Re: Re: Re: Re: PGP
Hi Alex,
The idle current was 150 mA in a single pair of output devices. The output stage distortion was measured looking into an 8 ohm load.
The 0.06% that I quoted corresponds well with my picture in Figure 10 of the paper. It is also within a swag of the closed loop distortion of 0.006% reduced by the NFB at 20 kHz of 40 dB, corresponding to a 2 MHz gain crossover.
However, there is a big nit here, and that is that the THD should normally be reduced by the amount of feedback at the harmonic frequencies, which will be 6 dB and 10 dB less for dominant second and thirds, respectively. So, it is possible that the output stage distortion in the final amplifier as measured was somewhat lower than 0.06 percent, as Edmond speculated. Maybe between 0.02 and 0.03%.
Incidentally, had I used three output pairs, each biased at 150 mA, the output stage distortion would have been reduced considerably.
Cheers,
Bob
x-pro said:
Hi Alex,
The idle current was 150 mA in a single pair of output devices. The output stage distortion was measured looking into an 8 ohm load.
The 0.06% that I quoted corresponds well with my picture in Figure 10 of the paper. It is also within a swag of the closed loop distortion of 0.006% reduced by the NFB at 20 kHz of 40 dB, corresponding to a 2 MHz gain crossover.
However, there is a big nit here, and that is that the THD should normally be reduced by the amount of feedback at the harmonic frequencies, which will be 6 dB and 10 dB less for dominant second and thirds, respectively. So, it is possible that the output stage distortion in the final amplifier as measured was somewhat lower than 0.06 percent, as Edmond speculated. Maybe between 0.02 and 0.03%.
Incidentally, had I used three output pairs, each biased at 150 mA, the output stage distortion would have been reduced considerably.
Cheers,
Bob
Re: Re: Re: Re: PGP
I have for some time used verticals and protected them with a circuit that I discussed awhile back. I share your distrust of protection circuits on the sound quality. This protection circuit is totally out of the picture except in the case of a short circuit.
You've got enough safe area in three output MOSFET pairs to not need SOA protection. All you need is catastrophic short circuit protection.
My circuit latches and immediately removes gate drive to all output MOSFETs in the event that the current in a MOSFET exceeds some threshold current for more than some threshold time. I usually set it for 10 Amps and whatever number of ms the SOA of the device will handle at the rail voltage (usually some number of tens of ms).
The power need only be cycled to turn the amplifier back on. It has saved me a number of times.
Cheers,
Bob
syn08 said:
I have for some time used verticals and protected them with a circuit that I discussed awhile back. I share your distrust of protection circuits on the sound quality. This protection circuit is totally out of the picture except in the case of a short circuit.
You've got enough safe area in three output MOSFET pairs to not need SOA protection. All you need is catastrophic short circuit protection.
My circuit latches and immediately removes gate drive to all output MOSFETs in the event that the current in a MOSFET exceeds some threshold current for more than some threshold time. I usually set it for 10 Amps and whatever number of ms the SOA of the device will handle at the rail voltage (usually some number of tens of ms).
The power need only be cycled to turn the amplifier back on. It has saved me a number of times.
Cheers,
Bob
Re: Re: Re: Re: Re: Re: Re: PGP
Hi Edmond - I am using the model from here:
http://www.diyaudio.com/forums/showthread.php?postid=1316183#post1316183
in my O/P stage only N-ch devices used. It was quite interesting to see a close level of performance to an EC circuit - after Bob mentioned it.
Cheers
Alex
P.S. - is there a large difefrence with a later model? I also got close performance results using HUF76639 models from Fairchild Semi.
Edmond Stuart said:
Hi Edmond - I am using the model from here:
http://www.diyaudio.com/forums/showthread.php?postid=1316183#post1316183
in my O/P stage only N-ch devices used. It was quite interesting to see a close level of performance to an EC circuit - after Bob mentioned it.
Cheers
Alex
P.S. - is there a large difefrence with a later model? I also got close performance results using HUF76639 models from Fairchild Semi.
Re: Re: Your amplifier
Hi Edmond, you are most welcome for the appreciative words, and your effort here is very deserving of it.
I must admit, I was confused enough about the bootstrapping that it was not a typo, but it was also not backed up by a lot of thought. Is it the VAS that is bootstrapped to the output signal?
Two things usually cause me to exercise caution when bootstrapping. I do use it, by the way, in some cases. The first thing that worries me is the possible appearance of some unwanted feedback, maybe even positive, at high frequencies. The second is that the circuit may behave in unanticipated ways under clipping or fault conditions.
