Pass DIY Addict
Joined 2000
Paid Member
William,
The first time I adjusted my absolute dc offset after making the change to JFETs and changing the McMillan resistors (R46/47) from 10k to 20k, I did see movement in my absolute DC offset as well. I had to bring absolute offset down from about 8v to 0v with VR2 and while doing so, I saw movement in relative offset from about -0.5v to about +0.4v. I was simultaneously using three DMMs on the output terminals and watching as I adjusted.
Given that it doesn't look like I've broken anything yet (which still surprises me) I'll just put it back together this evening and see what impacts judicious adjustment of VR2 can make. I created a simpler physical arrangement that doesn't have the thermal coupling shown above, so it's just a small PtP circuit that sits above the main PCB on short little wire "stilts."
Eric
The first time I adjusted my absolute dc offset after making the change to JFETs and changing the McMillan resistors (R46/47) from 10k to 20k, I did see movement in my absolute DC offset as well. I had to bring absolute offset down from about 8v to 0v with VR2 and while doing so, I saw movement in relative offset from about -0.5v to about +0.4v. I was simultaneously using three DMMs on the output terminals and watching as I adjusted.
Given that it doesn't look like I've broken anything yet (which still surprises me) I'll just put it back together this evening and see what impacts judicious adjustment of VR2 can make. I created a simpler physical arrangement that doesn't have the thermal coupling shown above, so it's just a small PtP circuit that sits above the main PCB on short little wire "stilts."
Eric
Pass DIY Addict
Joined 2000
Paid Member
Success! Though a few problems remain...
My third try to make things work with JFETs seems to be more successful. Adjusting VR2 (by a great amount) made all of the difference - this makes me think I was more successful last time around than I realized...
Over the course of 2 hours, I made continuous adjustments to VR2 to keep offset close to zero as it warmed up. I readjusted bias on the output mosfets to about 8A per side (~0.475v across Rsource). Rails were running about 22.8v. I measured 4.72v across R23 and 4.73v across R25 indicating about 12mA bias through the differential pairs (four SJ74s cascoded to four 550s).
10kHz sine and square waves into 8 ohms look pretty decent, though the square wave shows a little asymmetry on overshoot at the bottom:
Bandwidth into 4ohms is about 90kHz and into 8ohm measures about 105kHz. They are no longer backwards, but this bandwidth is about half of the 200+kHz I had been getting into 8 ohms with the 9610s.
The strange behavior happens at the onset of clipping. I can only swing 23Vrms (66w) into 8 ohms - this used to be ~95w with the 9610s. With a 4 ohm load, I can only get 16.9Vrms (71w) - this used to be closer to 110w with the 9610 in place. Further increasing the input signal into a 4ohm load past clipping causes output voltage to decline(!) to 14vrms or so - shown in right image below. I never saw an output voltage decline at clipping with the 9610s - just increasingly distorted sine and square waves.
Here is the output just before and after clipping:
The top trace is the -Speaker terminal and the bottom trace is the +Speaker terminal.
My JFETs are all matched to within 0.05mA at about 10.6mA. I remeasured them before I built this input differential.
I quickly re-measured offset this morning when the amp was cold. I got about 16.4v absolute offset and about 30mV relative offset with 22k for R46/R47. Absolute offset now declines VERY slowly making R1/4 and R44/45 rather warm. For the time being, I'll parallel another 22k with R46/47 to bring it back down to 11k.
Any ideas what is causing reduced output power and voltage decline beyond the onset of clipping? It seems to me that the input differential is running out of room...
My third try to make things work with JFETs seems to be more successful. Adjusting VR2 (by a great amount) made all of the difference - this makes me think I was more successful last time around than I realized...
Over the course of 2 hours, I made continuous adjustments to VR2 to keep offset close to zero as it warmed up. I readjusted bias on the output mosfets to about 8A per side (~0.475v across Rsource). Rails were running about 22.8v. I measured 4.72v across R23 and 4.73v across R25 indicating about 12mA bias through the differential pairs (four SJ74s cascoded to four 550s).
10kHz sine and square waves into 8 ohms look pretty decent, though the square wave shows a little asymmetry on overshoot at the bottom:
An externally hosted image should be here but it was not working when we last tested it.
