TGM2 amplifier

[Professor smith]

Gareth,

I'd guess the reason is that you have a phase reversal before the OLG has come back to unity - that is, you have not met the Bode-Nyquist criteria. The fact more degeneration seems to correct it is clear indication your lag compensation may be the problem. Increase Cdom!


Cheers,

Hugh

Dear Mr. Hugh,

would you like to explain further in mathematical terms the idea of the bode nyquist criteria?

I'm not sure what you mean by this.

Thanks.

 

[AKSA]

No, Mr Tamura,

I will not, even if I could. The maths of Bode-Nyquist is tricky, and while I understand the principle, I have neither the time, nor am I willing to negotiate the symbology on this forum as I have already explained to you in a personal email.

I can only assume that your request is another 'sting in the tail', a careful mischief. Why do you try to confront people, Kenji?

Pete, John, do you know the math on this? Perhaps you can help Kenji to understand more fully. My simple word explanation really should be adequate for someone with a basic understanding of electronics anyway. I believe it is fairly obvious that if the propagation delay through the amp equals or exceeds half the period of the waveform, then there will be a 180 degree phase inversion of output over input. If global feedback is applied, it will then be positive, not negative, and uncontrolled amplification leading to rail to rail oscillation will immediately result.

Hugh

 

[ostripper]

Here's some links , professor...
http://www.esr.ruhr-uni-bochum.de/rt1/syscontrol/node45.html

Easier and how it applies to amplifiers..

http://www.facstaff.bucknell.edu/mastascu/eControlHTML/Freq/Nyquist3.html#HowStable

Amplifier stability is no longer a issue , the finer aspects of the amps fidelity is the current dilemma..

If you want get your "hands dirty", download ...
http://www.linear.com/designtools/software/ltspice.jsp
Open the file enclosed in BTF50.zip (attached) , simulate - RUN ..
in empty plot pane , add expression ... -V(c)/V(a)

Then .. change values of R3-4 , C9 , R16 to see the plot change.
Its all set up to work.(no models needed)
OS

EDIT ... with the CFP , BG

 

[Bigun]

I didn't have time to get back to the CFP sims today, maybe this weekend. I did revisit the turn-on thump question as it seemed like that would be a quick lunchtime activity on my existing spice model.

But......

Keantoken,

In spice I see a turn-on hump at the output of around 1.6V to 2V, but equalizing the RC's on the LTP inputs was problematic. So I need to seek more guidance on my understanding of these things (bear in mind that the actual TGM1 amplifier doesn't have a loud turn-on thump problem so my sims are exagerating things).

Here goes: I just looked at a simple LTP by taking off the 'slave' CFP devices and adjusting the LTP feed resistor accordingly.

Power comes up at the output stages according to the RC of the psu (7ms) and at the front-end stages according to the on-board rail-caps that isolate the front-end of the amplifier (10ms). The LTP devices become fwd biassed and turn on pretty early (in my sim. it's 3ms but that's not universal) and as they do current starts to flow through the LTP. This current depends on the LTP feed resistor and the rail voltages which are still coming to life. Hence, we the voltage difference between the LTP collectors climbs with increasing rail voltage. As the base-emitter voltages stabilize the LTP output settles down to a steady level.

In the meantime, over at the VAS house, things are sleepy, the VAS is not yet forward biassed because the LTP output has settled already at a differential voltage that is too low. Current is flowing through R5 to charge the Bootstrap cap (RC = 220ms) with nothing much flowing down through the VAS. The VAS collector follows this voltage on the Bootstrap cap. [digression - thank goodness for this cap, without which the VAS collector would rise quickly and produce a nasty output spike]. As soon as the VAS collector voltage rises enough to fwd bias the driver (in my sim. it's around 30ms) the party starts and the amplifier output swings up towards the +ve rail and your speaker cone moves next door.

Before this gets out of hand, the neighbours have complained i.e. the feedback node at the LTP sees the output driving upwards. The LTP output sends out an error signal which climbs and eventually there's enough current flowing through the LTP to fwd bias the VAS. This turns on the VAS, lets the air out of the party and the good ol' sensible feedback loop steps in to restore order to the output. Now the LTP output settles out at a high enough level to keep the VAS turned on and everyone is happy.

