Changes and simulation should be trivial, and I could do it, but you would be the 1st builder, I'm afraid 🙂
Here is simple and 'low power' (relatively speaking) Beta Nirvana with smaller fets, and 15V rails.
Sim attached. Of course bigger fets also can be still used.
Sim attached. Of course bigger fets also can be still used.
I will look at your simulation options later. For the present there is a question over setting the output standing current for an actual build. The dissipation in the second of your two options is 44 W per output FET
Simulations are based on conditions at 25 degrees C normal room temperature. The negative temperature coefficient of FETs can be simulated in a Spice can be used to cover device temperatures - singly or sweep using .temp command. The difference may not be significant at some levels however Iq depends on the values for R8 and R9 and changing these could be awkward. Some experience with a prototype could be useful as guidance for prospective builders.
My interest in this amplifier is academic since I am happy with the JLH1996 Class A amplifier I built in that year.
By coincidence the heat sinks specified for 44 this amplifier were to have no more than 0.6 degrees C per W. For me that is somewhat shy of the real need to keep the heat below 60 degrees C. Mine were about 0.4 degrees per W and even at that level I felt they were not as good as I had hoped.
Simulations are based on conditions at 25 degrees C normal room temperature. The negative temperature coefficient of FETs can be simulated in a Spice can be used to cover device temperatures - singly or sweep using .temp command. The difference may not be significant at some levels however Iq depends on the values for R8 and R9 and changing these could be awkward. Some experience with a prototype could be useful as guidance for prospective builders.
My interest in this amplifier is academic since I am happy with the JLH1996 Class A amplifier I built in that year.
By coincidence the heat sinks specified for 44 this amplifier were to have no more than 0.6 degrees C per W. For me that is somewhat shy of the real need to keep the heat below 60 degrees C. Mine were about 0.4 degrees per W and even at that level I felt they were not as good as I had hoped.
Last edited:
I'm spicing up your latest, Minek, and will post it later today.
Right now I have to boot up my heart with my first coffee for the day!
Hugh
Right now I have to boot up my heart with my first coffee for the day!
Hugh
Hi Minek,
Your first analysis using LTSpice. I have changed the command lines to give distortion levels, and accommodate many different frequencies. First is the schematic, then the Bode plot which is remarkably flat attributed to the 1.5k and 22pF network from VAS collector to fb node.
However, at most frequencies and outputs the dominant harmonic is H3. At 1KHz @ 25W output (40Vpp//8R) H2 is -120dB and H3 -102dB. I like this situation to be reversed, as H2, albeit low at -100dB, can masked the H3. The PM of 89 degrees and GM of 20.8dB looks very good, and THD is very, very low however.
My schem and Bode plots attached.
Hugh
Your first analysis using LTSpice. I have changed the command lines to give distortion levels, and accommodate many different frequencies. First is the schematic, then the Bode plot which is remarkably flat attributed to the 1.5k and 22pF network from VAS collector to fb node.
However, at most frequencies and outputs the dominant harmonic is H3. At 1KHz @ 25W output (40Vpp//8R) H2 is -120dB and H3 -102dB. I like this situation to be reversed, as H2, albeit low at -100dB, can masked the H3. The PM of 89 degrees and GM of 20.8dB looks very good, and THD is very, very low however.
My schem and Bode plots attached.
Hugh
I compared FFT profiles (at 10kHz) of AN, and all the versions of BN in this thread, to see H2 vs H3,
and you are right. BN's H2 is at best on the same level at H3, sometimes slightly lower.
I think the difference in H2 level is less than 5% (compared to org AN). At the same time, Thd numbers are better.
Now, if there was a way to improve this H2/H3 ratio..
I just did couple of quick tweaks and simulations, and I see slight improvement (H2/H3) if shaping network (C9/R5) is taken from VAS
(not from OUT) - like in the original design. Hugh was right all along!
The reason to take feedback from OUT instead of VAS - I was unable to simulate OLG when it was taken from VAS.
And results were almost the same, I did not notice H2/H3 slightly different ratio..
Hugh, can you try and confirm this?
and you are right. BN's H2 is at best on the same level at H3, sometimes slightly lower.
I think the difference in H2 level is less than 5% (compared to org AN). At the same time, Thd numbers are better.
Now, if there was a way to improve this H2/H3 ratio..
