thinking, and I went back to Morgan Jones on transformers. He says to expect an 18dB/octave LPF based on the characteristics of a transformer, the frequency dependent on transformer design. Also has some suggestions regarding loading and/or zobels to improve performance.
I think the O-Netics are a good tranny, so I'll continue to burn in the amp (currently have about 10 hours on them) and see what happens in about 20 hours. I will swear by a break-in period with some Lundahl's I have (love them). Who knows, these may require the same.
Quickly checked the AC voltage across the cathode cap. It's essentially immeasurable from 1kHz to 20kHz; no change that I can see. The search continues.....
Hmm, those are not very happy measurements. Some quick questions - what is the Rp of the triode-connected 46? The 45 PP driver was specifically chosen for low distortion and low output impedance, as was the previous version using 5687, 7044, and 7119 drivers.
I was not entirely happy with the bandwidth of the O-Netics interstages. The Lundahls were going out to 80 khz without any peaking, while the O-Netics were making it out to 25 kHz with about 3 dB of peaking. I like the sound of the O-Netics better, but I've never tried the amorphous-core Lundahls, which many people say are very good.
The amplifier must be flat to 20 kHz at a minimum, with 50 khz a desirable target. RC loading of the secondaries on the interstages does remove the overshoot and peaking, but it also loads the tubes much more heavily at the highest frequencies. I found it sounded worse, but then my amplifiers didn't have the severe overshoots that your amplifier is showing.
The best way to optimize the impulse response is with a scope and a fast-risetime square wave. This can be injected at any stage of the amplifier, and the bad actors found. I'd start with the grids of the output tubes, work your way backward, and find out where the overshoot is coming from.
Since there is no global feedback and no shared grounds, it is easy to isolate and debug the amplifier stage by stage. If any given stage is the cause of the trouble, there isn't much to change except the tubes and the transformers.
Another quick question - what is the purpose of the 33 nF input cap on the primary of the input transformer? Do you have sources with several milliamps of DC offset coming from them? If so, you'll be hearing crackles as the volume control is rotated.
The reason I mention this is that small-value caps resonate with the transformer inductance and you can end up with very large LF peaking in the 10~50 Hz region. One of the annoying things about parallel-feed transformer coupling is that the required coupling cap almost always has to be 5 uF or larger to avoid the subsonic peaking - and these caps don't usually sound very good.
I would also remove the 0.068 uF caps shunting the VR tubes, at least during the debugging phase.
I'd also add 20~40 uF of capacitance between the virtual cathodes of the driver stage and the B+ of the driver stage and replace the negative-supply current source with a simple resistor to ground; what you have now is differential drive, and this has substantially higher 3rd-order harmonic content (2 to 3 times) and far less peak current drive capability than Class A PP (which is what the bypass cap gives you).
This one change alone might improve the frequency response and distortion substantially. (I'd also do it for the input stage as well; the improvement might actually be greater, but the driver stage MUST be Class A PP, not differential. I've tried differential, and the performance is substantially poorer in every respect - more distortion, less available current, higher Zout, and sounds worse, too.)
You can do this quickly with clip-leads; once you audition and measure Class A PP versus differential drive, I don't think you'll be going back.
I was not entirely happy with the bandwidth of the O-Netics interstages. The Lundahls were going out to 80 khz without any peaking, while the O-Netics were making it out to 25 kHz with about 3 dB of peaking. I like the sound of the O-Netics better, but I've never tried the amorphous-core Lundahls, which many people say are very good.
The amplifier must be flat to 20 kHz at a minimum, with 50 khz a desirable target. RC loading of the secondaries on the interstages does remove the overshoot and peaking, but it also loads the tubes much more heavily at the highest frequencies. I found it sounded worse, but then my amplifiers didn't have the severe overshoots that your amplifier is showing.
The best way to optimize the impulse response is with a scope and a fast-risetime square wave. This can be injected at any stage of the amplifier, and the bad actors found. I'd start with the grids of the output tubes, work your way backward, and find out where the overshoot is coming from.
Since there is no global feedback and no shared grounds, it is easy to isolate and debug the amplifier stage by stage. If any given stage is the cause of the trouble, there isn't much to change except the tubes and the transformers.