Bandwidth into 4ohms is about 90kHz and into 8ohm measures about 105kHz. They are no longer backwards, but this bandwidth is about half of the 200+kHz I had been getting into 8 ohms with the 9610s.
The strange behavior happens at the onset of clipping. I can only swing 23Vrms (66w) into 8 ohms - this used to be ~95w with the 9610s. With a 4 ohm load, I can only get 16.9Vrms (71w) - this used to be closer to 110w with the 9610 in place. Further increasing the input signal into a 4ohm load past clipping causes output voltage to decline(!) to 14vrms or so - shown in right image below. I never saw an output voltage decline at clipping with the 9610s - just increasingly distorted sine and square waves.
Here is the output just before and after clipping:
An externally hosted image should be here but it was not working when we last tested it.
The top trace is the -Speaker terminal and the bottom trace is the +Speaker terminal.
My JFETs are all matched to within 0.05mA at about 10.6mA. I remeasured them before I built this input differential.
I quickly re-measured offset this morning when the amp was cold. I got about 16.4v absolute offset and about 30mV relative offset with 22k for R46/R47. Absolute offset now declines VERY slowly making R1/4 and R44/45 rather warm. For the time being, I'll parallel another 22k with R46/47 to bring it back down to 11k.
Any ideas what is causing reduced output power and voltage decline beyond the onset of clipping? It seems to me that the input differential is running out of room...
Pass DIY Addict
Joined 2000
Paid Member
This is the circuit I am currently using:
An externally hosted image should be here but it was not working when we last tested it.
Hi Eric,
the decline is normally caused by oscillation in the active current source. You need to change C9 and C10 until this is normal again.
Since you are running 24mA of bias, 22k for the McMillans maybe a bit too much. I´m using 22k at the moment with 9.6mA of bias. Changing them back to 12k or so would change the startup voltage to 8V.
William
the decline is normally caused by oscillation in the active current source. You need to change C9 and C10 until this is normal again.
Since you are running 24mA of bias, 22k for the McMillans maybe a bit too much. I´m using 22k at the moment with 9.6mA of bias. Changing them back to 12k or so would change the startup voltage to 8V.
William
Pass DIY Addict
Joined 2000
Paid Member
Hi William,
Thanks for the insights! This is curious - I removed C9/C10 about a year ago and found it increased bandwidth from about 140kHz to 210kHz. Guess it's time to put them back in... With the change to JFETs, I now have an increased low-amplitude buzzing sound in my speaker. Is this the audible impact of oscillation? Is there a way that I can actually see the oscillation on my scope?
I'll change R46/47 and reduce them to 10k or so again. I'm hoping you'll share your servo mechanism in a little while (hint, hint)
I already have a small pile of KTY81-110 devices just waiting for me to be done with the JFET conversion...
As for the bias through the input differential, I presume your bias is lower because you are using two SJ74s where I am using four. What practical difference does using two vs four and different bias levels make?
Thanks again!
Eric
Thanks for the insights! This is curious - I removed C9/C10 about a year ago and found it increased bandwidth from about 140kHz to 210kHz. Guess it's time to put them back in... With the change to JFETs, I now have an increased low-amplitude buzzing sound in my speaker. Is this the audible impact of oscillation? Is there a way that I can actually see the oscillation on my scope?
I'll change R46/47 and reduce them to 10k or so again. I'm hoping you'll share your servo mechanism in a little while (hint, hint)
I already have a small pile of KTY81-110 devices just waiting for me to be done with the JFET conversion...As for the bias through the input differential, I presume your bias is lower because you are using two SJ74s where I am using four. What practical difference does using two vs four and different bias levels make?
Thanks again!
Eric
Hi Eric,
no problem. Don´t know about the buzzing sound but you can see the oscillation at the caps. Not sure how it looked cause it is long ago but it is very obvious. There must be a description in my AlephJ-X thread.
My servo is ready waiting to be put in the amp and will be tested soon.......
My bias is around 9mA (using 910R drain resistors). Having more JFets should increase the bandwidth as there is more current available to feed the input capacitance of the output fets.