The attached shows some of what I'm trying hard to understand. The red line is the ramp up of the supply rail.The green line shows the voltage difference between the two LTP collectors (it first starts moving after the LTP devices turn-on, then settles out at 300mV, then gets sharply pulled up by the influence of the fdbk), blue is the amplifier output.

So how do we fix this ? - Speeding up the fdbk loop by reducing C2 but this cuts into the bass response, we need to kick-start the turn-on of the VAS to get the fdbk servo running earlier. A current mirror would fix it, not a CCS.

 

[keantoken]

First, to clarify.

Most amplifiers will have some kind of turn-on thump, because of active devices turning on, etc. If you use bootstraps or a CCS you can change the cap values so that one part will turn on after or before another (what if you bootstrap the LTP and have it turn on after the output stage is on?). That said, most designers don't go crazy about trying to make it dead-silent when it turns on (If I'm not mistaken). Sometimes it is a price you have to pay for a given topology that has benefits in other places. If it's too bad, then you can always connect the speaker through a relay and have the relay turn on only after the amp is fully on.

I might have made a mistake mentioning the RC constants so urgently. I should have just said that I thought it would help. But now that you're interested, maybe you'll be happy to hear the full story.

In the schematic you posted:
The input cap is 470nF. The feedback cap is 22uF. That is about 50 times as large. Your gain is only somewhere around 20.

If the net input network resistance and feedback network resistances are the same, then your feedback cap should be about the value of your input cap multiplied by your gain. So,

Given that R1 = R9+R8 (notice however, this is not true in your schematic), then

Cfb should = Cin*gain

Your cap is about 2.5 times this value, which means your feedback loop has higher low-end bandwidth than is allowed through the input. I would say 8.6uF would be close to the value you want here.

Generally, lower value caps will be cheaper if you get the high-quality ones. Rather than getting an expensive 22uF cap where it is not really needed, you can get a 8.6uF cap to put in the same place for cheaper. (then again, every component adds distortion. By using a cap well within its ratings, you can be sure it doesn't contribute significant distortion - but this is the purist approach, maybe not important for you.)

Although this might not have any serious benefits, I feel happy knowing that both ends of my LTP are that much closer to being equal.

Next thing:

There is a high impedance between your output and your feedback device (R9). This means that lower signals can cause significant error. If you decrease the net resistance of your feedback network by 10, I think your circuit will be more stable and will also have insignificantly lower distortion (because the base currents of the LTP themselves cause error especially when you have high impedance near them)(this would make R9 4.7k and R8 270ohm). This probably won't matter as much with your CFP LTP because of the high gain, but I think you should try it and see.

If you lower the feedback network's net resistance by 10, you will have to multiply the feedback cap by 10, to keep the even RC constants, FYI.

Why don't you replace your input and feedback resistors and caps with the values of those in my LTP schematic. In my schematic there is a gain of exactly 10 (perfect for 20V rails, given 2V pk-pk maximum input) and no strange component values, so matching time constants is trivial. If you want higher gain you can use 33k and 2.2k resistors, which would give a gain of exactly 15 (perfect for 30V rails). Then when you feel you know the system well, you can use your own values and modify to your liking.

And C6: I have tried using such a cap in the simulator and it was unsuccessful in my designs and caused instant oscillation. Theoretically it should be very helpful for stability because it isolates the output stage from any HF input, while ensuring LTP stability. However, it might increase 20KHz distortion (which generally isn't considered important). With the modifications I've mentioned, it might not be necessary. Better coupling between the output and feedback can help with stability issues in my LTSpice experience.

I hope this helps,
- keantoken

 

[Bigun]

Originally posted by keantoken If you use bootstraps or a CCS you can change the cap values so that one part will turn on after or before another (what if you bootstrap the LTP and have it turn on after the output stage is on?).

