Hi Guys,
I finally got the boards populated. To my amazement, the first board fired up without issue. I was able to set the offset to zero and the bias control works properly. However, I haven't been able to find where it says what the bias should be set to. I have it set to 60mA per rail presently. If one of you could help with that I would appreciate it. The second channel fired up too but on that channel I have -1.2V offset so I have some searching to do. The offset pot has very little affect. Hopefully I will be able to track down the issue using the other channel as reference. Both channels will play sine and square waves. The sine waves look fine but the square waves have a little squiggle on the leading edge. Is that normal for this amp?
Thanks, Terry
I finally got the boards populated. To my amazement, the first board fired up without issue. I was able to set the offset to zero and the bias control works properly. However, I haven't been able to find where it says what the bias should be set to. I have it set to 60mA per rail presently. If one of you could help with that I would appreciate it. The second channel fired up too but on that channel I have -1.2V offset so I have some searching to do. The offset pot has very little affect. Hopefully I will be able to track down the issue using the other channel as reference. Both channels will play sine and square waves. The sine waves look fine but the square waves have a little squiggle on the leading edge. Is that normal for this amp?
Thanks, Terry
Attachments
Member
Joined 2009
Paid Member
Nice work Terry !
60mA of idle current through the output stage is fine (I'll check to see if I had an optimum written down).
I will have to check my notes, but generally I don't use square waves at such high frequencies and haven't noted any 'squigles'. You do have a load connected ? Interesting that they are not the same shape at the bottom and top of each transition i.e. the behaviour is asymmetric. Perhaps the compensation I'm using is a little light. Have you played any music yet ?
I would point out that the CFP output stage is susceptible to cross-conduction because the minority charge carries are not easily swept out of the bases of the power devices. If you read back a few pages you'll find I did some investigations and increased current through the drivers. In simulations there should be less cross-conduction than would be normal for the original P3a but up above the audio range there will be some. Be careful not to let the outputs get too hot when running square waves at 100kHz if you have a load.
60mA of idle current through the output stage is fine (I'll check to see if I had an optimum written down).
I will have to check my notes, but generally I don't use square waves at such high frequencies and haven't noted any 'squigles'. You do have a load connected ? Interesting that they are not the same shape at the bottom and top of each transition i.e. the behaviour is asymmetric. Perhaps the compensation I'm using is a little light. Have you played any music yet ?
I would point out that the CFP output stage is susceptible to cross-conduction because the minority charge carries are not easily swept out of the bases of the power devices. If you read back a few pages you'll find I did some investigations and increased current through the drivers. In simulations there should be less cross-conduction than would be normal for the original P3a but up above the audio range there will be some. Be careful not to let the outputs get too hot when running square waves at 100kHz if you have a load.
Last edited:
Hi Gareth,
Thanks for the info. The reason I have gotten into the habit of looking at square waves is they give a little more detail than sine waves. Almost every amp I have has normal looking sine waves. Almost all of them look a little different from one another when amplifying square waves. I learned my lesson about playing them at high volume. I normally set the output at 10Vac at 1k and leave it there. I haven't burned a zobel since I started that practice. That ripple is there even at 10K but just shows up as dashes. If you didn't play square waves through yours I guess you can't tell me if this is normal for this design of if I have something wrong.
I found the issue with the offset on the other channel. I had D1 in backward. Unfortunately, when removing it I lost the little cap I had soldered in parallel. So, that channel still has issues. It plays distorted and though I can zero the offset, when playing the offset fluctuates a lot. I've got some more searching to do. The good channel sounds fine.
Blessings, Terry
Thanks for the info. The reason I have gotten into the habit of looking at square waves is they give a little more detail than sine waves. Almost every amp I have has normal looking sine waves. Almost all of them look a little different from one another when amplifying square waves. I learned my lesson about playing them at high volume. I normally set the output at 10Vac at 1k and leave it there. I haven't burned a zobel since I started that practice. That ripple is there even at 10K but just shows up as dashes. If you didn't play square waves through yours I guess you can't tell me if this is normal for this design of if I have something wrong.
I found the issue with the offset on the other channel. I had D1 in backward. Unfortunately, when removing it I lost the little cap I had soldered in parallel. So, that channel still has issues. It plays distorted and though I can zero the offset, when playing the offset fluctuates a lot. I've got some more searching to do. The good channel sounds fine.
Blessings, Terry
Attachments
Member
Joined 2009
Paid Member
It's entirely possible that the square waves at high frequency have their foibles, the amplifier has some unusual compensation, it has high OLG, it has a hybrid BJT-FET CFP output and the FETs operate in Class C and probably don't turn on at exactly the same points in the waveform. But it does sound nice to my ears.
