I couldn't run that sim as is, because of missing models. I changed diodes to the 4148 available in ltspsice and looked for others like the bd139/40 and that zener.
I ran the sim with those but I don't find the same low thd. Perhaps because of the models.
I'm posting back the sim with the added models and a log.
Perhaps there are some ideas here to use for the project. I am wondering if using those regulators on the rails for the front end is worth the bunch of extra parts. Anyone can comment on that?
It's amazing how this one can reach 100khz with such low thd... How is that possible?
I'm still optimising a cascode driver, so won't say more on that at this time.
The performance simulated only had cascodes in the VAS stages. It's not a double slope Miller, as Bimo suggests. I use "sprog stopper" capacitors as described by Cherry and a global frequency compensation capacitor around the feedback resistor. This keeps the VAS impedance high, eliminating gm doubling distortion but as a consequence can be significantly improved by gain-matching the output transistors. Which I do every time, as this offers yet further improvements in performance, even in the basic "Blameless" amplifier described by Doug Self.
John
The performance simulated only had cascodes in the VAS stages. It's not a double slope Miller, as Bimo suggests. I use "sprog stopper" capacitors as described by Cherry and a global frequency compensation capacitor around the feedback resistor. This keeps the VAS impedance high, eliminating gm doubling distortion but as a consequence can be significantly improved by gain-matching the output transistors. Which I do every time, as this offers yet further improvements in performance, even in the basic "Blameless" amplifier described by Doug Self.
John
That's cool. I hope this will be a good improvement. I suspect that small signals could be used to drive those drivers in the cascode. Perhaps not the very small ones, but at least the same as the pre-drivers.
Would that be that nested feedback? Complex calculations. Once properly calculated it doesn't cost much in parts to implement.
I'm still trying to stabilize the complementary triple now. I had it oscillation free at some point and the thd was down below 0.001% at 1khz on 4 ohms load. So it should work, but for some reason it oscillates again and I couldn't stop that.
What I have now is a tiny and very high frequency oscillation, and the thd isn't really that bad at 0.062369% at 20khz despite that. I'm posting this now.
I'm using the MJ15015/6 models. I put snubbers on the drivers' bases and they have base stoppers.
One thing that brought a huge improvement was to switch to the higher beta small signals (BC550C/60C), so this is a simple and dirt cheap thing to do for better results. Some values had to be changed with that, so then the choice of parts is important at the time to build. All parts should really be matched the best possible when building.
Attachments
Hi Spookydd, I will build the design that I sent here, of course with modern transistors 
. I still learn how to design audio amplifier and how to use ltspice. I surprise when I change it using ancient transistor, it still give high performance. I use transistor model from Bob Cordell's website, except these:
*.model TIP3055 ako:NSC_4A NPN() ; case TO-220 (s)
.MODEL TIP3055 NPN(Is=457.5f Xti=3 Eg=1.11 Vaf=50 Bf=156.7 Ise=1.346p Ne=1.34
+ Ikf=3.296 Nk=.5961 Xtb=2.2 Br=7.639 Isc=604.1f Nc=2.168
+ Ikr=8.131m Rc=91.29m Cjc=278.7p Mjc=.385 Vjc=.75 Fc=.5 Cje=433p
+ Mje=.5 Vje=.75 Tr=1.412u Tf=37.34n Itf=35.68 Xtf=1.163 Vtf=10
+ Rb=.1)
*.model TIP2955 ako:NSC_5A PNP() ; case TO-220 (s)
.MODEL TIP2955 PNP(Is=66.19f Xti=3 Eg=1.11 Vaf=100 Bf=137.6 Ise=862.2f
+ Ne=1.481 Ikf=1.642 Nk=.5695 Xtb=2 Br=5.88 Isc=273.5f Nc=1.24
+ Ikr=3.555 Rc=79.39m Cjc=870.4p Mjc=.6481 Vjc=.75 Fc=.5
+ Cje=390.1p Mje=.4343 Vje=.75 Tr=235.4n Tf=23.21n Itf=71.33
+ Xtf=5.982 Vtf=10 Rb=.1)
To achieve low distortion in high frequency, we must use advanced compensation like TMC, TPC, etc.