I´m now having around 100 kHz (will test) as I disconnected 4 output fets (leaving 8) and reduced the bias as my Zu Druids don't need that much power.
William
no problem. Don´t know about the buzzing sound but you can see the oscillation at the caps. Not sure how it looked cause it is long ago but it is very obvious. There must be a description in my AlephJ-X thread.
My servo is ready waiting to be put in the amp and will be tested soon.......
My bias is around 9mA (using 910R drain resistors). Having more JFets should increase the bandwidth as there is more current available to feed the input capacitance of the output fets.
I´m now having around 100 kHz (will test) as I disconnected 4 output fets (leaving 8) and reduced the bias as my Zu Druids don't need that much power.
William
Pass DIY Addict
Joined 2000
Paid Member
Well, I have some mixed results which makes me think I need to try this again.
I started moving McMillan resistors back down to 10k and putting a 1nF cap in C9/C10. After a one hour warm up and adj absolute offset down to 0v, I had a relative offset of about 17mV. Bandwidth and max output are below:
8R: 210kHz, 82w (was 105kHz/66w with no cap)
4R: 160kHz, 95w (was 90kHz/71w with no cap)
This looked much better and I was happier (but was still looking for more output power like I get in my remaining amps)! I should have stopped here, but I went on to experiment with different values for C9/C10. What I got was very inconsistent and I'm wondering if it was because I was putting multiple small caps in parallel and series to explore the impacts of different values.
I tried 3n3, 2nF, 1nF, 680pF, 500pF, 434pF, 333pF. The 3n3 resulted in more rounded off corners on a 10kHz square wave. Values between 333pF and 1nF seemed to make the nicest 10kHz squares, but bandwidth seemed all over the place. There was no reliable pattern linking bandwidth to the value of C9/C10. When I originally measured 1nF, I got 210k/8R and 160k/4R. I tried other values that didn't produce a clear pattern and came back to 1nF and re-measured bandwidth at 140k/8R and 110k/4R. The amp was now on for about 2 hours so I went back and readjusted abs offset again to 0v and relative offset was at 130mV.
3n3: 160k/8R and 120k/4R (single cap)
2n2: 165k/8R and 140k/4R (parallel 1nF)
1nF: 210k/8R and 160k/4R (single cap)
680pF: 140k/8R and 115k/4R (single cap)
500pF: 160k/8R and 120k/4R (series 1nF caps)
434pF: 180k/8R and 135k/4R (series 1nF, 1nF and 39nF caps)
333pF: 145k/8R and 135k/4R (series 1nF caps)
288pF: 140k/8R and 110k4R (series 1nF, 1nF, 680pF caps)
I seem to have some instability somewhere - or just messiness associated with stacking caps together. No matter what value of C9/C10 I used, measuring maximum output into 8R resulted in curves with a little kink in them as the amp hit clipping (looks like the curves posted in your AlephJ-X thread). Beyond this, the curve just turned nasty:
A typical 4R clipped sine wave got really wide at the bottom and cleanly truncated at the top:
I also measured maximum clean output sinewaves into 8R and 4R at 1kHz, 5kHz, 10kHz and 15kHz. All 8R curves developed the kink just before clipping (at all frequencies) and then turned nasty. All 4R curves collapsed (voltage drop) beyond clipping, but ONLY at 15kHz.
Is there something more here, or should I just put in the 1nf for C9/C10 and call it done? I need to go back and re-read your AlephJ-X thread (again...).
I started moving McMillan resistors back down to 10k and putting a 1nF cap in C9/C10. After a one hour warm up and adj absolute offset down to 0v, I had a relative offset of about 17mV. Bandwidth and max output are below:
8R: 210kHz, 82w (was 105kHz/66w with no cap)
4R: 160kHz, 95w (was 90kHz/71w with no cap)
This looked much better and I was happier (but was still looking for more output power like I get in my remaining amps)! I should have stopped here, but I went on to experiment with different values for C9/C10. What I got was very inconsistent and I'm wondering if it was because I was putting multiple small caps in parallel and series to explore the impacts of different values.