I did find my notes. I also used a rail current of 60mA when setting the bias.
I tested mostly with sine waves, all the way up to 2MHz as it happens. For square waves I used the probe test output on the front of my O'scope. It's limited to 1kHz but it's a super clean looking square wave, much better than anything out of my signal generator. The transition is very fast and very clean. Any how my notes say:
"No - load, 1kHz sq. wave from scope with +/- 0.25V, O/P aprpox +/- 7.5V, gain approx 30 sharp transitions.
Sq. wave into 8R load. Quick test - see about 1/2A per rail. Expected current = 7.5/8 x duty cycle 50% is approx 1/2A - O.K.
Scope trace looks clean, slight rounding on corners, no ripple."
You have a nice chunky heatsink on your set up. I'm amazed at how many amplifiers you build.
Do I see one of the amplifier boards with one of the power rail caps being different than all the others ?
Edit: IF I remember correctly, the cap across D1 can be left out and it still works, so something else is bugging it.
I did find my notes. I also used a rail current of 60mA when setting the bias.
I tested mostly with sine waves, all the way up to 2MHz as it happens. For square waves I used the probe test output on the front of my O'scope. It's limited to 1kHz but it's a super clean looking square wave, much better than anything out of my signal generator. The transition is very fast and very clean. Any how my notes say:
"No - load, 1kHz sq. wave from scope with +/- 0.25V, O/P aprpox +/- 7.5V, gain approx 30 sharp transitions.
Sq. wave into 8R load. Quick test - see about 1/2A per rail. Expected current = 7.5/8 x duty cycle 50% is approx 1/2A - O.K.
Scope trace looks clean, slight rounding on corners, no ripple."
You have a nice chunky heatsink on your set up. I'm amazed at how many amplifiers you build.
Do I see one of the amplifier boards with one of the power rail caps being different than all the others ?
Edit: IF I remember correctly, the cap across D1 can be left out and it still works, so something else is bugging it.
Last edited:
Hi Gareth,
Please don't think I am questioning the design. I just wanted to make sure it was behaving as it should.
The one channel has issues. The voltages around Q4 are off. Now I am worried that I may D2 in wrong as well. I tried hard but I am still not a fan of SMD. Bad thing is D2 is buried below one of the outputs
The bad thing was that I didn't discover the components drawing until after I had already installed all the parts and the screen print was very hard to read. I'm going to have to go over the boards very closely now and see if I have anything else reversed.
Fun stuff. 😀
Please don't think I am questioning the design. I just wanted to make sure it was behaving as it should.
The one channel has issues. The voltages around Q4 are off. Now I am worried that I may D2 in wrong as well. I tried hard but I am still not a fan of SMD. Bad thing is D2 is buried below one of the outputs
The bad thing was that I didn't discover the components drawing until after I had already installed all the parts and the screen print was very hard to read. I'm going to have to go over the boards very closely now and see if I have anything else reversed.
Fun stuff. 😀
I did some subjective tests with original P3A and found that optimal bias sounds excellent. That is 15mA per output pair (10mV across two 0R33 resistors). I think it is one of the great advantages of CFP outputs that you can have super cool amp and at the same time get very good lab results and sound. It is possible to completely enclose inside of amp (including heatsinks), we do not need dust inside. Or if very loud listening is not expected we can just use 3mm alu chassis as heatsink.
Member
Joined 2009
Paid Member
Hi Terry - no issue even if you have questions about the design, we're all learning. Yes, I found the SMD diodes hard to install because the markings are difficult to see. The board is not an easy one to work with being so dense and repairs will not be beginner friendly!
Hi Ivan - I too like the cooler running of these amps. My TGM7 runs considerably hotter which is less convenient.
Hi Ivan - I too like the cooler running of these amps. My TGM7 runs considerably hotter which is less convenient.
No one has made comment on the oscillation of Stil's square waves.
pics 2 & 3 of post621 show it after the negative slope.
Does it get worse when the load is slightly reactive?
Post623 shows a very broad vertical. Is this an indication of instability?
i.e. lots of superimposed 10kHz starting at slightly different times.
pics 2 & 3 of post621 show it after the negative slope.
Does it get worse when the load is slightly reactive?
Post623 shows a very broad vertical. Is this an indication of instability?
i.e. lots of superimposed 10kHz starting at slightly different times.
Last edited:
Hi, Gareth,
Like I said, I really did try to be very careful when installing everything. I built both boards together so if one was wrong they both should have been. I was hating life while doing the initial install because I hadn't found the printout of the top of the board and the screen print was so faint I could barely read it. I was having to trace the foil circuits and compare them to the schematic to figure out which part was which. I didn't think about all the parts that would be buried behind the outputs. I'll be digging into that board this morning to see if I can find my error. I'll sit down with the schematic and write down all the voltages and go from there. Always something to learn. I always have a better understanding of a circuit after I have had to troubleshoot it.