This link, Super TIS is very interisting.
. I still learn how to design audio amplifier and how to use ltspice. I surprise when I change it using ancient transistor, it still give high performance. I use transistor model from Bob Cordell's website, except these:*.model TIP3055 ako:NSC_4A NPN() ; case TO-220 (s)
.MODEL TIP3055 NPN(Is=457.5f Xti=3 Eg=1.11 Vaf=50 Bf=156.7 Ise=1.346p Ne=1.34
+ Ikf=3.296 Nk=.5961 Xtb=2.2 Br=7.639 Isc=604.1f Nc=2.168
+ Ikr=8.131m Rc=91.29m Cjc=278.7p Mjc=.385 Vjc=.75 Fc=.5 Cje=433p
+ Mje=.5 Vje=.75 Tr=1.412u Tf=37.34n Itf=35.68 Xtf=1.163 Vtf=10
+ Rb=.1)
*.model TIP2955 ako:NSC_5A PNP() ; case TO-220 (s)
.MODEL TIP2955 PNP(Is=66.19f Xti=3 Eg=1.11 Vaf=100 Bf=137.6 Ise=862.2f
+ Ne=1.481 Ikf=1.642 Nk=.5695 Xtb=2 Br=5.88 Isc=273.5f Nc=1.24
+ Ikr=3.555 Rc=79.39m Cjc=870.4p Mjc=.6481 Vjc=.75 Fc=.5
+ Cje=390.1p Mje=.4343 Vje=.75 Tr=235.4n Tf=23.21n Itf=71.33
+ Xtf=5.982 Vtf=10 Rb=.1)
To achieve low distortion in high frequency, we must use advanced compensation like TMC, TPC, etc.
This link, Super TIS is very interisting.
I looked at that super tis stuff before, but this looks rather complex for the average diyer.
I have been trying to figure out a simple way to add a cascode on the diff amps, without adding too much complexity. This seems like one good way to increase gain and bandwidth, and for higher rails amps, it can allow using higher gain small signal transistors on the diff amps, while allowing those high voltage rails.
My simulations, although not very advanced (I'm still a newbie), have shown how much can be gained by using those high beta small signals transistors, and there aren't any that are really high gain with high vce0 at the same time.
This won't be an issue on the current 3055 amp project, but perhaps the other advantages of the cascoding could be useful, as long as we only add a few cheap parts as we do it with current mirrors for example, but without adding overly complex calculations.
One other thing that bothers me is the fact that it's hard to find designs that are fully stable without any compensation, and I hate to see so many compensation caps all over the place.
I was able to stabilize that simulation with the outputs replaced with the MJ15015/6 models, but not without too many compensation caps and a snubber on the drivers. And once stabilized, the performance isn't really there as much as it could've been, with overly high thd at high frequencies.
I'm afraid that with such an unstable design to begin with, that requires such tricky compensations throughout, once we add the other considerations from the pcb and other building issues, there would be additional sources for instability from a bad pcb layout and other wiring, so such a difficult to build amp by a diyer wouldn't be worth the attempt.
I really want to make sure we make something that bullet proof and that can be built easily by anyone without much experience. For one thing, the pcb design will require some serious scrutiny to make sure it's done right and with all the proper optimizations and precautions so there are no chances for instability and that the amp can be built with the minimum amount of precaution, perhaps just making sure the parts are properly matched.
Can anyone suggest a good and simple way to cascode the diff amp stages? On top of the current mirrors. I think high beta and low noise parts like the bc550C/60C could be nice, and then the higher voltage ones could be lower gain types like the 2N5401/5551 or mpsa42/92...