Another quick question - what is the purpose of the 33 nF input cap on the primary of the input transformer? Do you have sources with several milliamps of DC offset coming from them? If so, you'll be hearing crackles as the volume control is rotated.
The reason I mention this is that small-value caps resonate with the transformer inductance and you can end up with very large LF peaking in the 10~50 Hz region. One of the annoying things about parallel-feed transformer coupling is that the required coupling cap almost always has to be 5 uF or larger to avoid the subsonic peaking - and these caps don't usually sound very good.
I would also remove the 0.068 uF caps shunting the VR tubes, at least during the debugging phase.
I'd also add 20~40 uF of capacitance between the virtual cathodes of the driver stage and the B+ of the driver stage and replace the negative-supply current source with a simple resistor to ground; what you have now is differential drive, and this has substantially higher 3rd-order harmonic content (2 to 3 times) and far less peak current drive capability than Class A PP (which is what the bypass cap gives you).
This one change alone might improve the frequency response and distortion substantially. (I'd also do it for the input stage as well; the improvement might actually be greater, but the driver stage MUST be Class A PP, not differential. I've tried differential, and the performance is substantially poorer in every respect - more distortion, less available current, higher Zout, and sounds worse, too.)
You can do this quickly with clip-leads; once you audition and measure Class A PP versus differential drive, I don't think you'll be going back.
zigzagflux
I too would add a 10 k termination to both IT's for measurement purposes. Listening might be another story, as the terminating resistor will eat up depth of field info badly. These IT's simply do not have enough distributed capacitance, or leakage inductance, to provide flat performance much beyond 20 K. In their defense, they were designed for audible characteristics and not test equipment characteristics.
Your findings on the OPT do surprise me. Typical phase performance has been flat to about 35 k with a gradual roll into lag from there. They do require about 40 hours to charge the dielectric circuit in the coil however and the audible changes from that charging are not subtle at all.
Might be a good idea to get Gary Pimm involved here too as he has a wealth of detail experience making the Karna amps sing sweetly.
Bud
I too would add a 10 k termination to both IT's for measurement purposes. Listening might be another story, as the terminating resistor will eat up depth of field info badly. These IT's simply do not have enough distributed capacitance, or leakage inductance, to provide flat performance much beyond 20 K. In their defense, they were designed for audible characteristics and not test equipment characteristics.
Your findings on the OPT do surprise me. Typical phase performance has been flat to about 35 k with a gradual roll into lag from there. They do require about 40 hours to charge the dielectric circuit in the coil however and the audible changes from that charging are not subtle at all.
Might be a good idea to get Gary Pimm involved here too as he has a wealth of detail experience making the Karna amps sing sweetly.
Bud
Hi Lynn, (and Zigzag)
Took me a moment to realize what you were saying in regards to differential vs class A pp, because obviously the differential stage is operating in class A as well. If I understand you correctly you are talking about eliminating the differential local feedback in the cathode circuit of each of the preceding stages by getting rid of the CCS and independently biasing each tube in the pair. (Or bypassing the common bias point so there is no common ac between the two devices.) Clearly the mechanism here is that the less perfect balance results in some residual 2nd harmonic and perhaps slightly less 3rd. It seems like a good suggestion and I would further suggest the use of the existing negative supplies to provide fixed bias to these stages in lieu of cathode bias to avoid the need for very good cathode bypass caps.
You have another degree of freedom in that if the problem is truly caused primarily by the transformer's leakage inductance you may be able to improve the overshoot performance considerably by raising the effective rp of the tubes driving the transformer - local cathode feedback (just an unbypassed cathode resistor) will do this.. Note as well you can do this with your current ccs just by adding resistors between the cathodes and ccs. (Technically with a really good ccs only one resistor is needed, but this bothers me from a semantic if not technical perspective.)
IIRC the 46 in triode connection has roughly an rp of 2 - 2.2K , but I could be substantially off.
Stray winding capacitance is less of an issue with low rp triodes than leakage inductance in many instances IMO, but I am unfamiliar with the specific transformers.
Zigzag keep us posted, and good luck!