So are you guys saying optimal bias for this amp is just 5mV across one 0R33 resistor?
Thanks, Terry
Like I said, I really did try to be very careful when installing everything. I built both boards together so if one was wrong they both should have been. I was hating life while doing the initial install because I hadn't found the printout of the top of the board and the screen print was so faint I could barely read it. I was having to trace the foil circuits and compare them to the schematic to figure out which part was which. I didn't think about all the parts that would be buried behind the outputs. I'll be digging into that board this morning to see if I can find my error. I'll sit down with the schematic and write down all the voltages and go from there. Always something to learn. I always have a better understanding of a circuit after I have had to troubleshoot it.
So are you guys saying optimal bias for this amp is just 5mV across one 0R33 resistor?
Thanks, Terry
The drivers need to have some bias.
The outputs should also have some bias.
The total bias for both these pairs can be quite low, but ensure that neither reaches zero.
It has not been spelled out by D.Self, who is usually pretty thorough, on what proportion of the total bias should flow through the drivers or even whether there is a particular driver bias that is optimal.
If one copied the CFP of D.Self and set the stage to optimal bias for those particular values, then you do find that the bias current and the bias voltage are very low cf. an EF stage.
But what happens to optimal bias when the component values are varied? Self did spell out the EF variations for optimal bias.
The outputs should also have some bias.
The total bias for both these pairs can be quite low, but ensure that neither reaches zero.
It has not been spelled out by D.Self, who is usually pretty thorough, on what proportion of the total bias should flow through the drivers or even whether there is a particular driver bias that is optimal.
If one copied the CFP of D.Self and set the stage to optimal bias for those particular values, then you do find that the bias current and the bias voltage are very low cf. an EF stage.
But what happens to optimal bias when the component values are varied? Self did spell out the EF variations for optimal bias.
That is why I posted the pics. I haven't seen these shapes before so I hoped someone would recognize what was going on. I included all of the parts. It may be that I was supposed to make some decisions along the way and didn't. Both boards show that spotted up slope. The sine waves look good except they are a little thicker than usual.
I have been following this thread from the beginning but dont remember all the details along the way. I have to be careful when searching back through the thread because the design was being worked out as the thread progressed so all information is not still accurate.
Blessings, Terry
Member
Joined 2009
Paid Member
The driver current is set, more or less, by the driver collector resistors. The bias does have a different optimal point than the more common EF output stage but I don't off the top of my head know what the optimal point for CFP is. What I can say up front is that 60mA total rail current for the amplifier of 60 mA sounds very good. If I get time, I'll try to provide some better guidance. Maybe Christian can comment on his experience ?
The blips on the rising slope have been commented on earlier. Maybe in a different Thread.
They are a small oscillation superimposed on the fast slope.
I chose to ignore them in my amp re-build.
It's the oscillation/ringing at the end of the falling slope that I see and wonder if it means much, or little.
I know that my knowledge of AC behaviour and compensation tuning is very lacking.
I hope some one with much more knowledge will come in a tell us to ignore or "do this" to solve.
But adding some capacitance across the dummy load might be more revealing of what is trying to show it's character.
They are a small oscillation superimposed on the fast slope.
I chose to ignore them in my amp re-build.
It's the oscillation/ringing at the end of the falling slope that I see and wonder if it means much, or little.
I know that my knowledge of AC behaviour and compensation tuning is very lacking.
I hope some one with much more knowledge will come in a tell us to ignore or "do this" to solve.
But adding some capacitance across the dummy load might be more revealing of what is trying to show it's character.
Last edited:
Audio Precision analyzer is out of reach of most diy-ers. So we are unable to set proper bias for any combination of circuit values. I tried 15mA for P3A and was satisfied that it sounds good even with such low bias.
Is there any method to calculate proper bias for CFP outputs that mere mortals can use instead of just dreaming to buy AP system?
D.Self gives a lot of information for his specific component values.
It is well worth a thorough read.
Then you can think about what might need altering if you change a few component values.
BUT,
Self arrived at his conclusion by distortion analysis, not by simulation. He used the sim to help confirm that his experimental analysis was correct.
It is well worth a thorough read.
Then you can think about what might need altering if you change a few component values.
BUT,
Self arrived at his conclusion by distortion analysis, not by simulation. He used the sim to help confirm that his experimental analysis was correct.
Member
Joined 2009
Paid Member
My advice is to err on the higher side of bias - I know the CFP has traditionally been seen as operating at lower bias than the EF, but switching artifacts from the CFP are nasty and I like to have a bit of extra current. The thermal control tends to err on the side of overcompensation so when the amp gets hot the bias reduces slightly and you want to avoid it getting too low.