If my sims are right, the issue is not how fast the output stage turns on, it's on already. The issue is getting the VAS turned on so that the fdbk servo is up and running. This means ensuring that the LTP output is up quickly, not slowly ?

If the net input network resistance and feedback network resistances are the same, then your feedback cap should be about the value of your input cap multiplied by your gain. So, I would say 8.6uF would be close to the value you want here.

I agree, I could drop the 22uF down, but I have already bought them so it would actually be less trouble to leave it as-is. I know that's not in the spirit of experimentation but it does play nicely into your next suggestion....

If you decrease the net resistance of your feedback network, I think your circuit will be more stable and will also have insignificantly lower distortion. If you lower the feedback network's net resistance you will have to multiply the feedback cap, to keep the even RC constants, FYI.

With the 22uF cap I can lower the fdbk loop resistance, perhaps only marginal in this case but an easy change (I'll check the sim).

Why don't you replace your input and feedback resistors and caps with the values of those in my LTP schematic. Then when you feel you know the system well, you can use your own values and modify to your liking.

I'll certainly try it out in the sim as a minimum and depending on how lazy I get the pcb can also be used to make a real life comparison.

And C6: I have tried using such a cap in the simulator and it was unsuccessful in my designs and caused instant oscillation.

I have found this also, in the sim, C6 feeds back too much high-freq. In TGM1 the LTP gain is lower so it was a good value. For TGM2 I will lower C6.[/B][/QUOTE]

Thanks for the input - very helpful !

 

[Bigun]

Improving the VAS

Whilst playing around...

I tried the quasi-differential input mentioned on my old thread.

The results of this weren't compelling, if anything I saw an increase in H3.

Cross this one off the list.

 

[Bigun]

Improving the VAS part 2

But since the 'theme' of TGM2 is the use of CFP , so the natural thing is to put a CFP into the VAS. It's more in keeping with the approach so far than the Cascode (Nelsonvandal - maybe CFP VAS is still good for the 'memory effect' ?). Now we can make pretty much all of the front end prone to oscillation, the VAS included !

Will the output stages also need to be CFP ?

The good news is that this cleans up the distortion spectra quite nicely, particularly H3. I assume that it also provides the opportunity to buffer some of the non-linear loading from the drivers.

[no current mirrors were harmed or used in this design]

BUT: will it sound better, or have I taken away something here, perhaps the Bootstrap has lost its influence over the VAS with the CFP version ???

 

[Bigun]

Originally posted by keantoken If you decrease the net resistance of your feedback network by 10, I think your circuit will be more stable and will also have insignificantly lower distortion

OK, I gave this a spin too:

by reducing the fdbk loop resistance whilst keeping the gain roughly the same. As you predicted there is a clear improvement in the overall distortion - in that the 'floor' moves down by several dB compared to my starting point. However, as the floor moves down, it uncovers harmonics. So, will we have any sonic benefit from making this change ????

The frequency response also changed, as could be expected. The lower resistance cuts into the base response and extends the hf response. The phase margin might also be improved, especially if I make some other adjustments.

Attached shows 3 different combinatinos of fdbk series/shunt resistance (red: 47k/2.7k, blue: 32k/1.8k, green 17.4k/1k)

p.s. this was done with the CFP VAS.
p.p.s. Bootstrap was bad for distortion, crossing that one off the list.

 

[keantoken]

OK, I gave this a spin too:

by reducing the fdbk loop resistance whilst keeping the gain roughly the same. As you predicted there is a clear improvement in the overall distortion - in that the 'floor' moves down by several dB compared to my starting point. However, as the floor moves down, it uncovers harmonics. So, will we have any sonic benefit from making this change ????

You are simply uncovering what was already there in the first place. I don't know if the SPICE noise floor is the same as the real life noise floor, because I know nothing about noise. Personally, I would keep this if it improves the circuit and then see if I can make the distortions that are unavoidable more "favorable". Perhaps you should do a personal listening comparison to see and decide for yourself.

Try a metaphor. As the commercial goes, "other fresheners just cover up bad scents". I would rather have an honest amplifier than one where the bad parts are covered up or disguised.

The frequency response also changed, as could be expected. The lower resistance cuts into the base response and extends the hf response. The phase margin might also be improved, especially if I make some other adjustments.

Attached shows 3 different combinatinos of fdbk series/shunt resistance (red: 47k/2.7k, blue: 32k/1.8k, green 17.4k/1k)

p.s. this was done with the CFP VAS.
p.p.s. Bootstrap was bad for distortion, crossing that one off the list.

The FFT looks interesting. Those HF harmonics trailing to the horizon? Supposedly they sound bad. Were these harmonics simply covered up or did we produce them by the changes we made?

MikeB on this forum has mentioned the harmonics "upfolding" affect. Basically, this states that when your VAS and/or LTP are not linear, and you use a large Cdom size, the lower harmonics get circulated through the feedback system. Now you have distortion harmonics of the first set of harmonics added in. The distortion is distorted and gets reverberated up in frequency like folding a piece of paper.

So you want your VAS to be very linear. Your CFP LTP is already extremely linear, it seems.

Here is my observation: The VAS transistor has to handle the full output voltage across its collector. As we know, a transistor's gain changes with collector current and collector voltage. Run your amplifier at 1KHz full output, and plot the gain of your VAS transistor (use a single transistor VAS, not CFP). You will see something awful. Do it. Run it again at 20KHz and see what happens then.

Using a CFP VAS helps this, but only by adding gain. It doesn't solve the problem. This has obvious implications for stability. And since you need such a large Cdom to control this high gain, your distortion will start to rise at a lower frequency, and your THD at 20KHz will increase.

My solution to this has been to cascode the VAS, as illustrated in my schematics in the below link (Q4 is the cascode transistor). Try this and then look at the VAS gain at 1 and 20KHz. Some say you don't need this, since no one hears above 16KHz nowadays (but what about your children?). Others say frequencies higher than 20KHz can affect the tonal qualities of the music. I have found that if I do this, I don't need super-high gain in order to get low distortion. I will let you decide what you want.

http://www.diyaudio.com/forums/showthread.php?postid=1826545#post1826545

One possible bad side-effect, is that since you will have to set the cascode voltage low in order to get full output swing, the Cob of the VAS will be higher on average (Cob or collector-base junction capacitance decreases with higher voltage, just like a Varactor. This goes with all semiconductors). This hasn't been a problem for me, and it only seems to amount to lower HF bandwidth.

Next, I've heard it mentioned that some B and AB designs use unusually high bias on the driver transistors. This makes it basically a Class-A amp driving a class-B one, buffering the high harmonics produced during crossover. You might want to try this by making R15 22ohms.

3rd suggestion: I think you can do better with your choice of transistors.
I think you can do better than the 2N5401/5551 with 26V rails. I have seen many times a person will compare different designs on the simulator, one using inferior transistors and the other using transistors up to spec. It's a no-brainer which circuit will win. USE GOOD TRANSISTORS, BEFORE YOU DISMISS A DESIGN. A SINGLE TRANSISTOR CAN BE EXTREMELY IMPORTANT TO A DESIGN'S SUCCESS, CONTRARY TO CLAIMS THAT "most transistors will have generally the same performance in a certain position".

Here are some good transistors that I use exclusively and I think will serve you well:

2N5089: This is 25V and has decent gain. The 5088 is the same thing but with 30V. The 2N5210 is again similar but has 50V max voltage. Your LTP transistors only take half the full voltage, so you don't have to choose transistors to take full rail-to-rail voltage.

2N5087: This is basically the PNP version of the 5089 and has a max voltage of 50V.

Bc546/556: these one come in A,B and C varieties for different gain ratings and are used very often. You can get PNP and NPN versions with similar Hfe and have a balanced complimentary design.

MPSA18: This is a low-noise transistor with very high gain (2k) and high voltage (45V). I often use this as a VAS in a PNP LTP amp, or in the LTP in an amp with a PNP VAS. It's cheap and performs well. This is my first choice for a higher voltage transistor if I can't get by with a 5089 or 5210 (this could phase those two out completely, though).