I just did couple of quick tweaks and simulations, and I see slight improvement (H2/H3) if shaping network (C9/R5) is taken from VAS
(not from OUT) - like in the original design. Hugh was right all along!
The reason to take feedback from OUT instead of VAS - I was unable to simulate OLG when it was taken from VAS.
And results were almost the same, I did not notice H2/H3 slightly different ratio..
Hugh, can you try and confirm this?
Last edited:
Minek,
There are lots of moving parts in this. If we connect the profile shaper from VAS to fb node, we have much reduced global fb, much more nested fb, and almost as good a GM (20.2dB) and PM (82.5 degrees) but higher harmonics, mostly lower orders like H2, H3 and H4. Bode plot is below, note the profile shaper has a 15pF across the network, which smoothes the curve considerably.
The profile of the harmonics is also evident, with a skew towards low order, giving H2 dominant at all times. This is what I'm always concerned about as it delivers a Hiraga profile which I found sounds best. This does NOT give a lot THD however; it's higher, but not much higher, and still far, far less than a tube amp. Yet the profile is tube like, and that is the rub.
At 1KHz, H2 -82dB H3 -97dB H5 -107dB THD 0.00837% Notice this is little higher than 0.0017%, but in truth the extra is basically H2, H3 and H4, which are musical.
At 20KHz, H2 -75dB H3 -77dB H5 -83dB THD 0.026860%, almost 3.2 times more, but the profile is maintained although H3 is about to overtake H2 and you cannot hear H2 at this frequency at all, and even less at higher orders although there is some evidence that the perceives this less than natural. Furthermore, the very high frequencies do not attract a lot of high power; they would be in the milliwatts, not watts, so their distortion would be far, far less.
Schematic:
Schematic is first, then 1KHz harmonic profile, and last Bode plot for the first schematic.
CONCLUSION:
Conventional global fb reduces THD to minimal levels, as is well known. However, it works less at higher orders; so H2, H3 and H4, all musical, are more reduced than the H5 orders and beyond. These are more objectionable to the ear, so whilst THD, which takes no more account of higher order harmonics than lower, at the very low levels of harmonics conventional fb dominates the design landscape. This effectively strips some of the lower harmonics from the original signal, leaving a very low level of higher orders harmonics in the output. This presents as 'unnatural' sound.
But by reducing conventional fb we increase THD considerably, with less reduction of low orders. But we can reduce the higher distortion far more by then using nested fb, which delivers a cleaner signal to the output stage. And so using nested across two or three stages only, not including the output stage, we can reduce the production of higher harmonics produced in the output stage. If the output stage is Class A, its dominant harmonic should be H2, and this too can be used to mask some of the higher orders. We can also shape the nested fb so that the global fb, and hence the damping factor, by using the nfb through a small cap - C7 and R20 on this schematic. By using phase lead - C3 - we can smooth the Bode plot so that the 180 degree turnover occurs at very high frequencies, 4.85MHz on this circuit, improving the gain margin considerably.
HD
There are lots of moving parts in this. If we connect the profile shaper from VAS to fb node, we have much reduced global fb, much more nested fb, and almost as good a GM (20.2dB) and PM (82.5 degrees) but higher harmonics, mostly lower orders like H2, H3 and H4. Bode plot is below, note the profile shaper has a 15pF across the network, which smoothes the curve considerably.
The profile of the harmonics is also evident, with a skew towards low order, giving H2 dominant at all times. This is what I'm always concerned about as it delivers a Hiraga profile which I found sounds best. This does NOT give a lot THD however; it's higher, but not much higher, and still far, far less than a tube amp. Yet the profile is tube like, and that is the rub.
At 1KHz, H2 -82dB H3 -97dB H5 -107dB THD 0.00837% Notice this is little higher than 0.0017%, but in truth the extra is basically H2, H3 and H4, which are musical.
At 20KHz, H2 -75dB H3 -77dB H5 -83dB THD 0.026860%, almost 3.2 times more, but the profile is maintained although H3 is about to overtake H2 and you cannot hear H2 at this frequency at all, and even less at higher orders although there is some evidence that the perceives this less than natural. Furthermore, the very high frequencies do not attract a lot of high power; they would be in the milliwatts, not watts, so their distortion would be far, far less.
Schematic:
Schematic is first, then 1KHz harmonic profile, and last Bode plot for the first schematic.