I tried 3n3, 2nF, 1nF, 680pF, 500pF, 434pF, 333pF. The 3n3 resulted in more rounded off corners on a 10kHz square wave. Values between 333pF and 1nF seemed to make the nicest 10kHz squares, but bandwidth seemed all over the place. There was no reliable pattern linking bandwidth to the value of C9/C10. When I originally measured 1nF, I got 210k/8R and 160k/4R. I tried other values that didn't produce a clear pattern and came back to 1nF and re-measured bandwidth at 140k/8R and 110k/4R.
The amp was now on for about 2 hours so I went back and readjusted abs offset again to 0v and relative offset was at 130mV. 3n3: 160k/8R and 120k/4R (single cap)
2n2: 165k/8R and 140k/4R (parallel 1nF)
1nF: 210k/8R and 160k/4R (single cap)
680pF: 140k/8R and 115k/4R (single cap)
500pF: 160k/8R and 120k/4R (series 1nF caps)
434pF: 180k/8R and 135k/4R (series 1nF, 1nF and 39nF caps)
333pF: 145k/8R and 135k/4R (series 1nF caps)
288pF: 140k/8R and 110k4R (series 1nF, 1nF, 680pF caps)
I seem to have some instability somewhere - or just messiness associated with stacking caps together. No matter what value of C9/C10 I used, measuring maximum output into 8R resulted in curves with a little kink in them as the amp hit clipping (looks like the curves posted in your AlephJ-X thread). Beyond this, the curve just turned nasty:
An externally hosted image should be here but it was not working when we last tested it.
A typical 4R clipped sine wave got really wide at the bottom and cleanly truncated at the top:
An externally hosted image should be here but it was not working when we last tested it.
I also measured maximum clean output sinewaves into 8R and 4R at 1kHz, 5kHz, 10kHz and 15kHz. All 8R curves developed the kink just before clipping (at all frequencies) and then turned nasty. All 4R curves collapsed (voltage drop) beyond clipping, but ONLY at 15kHz.
Is there something more here, or should I just put in the 1nf for C9/C10 and call it done? I need to go back and re-read your AlephJ-X thread (again...).
Last edited:
Pass DIY Addict
Joined 2000
Paid Member
William,
Did you implement the servo that Salomon pointed us toward (a few posts ago), or did you go in a different direction?
Now that I have my VR2 tethered on wires instead of stuck on the PCB, it is much easier to adjust it immediately on powerup for low offset and measure the resistance and then re-adjust/re-measure after the amp is warmed up to see what resistance swing is necessary to keep offset closer to zero across the cold to fully warmed up continuum.
My poor Frankenstein-amplifier: Wires, scars, extra parts grafted on the PCB One day I'll get all of this straightened out and move it from the living room floor back into the theater...
Eric
Did you implement the servo that Salomon pointed us toward (a few posts ago), or did you go in a different direction?
Now that I have my VR2 tethered on wires instead of stuck on the PCB, it is much easier to adjust it immediately on powerup for low offset and measure the resistance and then re-adjust/re-measure after the amp is warmed up to see what resistance swing is necessary to keep offset closer to zero across the cold to fully warmed up continuum.
My poor Frankenstein-amplifier: Wires, scars, extra parts grafted on the PCB
One day I'll get all of this straightened out and move it from the living room floor back into the theater...Eric
Hi Eric,
I used my own servo but combined with the PTC. The servo only compensates for the quite strong variations in AC voltage. Yes, it is in one amp and works fine. Keeps the output at zero volts between 215 and 245 volts AC.
It is based upon an OP77GP opamp and fed from the rails with a bit of filtering and a 78 and 7918 (the small TO-92 ones). Input is from one output and output goes to R24/R26. This is also why it didn´t work yesterday as the voltage here is almost 17V and I connected it on the wrong side of R24 where the voltage was around 19,5V. (above the rail voltage of the servo).
Measured power and bandwidth and looked for oscillations.