2N5771/5769: Everyone eventually needs (or wants) a good switching transistor. Sometimes they can be useful in Audio as well. There can be sweet benefits from paralleling a bunch of switching transistors to get a higher-gain and superfast output.

Note: Generally, using fast transistors in amplifiers will get you lower distortion and higher stability within the audio range.

Higher voltage transistors will generally have higher Cob. Instead of using a 150V transistor with high Cob, you can use a lower voltage version with lower Cob, which will give you more stability, and better distortion (and less "upfolding" where it is unavoidable).

But be careful. You should choose transistor with high enough voltage to withstand the higher voltage that can come from your power supply before there is enough current running to pull the voltage down.

Now I should emphasize that nothing I say can hold up to what you learn through your own experience. You should test what I say and see if it's that way for you. I'm interested in your results. Reality has a tendency to be different sometimes for different people...

Good luck,
- keantoken

 

[ostripper]

The FFT looks interesting. Those HF harmonics trailing to the horizon? Supposedly they sound bad. Were these harmonics simply covered up or did we produce them by the changes we made?

Those are "artifacts" caused by too low a number of FFT points.

At the default (LT decides) or this... .param num_fft_pts 1024 , a "herringbone" plot emerges. with increased sampling ...
.param num_fft_pts 32768 , a pure plot will appear.
OS

 

[keantoken]

For 1KHz THD, I use a 1uS timestep. For 20KHz, 100nS timestep has done me well.

If I want more, though, I have a set of options that I paste into my circuit that I could share with you. It computes the timestep according to FFT size, your test signal frequency, and the length of your simulation.

- keantoken

 

[Bigun]

Originally posted by keantoken MikeB has mentioned the harmonics "upfolding" affect... the lower harmonics get circulated through the feedback system.

Yes - In fact I have seen this in the sims already, if the design isn't linear high signal amplitudes generate IM products. In one version I looked at, these products were all over the place, a real mess and no doubt the sound will get all muddy, especially the LF where amplitudes tend to be larger.

Run your amplifier at 1KHz full output, and plot the gain of your VAS transistor (use a single transistor VAS, not CFP). You will see something awful.

Using a CFP VAS helps this, but only by adding gain. It doesn't solve the problem. And since you need such a large Cdom to control this high gain, your distortion will start to rise at a lower frequency, and your THD at 20KHz will increase.

I have yet to look specifically at the VAS gain but I have looked at the overall amplifier FFT at 200Hz, 2kHz and 20kHz and as you say, the amplifier does have an easier time of it at lower freq.

Here's where my understanding is still evolving. I had thought that the CFP provides a benefit in a similar fashion to a Cascode - in that the slave device helps control the voltage across the the collector-emitter of the master device, thus linearizing it. The CFP has a level of local nfb within the pairing, meaning that it trades off some of the higher gain for lower distortion. Now I haven't looked at the stability, but Cdom must surely serve this purpose admirably without a need to increase it from the non-CFP version. Another benefit of the CFP is that the slave device buffers the master device from the non-linearity of the output stage. Whether it does this any more effectively than a Cascode I'm not sure.

Does my understanding look right ??

My reluctance to use a cascode is purely subjective. I suspect that it changes the complex interaction between the VAS and Bootstrap which is responsible for the characteristic 'sound' of Bootstrap amplifiers. I want to retain that sound. In your schematic ou use a CCS for the VAS (looks very interesting by the way, we just take your CFP output and my CFP front end to make an all-star CFP amp !).

high bias on the driver transistors makes it a Class-A amp driving a class-B, buffering the high harmonics produced during crossover. You might want to try making R15 22ohms.

Definitely worth looking at on the sim, and doing the experiment on the pcb would be easy too. My attention hasn't moved to the drivers yet, no doubt there's more to learn here too.

3rd suggestion: I think you can do better with your choice of transistors.