Having simulated this 3055 amp with added current mirrors, 2 transistors vas plus cascode, followed by a triple output stage, I can see that we can achieve a very decent amp, with very low thd, despite making use of old style parts like the 3055/2955. The only thing that I'm now trying to reach is better stability and lower thd at high frequencies. The thd at 1khz is outstanding already, as I was able to reach below 0.0008% at full rated power of 50W on 8ohms. The 4 ohms thd is a bit higher but still very respectable and would likely get even better if the higher frequencies thd was under better control.
If I get this right, the high frequencies thd is high because of insufficient slew rate, so I'm thinking we can help this with the use of cascoded diff amps. We could increase the input stage's gain, which would allow more degeneration and the slew rate would get better. Am I wrong?
I have been trying to figure out a simple way to add a cascode on the diff amps, without adding too much complexity. This seems like one good way to increase gain and bandwidth, and for higher rails amps, it can allow using higher gain small signal transistors on the diff amps, while allowing those high voltage rails.
My simulations, although not very advanced (I'm still a newbie), have shown how much can be gained by using those high beta small signals transistors, and there aren't any that are really high gain with high vce0 at the same time.
This won't be an issue on the current 3055 amp project, but perhaps the other advantages of the cascoding could be useful, as long as we only add a few cheap parts as we do it with current mirrors for example, but without adding overly complex calculations.
One other thing that bothers me is the fact that it's hard to find designs that are fully stable without any compensation, and I hate to see so many compensation caps all over the place.
I was able to stabilize that simulation with the outputs replaced with the MJ15015/6 models, but not without too many compensation caps and a snubber on the drivers. And once stabilized, the performance isn't really there as much as it could've been, with overly high thd at high frequencies.
I'm afraid that with such an unstable design to begin with, that requires such tricky compensations throughout, once we add the other considerations from the pcb and other building issues, there would be additional sources for instability from a bad pcb layout and other wiring, so such a difficult to build amp by a diyer wouldn't be worth the attempt.
I really want to make sure we make something that bullet proof and that can be built easily by anyone without much experience. For one thing, the pcb design will require some serious scrutiny to make sure it's done right and with all the proper optimizations and precautions so there are no chances for instability and that the amp can be built with the minimum amount of precaution, perhaps just making sure the parts are properly matched.
Can anyone suggest a good and simple way to cascode the diff amp stages? On top of the current mirrors. I think high beta and low noise parts like the bc550C/60C could be nice, and then the higher voltage ones could be lower gain types like the 2N5401/5551 or mpsa42/92...
Having simulated this 3055 amp with added current mirrors, 2 transistors vas plus cascode, followed by a triple output stage, I can see that we can achieve a very decent amp, with very low thd, despite making use of old style parts like the 3055/2955. The only thing that I'm now trying to reach is better stability and lower thd at high frequencies. The thd at 1khz is outstanding already, as I was able to reach below 0.0008% at full rated power of 50W on 8ohms. The 4 ohms thd is a bit higher but still very respectable and would likely get even better if the higher frequencies thd was under better control.
If I get this right, the high frequencies thd is high because of insufficient slew rate, so I'm thinking we can help this with the use of cascoded diff amps. We could increase the input stage's gain, which would allow more degeneration and the slew rate would get better. Am I wrong?
I took a MJ2955/2N3055 based Maplin 225WRMS amplifier and redesigned it with constant current sources and a current mirror and modern transistors.
The result was I cant tell the difference between the sound of the original Maplin amplifier and my own redesign !
Looks like they got it right the first time or my redesign is poor !
Hi
For a newbie the simplest method of stabilisation is to use Miller. I'd recommend, however, inclusive Miller feedback. Here is such a circuit based on the previous one. Changes are basically (1) delete the 680pF roll-off capacitors across the bases of the Darlington-cascode VAS stages (2) replace the emitter input resistors in the diff pairs by 330 ohms and (3) run the input stages at 6mA instead of 2 and (4) use 47pF inclusive Miller capacitors (taken from the output rail) to the bases of the VAS inputs.