24V rms into 8R, 72Watts
12V rms into 4R, 36Watts
135kHz -3dB 4R
145kHz -3dB 8R
No oscillations
So now I´ve got one amp with and one without the servo. Both have 27k McMillans and are slowly turning into Frankenstein amps (which is appropriate as he once lived here in Ingolstadt)
Now I will take it downstairs and have a listen
William
I used my own servo but combined with the PTC. The servo only compensates for the quite strong variations in AC voltage. Yes, it is in one amp and works fine. Keeps the output at zero volts between 215 and 245 volts AC.
It is based upon an OP77GP opamp and fed from the rails with a bit of filtering and a 78 and 7918 (the small TO-92 ones). Input is from one output and output goes to R24/R26. This is also why it didn´t work yesterday as the voltage here is almost 17V and I connected it on the wrong side of R24 where the voltage was around 19,5V. (above the rail voltage of the servo).
Measured power and bandwidth and looked for oscillations.
24V rms into 8R, 72Watts
12V rms into 4R, 36Watts
135kHz -3dB 4R
145kHz -3dB 8R
No oscillations
So now I´ve got one amp with and one without the servo. Both have 27k McMillans and are slowly turning into Frankenstein amps (which is appropriate as he once lived here in Ingolstadt
)Now I will take it downstairs and have a listen
William
First impressions
Well,
it´s still very early but sound wise nothing bad has happened. The only negative thing is some hum in the servoed channel. Only heard with one ear next to the (101dB) speaker. Here it probably would be good to filter the opamp output a bit.
Maybe it would be better to see if I can get some really nice and constant 230V out of the wall.........has anybody tried USV´s? I found some from AEG (Protect type D) which make there own sinus and cost a lot less than a PS Audio Powerplant.
William
Well,
it´s still very early but sound wise nothing bad has happened. The only negative thing is some hum in the servoed channel. Only heard with one ear next to the (101dB) speaker. Here it probably would be good to filter the opamp output a bit.
Maybe it would be better to see if I can get some really nice and constant 230V out of the wall.........has anybody tried USV´s? I found some from AEG (Protect type D) which make there own sinus and cost a lot less than a PS Audio Powerplant.
William
Pass DIY Addict
Joined 2000
Paid Member
The servo implementation sounds promising so far, William. Impressive that you have so little hum into 101dB speakers. Mine are probably somewhere in the 87-90dB range and I can hear hum within about 1 foot of the speaker. I still need to fiddle around a bit to remove this. I don't get the mains voltage swings quite to the same extent that you do, thus I was thinking of using the servo just to bring down the absolute offset more quickly. I'm wondering how a servo will be affected by summer (19c) vs winter (13c) temperatures in my basement.
I think I still need to play around with values for C9/C10 in my JFET amp to better control oscillation. I think I'll just order a wide variety of caps in the pF range and seeing which ones seem to work better. Right now, I have 680pF cap in place and behavior is clearly better than with no cap here, but still have some collapse in power output beyond clipping, but only at the top end of the audio spectrum (20kHz) and only into 4R loads.
Directly comparing amps with JFETs and 9610's, I am getting more stable relative offset over time, greater power output, and wider bandwidth with the 9610's. With JFETs, maximum power is a bit reduced and waveform deterioration with the JFETs gets nastier beyond clipping. 10kHz square waves at 10Vrms are identical between the 9610s and the JFETs.
With 9610's and ~22v rails I can get 4mV relative offset and:
8.8A bias provides 120w/4R and 78w/8R
9.25A bias provides 144w/4R and 90w/8R
9.6A bias provides 156w/4R and 98w/8R
I don't quite understand why output into 8R increases with bias- I thought 8R output was voltage limited and 4R output was current limited...
With JEFTs and ~22v rails, I can get ~30mV relative offset cold and ~100mV of relative offset when warm and:
8.8A bias provides 90w/4R and 80w/8R
9.25A bias provides 115w/4R and 83w/8R
9.6A bias provides 119w/4R and 83w/8R
Putting the amp back down in my theater room, the obvious difference with the new JFET front end is that I get more treble and more balanced midrange. I find these to be welcome improvements, though the magnitude of the change is small. At first, I wondered if this is worth it *for me* since my theater setup makes use of EQ in the preamp - thus, this is something I could largely duplicate with EQ instead of changes to the amp. For someone who uses the amp in a more "purist" form (source, volume control, amp, speaker, no EQ), the change to JFETs is clearly valuable.