Took me a moment to realize what you were saying in regards to differential vs class A pp, because obviously the differential stage is operating in class A as well. If I understand you correctly you are talking about eliminating the differential local feedback in the cathode circuit of each of the preceding stages by getting rid of the CCS and independently biasing each tube in the pair. (Or bypassing the common bias point so there is no common ac between the two devices.) Clearly the mechanism here is that the less perfect balance results in some residual 2nd harmonic and perhaps slightly less 3rd. It seems like a good suggestion and I would further suggest the use of the existing negative supplies to provide fixed bias to these stages in lieu of cathode bias to avoid the need for very good cathode bypass caps.
You have another degree of freedom in that if the problem is truly caused primarily by the transformer's leakage inductance you may be able to improve the overshoot performance considerably by raising the effective rp of the tubes driving the transformer - local cathode feedback (just an unbypassed cathode resistor) will do this.. Note as well you can do this with your current ccs just by adding resistors between the cathodes and ccs. (Technically with a really good ccs only one resistor is needed, but this bothers me from a semantic if not technical perspective.)
IIRC the 46 in triode connection has roughly an rp of 2 - 2.2K , but I could be substantially off.
Stray winding capacitance is less of an issue with low rp triodes than leakage inductance in many instances IMO, but I am unfamiliar with the specific transformers.
Zigzag keep us posted, and good luck!
To clarify the discussion of differential vs Class A PP: Differential is series operation; if one tube is pulled out, clips, or runs out of current, the whole stage shuts down, just like old-fashioned Christmas tree lights. By contrast, what I'm calling "Class A PP" operates in parallel; if one tube is pulled out, clips, or runs out of current, the other tube takes over, and keeps on going. The stage slides into Class AB if it needs to, while the differential stage just flat-out clips.
This may seem like semantics, but the transition into Class AB with vacuum tubes is actually surprisingly broad, not the 0.7V diode drop we see in transistor output stages, which hard-switch on and off. I was still seeing residual transition effects with many tens of volts of negative bias on the DHT grids, which is why the output-tube distortion is sensitive to drive impedance.
The impedance between the center-tap of the interstage (or output) transformer and the virtual cathodes of the pair of tubes is what controls the ratio between the two modes of operation. John Atwood made a detailed series of measurements by varying this impedance and did find with medium-distortion tubes the best place of operation was fairly close (but not the same as) as the traditional Class A PP mode - that is, parallel operation.
He also measured it again with lower distortion tubes in the DHT family and the optimum, lowest-distortion mode was the Class A PP flavor. When the potentiometer was moved away from the lowest-distortion point, what was significant was the higher-order distortion came up first - 3rd, 5th, 7th, all the ugly ones. In addition, and probably closely related to the prior finding, was the linear current delivery dropped rapidly as well. The ability to deliver linear current is closely related to the absence of high-order distortion content, and this is where the differential circuit falls down the worst.
Returning the previous post, the decision to switch the input and driver stage to pure differential operation is almost certainly the reason the amplifier is grossly underperforming compared to the Karna. Although I didn't mention this on the Web-page, the decision to avoid differential operation was based on a series of measurements and audition.
On direct A/B comparison, the differential circuit sounded thin, scrawny, and with "pinched" and closed-in HF, with poor dynamic reserves. Measurement confirmed this: distortion was several times higher, the proportion of upper harmonic to lower harmonics was unfavorable, and most important, peak current delivery was curtailed by severalfold. This was the real source of the problem: the driver stage was unable to provide peak current to the 300B grids when the output tubes were getting near Class A2 conditions or Miller capacitance + interstage stray capacitance were demanding more and more current from the driver at high frequencies. When it comes to driver stage design, it's all about linear current delivery, particularly at high frequencies.
This is not the same as just throwing in a cathode-follower; that lowers Zsource, but has NO effect on distortion or the ability to deliver peak current into a load with complex (and nonlinear) characteristics. In effect, the driver really needs to be a small power amplifier in its own right, or at least able to drive headphones.
One of the most revealing tests of the Karna was a test I did a while ago to see if I could slew the amplifier; although the HF rolls off above 40~50 kHz, there's nothing to stop me from increasing the input level. So I fed in a sinewave, increased the gain to just below clipping (16 watts output), and kept raising the frequency, and then level, keeping the amplifier just below clipping the whole time.