CONCLUSION:
Conventional global fb reduces THD to minimal levels, as is well known. However, it works less at higher orders; so H2, H3 and H4, all musical, are more reduced than the H5 orders and beyond. These are more objectionable to the ear, so whilst THD, which takes no more account of higher order harmonics than lower, at the very low levels of harmonics conventional fb dominates the design landscape. This effectively strips some of the lower harmonics from the original signal, leaving a very low level of higher orders harmonics in the output. This presents as 'unnatural' sound.
But by reducing conventional fb we increase THD considerably, with less reduction of low orders. But we can reduce the higher distortion far more by then using nested fb, which delivers a cleaner signal to the output stage. And so using nested across two or three stages only, not including the output stage, we can reduce the production of higher harmonics produced in the output stage. If the output stage is Class A, its dominant harmonic should be H2, and this too can be used to mask some of the higher orders. We can also shape the nested fb so that the global fb, and hence the damping factor, by using the nfb through a small cap - C7 and R20 on this schematic. By using phase lead - C3 - we can smooth the Bode plot so that the 180 degree turnover occurs at very high frequencies, 4.85MHz on this circuit, improving the gain margin considerably.
HD
Last edited:
Hugh, if shaping RC network is taken from VAS, OLG can not be reliably measured/plotted using simple methods.
In one of the previous posts (post #76), Paul proposed OLG measuring method with 2 probes that will work for this case.
PM/GM numbers measured by your/mine method are meaningless (when feedback comes from VAS),
so we don't need to bother with these flat Bode plots, I don't think they show anything useful.
Paul indicated that using his method he was getting better results when feedback was taken from VAS, so we can ASSUME
that if PM/GM plots look good when feedback is taken from OUT, it will be at least NO WORSE than when taken from VAS.
I'll try to understand/reproduce his method of measurement with 2 probes...
My conclusion so far:
a) when simulating OLG/PM/GM - use feedback from OUT
b) when simulating FFT profile shaping (H2/H3), connect RC network to VAS
So, in case of feedback from VAS - if Thd and profile (H2/H3) look good, and OLG/PM/GM looked good when feedback was taken from OUT,
we should be good.
I'll use your latest schematic, and try it (squares, etc) tomorrow. I think that should be it.
In one of the previous posts (post #76), Paul proposed OLG measuring method with 2 probes that will work for this case.
PM/GM numbers measured by your/mine method are meaningless (when feedback comes from VAS),
so we don't need to bother with these flat Bode plots, I don't think they show anything useful.
Paul indicated that using his method he was getting better results when feedback was taken from VAS, so we can ASSUME
that if PM/GM plots look good when feedback is taken from OUT, it will be at least NO WORSE than when taken from VAS.
I'll try to understand/reproduce his method of measurement with 2 probes...
My conclusion so far:
a) when simulating OLG/PM/GM - use feedback from OUT
b) when simulating FFT profile shaping (H2/H3), connect RC network to VAS
So, in case of feedback from VAS - if Thd and profile (H2/H3) look good, and OLG/PM/GM looked good when feedback was taken from OUT,
we should be good.
I'll use your latest schematic, and try it (squares, etc) tomorrow. I think that should be it.
Last edited:
First point the second version was executed by Paul following ideas and suggestions I mentioned in posts 70 and 72. I was grateful he picked up on these and am glad his work has widened the focus on this project.
I had done my own simulations and following the above posts was asked to post the details - I did not respond to these requests immediately due to pressures of time. My wife who has a disability had been in hospital from late January until Wednesday this week and I was visiting her daily which took a lot of my time. I wanted to avoid getting into any involved explanations or questions from posting the circuit that would get in my way. I wish I could get my thoughts on paper in order without having to revise them, this is not a gift I have been endowed with.
Anyway Paul acknowledged my ideas and was faster off the mark than I anticipated.
I don't support version 1 as there are two poles one of which is in parallel with the feedback resistor.FETs have exceedingly high frequency characteristics and the higher you go up the ladder the lower the impedance of the capacitor will be. While a coil can mitigate return signals from a speaker there a lot of emi in the ether due to device switching etc.
I have seen schemes like the one in the disconnected block - I built a Leach Low TIM amplifier which used these. These can work in BJT stages with fT of around 2MHz because that will tend to form part of the scheme of getting the circuit gain down to unity. FETs having much higher frequency characteristics defy that logic. You can see what happened to the square wave from asking Paul to connect the one of these networks to see the FFT results.
