Patchwork Reloaded: Circuit Optimization and Board Layout.

[GK]

Hi homemodder,
That is the overall objective - to improve performance, but within reason. I'm picking up bits and pieces as I go through this and become more and more informed. I will run the simulation with higher gain devices and see how the results look. Like I said before, the original amp sounded pretty darn good and measured very well, this had the 2N's in circuit.

Just out of curiosity, why are you persisting with the bootstrapped VAS?

Cheers,
Glen

 

[MJL21193]

How are the listening tests going??

Very good. Running on my +/-30V lab supply mono (I only have one prototype) most of today.

Just out of curiosity, why are you persisting with the bootstrapped VAS?

Hi Glen,
First, answer me this: Can I get away with lower voltage transistors in the cascode, mirror and EF? 80V devices ok here?See schematic below.

I stick with the bootstrap because it works fine and I like it. Do you think I will see a significant performance improvement with a CCS?

 

[GK]

Very good. Running on my +/-30V lab supply mono (I only have one prototype) most of today.

Hi Glen,
First, answer me this: Can I get away with lower voltage transistors in the cascode, mirror and EF? 80V devices ok here?See schematic below.

I stick with the bootstrap because it works fine and I like it. Do you think I will see a significant performance improvement with a CCS?

The diode connected mirror transistor only ever sees a Vce of about 0.7V and the other mirror transistor (VbeQ7+VbeQ8)-VR2 ; only a volt or so. You can use just about any small signal transistor here.

If you want to use a lower voltage transistor with higher beta and fT for Q7, then can just clamp its collector volatge with a 20V zener diode connected between the +rail and the collector of Q7.
Watch the dissipation in R7 though.

The LTP cascode devices should be able to withstand the full rail voltage. Personally, I like to run fast, low Cob, TO-126 CRT driver transistors here with a little heatsink and a higher LTP tail current (like 10mA). This a proportional increase in slewrate (provided that the VAS current delivery to the driver capacitace can keep up.
A triple EF can come in handy here.

A bootstrapped VAS will never be as linear as a current source loaded VAS. Poor PSR is also an issue.

Cheers,
Glen

 

[ferencz]

A bootstrapped VAS will never be as linear as a current source loaded VAS. Poor PSR is also an issue.

You are right about linearity, but I've found that bootstrapped VAS gives better sound. It's simpler too! (I prefer the resistor tail for the LTP too!)
There must be a reason that many designer are using bootstrapped VAS.
PSR is not an issue here, John is using an RC filter on the psu rail.

Anyway... try it!

 

[MJL21193]

The LTP cascode devices should be able to withstand the full rail voltage.

Do you mean rail to rail voltage? In this case, with 56V rails, I'd need 120V devices? Thanks, It's nice to know these things for sure (you know, without actually reading and studying )

A bootstrapped VAS will never be as linear as a current source loaded VAS. Poor PSR is also an issue.

Any shortcomings in linearity are not showing in either the sims THD or the actual amps THD performance. I like the simple precision of the bootstrap. Anywhere that an active component can be eliminated to give comparable results, works for me.

In the end I want it to be as simple as possible without sacrificing performance, but keeping all things in perspective.

You are right about linearity, but I've found that bootstrapped VAS gives better sound. It's simpler too! (I prefer the resistor tail for the LTP too!)

Hi ferencz,
I can't comment of the difference in sound from a bootstrapped or CCS VAS, but I do like the simplicity.
How is the performance on your simulation now? Did you update your models?

I have a schematic update. The LTP devices have been changed plus the EF on the VAS.
I have ordered a pile of 2SA970/2SC2240 to use here (150 of each ).
Does anyone have known good models for these transistors?

 

[andy_c]

Any shortcomings in linearity are not showing in either the sims THD or the actual amps THD performance. I like the simple precision of the bootstrap. Anywhere that an active component can be eliminated to give comparable results, works for me.

Hi John,

If there is a detrimental effect from the bootstrap, the simulated distortion at low frequencies should show it. This would be due to the combination of two factors:

1) The loop gain will begin to decrease at low frequencies due to an increase in the impedance of the bootstrapping capacitor. The increasing bootstrap capacitor impedance causes the effective load impedance seen by the VAS to decrease. The decreasing effective impedance causes the open-loop gain to decrease proportionally, meaning that there's less feedback available at low frequencies to reduce the distortion.

2) The open-loop distortion at low frequencies will increase, as the VAS will now see a lower impedance at its collector. Because of this, for the same voltage swing, the VAS will need to put out more current.

Since the closed-loop distortion is approximately the open-loop distortion divided by (one plus the loop gain) the combined effect can be a kind of "double whammy" if the frequency becomes low enough.

How can you tell if it's a problem? Try simulating distortion at full power at 20 Hz and 200 Hz. If the distortion at 20 Hz is higher than at 200 Hz, the bootstrapping is likely the culprit.

I agree that the bootstrap is attractive from the point of view of simplicity.

 

[MJL21193]

How can you tell if it's a problem? Try simulating distortion at full power at 20 Hz and 200 Hz. If the distortion at 20 Hz is higher than at 200 Hz, the bootstrapping is likely the culprit.

I agree that the bootstrap is attractive from the point of view of simplicity.

Thanks Andy,
Knowing where to look goes a long way. You are dead on - with the simulated THD between 20Hz and 200Hz 0.001% and 0.000029% respectively.
In your opinion, how much of an issue is this? Is it worth changing for this improved linearity when THD is so low already?
When I modeled a CCS, both LED bias and actively biased, I wound up with overall higher distortion for equal VAS current.
A trade off - slightly better linearity for slightly higher overall THD?

BTW, thanks for the effort you have put into providing accurate models for us schmucks to play with.
What do you think of the following for the 2SA1943/2SC5200?

model F5200 NPN
+ IS = 3.0463E-11 BF = 96.20 VAF = 100
+ IKF = 15.04256 ISE = 5.6190E-11 NE = 2.0
+ BR = 4.849 IKR = 1.05012 VAR = 100
+ ISC = 7.18E-8 NC = 1.5 RE = 0.0025
+ RB = 20.18 RBM = 0.0014 IRB = 1.0E-7
+ RC = 0.01137 CJE = 4.5000E-10 CJC = 8.4915E-10
+ VJC = 0.68977 MJC = 0.54081 TF = 6.8583E-10
+ XTF = 9.5721 VTF = 10.425 ITF = 6.8697E-2
+ TR = 1.000E-8 XTB = 1.45 EG = 0.82
+ FC = 0.5

.model F1943 PNP
+ IS=1.30E-10 BF=91.42 VAF=100
+ IKF=4.480 ISE=1.02E-10 NE=2.0
+ VAR=100 ISC=5.0900E-9 NC=1.5
+ BR=0.882 IKR=2.9015 RE=0.0011
+ RC=0.0553 RB=140.05 RBM=0.0041
+ IRB=8.5e-9 CJE=2.00E-10 FC=0.5
+ CJC=9.45E-10 VJC=0.48 MJC=0.28
+ TF=9.250E-10 XTF=10 VTF=10
+ ITF=1 TR=1.00E-8 EG=0.76
+ XTB=2.68

 

[homemodder]

.model 2sc2240 NPN(Is=99.13f Xti=3 Eg=1.11 Vaf=422.2 Bf=352.8 Ise=1.179p
+ Ne=1.782 Ikf=.4704 Nk=.9631 Xtb=1.5 Var=100 Br=1.663 Isc=555.1p
+ Nc=1.796 Ikr=5.85 Rc=.2032 Cjc=7.561p Mjc=.2472 Vjc=.3905 Fc=.5
+ Cje=5p Mje=.3333 Vje=.75 Tr=10n Tf=1.295n Itf=1 Xtf=0 Vtf=10)


.model 2sa970 PNP(Is=465.4f Xti=3 Eg=1.11 Vaf=57 Bf=407.6 Ise=4.683p Ne=2.051
+ Ikf=.3998 Nk=1.192 Xtb=1.5 Var=100 Br=1 Isc=465.4f Nc=1.048
+ Ikr=6.032 Rc=2.343 Cjc=11.59p Mjc=.4014 Vjc=1.155 Fc=.5 Cje=5p
+ Mje=.3333 Vje=.75 Tr=10n Tf=1.252n Itf=1 Xtf=0 Vtf=10)

 

[homemodder]

MJL21193

I have extensive Jap library, but seems you have one I dont have, can you put it up please, 2sa1011 and comp.

 

[andy_c]

Thanks Andy,
Knowing where to look goes a long way. You are dead on - with the simulated THD between 20Hz and 200Hz 0.001% and 0.000029% respectively.
In your opinion, how much of an issue is this?

I'd say it's a non-problem. It's one thing if you're targeting the ultimate low distortion design as some are. But you are targeting "as simple as possible, consistent with very good measured performance". What you have seems very consistent with those goals.

Is it worth changing for this improved linearity when THD is so low already?
When I modeled a CCS, both LED bias and actively biased, I wound up with overall higher distortion for equal VAS current.
A trade off - slightly better linearity for slightly higher overall THD?

Well, better linearity should correspond to lower distortion, so something else is going on here. When the VAS is driving the drivers of the output stage directly, this presents a somewhat nonlinear load to the VAS. When this effect is not negligible, you'll often see unpredictable simulated results with VAS loading. For example, some people have noticed that resistive loading of the VAS reduces distortion in some cases - as long as that resistor is not too small. Eliminating this quirk involves an additional EF to buffer the VAS from the nonlinear input impedance of the drivers. This is probably inconsistent with your goal of simplicity though. It sounds to me like you've got this topology pretty near its limits, to the point where you're seeing these quirks. I'd be inclined to leave it as is.

BTW, thanks for the effort you have put into providing accurate models for us schmucks to play with.
What do you think of the following for the 2SA1943/2SC5200?

Thanks. I'm not sure about the quality of the model. It takes many hours to properly evaluate a model and tweak it to give an optimum match to the data sheet. If I consider the model integrity to be very important to a project I'm working on, I will spend a bunch of time with the model, then share my results. So whatever models I provide end up being something I'm using, or considering using, in a project. Doing this with every model floating around isn't practical due to the time involved.

 

[Nico Ras]

Hi John,

I think you have come to the point where the digits in THD are becoming purely academic and would like to inspire you to put your mechanical and woodworking genius into designing and making an aesthetically beautiful cabinet and share this with us as well.

Nico

 

[homemodder]

Nico,
Do you have connections in JHB?? I know some parts dealers there and wouldnt mind getting my hands on some extra supplies, very cheap, was anyway till I left.

Can I mail you privately??

 

[MJL21193]

MJL21193

I have extensive Jap library, but seems you have one I dont have, can you put it up please, 2sa1011 and comp.

Thanks for the models! Excellent!

I couldn't find a model anywhere for the 2SA1011 when I originally got it for the VAS. I did a Digikey search for similar transistors, and the best match that I could find a model for was the Rohm 2SB1275. Similar speed, gain, Cob, voltage, power. I figured it was better than nothing for this unit.
Here is the model I used:

* Q2SB1275 PNP BJT model
* Date: 2006/11/20
.MODEL Q2SB1275 PNP
+ IS=450.00E-15
+ BF=75.284
+ VAF=6.5824
+ IKF=1.0712
+ ISE=450.00E-15
+ NE=1.3466
+ BR=318.62
+ VAR=100
+ IKR=4.1919
+ ISC=1.4330E-9
+ NC=1.6223
+ NK=.57797
+ RE=.1
+ RB=1.4759
+ RC=.16193
+ CJE=327.67E-12
+ MJE=.58268
+ CJC=103.96E-12
+ MJC=.46676
+ TF=2.9216E-9
+ XTF=199.17
+ VTF=22.382
+ ITF=39.180
+ TR=273.01E-9
+ XTB=1.5000

I didn't use it much and I don't use it at all now. I didn't model the 2SA1011/2344 in the amp as the drivers - I used the MJE15030/31.

 

[MJL21193]

I'd say it's a non-problem. It's one thing if you're targeting the ultimate low distortion design as some are. But you are targeting "as simple as possible, consistent with very good measured performance". What you have seems very consistent with those goals.

Thanks,
Yes, it's not my goal to drive distortion lower at the expense of circuit complexity. Respectable numbers is what I want and stability. I think I have both here.

I think there should be a thread for discrete spice models only. The one on simulation we have is loaded with other stuff. It would be nice to go to one place to get these hard to track down models and updated or optimized ones. Quick to copy and paste from one convenient place.

Hi John,

I think you have come to the point where the digits in THD are becoming purely academic and would like to inspire you to put your mechanical and woodworking genius into designing and making an aesthetically beautiful cabinet and share this with us as well.

Ah Nico, you're making me blush again...
Yes, the time has come to turn my attention to the physical portion of this project. Board layout is next though. I'm working on a tight, one sided layout. Details soon.
As to the housing, I am rolling ideas around in my head as I go through this. With 6 channels and 840 total watts, this needs to be a large enclosure.
I have one of the heatsinks nearly finished.

 

[GK]

Do you mean rail to rail voltage? In this case, with 56V rails, I'd need 120V devices? Thanks, It's nice to know these things for sure (you know, without actually reading and studying )

Any shortcomings in linearity are not showing in either the sims THD or the actual amps THD performance. I like the simple precision of the bootstrap. Anywhere that an active component can be eliminated to give comparable results, works for me.

In the end I want it to be as simple as possible without sacrificing performance, but keeping all things in perspective.

The LTP cascode transistors only need to withstand the positive rail voltage with a little safety headroom.

Andy has covered the issues with the bootstrapped VAS linearity. I'd also point out the lousy PSR at low frequencies too (this is where the rail RC filters are least effective, unlike WRT the miller compensation cap which degrades high frequency PSR).

As for the effect on overall performance, it will be small, but you did say that you wanted to optimise the design!
It would make more sense to get rid of the VAS bootstrap before sourcing a fancy transistor for the VAS or bothering to bootstrap the LTP cascode transistors with the common-mode signal.

I won't argue that it cannot be made to work just fine, but I don't buy half baked arguments with regards to the "superior sound" of the bootstrapped VAS.

Cheers,
Glen

 

[MJL21193]

The LTP cascode transistors only need to withstand the positive rail voltage with a little safety headroom.

As for the effect on overall performance, it will be small, but you did say that you wanted to optimise the design!

I won't argue that it cannot be made to work just fine, but I don't buy half baked arguments with regards to the "superior sound" of the bootstrapped VAS.

Thanks Glen. I kinda thought that would be the case, but it's nice to hear it.

I did want to optimize the design and it has come a long way. I can see your point, but the fancy VAS transistor is just a 36 cent, off the shelf Fairchild and by all accounts, one heck of a VAS. The re-configuring of the cascode bias was a two for one - lower component count and better performance. Plus I got rid of those garish LED's.
There's more to come. I need to get my final board layout done, populated with the final component choices and do some full power tests. Hopefully all goes well.

Like i said before, I'm not looking for record breaking numbers - any amp that can keep distortion below .01% is doing as much as it needs to IMO. I want the power, stability and simplicity that this seems to have. Fairly low cost too.



...And who knows, maybe the mixture of these different disciplines and levels of complexity elevates it to a sublime plateau, achieving the magical "golden" sound that so few solid state amplifiers possess....




BTW, I have the final piece of the K10A puzzle - the dual 30VAC transformer. In very short order I'll be finishing that project up.
Ummm warm, glowing tubes.

Double BTW, My spell check wants to correct your name to "Consummated"

 

[gtforme00]

Brilliant heat sink John! Kudos for your resourcefulness. How did you get the surface so flat with all those pieces bolted together? That looks like an awful lot of sanding.

-David

 

[MJL21193]

Hi David,
Thanks. My bolt holes were pretty precise, location wise and they were slightly oversize. This gave it the flexibility to be put together evenly - I laid it on a flat surface after it was assembled and before it was bolted tight.
I went to town on it with the belt sander driving a 60 grit belt. It's still not as smooth as I want it yet.
I have all of the parts for the other one, just need to put it together.
These are 6" high and 17" long. Hefty.

 

[MJL21193]

Making progress....

 

[GK]

Thanks Glen. I kinda thought that would be the case, but it's nice to hear it.

I did want to optimize the design and it has come a long way. I can see your point, but the fancy VAS transistor is just a 36 cent, off the shelf Fairchild and by all accounts, one heck of a VAS. The re-configuring of the cascode bias was a two for one - lower component count and better performance. Plus I got rid of those garish LED's.
There's more to come. I need to get my final board layout done, populated with the final component choices and do some full power tests. Hopefully all goes well.

Like i said before, I'm not looking for record breaking numbers - any amp that can keep distortion below .01% is doing as much as it needs to IMO. I want the power, stability and simplicity that this seems to have. Fairly low cost too.



...And who knows, maybe the mixture of these different disciplines and levels of complexity elevates it to a sublime plateau, achieving the magical "golden" sound that so few solid state amplifiers possess....




BTW, I have the final piece of the K10A puzzle - the dual 30VAC transformer. In very short order I'll be finishing that project up.
Ummm warm, glowing tubes.

Double BTW, My spell check wants to correct your name to "Consummated"

Welllll....... I guess it all depends on what one views as complicated. I'd personally occupy my PCB real estate with a pair of BJT's as current sources for the LTP tail and VAS bias over some power resistors and chunky electrolytic capacitors, and reap the much better low frequency PSR.


Consummated

–adjective
4. complete or perfect; supremely skilled; superb........


Cheers,
Glen