I finally lost my nerve at 500 kHz, just below the AM radio band - the sine wave was still looking very close to a sine-wave, no sign of flat-topping or starting to turn into triangles, and the amplifier was acting like a 16-watt radio transmitter putting out a carrier wave. I never did see slewing, although I shudder to think how much current was going into the plates of the driver stage (the driver was looking into a nearly pure reactance, so all the power was getting bounced right back into the driver-stage plates).
A test like this would destroy most transistor amplifiers, and I very much doubt a differential driver would have cheerfully slid into Class AB to deliver the amount of current the 300B grids were demanding.
So ditch the negative-voltage current sources for the input and drive stages, replace them with wirewound resistors, and shunt the cathodes to the B+ supply with a good-quality 40 uF oil cap. One thing to be careful of: the VR tubes do NOT like to see capacitive loads (they will oscillate), so either put the bypass caps between the cathodes and ground, or cathodes and B+, but not both.
This may seem like semantics, but the transition into Class AB with vacuum tubes is actually surprisingly broad, not the 0.7V diode drop we see in transistor output stages, which hard-switch on and off. I was still seeing residual transition effects with many tens of volts of negative bias on the DHT grids, which is why the output-tube distortion is sensitive to drive impedance.
The impedance between the center-tap of the interstage (or output) transformer and the virtual cathodes of the pair of tubes is what controls the ratio between the two modes of operation. John Atwood made a detailed series of measurements by varying this impedance and did find with medium-distortion tubes the best place of operation was fairly close (but not the same as) as the traditional Class A PP mode - that is, parallel operation.
He also measured it again with lower distortion tubes in the DHT family and the optimum, lowest-distortion mode was the Class A PP flavor. When the potentiometer was moved away from the lowest-distortion point, what was significant was the higher-order distortion came up first - 3rd, 5th, 7th, all the ugly ones. In addition, and probably closely related to the prior finding, was the linear current delivery dropped rapidly as well. The ability to deliver linear current is closely related to the absence of high-order distortion content, and this is where the differential circuit falls down the worst.
Returning the previous post, the decision to switch the input and driver stage to pure differential operation is almost certainly the reason the amplifier is grossly underperforming compared to the Karna. Although I didn't mention this on the Web-page, the decision to avoid differential operation was based on a series of measurements and audition.
On direct A/B comparison, the differential circuit sounded thin, scrawny, and with "pinched" and closed-in HF, with poor dynamic reserves. Measurement confirmed this: distortion was several times higher, the proportion of upper harmonic to lower harmonics was unfavorable, and most important, peak current delivery was curtailed by severalfold. This was the real source of the problem: the driver stage was unable to provide peak current to the 300B grids when the output tubes were getting near Class A2 conditions or Miller capacitance + interstage stray capacitance were demanding more and more current from the driver at high frequencies. When it comes to driver stage design, it's all about linear current delivery, particularly at high frequencies.
This is not the same as just throwing in a cathode-follower; that lowers Zsource, but has NO effect on distortion or the ability to deliver peak current into a load with complex (and nonlinear) characteristics. In effect, the driver really needs to be a small power amplifier in its own right, or at least able to drive headphones.
One of the most revealing tests of the Karna was a test I did a while ago to see if I could slew the amplifier; although the HF rolls off above 40~50 kHz, there's nothing to stop me from increasing the input level. So I fed in a sinewave, increased the gain to just below clipping (16 watts output), and kept raising the frequency, and then level, keeping the amplifier just below clipping the whole time.
I finally lost my nerve at 500 kHz, just below the AM radio band - the sine wave was still looking very close to a sine-wave, no sign of flat-topping or starting to turn into triangles, and the amplifier was acting like a 16-watt radio transmitter putting out a carrier wave. I never did see slewing, although I shudder to think how much current was going into the plates of the driver stage (the driver was looking into a nearly pure reactance, so all the power was getting bounced right back into the driver-stage plates).
A test like this would destroy most transistor amplifiers, and I very much doubt a differential driver would have cheerfully slid into Class AB to deliver the amount of current the 300B grids were demanding.
So ditch the negative-voltage current sources for the input and drive stages, replace them with wirewound resistors, and shunt the cathodes to the B+ supply with a good-quality 40 uF oil cap. One thing to be careful of: the VR tubes do NOT like to see capacitive loads (they will oscillate), so either put the bypass caps between the cathodes and ground, or cathodes and B+, but not both.