In the second version the stability scheme is a step network and in a FET stage the response will level out and not get down to unity. You have stated this is stable on square waves with a capacitor of 100n in parallel with the speaker load. In the simulation the value is shown as zero n. The 1k5 resistor in series with 22p needs to be shorted out. With this done the circuit is more in line with my take. I have updated this with a step network in parallel with the VAS transistor emitter resistor.
I have checked stability at Ctest of 100nF, in my experience a better test than 1uF or 2uF, and it just meets the standard with around 4dB GM at 1.8MHz.
The designations have been tidied up, and prettied up to make it a base document. Note this is for 27V rails, but the other version for 20V rails is in this thread for use as well, as needed.
The designations have been tidied up, and prettied up to make it a base document. Note this is for 27V rails, but the other version for 20V rails is in this thread for use as well, as needed.
I have not found any threads other than this one to be of interest to me. In that sense it is a winner but all good things come to an end and seeing that fruition is imminent and I am not committed to building any projects at present, it is time for me to get out.
44W is in line with original AN, so I guess it makes sense to keep the same number (and definitely not higher).
It's already running hot.
My AN is built with active cooling, and this one will be the same. With passive cooling, the size of this amp will be enormous... (and expensive).
Micheal, I think we all value your input a lot, and even as you are not planning a build, I see no reason why not stay around and contribute..
I'm trying to learn something new with each new project from you guys.
Hope your wife is doing better... 👍
Last edited:
Two days ago I sent my 'small' PCB design for production (dirtypcbs.com), and I'm eagerly waiting to start the 1st build 🙂
I guess this will take some time.
Now, as a I need some distraction, here is something new. Any ideas if it may work?
My intention was A-class (or a very hot AB-Class) version of Wiederhold amp from 1977.
I did build AB-class quasi Wiederhold with TIP3055 outputs (and a LatFet version of it) 2 years ago,
and was amazed by how good did it sound.
By good I mean I couldn't tell it from any other amp in my stable (14 built/finished amps and counting).
And it measured very good too (for a quasi).
Original Wiederhold article from 'Radio Fernsehen Electronic' also attached.
So why not try 'alpha' treatment for it?
Here is the profile:
Thd at 1kHz:
I guess this will take some time.
Now, as a I need some distraction, here is something new. Any ideas if it may work?
My intention was A-class (or a very hot AB-Class) version of Wiederhold amp from 1977.
I did build AB-class quasi Wiederhold with TIP3055 outputs (and a LatFet version of it) 2 years ago,
and was amazed by how good did it sound.
By good I mean I couldn't tell it from any other amp in my stable (14 built/finished amps and counting).
And it measured very good too (for a quasi).
Original Wiederhold article from 'Radio Fernsehen Electronic' also attached.
So why not try 'alpha' treatment for it?
Here is the profile:
Thd at 1kHz:
Sorry minek to bring you back to this version again. I'm still into optimising it.
Thought I'd check out Hugh's version.
The stability checks out but have found a small "improvement" can be made by changing C7 (680p, on his schematic) to 330p. Gives a better balance to the VAS loop stability.
Checked the stability margins with the 100nF test capacitor connected and they looked ok too.
But when the circuit is run .trans mode it can't handle square waves into 8R//100n.
Has anyone else tried this? Maybe I have something wrong.
Or is there something more subtle going on?
Wonder if we should add a little series R to the test capacitor to make it a little more realistic?
Mjona, your ideas were the catalyst for version 2. Thought I'd have a go at interpreting them.
Thank you for sharing them 🙂
Thought I'd check out Hugh's version.
The stability checks out but have found a small "improvement" can be made by changing C7 (680p, on his schematic) to 330p. Gives a better balance to the VAS loop stability.
Checked the stability margins with the 100nF test capacitor connected and they looked ok too.
But when the circuit is run .trans mode it can't handle square waves into 8R//100n.
Has anyone else tried this? Maybe I have something wrong.
Or is there something more subtle going on?
Wonder if we should add a little series R to the test capacitor to make it a little more realistic?
Mjona, your ideas were the catalyst for version 2. Thought I'd have a go at interpreting them.
Thank you for sharing them 🙂
Last edited: