This is an evolution of an amplifier I made some time ago. Some of it's properties are monstrous and I must therefore give it a proper name – Godzilla. I have only simulated it on my computer.
The AC measurements are from the base of Q4
First of all, many things with it are quite non monstrous. It's a class AB type of amp and the THD figures are good but not spectacular.
But two things are maximized in a grotesque manner. First of all, the amount of NFB is roughly 2000 times – 65db. Secondly, this heavy NFB is almost constant up to a couple of hundred kilohertz. So this is an amp for bats. The funny thing is that due to the very short feedback path, the phase margin seems to be excellent up to several megahertz – typically more than 140 degrees. Look at the AC figures measured at the base of Q4.
Then, the damping factor is more than 4000 up to at least 500khz. ( I haven't measured at higher frequencies ). The bandwidth is roughly 3mhz.
Below 20khz, the distortion is typically less than 0.005% except the second order which is very large – around 0.2%. But that comes mainly from the input mosfet which runs in single ended class A. Normally, 2:nd order dist is harmless, but it can easily be removed by applying a short local feedback loop around Q5 together with some small signal transistor.
It seems to me that most of the THD actually comes from Q4 and Q5.
Another unusual thing is that the bias of both Q4 and Q5 is higher than the output transistors. These can be run with a very low bias. The heavy NFB will take away most of the cross over artifacts.
But the amp will dissipate some heat. Q4 will burn approximately 15W with +-50 volt rails. Q5 can be biased with a separate low voltage supply. Biased to 1A the losses will be 10W. So one channel will consume roughly 30W idle, and that's indeed not much compared with a class A amp.
There are a lot of practical things left to figure out, such as realization of proper current generators, and a DC servo. Then, if two units are used in a balanced way, the output capacitor may be omitted.
One more thing, with Q4 biased at 150mA, the max output current will be around 10A. In my simulation I'm using the only power devices that my emulator supports.
Fairchild has more suitable devices, FJAF4210Y/ 4310Y, wich has a beta of typically 150.
What do you say folk's? Is it worth developing? Maybe bats will nod their heads.
If you would like to simulate the circuit, you can simplify things by connecting a current signal generator to the base of Q4 and omit Q5.
One more detail. If the amp must be stable with 1uf hooked on the output, C5 must be increased to around 5nf.
Finally, I must anticipate some comments about the fact that this thing really don't give any better THD figures than any ordinary blameless apmlifier and therefore must be regarded as some sort of freak.
Well, it is a freak, that's the whole idea.😀
But not just so. Many think that a circuit made up by few active elements usually sounds better subjectively, since the THD will have less complexity even though the actual level is higher. And it seems to be very stable.
An externally hosted image should be here but it was not working when we last tested it.
An externally hosted image should be here but it was not working when we last tested it.
The AC measurements are from the base of Q4
First of all, many things with it are quite non monstrous. It's a class AB type of amp and the THD figures are good but not spectacular.
But two things are maximized in a grotesque manner. First of all, the amount of NFB is roughly 2000 times – 65db. Secondly, this heavy NFB is almost constant up to a couple of hundred kilohertz. So this is an amp for bats. The funny thing is that due to the very short feedback path, the phase margin seems to be excellent up to several megahertz – typically more than 140 degrees. Look at the AC figures measured at the base of Q4.
Then, the damping factor is more than 4000 up to at least 500khz. ( I haven't measured at higher frequencies ). The bandwidth is roughly 3mhz.
Below 20khz, the distortion is typically less than 0.005% except the second order which is very large – around 0.2%. But that comes mainly from the input mosfet which runs in single ended class A. Normally, 2:nd order dist is harmless, but it can easily be removed by applying a short local feedback loop around Q5 together with some small signal transistor.
It seems to me that most of the THD actually comes from Q4 and Q5.
Another unusual thing is that the bias of both Q4 and Q5 is higher than the output transistors. These can be run with a very low bias. The heavy NFB will take away most of the cross over artifacts.
But the amp will dissipate some heat. Q4 will burn approximately 15W with +-50 volt rails. Q5 can be biased with a separate low voltage supply. Biased to 1A the losses will be 10W. So one channel will consume roughly 30W idle, and that's indeed not much compared with a class A amp.
There are a lot of practical things left to figure out, such as realization of proper current generators, and a DC servo. Then, if two units are used in a balanced way, the output capacitor may be omitted.