The big advantage of the CFP is that the temperature of output devices do not significantly affect the crossover distortion of the stage.
The drivers do most of the setting of the bias requirement and the drivers do not vary in temperature that much.
If the drivers have a decent heatsink and have an appropriate bias current then you will find that crossover distortion does not vary much as the output pair change in temperature due to load currents.
The CFP is quite different from an EF stage in this respect.
Take a typical driver collector resistor @ 220r. The output device when biased on will apply ~ 660mVdc (1*Vbe rounded to an easy arithmetical number).
This gives a driver current of 3mA.
If the driver current changes to 6mA (due to it's output changing by +-100% of Ibias), then the Vbe of the output becomes 2 * 660mVbe = 1.2Vbe
Look at the output device datasheet and the collector current will be off the top of the plot.
This tells us the the driver current changes by < 100% of Ibias. +-30% of Ibias is probably a more typical maximum output for the drivers. Simulation experts can look into this. Leach gave some numbers for driver currents in his Lo Tim paper (but it was for an EF).
The result is that the driver current changes from 3mA to a range of 2mApk to 4mApk over a maximum signal cycle.
That, for a +-35Vdc supply approximates to Pq = ~100mW and transient (instantaneous) peaks varying from 70mW to 130mW.
The drivers essentially don't change in temperature with output current. Their Vbe do drift downwards as the internal temperature rises from cold to fully warmed up, but this is adjusted out by appropriate bias tempco.
The drivers do most of the setting of the bias requirement and the drivers do not vary in temperature that much.
If the drivers have a decent heatsink and have an appropriate bias current then you will find that crossover distortion does not vary much as the output pair change in temperature due to load currents.
The CFP is quite different from an EF stage in this respect.
Take a typical driver collector resistor @ 220r. The output device when biased on will apply ~ 660mVdc (1*Vbe rounded to an easy arithmetical number).
This gives a driver current of 3mA.
If the driver current changes to 6mA (due to it's output changing by +-100% of Ibias), then the Vbe of the output becomes 2 * 660mVbe = 1.2Vbe
Look at the output device datasheet and the collector current will be off the top of the plot.
This tells us the the driver current changes by < 100% of Ibias. +-30% of Ibias is probably a more typical maximum output for the drivers. Simulation experts can look into this. Leach gave some numbers for driver currents in his Lo Tim paper (but it was for an EF).
The result is that the driver current changes from 3mA to a range of 2mApk to 4mApk over a maximum signal cycle.
That, for a +-35Vdc supply approximates to Pq = ~100mW and transient (instantaneous) peaks varying from 70mW to 130mW.
The drivers essentially don't change in temperature with output current. Their Vbe do drift downwards as the internal temperature rises from cold to fully warmed up, but this is adjusted out by appropriate bias tempco.
Last edited:
Success! I found the issue with the second channel. I did indeed have D2 reversed. I swapped it around but that didn't solve the problem. I still had rail voltage on the gate of Q8. Q8 measured fine with the amp powered down but Q6 didn't. I evidently stressed it when I bolted it to Q5 because the base pin just fell out of it when I removed it from the board. A new Q6 and it is playing perfectly. I have it on my A/B setup along with my favorite VSSA and they are like twins aside from the TMG8 having slightly more gain. You done good Gareth! I can see why you were so satisfied. I've got some Gerry Mulligan playing through it right now and it is beautiful. Thanks for sharing this with me.
Blessings, Terry
Blessings, Terry
Member
Joined 2009
Paid Member
Terry - glad you sorted it out - it's not an easy build and you have things all working in very little time. Your praise of the amp carries a lot of credibility with me; you have a lot of experience building and listening to DIY amps and if you say it's a good, then it is good 🙂
Well ..Gareth ...thread is quiet long ...it will take too much time for me to read all ...
Here is a couple of Questions
lets suppose that we make a blind test between a P3a and a your amp in a given power of lets say 10 W which amplifier will sound better ?
Second Andrew T got a point as always square wave and resistive load + capacitor is in order here obviously measuring above 10K will present more ( if any ) problems i think one should do that ...
May Andrew said "dont bother for this type of oscillation "" but sorry not me .... i would like to have it cleared out ..
Kind regards
Sakis
Here is a couple of Questions
lets suppose that we make a blind test between a P3a and a your amp in a given power of lets say 10 W which amplifier will sound better ?
Second Andrew T got a point as always square wave and resistive load + capacitor is in order here obviously measuring above 10K will present more ( if any ) problems i think one should do that ...
May Andrew said "dont bother for this type of oscillation "" but sorry not me .... i would like to have it cleared out ..
Kind regards
Sakis