Regarding the d.c. servo, I always seem to have +/- 15 V lying around, so that is not a problem. Audio-grade op amps should be used in servo circuits, but it is a misconception that when properly used in a servo circuit they need to be as good as the signal path. In a poorly designed servo, where the output of the op amp was only minimally attenuated before being applied to the amplifier input, one would have to have an exceptionally good one. If, on the other hand, its output is attenuated by 100:1 or even 1000:1, depending on amount of correction needed, it need not be as good (i.e., not garbage, but good audio grade, at least as good as an OPA604, for example).
Although there are some pretty good electrolytics out there, there is enough anecdotal information on capacitors so as to avoid electrolytics. The dc servo integrating capacitor should be decent, such as a polypropylene or better. My view is that if you believe that capacitors in the signal path can make a difference, then you virtually never put an electrolytic in the signal path. Of course, some people mistakenly think that a shunt capacitor in the feedback path is not in the signal path
.
I think the coil discussion was not without value. I'm not sure to what extent I buy into the coil thing either, but I do like to be on the safe side. To me, the safe side means two things: 1) there is a coil if it is necessary to make the design unconditionally stable; 2) the coil is as small as possible.
This, in my view, is not that much different than believing that 0.0001% is audibly superior to 0.001% distortion. There is a little bit of faith and belief system in each.
Here's another way of looking at it. Lets say we offered the community here their choice of two different amplifiers. Amplifier A has 0.0001% THD-20 and uses a 2.5 uH output coil. Amplifier B has 0.001% THD-20 and uses only a 0.5 uH output coil. I'm guessing that some would choose A and some would choose B.
In any case, the sweat and intellect that you have put into this amplifier has stimulated much very worthwhile discussion.
Cheers,
Bob
Edmond Stuart said:
Hi Edmond, you are most welcome for the appreciative words, and your effort here is very deserving of it.
I must admit, I was confused enough about the bootstrapping that it was not a typo, but it was also not backed up by a lot of thought. Is it the VAS that is bootstrapped to the output signal?
Two things usually cause me to exercise caution when bootstrapping. I do use it, by the way, in some cases. The first thing that worries me is the possible appearance of some unwanted feedback, maybe even positive, at high frequencies. The second is that the circuit may behave in unanticipated ways under clipping or fault conditions.
Regarding the d.c. servo, I always seem to have +/- 15 V lying around, so that is not a problem. Audio-grade op amps should be used in servo circuits, but it is a misconception that when properly used in a servo circuit they need to be as good as the signal path. In a poorly designed servo, where the output of the op amp was only minimally attenuated before being applied to the amplifier input, one would have to have an exceptionally good one. If, on the other hand, its output is attenuated by 100:1 or even 1000:1, depending on amount of correction needed, it need not be as good (i.e., not garbage, but good audio grade, at least as good as an OPA604, for example).
Although there are some pretty good electrolytics out there, there is enough anecdotal information on capacitors so as to avoid electrolytics. The dc servo integrating capacitor should be decent, such as a polypropylene or better. My view is that if you believe that capacitors in the signal path can make a difference, then you virtually never put an electrolytic in the signal path. Of course, some people mistakenly think that a shunt capacitor in the feedback path is not in the signal path
.I think the coil discussion was not without value. I'm not sure to what extent I buy into the coil thing either, but I do like to be on the safe side. To me, the safe side means two things: 1) there is a coil if it is necessary to make the design unconditionally stable; 2) the coil is as small as possible.
This, in my view, is not that much different than believing that 0.0001% is audibly superior to 0.001% distortion. There is a little bit of faith and belief system in each.
Here's another way of looking at it. Lets say we offered the community here their choice of two different amplifiers. Amplifier A has 0.0001% THD-20 and uses a 2.5 uH output coil. Amplifier B has 0.001% THD-20 and uses only a 0.5 uH output coil. I'm guessing that some would choose A and some would choose B.
In any case, the sweat and intellect that you have put into this amplifier has stimulated much very worthwhile discussion.
Cheers,
Bob
Glen, you have been asking a lot of very good questions. I am trying to work out what is actually going on in this design and in Bob Cordell's. It looks something like this to me so far:
PGP:
output stage THD20 -82dB
overall THD -120dB
=> main loop NFB of 38dB at 20kHz.
Measured at 400W avg into 4 ohms resistive
BC:
output stage THD -64dB
overall THD -104dB
=> main loop NFB of 40dB at 20kHz.
Measures at 50W avg into 8 ohms resistive
One of the key parameters that seems to be missing in all this is the relative stability margin of the two circuits. There is some evidence that the PGP is less stable than BC both from the size of output inductor and the under-damped pulse response that Bob has remarked on earlier. As Bob mentioned in the EC thread, the distortion performance cannot be fairly compared if the stability margins are different.