One more thing, with Q4 biased at 150mA, the max output current will be around 10A. In my simulation I'm using the only power devices that my emulator supports.
Fairchild has more suitable devices, FJAF4210Y/ 4310Y, wich has a beta of typically 150.
What do you say folk's? Is it worth developing? Maybe bats will nod their heads.
If you would like to simulate the circuit, you can simplify things by connecting a current signal generator to the base of Q4 and omit Q5.
One more detail. If the amp must be stable with 1uf hooked on the output, C5 must be increased to around 5nf.
Finally, I must anticipate some comments about the fact that this thing really don't give any better THD figures than any ordinary blameless apmlifier and therefore must be regarded as some sort of freak.
Well, it is a freak, that's the whole idea.😀But not just so. Many think that a circuit made up by few active elements usually sounds better subjectively, since the THD will have less complexity even though the actual level is higher. And it seems to be very stable.
I'm taking myself the liberty to be the first to reply on my own thread.
I don't want it to subside too far down to diyaudio oblivion before someone who finds it funny or interesting has a chance discovering it.
I want to hear what you think, guys!
Actually, this construction is some sort of homage to mr Nelson Pass. ( I don't know if he will be flattered if he reads this).
I don't want it to subside too far down to diyaudio oblivion before someone who finds it funny or interesting has a chance discovering it.
I want to hear what you think, guys!
Actually, this construction is some sort of homage to mr Nelson Pass. ( I don't know if he will be flattered if he reads this).
Why not! 🙂
I wonder how does the harmonics pattern look like if you feed Q4 directly so we can exclude Q5 for a while, perhaps adding a series base resistor so the NFB can work.
And if we increase degeneration resistor for Q4 a little bit, global gain goes down and output impedance increases, maybe some interesting changes to the harmonics, phase, GBW etc., keep new simulation findings coming, I am here in my armchair ready with popcorn and everything, cheers to the Pass philosophy.
I wonder how does the harmonics pattern look like if you feed Q4 directly so we can exclude Q5 for a while, perhaps adding a series base resistor so the NFB can work.
And if we increase degeneration resistor for Q4 a little bit, global gain goes down and output impedance increases, maybe some interesting changes to the harmonics, phase, GBW etc., keep new simulation findings coming, I am here in my armchair ready with popcorn and everything, cheers to the Pass philosophy.
Yes. As I mentioned, the whole part to the left of the base of Q4 may be substituted by a signal generator in serial with a high ohmic resistor to simplify simulations.
100V AC input with 2200 ohm will result in an output amplitude of 10V. One important thing wich makes the amp have such high NFB even at high frequencies is that the base of Q4 is driven with high impedance, in this case > say, 200 ohms.
But I actually made a calculation error. The NFB loop is not currently 2000 times, rather 700, since I forgot to take the base impedance of Q4 in account, wich is around 25 ohm.
But after I substituted Q4 to a medium power device with a beta of 200, the NFB rised to 1500 times, that is 63db or so. The phase margin didn't suffer.
The feedback of this thing can't really be analyzed as we are used to. Since Q4 is driven high ohmic, the feedback current isn't divided by any resistor so virtually every electron passing R6 will be used for error correction.
I will probably work with refining this amp and i will certainly post new versions as they develops.
Such as Dc regulation. And some means to use ground as signal ground in a normal way so that we may omit the output cap.
Comments are welcomed.
100V AC input with 2200 ohm will result in an output amplitude of 10V. One important thing wich makes the amp have such high NFB even at high frequencies is that the base of Q4 is driven with high impedance, in this case > say, 200 ohms.
But I actually made a calculation error. The NFB loop is not currently 2000 times, rather 700, since I forgot to take the base impedance of Q4 in account, wich is around 25 ohm.
But after I substituted Q4 to a medium power device with a beta of 200, the NFB rised to 1500 times, that is 63db or so. The phase margin didn't suffer.
The feedback of this thing can't really be analyzed as we are used to. Since Q4 is driven high ohmic, the feedback current isn't divided by any resistor so virtually every electron passing R6 will be used for error correction.
I will probably work with refining this amp and i will certainly post new versions as they develops.
Such as Dc regulation. And some means to use ground as signal ground in a normal way so that we may omit the output cap.
Comments are welcomed.
The whole amp from Q4 on basically is a big fancy VAS (transimpedance amplifier), with cascode and follower.
Is3 makes no sense IMHO. You'll also have to elaborate on Vs1.
Yes, isn't it nice with a heavy VAS!