Hi Bud:
Very thankful you have elected to chime in. Two hours and two beers later, I have some more data that should help the process.
I've held off on the phase shift issues for now, and just investigated the peaking over about 5K. The gain across each differential stage is fairly constant, so that doesn't seem to be the problem. It is each IT that is contributing to the rise in response.
I've experimented with resistor loading each IT, and around 30K for the first stage and 22K for the second stage are fairly happy values that tame the peak. These values could probably be increased a little, and a series cap used to bring it in near 5K, for a nice smooth response.
However, based on Lynn's issues with a current source loaded differential stage, this peaking may all be moot anyway. My prior experience with the Lundahl's was similar to your statement; measurements don't always make for good sound. It was Kevin Carter's recommendation to NOT add zobels to transformers (or very rarely), as they don't improve the sound. At least in my amorphous transformer coupled preamp, he was right. So before I go throwing zobels everywhere, I'll have to address Lynn's recommendations in the next post. I suspect (hope) everything will fix itself.
Happy to report the output transformer, as you stated, does not exhibit peaking. This is only affecting the IT's. Why the output stage is distorting, I guess that's a third issue, but one step at a time.
Very thankful you have elected to chime in. Two hours and two beers later, I have some more data that should help the process.
I've held off on the phase shift issues for now, and just investigated the peaking over about 5K. The gain across each differential stage is fairly constant, so that doesn't seem to be the problem. It is each IT that is contributing to the rise in response.
I've experimented with resistor loading each IT, and around 30K for the first stage and 22K for the second stage are fairly happy values that tame the peak. These values could probably be increased a little, and a series cap used to bring it in near 5K, for a nice smooth response.
However, based on Lynn's issues with a current source loaded differential stage, this peaking may all be moot anyway. My prior experience with the Lundahl's was similar to your statement; measurements don't always make for good sound. It was Kevin Carter's recommendation to NOT add zobels to transformers (or very rarely), as they don't improve the sound. At least in my amorphous transformer coupled preamp, he was right. So before I go throwing zobels everywhere, I'll have to address Lynn's recommendations in the next post. I suspect (hope) everything will fix itself.
Happy to report the output transformer, as you stated, does not exhibit peaking. This is only affecting the IT's. Why the output stage is distorting, I guess that's a third issue, but one step at a time.
Hello Lynn:
Good to hear from the amp's creator. It's truly honorable to have you spend the effort posting. As I know you understand, building one of these isn't exactly cheap, so your help is greatly appreciated.
Well (*(insert profanity here)*) !! And here I thought I was improving the situation. Just to explain myself, I got this idea from Gary Pimm's site, where he was using the resistor in the cathode, and by replacing it with a CCS, was able to remove the capacitor with no sonic loss. Getting rid of a capacitor is always a good thing. Wrong was I. I got a bunch of oil cans and soviet teflons, I'll give this a try.
If you are right, it is very possible all other problems will be solved (crosses fingers). One question though; at around 20kHz 5W output the output stage is clearly distorted, while the signal at the grids is nice and clean. This would indicate my driver is capable, but the output stage itself is distorting, wouldn't it?
Now you tell me !
Yep, that would be a very good description. Closed in. Frustrating, in that the midrange really is beautiful. Highs is where everything starts to suffer. What is more interesting is that even though there is peaking of the HF range, the shimmer and glitter of cymbals has just disappeared, and actually sounds quieter in a relative sense.
I guess I deserve this. Thank you for the marching orders. I'll pursue all the other 'stuff' (peaking, phase shift, distortion) once I make those changes.
Good to hear from the amp's creator. It's truly honorable to have you spend the effort posting. As I know you understand, building one of these isn't exactly cheap, so your help is greatly appreciated.
Rp is about 2300K, compared to 1700K of the 45. I chose to use it because it costs less than half the 45, and a number of people have had very good things to say about it as a driver. Supposed to sound very similar to the 45. Hope I'm not shooting myself in the foot?Lynn Olson said:
It is serving as my high pass filter, around 115Hz. I admit this is a questionable implementation, but I couldn't come up with a better method for a fully balanced system. With test equipment, it actually works pretty well; sonically, we'll see.