But after a day of listening to music (in pure direct mode, which bypasses any digital EQ in my preamp), I am detecting another interesting improvement: It's hard to describe it well, but the impression I have is that the speaker driven by the 9610 amp seems to "call more attention to itself." Music seems to be coming more directly from the speaker, so I'd call that speaker more localizable during listening. The speaker driven by the JFET amp seems to have "disappeared" more into the room. I can't as easily identify the physical location of the speaker while listening.
At first, I thought this was an SPL imbalance between the amps due to the change, but a few minutes with my meter proves this theory wrong.
With the JFETs, the speaker seems to have completely disappeared, leaving behind a wider soundstage and more focused imaging. This is clear while listening to both music and movies.
I will have to spend some time with higher SPL movie playback to see if I can detect any signs of earlier clipping due to the reduced power output with JFETs. Movie nights with friends and family often "redlines" the audio chain, so power output is also important to me.
I think I need to spend some more time listening before I change anything else...
Edit: the little "kink" in the sine wave trace posted just above is present with BOTH the 9610 and the JEFT differential just prior to clipping into 8 ohms...
I think I still need to play around with values for C9/C10 in my JFET amp to better control oscillation. I think I'll just order a wide variety of caps in the pF range and seeing which ones seem to work better. Right now, I have 680pF cap in place and behavior is clearly better than with no cap here, but still have some collapse in power output beyond clipping, but only at the top end of the audio spectrum (20kHz) and only into 4R loads.
Directly comparing amps with JFETs and 9610's, I am getting more stable relative offset over time, greater power output, and wider bandwidth with the 9610's. With JFETs, maximum power is a bit reduced and waveform deterioration with the JFETs gets nastier beyond clipping. 10kHz square waves at 10Vrms are identical between the 9610s and the JFETs.
With 9610's and ~22v rails I can get 4mV relative offset and:
8.8A bias provides 120w/4R and 78w/8R
9.25A bias provides 144w/4R and 90w/8R
9.6A bias provides 156w/4R and 98w/8R
I don't quite understand why output into 8R increases with bias- I thought 8R output was voltage limited and 4R output was current limited...
With JEFTs and ~22v rails, I can get ~30mV relative offset cold and ~100mV of relative offset when warm and:
8.8A bias provides 90w/4R and 80w/8R
9.25A bias provides 115w/4R and 83w/8R
9.6A bias provides 119w/4R and 83w/8R
Putting the amp back down in my theater room, the obvious difference with the new JFET front end is that I get more treble and more balanced midrange. I find these to be welcome improvements, though the magnitude of the change is small. At first, I wondered if this is worth it *for me* since my theater setup makes use of EQ in the preamp - thus, this is something I could largely duplicate with EQ instead of changes to the amp. For someone who uses the amp in a more "purist" form (source, volume control, amp, speaker, no EQ), the change to JFETs is clearly valuable.
But after a day of listening to music (in pure direct mode, which bypasses any digital EQ in my preamp), I am detecting another interesting improvement: It's hard to describe it well, but the impression I have is that the speaker driven by the 9610 amp seems to "call more attention to itself." Music seems to be coming more directly from the speaker, so I'd call that speaker more localizable during listening. The speaker driven by the JFET amp seems to have "disappeared" more into the room. I can't as easily identify the physical location of the speaker while listening.
At first, I thought this was an SPL imbalance between the amps due to the change, but a few minutes with my meter proves this theory wrong.
With the JFETs, the speaker seems to have completely disappeared, leaving behind a wider soundstage and more focused imaging. This is clear while listening to both music and movies.
I will have to spend some time with higher SPL movie playback to see if I can detect any signs of earlier clipping due to the reduced power output with JFETs. Movie nights with friends and family often "redlines" the audio chain, so power output is also important to me.
I think I need to spend some more time listening before I change anything else...
Edit: the little "kink" in the sine wave trace posted just above is present with BOTH the 9610 and the JEFT differential just prior to clipping into 8 ohms...
Last edited:
Hi Eric,
servo works fine. I made a second one with a lot more filtering on in- and outputs and the power supply. This is more silent now but I can still hear a slight hum with my ear to the speaker, the other side is completely silent.