Sgrossklass, I will certainly elaborate on many things. This amp is only sketched out, but I will gradually make it more usable.
Is3 makes sense since Q5 needs a very large bias to keep THD at reasonable levels. Vs1 will certainly be properly implemented.
Here is a simplified version with Q5 replaced with a high impedance signal gen. ( for simulating purposes ) As you can see, the hefty 2:nd order harm is now gone more or less. If we don't want 2:nd order we will have to apply a local NFB loop around Q5, but I will come to that later.
Next time I will be back with a DC servo mechanism.
Sgrossklass, I will certainly elaborate on many things. This amp is only sketched out, but I will gradually make it more usable.
Is3 makes sense since Q5 needs a very large bias to keep THD at reasonable levels. Vs1 will certainly be properly implemented.
Here is a simplified version with Q5 replaced with a high impedance signal gen. ( for simulating purposes ) As you can see, the hefty 2:nd order harm is now gone more or less. If we don't want 2:nd order we will have to apply a local NFB loop around Q5, but I will come to that later.
Next time I will be back with a DC servo mechanism.
Set the vertical scale to decibels to get a more standardized distortion measurement and something others can better interpret. Right click on waveform display, choose "scaling", select "magnitude in decibels" in primary box, choose "none" for secondary, hit "ok". On the left tab, click AutoY then FitX buttons,in the "scaling" section, change Y Division to 25, change Y Offset to 50.
Last edited:
I'm using an old simulator - Circuitmaker. I like it and I will never change to anything else, but the fourier section is crappy. The DB option gives absurd figures.
But since the output currently is 10V peak, for example the 5:th order is around 0.001%. Or -100db if you like.
But to me, the main thing is that the most disturbing things - cross over dist - are more or less gone due to the large amount of NFB. It seems that the above THD figures mainly comes from the VAS.
But since the output currently is 10V peak, for example the 5:th order is around 0.001%. Or -100db if you like.
But to me, the main thing is that the most disturbing things - cross over dist - are more or less gone due to the large amount of NFB. It seems that the above THD figures mainly comes from the VAS.
I have now made an input stage that distorts less. The 2:nd order THD is now much lower.
The DC servo is now more or less complete - it will operate on Is2. I will be back with that in a day or so.
It seems that we will have to live with connecting negative rail to signal ground. So a DC voltage of 30 to 50 volts will be present at the negative speaker terminal, which has a potential to be hazardous. I tried various methods to transfer the input signal from ground, but that would have made it necessary to regulate the various voltages to an extreme level.
Another solution is an input transformer. This also isolates the ground path and makes it possible to drive the thing balanced.
I think you should be excited by this amp. One thing: The THD measured at 100khz is almost the same as at 10khz. ( an increase with some 50% ) Isn't that an interesting property?
The DC servo is now more or less complete - it will operate on Is2. I will be back with that in a day or so.
It seems that we will have to live with connecting negative rail to signal ground. So a DC voltage of 30 to 50 volts will be present at the negative speaker terminal, which has a potential to be hazardous. I tried various methods to transfer the input signal from ground, but that would have made it necessary to regulate the various voltages to an extreme level.
Another solution is an input transformer. This also isolates the ground path and makes it possible to drive the thing balanced.
I think you should be excited by this amp. One thing: The THD measured at 100khz is almost the same as at 10khz. ( an increase with some 50% ) Isn't that an interesting property?
That buffer gain stage was a pure H2 ear balsam generator. 😛
But what surprised me a bit is the amount of H3 left over after leaving the first stage out, I wouldn't have expected that, maybe it's the cross conduction contributing to it and maybe it vanishes with larger output signals?
Indeed, the last statement explains the beauty with these simple amp circuits.
But what surprised me a bit is the amount of H3 left over after leaving the first stage out, I wouldn't have expected that, maybe it's the cross conduction contributing to it and maybe it vanishes with larger output signals?
Indeed, the last statement explains the beauty with these simple amp circuits.
Yes, Maiko, you can actually vary H2 by changing R9. A value approaching 1 ohm starts rasing it.
Yes, the H3 is puzzling. I think it has to do with the fact that the output devices are driven with high impedance. That is contrary to most designs where the usual exponential transfer function is in action. In this case the THD is set by hfe. How hfe varies under load, I don't know. Theoretically it is a constant but my simulator says otherwise.
Anyone who can illuminate this?