Well (*(insert profanity here)*) !! And here I thought I was improving the situation. Just to explain myself, I got this idea from Gary Pimm's site, where he was using the resistor in the cathode, and by replacing it with a CCS, was able to remove the capacitor with no sonic loss. Getting rid of a capacitor is always a good thing. Wrong was I. I got a bunch of oil cans and soviet teflons, I'll give this a try.
If you are right, it is very possible all other problems will be solved (crosses fingers). One question though; at around 20kHz 5W output the output stage is clearly distorted, while the signal at the grids is nice and clean. This would indicate my driver is capable, but the output stage itself is distorting, wouldn't it?
Now you tell me !
Yep, that would be a very good description. Closed in. Frustrating, in that the midrange really is beautiful. Highs is where everything starts to suffer. What is more interesting is that even though there is peaking of the HF range, the shimmer and glitter of cymbals has just disappeared, and actually sounds quieter in a relative sense.
I guess I deserve this. Thank you for the marching orders. I'll pursue all the other 'stuff' (peaking, phase shift, distortion) once I make those changes.
Gary Pimm actually built the Karna, and I took them back to Gary for a second go-round to further optimize the circuit. But the design is mine, with Gary acting as mentor for the instrumentation grounding techniques. If the grounding is carried out as drawn, the Karna should be entirely free of buzz and hash. With DC supplies for the DHT's (not tried yet), the output noise should down in the microvolt range. Building a good-sounding DC supply is not trivial, though - Gary Pimm's DC supply for his 300B SET was elaborate as any I've ever seen (very high speed shunt and series regulation), and he STILL found that battery-powered filaments outperformed them.
Gary Pimm's amp is headed in the same direction as mine, but he is using a completely different approach. In fact, the two don't commingle that well, although I have certainly contemplated using his front end instead of mine. We found that the most of the distortion in the Karna isn't in the 300B's, nor the 45's, but the input tube!!! To say the least, this is a comment on the linearity of the driver and output stage.
Gary gets rid of the Miller capacitance by driving pentodes, which have very low capacitance. I use a less elegant brute-force approach, with lots of linear current from transformer-coupled Class A PP DHT drivers. One of the biggest differences between this amp and a conventional RC-coupled PP amplifier is that the current from both sides of the driver is available when one of the 300B grids demands grid-current (they take turns going into grid-current, of course). So the transition into Class A into Class A2 happens with no visible "bump" on the scope at all - in fact, Gary found the 300B's operate quite happily with the grid voltage going 30 volts positive on the peaks!
With a conventional RC-coupled PP amplifier, as soon as the grid draws current, it discharges the coupling cap, and the amplifier is seriously misbiased for the length of time it takes to recharge the coupling cap - this commonly takes a half-second or longer, stretching out what would have been a millisecond overload into a lengthy fraction of second. This is a good reason to keep the RC time constant as short as possible consistent with acceptable bass response. By contrast, the transformer-coupled amplifier enters into Class A2 with no visible transition, and recovers instantly from overload, with no coupling caps to discharge.
That is why the modest-looking 16-watt amplifier (30 watts in Class A2) sounds as loud as a 60-watt Citation II (with much harder clipping thanks to feedback), or a 100-watt transistor amplifier (even harder clipping and a Class AB transition at a few watts).
It is true that there are still time constants in there - a cap-bypassed cathode is not the same as dedicated negative DC supplies for the grids. But then there is still a coupling cap lurking in the grid circuit (connecting the two halves of the secondaries together), although this can be a lot smaller than a cathode-bypass cap, and doesn't carry as much current. But this approach is certainly more complicated, since there then has to individual bias adjustments (and switchable metering) for each DHT to prevent it from running away.
If you really want to spend a lot of time twiddling with the circuit, a good-sounding DC supply for the filaments should meet the bill. You can spend many months or even years on that. But for now, get some clip-leads, a collection of GE or similar oil caps, and bypass those cathodes (I'd suggest to B+, it's a shorter current loop).
Here's the presentation on current loops.
Gary Pimm's amp is headed in the same direction as mine, but he is using a completely different approach. In fact, the two don't commingle that well, although I have certainly contemplated using his front end instead of mine. We found that the most of the distortion in the Karna isn't in the 300B's, nor the 45's, but the input tube!!! To say the least, this is a comment on the linearity of the driver and output stage.