Soundwise I can´t hear any difference.
I watched the other channel a bit longer, it varies between +3 and -1volt absolute offset depending on the mains voltage (with 27k matched McMillans).
If I keep the mains voltage constant it varies less than a volt from startup. (starts at 11V and then goes down to 1V within 15 seconds)
So if your mains is reasonably constant I would go for the PTC without the dc-servo. This will also manage your temperature differences without a problem.
I will try to remove the McMillans and see what happens with the sound. This will take a while though as I'm away for work almost constantly for the next 3 weeks and need to build a second one for the other channel. Then I can probably also post a schematic and some pics.
Not sure what´s wrong with your relative offset. Mine is around 3mV and stays there independent of temperature.
Don't know what causes the power raise into 8R and difference between 9610 and Jfets. I would try measuring the output voltage without a load and see where you land. With 8R and your bias the max voltage should be very close to that without load (you just loose some voltage over the output and source resistors)
I remember having 100 watts into 8R and 4R with my 7.5A bias and around 23V rails. Did you calculate the power and compared it to your results?
I didn't see any difference in power between 9610 and JFets, only bandwidth changed.
If you calculate the power difference in dB I don't think you will hear any difference in clipping behavior.
The little kink belongs to the Aleph-X
William
servo works fine. I made a second one with a lot more filtering on in- and outputs and the power supply. This is more silent now but I can still hear a slight hum with my ear to the speaker, the other side is completely silent.
Soundwise I can´t hear any difference.
I watched the other channel a bit longer, it varies between +3 and -1volt absolute offset depending on the mains voltage (with 27k matched McMillans).
If I keep the mains voltage constant it varies less than a volt from startup. (starts at 11V and then goes down to 1V within 15 seconds)
So if your mains is reasonably constant I would go for the PTC without the dc-servo. This will also manage your temperature differences without a problem.
I will try to remove the McMillans and see what happens with the sound. This will take a while though as I'm away for work almost constantly for the next 3 weeks and need to build a second one for the other channel. Then I can probably also post a schematic and some pics.
Not sure what´s wrong with your relative offset. Mine is around 3mV and stays there independent of temperature.
Don't know what causes the power raise into 8R and difference between 9610 and Jfets. I would try measuring the output voltage without a load and see where you land. With 8R and your bias the max voltage should be very close to that without load (you just loose some voltage over the output and source resistors)
I remember having 100 watts into 8R and 4R with my 7.5A bias and around 23V rails. Did you calculate the power and compared it to your results?
I didn't see any difference in power between 9610 and JFets, only bandwidth changed.
If you calculate the power difference in dB I don't think you will hear any difference in clipping behavior.
The little kink belongs to the Aleph-X
William
Pass DIY Addict
Joined 2000
Paid Member
Choky, I think I'm beginning to understand what you've been telling me... Thanks for the additional link, I'll have to read it a few more times for it to sink in more. DZ in Graeme's schematic is just just a single 9.1v Zener in my amp right now. I'll add another one in series along with a small cap and see if things improve.
When cold, my JFET amp is quieter at the speaker than the 9610, but when warm they are the same.
Pass DIY Addict
Joined 2000
Paid Member
Choky: The zener in Newby's nice writeup is a 5.6v. I presume the point is to control the swing of the input signal to keep it from overwhelming the current source. Thus, are "normal" 9.1v zeners too high for this task? My amp starts to clip with an input signal of about 2.1Vrms.
I have a pile of 9.1v zeners in my box. I would need to order 5.6v ones...
Thanks,
Eric
I have a pile of 9.1v zeners in my box. I would need to order 5.6v ones...
Thanks,
Eric
Hi Eric,
the 5.6V Zener has nothing to do with the input voltage to the diff pair. It keeps the current source fets from switching completely on by limiting the gate-source voltage to a value just above the normal "clipping" gate voltage.
I have tried those "clamping" zeners in my amps but in my case 5,6V didn´t help against the kink. Probably because the voltage depends on the source resistors used. I never looked deeper into this but it probably is worth another try.