Anyhow, the NFB takes care of the situation regardless of this. But simulations indicates that the THD is slightly higher driven high ohmic. But it seems that the THD in the cross over section is slightly less, so it may be so that this amp will sound a bit more like a class A, but I don't really know.
I have actually built a baby version of this amp and it really sounds good. Extremely holographic, yet lean and involving.
But I don't know how long I will be nurturing this thread. People doesn't seem overly enthusiastic. It's a pity since I truly believe it has great potential.
But if you want to see my other stuff, visit my very simple site:
- Embuddy
Yes, the H3 is puzzling. I think it has to do with the fact that the output devices are driven with high impedance. That is contrary to most designs where the usual exponential transfer function is in action. In this case the THD is set by hfe. How hfe varies under load, I don't know. Theoretically it is a constant but my simulator says otherwise.
Anyone who can illuminate this?
Anyhow, the NFB takes care of the situation regardless of this. But simulations indicates that the THD is slightly higher driven high ohmic. But it seems that the THD in the cross over section is slightly less, so it may be so that this amp will sound a bit more like a class A, but I don't really know.
I have actually built a baby version of this amp and it really sounds good. Extremely holographic, yet lean and involving.
But I don't know how long I will be nurturing this thread. People doesn't seem overly enthusiastic. It's a pity since I truly believe it has great potential.
But if you want to see my other stuff, visit my very simple site:
- Embuddy
Btw, here is a picture of my baby Godzilla.
Speaker terminals on front, inputs on the back. Made up of massive alu beams and brass and copper. I little inspired of the steampunk aesthetic.
Speaker terminals on front, inputs on the back. Made up of massive alu beams and brass and copper. I little inspired of the steampunk aesthetic.
An externally hosted image should be here but it was not working when we last tested it.
Last edited:
How do you come up with a DF of more than 4000?
It's a nice toy. Well, tbh, the bandwidth is nice but not neccessarily desirable. Even if your amp is stable, the input probably isn't. Once you get some kind of HF into your system, you'll instantly be promoted to a ham radio operator. 😀
Ah, well, I doubt that. I mean, unless you've got some ultrasonic tweeters too. 😉
How come? When the NFB is big enough it assures high DF. The simulator reports that the output level at 500 khz drops less than one mV when the load decreases from 9 to 8 ohm. Thats approximately 2mohm.
Of course it's overkill, therefore the name. But I can say one thing. This assures proper phase behavior above 20khz and the THD is constant up to 100khz or so. The short feedback path assures that the correction of the output devices comes instantly - no smearing.
But I admit, the amp needs some HF domestication from a practical point of view, but a NFB of 1500 up to 300khz is a very good advertising.
But as I mentioned, I have built a baby version that sounds great. I tried to put up a photo of it in my last reply, but something got wrong.
Of course it's overkill, therefore the name. But I can say one thing. This assures proper phase behavior above 20khz and the THD is constant up to 100khz or so. The short feedback path assures that the correction of the output devices comes instantly - no smearing.
But I admit, the amp needs some HF domestication from a practical point of view, but a NFB of 1500 up to 300khz is a very good advertising.
But as I mentioned, I have built a baby version that sounds great. I tried to put up a photo of it in my last reply, but something got wrong.
Attachments
OK. How did I calculate the NFB factor?
Q4 has a transconductance of around 150mA*40(ohm)= 6S. The gain will be 6*700ohm= 4200. The figure 700 is taken from the base impedance of the output devices and with 8 ohm load.
Then, the base impedance of Q4 is rougly 1/6*200(beta) = 30 ohm. But my simulator reports around 50 ohm. Anyhow, this impedance is half of r5 and the feedbacked signal will be divided by a factor of 3.5 or so. Therefore NFB = 4200/3.5 = 1200. Not 1500 but quite much.
These calculations are very rough and I have simulated values of between 1000 and 2000. It depends of the actual transistor and others.
Q4 has a transconductance of around 150mA*40(ohm)= 6S. The gain will be 6*700ohm= 4200. The figure 700 is taken from the base impedance of the output devices and with 8 ohm load.
Then, the base impedance of Q4 is rougly 1/6*200(beta) = 30 ohm. But my simulator reports around 50 ohm. Anyhow, this impedance is half of r5 and the feedbacked signal will be divided by a factor of 3.5 or so. Therefore NFB = 4200/3.5 = 1200. Not 1500 but quite much.
These calculations are very rough and I have simulated values of between 1000 and 2000. It depends of the actual transistor and others.