Gary gets rid of the Miller capacitance by driving pentodes, which have very low capacitance. I use a less elegant brute-force approach, with lots of linear current from transformer-coupled Class A PP DHT drivers. One of the biggest differences between this amp and a conventional RC-coupled PP amplifier is that the current from both sides of the driver is available when one of the 300B grids demands grid-current (they take turns going into grid-current, of course). So the transition into Class A into Class A2 happens with no visible "bump" on the scope at all - in fact, Gary found the 300B's operate quite happily with the grid voltage going 30 volts positive on the peaks!
With a conventional RC-coupled PP amplifier, as soon as the grid draws current, it discharges the coupling cap, and the amplifier is seriously misbiased for the length of time it takes to recharge the coupling cap - this commonly takes a half-second or longer, stretching out what would have been a millisecond overload into a lengthy fraction of second. This is a good reason to keep the RC time constant as short as possible consistent with acceptable bass response. By contrast, the transformer-coupled amplifier enters into Class A2 with no visible transition, and recovers instantly from overload, with no coupling caps to discharge.
That is why the modest-looking 16-watt amplifier (30 watts in Class A2) sounds as loud as a 60-watt Citation II (with much harder clipping thanks to feedback), or a 100-watt transistor amplifier (even harder clipping and a Class AB transition at a few watts).
It is true that there are still time constants in there - a cap-bypassed cathode is not the same as dedicated negative DC supplies for the grids. But then there is still a coupling cap lurking in the grid circuit (connecting the two halves of the secondaries together), although this can be a lot smaller than a cathode-bypass cap, and doesn't carry as much current. But this approach is certainly more complicated, since there then has to individual bias adjustments (and switchable metering) for each DHT to prevent it from running away.
If you really want to spend a lot of time twiddling with the circuit, a good-sounding DC supply for the filaments should meet the bill. You can spend many months or even years on that. But for now, get some clip-leads, a collection of GE or similar oil caps, and bypass those cathodes (I'd suggest to B+, it's a shorter current loop).
Here's the presentation on current loops.
Great thread! Thanks Lynn for the additional rationale on differential versus parrellel PP.
I too am a Karna builder. I am much earlier on in my build...finishing up the remote power supplies. I took a couple years to collect parts, build a couple of teething projects (which I am finishing final touches now) and read/learn all I could. I spent so much on this project (all amorphous Lundahl iron) I wanted to bring my understanding to a level that would exact the potential.
I finally finished Morgan Jones Valve Amplifiers this weekend. I am more than halfway done with Building Valve Amplifiers.
I too am a Karna builder. I am much earlier on in my build...finishing up the remote power supplies. I took a couple years to collect parts, build a couple of teething projects (which I am finishing final touches now) and read/learn all I could. I spent so much on this project (all amorphous Lundahl iron) I wanted to bring my understanding to a level that would exact the potential.
I finally finished Morgan Jones Valve Amplifiers this weekend. I am more than halfway done with Building Valve Amplifiers.
Good for you, at least you can learn from my mistakes. I also found this project to take somewhere around 3 years, and you and I didn't design a thing !! Happy there are people like Lynn who take the time to document and share their designs. No doubt he has spent 3 times as much as us in order to reach the final state.
You also using AC heating? I'm blown away as to how silent it is, though I admit this is my first DHT amp.
Let us know how the Lundahls work out for you. I should have my repairs completed sometime before the weekend, and I'll report back. Got the GE 97F caps, just waiting for the right size WW's to come in.
You also using AC heating? I'm blown away as to how silent it is, though I admit this is my first DHT amp.
Let us know how the Lundahls work out for you. I should have my repairs completed sometime before the weekend, and I'll report back. Got the GE 97F caps, just waiting for the right size WW's to come in.
I think I have some progress on at least one of the root causes.
I changed all the stages out per Lynn's request: 40uF GE 97F oilers plus 0.1uF teflon bypass from B+ to the cathode. Bias for the 12A4 stage is about 18V grid with 18mA per triode. Bias for the 46 driver is 28V grid at 28mA per triode.
Listening session was improved; midrange was definitely better, lows need help due to my selection of the HPF; I'll tweak with that in a few days. But the highs still were not satisfying, regardless of volume level.