The 220pF sprog-stoppers are still required. Is the input stage noisier? Yes, but I've never heard anything coming from hifi speakers with moderately high input bias currents. The design is very flexible, however, and if you want to keep 2mA bias you could use 1k emitter degen resistors. This is somewhat large and will increase distortion compared with 330 ohms. (but if you do this remember to reduce the Miller capacitors to 15pF). The whole point of having a critical upper frequency point set by the input stage is to prevent cut-off and severe transient intermodulation distortion.
For a newbie the simplest method of stabilisation is to use Miller. I'd recommend, however, inclusive Miller feedback. Here is such a circuit based on the previous one. Changes are basically (1) delete the 680pF roll-off capacitors across the bases of the Darlington-cascode VAS stages (2) replace the emitter input resistors in the diff pairs by 330 ohms and (3) run the input stages at 6mA instead of 2 and (4) use 47pF inclusive Miller capacitors (taken from the output rail) to the bases of the VAS inputs.
The 220pF sprog-stoppers are still required. Is the input stage noisier? Yes, but I've never heard anything coming from hifi speakers with moderately high input bias currents. The design is very flexible, however, and if you want to keep 2mA bias you could use 1k emitter degen resistors. This is somewhat large and will increase distortion compared with 330 ohms. (but if you do this remember to reduce the Miller capacitors to 15pF). The whole point of having a critical upper frequency point set by the input stage is to prevent cut-off and severe transient intermodulation distortion.
Attachments
I think we do not need cascode at LTP if we use low voltage power supply. Then we can use high hFE transistor like BC550C/BC560C.
To increase slew rate we can make high current CCS at LTP.
I preferred to use low Cob transistor for VAS, it can reduce distortion.
To reduce distortion in OPS, we should use 3EF because hFE of 2N3055 drop to much at high current. But 3EF is difficult to stabilize than 2EF. Or, if we insist to use 2EF, we should parallel more 2N3055.
That's true, but there are a few useful enhancements that do bring major improvements, so why not use them?
Without going to the extremes, there are basic things that can be done. Then I guess the rest can be done with various compensations, and that's probably there that we can get the remaining stability and performance.
That probably depends on what type of sound is to be reproduced. If it's not full range, or for instruments, or in an active configuration, or even just used on substandard speakers...
In any case, it does take a fine ear to detect distortions below 0.1%, however it seems that tim is a harsher type of distortion and it doesn't take a lot of it present to cause detection.
Once an amp is built, with all the limitations from hardware, it will likely not be as "good" as it seemed in simulations, so we can aim for the best we can in simulations, and then hope we can approach that in reality.
I think in the final realization, if the thd can still be 0.00x% something, then it's good enough, and that should probably apply to tim as well.
Is that something a regular diyer can tackle? The calculations are perhaps more complex than the simpler miller, but they must be less complex than cherry's ndfl concept. For the math poor average diyer, can for example a spreadsheet be put together to make this somewhat more automatic?
So you would run the ltp "rich", like our friend bonsai does on his ovation amps then.
Wouldn't that push those ltp stages out of their linear range?
Are those the ones located on the bases of the pre-drivers? Going to ground?
How would this work in the case of a grounded bridge, where the rails fly all the time around that ground?
It's true that on this current bridge design, we're only aiming for 35V rails, but I'm a little wary of the vce0 limits of those 550/560. Perhaps the current mirrors are bringing some effect like a cascode, by reducing the exposure, but those mirrors don't bring that much of a delta from the rails. We're only somewhere around 3V away from the rails where the vas taps. So although the excursion on the ltp stages isn't huge, I am wondering how close we come to the vce0 limits.
I am just wondering how this moves the operating point and what happens with the linear area.
Can you suggest any transistor types for this? Not too exotic, so anyone can find them.
We already are using 3 pairs, but I did notice how the 2ohms load would be a strain, although we don't really to make it work under 2ohms, we could go to 4 pairs. The drivers used are those beefy MJE1503x, and being placed on the heatsinks with the outputs, they shouldn't be overheating and they can handle quite a high current, far above what is required by several pairs of outputs.