Some points were made about the improvement in raw OS distortion that has been possible in the past 23 years due to improved transistors (higher speed allowing more feedback and better linearity). I conservatively estimate that the BC OS circuit ought to be able to manage in the range of -70dB to -80dB with modern parts, or 10dB better overall THD, assuming the front end has low enough distortion.
Brian
Re: Re: Re: Re: Re: Re: Re: Re: PGP
Hi Alex,
If you are using only the n-channel model, no problem at all. Using the p-channel model is another story. I twas very difficult for Andy to "connect" the data from the (real) measurements of the weak inversion to the higher drain current characteristics obtained from the data sheet. Therefor I dropped a modified version, see also:
http://www.diyaudio.com/forums/showthread.php?postid=1322569#post1322569
and:
http://www.diyaudio.com/forums/showthread.php?postid=1323417#post1323417
> is there a large difference with a later model?
Yes, in regard of crossover distortion, it does.
Cheers, Edmond.
x-pro said:
Hi Alex,
If you are using only the n-channel model, no problem at all. Using the p-channel model is another story. I twas very difficult for Andy to "connect" the data from the (real) measurements of the weak inversion to the higher drain current characteristics obtained from the data sheet. Therefor I dropped a modified version, see also:
http://www.diyaudio.com/forums/showthread.php?postid=1322569#post1322569
and:
http://www.diyaudio.com/forums/showthread.php?postid=1323417#post1323417
> is there a large difference with a later model?
Yes, in regard of crossover distortion, it does.
Cheers, Edmond.
Re: Re: Re: Re: Re: Re: Re: PGP
Yes (only local NFB)
I attach the simulated FFT graph and circuit I've used:
potentially used P-channel models (from LTspice) may give somewhat inflated figures. That is why I would like to try it live .
Cheers
Alex
P.S. Conditions as I've described (idle current 150mA, output voltage 80V p-p, on the graph 20kHz is at +29dB)
Bob Cordell said:
Yes (only local NFB)
I attach the simulated FFT graph and circuit I've used:
potentially used P-channel models (from LTspice) may give somewhat inflated figures. That is why I would like to try it live
.Cheers
Alex
P.S. Conditions as I've described (idle current 150mA, output voltage 80V p-p, on the graph 20kHz is at +29dB)
Attachments
Re: Re: Re: Your amplifier
Hi again Bob:
Well, identifying the Amber auto settings was less difficult than I was expecting. OTOH I guess I'm not really famous for procrastinating
I did some extra measurements re: the Amber 5500 connected back to back and created a temporary web page.
http://www.synaesthesia.ca/annex.html
As soon as we'll figure out how to integrate this data in the main site we'll publish this as part of the Distortion measurement page.
Bob Cordell said:
Hi again Bob:
Well, identifying the Amber auto settings was less difficult than I was expecting. OTOH I guess I'm not really famous for procrastinating
I did some extra measurements re: the Amber 5500 connected back to back and created a temporary web page.
http://www.synaesthesia.ca/annex.html
As soon as we'll figure out how to integrate this data in the main site we'll publish this as part of the Distortion measurement page.
Edmond Stuart said:
Thanks Edmond.
I think that the reason your version initially had a fighting VAS issue is because you tried to lighten the load on the current mirrors too much.
That's why I stuck with 1k loads for the 12W (not-so-serious) amp. Because of this, the 12W amp doesn't have a fighting VAS issue at all. Although this has attracted some criticism, when using BJT LTP's a 1k load really isn't that bad (With JFETs for the LTP's it would be an entirely different story).
Due to the high transconductance of SS BJTs, enough LTP gain can still be had (which is double with the I mirrors) and the linearity is still excellent - and lightening the load even by 10 times really does surprisingly little for the amplifiers overall THD performance; the majority of the distortion comes from the VAS and the OPS, and this is where I am concentrating my efforts in my next version of the design - the 300W monster.
Incidentally though, I am using lighter loads this time for close to 40dB gain in each LTP, but I am also using dual MATXX trannies for the LTP's and SMD high ft/low Cob I-mirror configured dual pairs for the current mirrors and resistors matched for 0.1%.
Good luck with the RET OPS. Let us know how it goes.
Cheers,
Glen
Re: Re: Re: Re: Your amplifier
/slightly OT
That Amber is a well-kept secret. More than 10 or maybe even 15 years ago I was looking for an audio analyzer and stumbled on the Amber. Around that time, the company was purchased by Audio Precision and it's owner Wayne Jones was, IIRC made AP VP of Product development. I've never been able to locate an Amber for sale.