William
the 5.6V Zener has nothing to do with the input voltage to the diff pair. It keeps the current source fets from switching completely on by limiting the gate-source voltage to a value just above the normal "clipping" gate voltage.
I have tried those "clamping" zeners in my amps but in my case 5,6V didn´t help against the kink. Probably because the voltage depends on the source resistors used. I never looked deeper into this but it probably is worth another try.
William
Pass DIY Addict
Joined 2000
Paid Member
I've been playing around with various CCS configurations (I copied Graeme's directly) for the front end differential and haven't really had any more "success" in restoring the lost power (yes, probably only 1.0-1.5dB as William points out) when I left the 9610's behind.
I am now getting more stable output into 4R loads, but still get some voltage collapse beyond 20kHz clipping. Output into 8R shows more "regular" clipping patterns and power output continues to increase right up until full rail voltage hits the outputs - though this is heavily clipped. As frequency increases, power output falls, so I need to play around with a greater variety of compensation caps for C9/C10 and C2/C4.
I tried Zen's back to back zener trick and this didn't really change the little "kink" on the downside of the sinewave at clipping.
Relative offset still creeps up slowly, starting about -13mV and seems to settle around 84mV or so. This seems closely tied to the ZTX550 parts. I have four of them all in a row and if I pinch them together with my fingers, relative offset drops to single digits of mV - which made me very happy. So, I put some thermal goo between them and pulled them tightly together with a plastic zip tie. This didn't do as much as I had hoped to keep relative offset from drifting upward... After a little while, it still climbed to ~100mV. I wonder if I should try to wrap a little copper band around them and tie this off as well.
Still playing around. After a few weeks the frustration is wearing off and being replaced by a spirit of adventure
What are the (dis)advantages of removing the 550's and just going with the 2SJ74's? In the Aleph-J schematic, Nelson just used 109's directly... This looks more like the approach that William used...
I am now getting more stable output into 4R loads, but still get some voltage collapse beyond 20kHz clipping. Output into 8R shows more "regular" clipping patterns and power output continues to increase right up until full rail voltage hits the outputs - though this is heavily clipped. As frequency increases, power output falls, so I need to play around with a greater variety of compensation caps for C9/C10 and C2/C4.
I tried Zen's back to back zener trick and this didn't really change the little "kink" on the downside of the sinewave at clipping.
Relative offset still creeps up slowly, starting about -13mV and seems to settle around 84mV or so. This seems closely tied to the ZTX550 parts. I have four of them all in a row and if I pinch them together with my fingers, relative offset drops to single digits of mV - which made me very happy. So, I put some thermal goo between them and pulled them tightly together with a plastic zip tie. This didn't do as much as I had hoped to keep relative offset from drifting upward... After a little while, it still climbed to ~100mV. I wonder if I should try to wrap a little copper band around them and tie this off as well.
Still playing around. After a few weeks the frustration is wearing off and being replaced by a spirit of adventure
What are the (dis)advantages of removing the 550's and just going with the 2SJ74's? In the Aleph-J schematic, Nelson just used 109's directly... This looks more like the approach that William used...
Last edited:
Pass DIY Addict
Joined 2000
Paid Member
I forgot to include in above, when I moved to Graeme's CSS for the input differential, I had to change VR2 from a 200R to a 1K pot in order to adjust absolute offset to 0v. The 200R pot didn't give me enough room to bring offset down to zero.
Using Graeme's schematic for labels, I have the following:
R18/R28 (only resistor between input signal and jfets) = 10k
R19/R29 (input to ground) = 10k - this seems to be a decade higher than input resistor in most other designs... Why is it 10k here?
R16/R30 (feedback) = 100k
R47/R47 (McMillan) = 10k
R23/R25 (connects ZTX550 collector to -Rail) = 392R
Using Graeme's schematic for labels, I have the following:
R18/R28 (only resistor between input signal and jfets) = 10k
R19/R29 (input to ground) = 10k - this seems to be a decade higher than input resistor in most other designs... Why is it 10k here?
R16/R30 (feedback) = 100k
R47/R47 (McMillan) = 10k
R23/R25 (connects ZTX550 collector to -Rail) = 392R