If we can manage a good compensation scheme to keep it stable, while keeping a high slew rate and low enough thd, we should be able to get this working well. Let's see how we can apply this compensation scheme proposed by john_ellis
I do have questions about this ltp stage design (post #409).
What I'm wondering about is what do those transistors used as diodes do on one of the legs of the ltp stages? (q24/25 I believe)
And what about those extra resistors (10k) across the ltp collectors? Are they part of that compensation scheme? What do they do? what do they bring?
And also on the vas cascode. Isn't using only 2 diodes there keeping it rather close to the rails? In bob cordell's amps, he uses an led with an other diode, so he gets about 2.5V from the rails. Using only 2 ordinary diodes would get maybe 2V max.
What I'm wondering about is what do those transistors used as diodes do on one of the legs of the ltp stages? (q24/25 I believe)
And what about those extra resistors (10k) across the ltp collectors? Are they part of that compensation scheme? What do they do? what do they bring?
And also on the vas cascode. Isn't using only 2 diodes there keeping it rather close to the rails? In bob cordell's amps, he uses an led with an other diode, so he gets about 2.5V from the rails. Using only 2 ordinary diodes would get maybe 2V max.
Alright, let's see how this compensation scheme would work on our design.
I disabled the millers and to make it clearer I deleted all the protections. I disabled a few other things and this simulation uses the MJ15015/6 instead of the 2N3055/MJ2955 for the complementary EF3.
This triple stage output is as standard as it can be, and perhaps there is something to do to the bias spreader.
However it is now stable and there is no apparent signs of oscillations. That is not unless we look at the signals on the diff amps output and inside the vas, but what I don't understand is how the signal can be so clean coming out of the vas when it's so uggly coming in???
I used the same values for the caps, but obviously this needs to be calculated for this setup.
Increasing the tail currents on the ltps beyond about 3mA is very uggly, the signal looks totally clipped at the top.
I embedded the models in the sim file, for convenience and to make sure the same ones are used when comparing. Most come from cordell's collection, plus some from onsemi and others.
The signal looks quite clean on the output, at about 20Vrms and I left the load at 4 ohms. The thd doesn't look so good, although at 20khz, it could be better.
Let's see how those different compensations can be adjusted to make this work better and take a look at an inclusive miller tapped from the output bus. Maybe we can also take a look at splitting the pre-drivers' 220ohms into 2 and tap an inner feedback from there as leach did it. I wonder how complex that is in calculations.
The slew rate increase would be good to reduce tim and the thd at high frequencies.
Attachments
I use BC550C/BC560C with +-35V power supply (unregulated).
Dadod successfully using high current CCS on LTP with his TT amp. You can search his thread here. Ic max. BC550C/BC560C is 100mA, we can use 10-11mA CCS and it still linear
I use KSC3503 and KSA1318 for VAS and plan to use 2SA1360 and 2SC3423 (I am already have them) in my new amplifier.
That's what I'm using for all simulations now on this project. We're using 35V rails, so as long as the 550/560 can handle it, that'll work, however some of the vas ones may need changed.
Dadod successfully using high current CCS on LTP with his TT amp. You can search his thread here. Ic max. BC550C/BC560C is 100mA, we can use 10-11mA CCS and it still linear
I'd like to see how this can work. Proper calculations required. I tried increasing that on my sims and beyond about 3mA it gets quite ugly and the thd climbs real fast well before we can see it.
The way I have it now the tail current's sweet spot is somewhere around 1mA, more or less.
There is no doubt the 550/560 can handle some more current. We're nowhere near their limits and we can't push them so hard as to think about dissipation issues. It's just that with the current setup my sims have shown an steep increase in thd when tail current goes up.
I don't know those 2SA1360 and 2SC3423 and I suspect they may not be so easy to get for many. They must be japanese and I consider those exotic.
However some have their counterparts in western countries, and that's the case for those KSC/A types, and I've located those models and have been using them for vas in some sims. Maybe those could be used on our 3055 project.
I "sim" your file. The THD is low enough. Why do you not build it? Sometime implementation is more difficult. If you get THD 2x or 3x than simulator, I think you success.