/OT off
Jan Didden
syn08 said:
Hi again Bob:
Well, identifying the Amber auto settings was less difficult than I was expecting. OTOH I guess I'm not really famous for procrastinating
I did some extra measurements re: the Amber 5500 connected back to back and created a temporary web page.
http://www.synaesthesia.ca/annex.html
As soon as we'll figure out how to integrate this data in the main site we'll publish this as part of the Distortion measurement page.
/slightly OT
That Amber is a well-kept secret. More than 10 or maybe even 15 years ago I was looking for an audio analyzer and stumbled on the Amber. Around that time, the company was purchased by Audio Precision and it's owner Wayne Jones was, IIRC made AP VP of Product development. I've never been able to locate an Amber for sale.
/OT off
Jan Didden
I must be in competent. I have done what you asked and am still no wiser, except for a power figure which I said was 400W rather than 200W.
PGP:
output stage THD20 -82dB (Ovidiu's measurement)
overall THD -120dB (Ovidiu's measurement)
=> main loop NFB of 38dB at 20kHz.
Measured at 200W avg into 4 ohms resistive
BC:
output stage THD20 -64dB (Bob's measurement)
overall THD -104dB (Bob's measurement)
=> main loop NFB of 40dB at 20kHz.
Measured at 50W avg into 8 ohms resistive
Which of these numbers is totally wrong?
Also, have you got the stability difference between the PGP and the BC amp? I believe you have simulated both circuits.
Thx.
Re: Re: Re: Re: Your amplifier
Thanks! Keep up the good work.
BTW, the noise of your amplifier must be pretty low to have gotten that earlier THD-20 picture with little more fuzz than the back-to-back in the 80 kHz bandwidth. Have you tried to calculate your amplifier's input-referred noise in nV/rt Hz?
Cheers,
Bob
syn08 said:
Hi again Bob:
Well, identifying the Amber auto settings was less difficult than I was expecting. OTOH I guess I'm not really famous for procrastinating
I did some extra measurements re: the Amber 5500 connected back to back and created a temporary web page.
http://www.synaesthesia.ca/annex.html
As soon as we'll figure out how to integrate this data in the main site we'll publish this as part of the Distortion measurement page.
Thanks! Keep up the good work.
BTW, the noise of your amplifier must be pretty low to have gotten that earlier THD-20 picture with little more fuzz than the back-to-back in the 80 kHz bandwidth. Have you tried to calculate your amplifier's input-referred noise in nV/rt Hz?
Cheers,
Bob
Re: Re: Re: Re: Re: Your amplifier
Jan:
Indeed, Amber Electro Design was a Canadian company founded by the fellow Canuck Wayne Jones in 1974. AED was purchased by AP in 1990 and Wayne was with AP first as Director of Technical Support, then Director of Product Development and ended (2000-2003) as VP of Application Engineering. He is currently Operations Manager at Intel Corp. If I recall correctly, Wayne has served on Standards Committees for the Audio Engineering Society.
The Amber analyzers were among the best in the 80's. They are built with standard parts, so they are very easy to mantain and repair. Even today the performance delivered by these instruments is amazing, AP has certainly learned a lot from these guys! I would heartly recommend getting one, if the budget is limited. They ocasionally pop on EBay for very low prices (compared to AP or HP or Boonton). The only problems I am having with the Amber 5500 (which was the top of the line analyzer) are the limited software support (National Instruments GPIB drivers are available but you still have to write every bit of GPIB code if you want any degree of automated measurements) and (most annoying) the Options availability. I am looking for quite some time for the IMD Option (I think it's 002) board and never found one. If anybody sees one around, I'm ready to pay top dollars for that.
janneman said:
Jan:
Indeed, Amber Electro Design was a Canadian company founded by the fellow Canuck Wayne Jones in 1974. AED was purchased by AP in 1990 and Wayne was with AP first as Director of Technical Support, then Director of Product Development and ended (2000-2003) as VP of Application Engineering. He is currently Operations Manager at Intel Corp. If I recall correctly, Wayne has served on Standards Committees for the Audio Engineering Society.
The Amber analyzers were among the best in the 80's. They are built with standard parts, so they are very easy to mantain and repair. Even today the performance delivered by these instruments is amazing, AP has certainly learned a lot from these guys! I would heartly recommend getting one, if the budget is limited. They ocasionally pop on EBay for very low prices (compared to AP or HP or Boonton). The only problems I am having with the Amber 5500 (which was the top of the line analyzer) are the limited software support (National Instruments GPIB drivers are available but you still have to write every bit of GPIB code if you want any degree of automated measurements) and (most annoying) the Options availability. I am looking for quite some time for the IMD Option (I think it's 002) board and never found one. If anybody sees one around, I'm ready to pay top dollars for that.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.