Repeated the measurements of gain/phase from 20 to 20K. Response was essentially identical to that done before the modifications. I would say the phase shift and peaking must be the IT bandwidth, which may or may not be audible.
Bigger revelation was when I looked at the output near 20KHz. Blue is input, red is grid-grid at the 300B, and yellow is the speaker output:
Distortion at the output is clearly visible. This is running almost 20W at 19kHz. Grids don't look too bad, so I figured this was the output stage distorting. Then, I took a look at the individual grids of the 46's:
O great. Correct me if I'm wrong, but I'm operating my 46 at 0V grid, and the driver ultimately distorts, as the first stage is unable to supply the required grid current to the 46. The differential grid-grid looks okay, but this is deceiving, as each grid distorts somewhat evenly.
I was considering the risks of using the 46 when dealing with the idle conditions. The 45 looks to bias over 50V, while the 46 less than 30V. So I get less allowable signal swing at the input.
So am I back to having to consider using the 45 instead of the 46? In the interim, I would suspect I get no more than about 15W before distortion really sets in.
This still, however, does not explain to me why the highs sound washed out, regardless of volume. I mostly run the amp at less than half the signal strength of these tests, and still am unsatisfied with the highs. Could the 33nF K40Y PIO be the reason for this?
I changed all the stages out per Lynn's request: 40uF GE 97F oilers plus 0.1uF teflon bypass from B+ to the cathode. Bias for the 12A4 stage is about 18V grid with 18mA per triode. Bias for the 46 driver is 28V grid at 28mA per triode.
Listening session was improved; midrange was definitely better, lows need help due to my selection of the HPF; I'll tweak with that in a few days. But the highs still were not satisfying, regardless of volume level.
Repeated the measurements of gain/phase from 20 to 20K. Response was essentially identical to that done before the modifications. I would say the phase shift and peaking must be the IT bandwidth, which may or may not be audible.
Bigger revelation was when I looked at the output near 20KHz. Blue is input, red is grid-grid at the 300B, and yellow is the speaker output:
An externally hosted image should be here but it was not working when we last tested it.
Distortion at the output is clearly visible. This is running almost 20W at 19kHz. Grids don't look too bad, so I figured this was the output stage distorting. Then, I took a look at the individual grids of the 46's:
An externally hosted image should be here but it was not working when we last tested it.
O great. Correct me if I'm wrong, but I'm operating my 46 at 0V grid, and the driver ultimately distorts, as the first stage is unable to supply the required grid current to the 46. The differential grid-grid looks okay, but this is deceiving, as each grid distorts somewhat evenly.
I was considering the risks of using the 46 when dealing with the idle conditions. The 45 looks to bias over 50V, while the 46 less than 30V. So I get less allowable signal swing at the input.
So am I back to having to consider using the 45 instead of the 46? In the interim, I would suspect I get no more than about 15W before distortion really sets in.
This still, however, does not explain to me why the highs sound washed out, regardless of volume. I mostly run the amp at less than half the signal strength of these tests, and still am unsatisfied with the highs. Could the 33nF K40Y PIO be the reason for this?
One other thought, upon further reflection.
I did some calculations on the phase shift issue. From about 500Hz and up, the phase shift from input to output is a CONSTANT TIME DELAY, around 15 microseconds of lag. It is constant as frequency increases. This leads me to think maybe the phase shift is not much of an issue in the audible realm, as what you hear from the speaker, though shifted in time, is relatively constant.
Sound reasonable?
BTW, took out that 33nF PIO capacitor and ran the speakers full range (not my long term intention), and the highs are back, sounding very nice and detailed. Frustrating when I can't measure what I hear. Now I'll have to experiment with teflon or other exotics to get the sound I know this amp is capable of.
I did some calculations on the phase shift issue. From about 500Hz and up, the phase shift from input to output is a CONSTANT TIME DELAY, around 15 microseconds of lag. It is constant as frequency increases. This leads me to think maybe the phase shift is not much of an issue in the audible realm, as what you hear from the speaker, though shifted in time, is relatively constant.
Sound reasonable?
BTW, took out that 33nF PIO capacitor and ran the speakers full range (not my long term intention), and the highs are back, sounding very nice and detailed. Frustrating when I can't measure what I hear. Now I'll have to experiment with teflon or other exotics to get the sound I know this amp is capable of.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.