> How hfe varies ...Theoretically it is a constant but my simulator says otherwise.
Theory says it must decline to zero at zero current, and faster due to recombination ("leakage").
And it probably goes-away when current is so high the crystal is full of space-charge.
Self has written at length on hFE fall-off for high current. Any proper data-sheet shows low-current fall-off.
SPICE has an in-built assumption (unless specifically declared) on low-I fall-off which seems to be pessimistic for many modern device processes (recombination is not as bad as the old days).
Theory says it must decline to zero at zero current, and faster due to recombination ("leakage").
And it probably goes-away when current is so high the crystal is full of space-charge.
Self has written at length on hFE fall-off for high current. Any proper data-sheet shows low-current fall-off.
SPICE has an in-built assumption (unless specifically declared) on low-I fall-off which seems to be pessimistic for many modern device processes (recombination is not as bad as the old days).
PRR, it would be interesting if there were some transfer function for hfe. The Vbe function is exponential and produces quite a lot of high order harmonics even if they have a low level. But what kind of harmonics are generated when the bjt is driven high ohmic? Perhaps there are no mathematical law. Perhaps it varies due to a lot of thermal variation and others. Probably the THD behavior is less mathematically describable.
ICG: My function generator stops at 22.05 khz ( DAC output ). But these things are very simple for the spice engine to calculate. I have also made my own calculations and they supports those figures.
The main reason for this good hf behavior is that the feedback resistor has a very low value that acts directly on the base of Q4 and therefore makes all the various capacitances have less impact.
The stability seems to come from the fact that this really is a current driven thing. Corrections passes C5 at HF and the current will be amplified and fed directly to the output devices which are resistive towards the VAS output ( about 700 ohm).
The beauty of it is that even when C5 starts to attenuate at HF, the strength of the NFB is exactly as powerful, in contrast to normal amps where the NFB declines at high frequencies.
This amp will probably sound very good since the NFB corrections comes instantly with almost no lag
ICG: My function generator stops at 22.05 khz ( DAC output ). But these things are very simple for the spice engine to calculate. I have also made my own calculations and they supports those figures.
The main reason for this good hf behavior is that the feedback resistor has a very low value that acts directly on the base of Q4 and therefore makes all the various capacitances have less impact.
The stability seems to come from the fact that this really is a current driven thing. Corrections passes C5 at HF and the current will be amplified and fed directly to the output devices which are resistive towards the VAS output ( about 700 ohm).
The beauty of it is that even when C5 starts to attenuate at HF, the strength of the NFB is exactly as powerful, in contrast to normal amps where the NFB declines at high frequencies.
This amp will probably sound very good since the NFB corrections comes instantly with almost no lag
svitjod, I was thinking for a while here how the H3 contribution is coming about for your circuit, maybe an incorrect assumption but I think maybe it is simply because the gain and output stage is inverting the signal, so when the feedback signal is fed back to the base it creates H3 because the H2 initially created gets once again inverted/summed.
It's a known fact if 2 gain stages produces pure H2 but are 180 degree out of phase, when summed together it creates H3, which is also symptomatic for the LTP circuit if I recall it correct.
Maybe in order to to get H3 lower is to trim and linearize Q4's operation for improved THD signature.
Your baby godzilla looks really nice with the transparent case and side grills, I can imagine if we add some ambient light with colorful LED's inside the case how beautiful it must look at dark hours, I noticed your ingenious loudspeaker cable arrangement in the picture, is it transformer wire?
Keep up posting once in a while, even though I can agree to some extent we should use simulators only to get a quick clue, take the results with a grain of salt and backing up with real world measurement results, don't let the stiff upper lip nay-Sayers discourage you.
TOP
It's a known fact if 2 gain stages produces pure H2 but are 180 degree out of phase, when summed together it creates H3, which is also symptomatic for the LTP circuit if I recall it correct.
Maybe in order to to get H3 lower is to trim and linearize Q4's operation for improved THD signature.
Your baby godzilla looks really nice with the transparent case and side grills, I can imagine if we add some ambient light with colorful LED's inside the case how beautiful it must look at dark hours, I noticed your ingenious loudspeaker cable arrangement in the picture, is it transformer wire?
Keep up posting once in a while, even though I can agree to some extent we should use simulators only to get a quick clue, take the results with a grain of salt and backing up with real world measurement results, don't let the stiff upper lip nay-Sayers discourage you.
TOP
- Status
- Not open for further replies.