I hadn't thought about this at all, a good point !
In fact, I had simply taken it for granted that I would play with the topology and use the devices I specified without a 2nd thought - they being a 'safe' choice. Looking at alternative devices opens up a whole new avenue to explore. That may have to wait until I have built one with the existing devices (which I bought already).
Thanks for your help and showing interest, Keantoken.

Originally posted by OS Those are "artifacts" caused by too low a number of FFT points.

Nicely spotted! I wouldn't be surprised at all if some further tweaking is needed.

 

[Bigun]

That may have to wait until I have built one with the existing devices (which I bought already).

I wasn't quite right on that, I haven't yet purchased the complementary device for the VAS, although I have plenty of the BD139's already. I'll have to get some BD140's.

the only JFET input transistors I've used are SK170 and SJ74

I don't see these at Digikey - are they still available ?
I am concerned that the particular JFET is important for the sound quality. I certainly don't want 'harsh'

Also, I see from your posts that you have built diamond buffers. In terms of improving the driver stage of an amplifier ? there are still EFs in the output no ?

The reason I ask is that I came across the Bromley output structure (from 1969) and have since found it to be similar to the Sijosae buffer and well, it's similar to the diamond buffer !

Maybe the advantage is that the buffer is operating in Class A, like Keantoken suggests ?

(I think Lumba also prefers these over the 'regular' push-pull ...?)

 

[Bigun]

sorry - not 'Bromley', but 'Blomley'

 

[keantoken]

I hope I'm not tiring your eyes by writing so much. Tell me if I'm getting exhausting.

Here's where my understanding is still evolving. I had thought that the CFP provides a benefit in a similar fashion to a Cascode - in that the slave device helps control the voltage across the the collector-emitter of the master device, thus linearizing it. The CFP has a level of local nfb within the pairing, meaning that it trades off some of the higher gain for lower distortion. Now I haven't looked at the stability, but Cdom must surely serve this purpose admirably without a need to increase it from the non-CFP version. Another benefit of the CFP is that the slave device buffers the master device from the non-linearity of the output stage. Whether it does this any more effectively than a Cascode I'm not sure.

Does my understanding look right ??

I thought this too. The slave device keeps the current through the master very constant, but it does nothing to keep the voltage in check - because there is always about .69V across the Vbe junction. The Vcb of the master device will simply be the output voltage minus .69V. The extra gain only keeps the master operating within a more linear current range, but the voltage range is untouched. The CFP lowers distortion simply because the slave device is trying to keep the master's Vbe at a very constant value while driving the load. In a way you have problems stacked on top of each other because you not only have to worry about the master's Vcb, but the Vcb of the slave's also.

You could use a cascoded CFP for the VAS. That covers the master device on all levels. There are surely sweet benefits from doing this!

My reluctance to use a cascode is purely subjective. I suspect that it changes the complex interaction between the VAS and Bootstrap which is responsible for the characteristic 'sound' of Bootstrap amplifiers. I want to retain that sound. In your schematic ou use a CCS for the VAS (looks very interesting by the way, we just take your CFP output and my CFP front end to make an all-star CFP amp !).

Well, subjectively, you'll have to try it out and see (I will make no statements regarding subjectivity vs. objectivity). Objectively, yes there might be some interaction. The parasitic capacitance of the cascode device might affect slewrate, etc. If you are worried about this you might want to use a 2N5769 as a cascode transistor, because of it's low Cob. But it's voltage regulation will be less because of low gain (that's exactly what a cascode is - a voltage regulator, usually a capacitance multiplier to be exact).

The bootstrap will affect current only, not voltage. This is because the input to the output stage will always be the same voltage and current to produce the same output. Whether we use a cascode or not, the current through the VAS will not change, only it's collector voltage. This will change its linearity characteristics. Whether you think the nonlinear VAS sounds better or worse than the more linear one is entirely up to your ears. I think building a board with extra slots for comparison is a good idea.

Thanks for your help and showing interest, Keantoken.

Your welcome. But if this weren't at least as fun for me as it is for you, I might not be writing. The thanking is mutual.

sorry - not 'Bromley', but 'Blomley'

Thank you for the info. Usually, the method for output stages is to use drivers to "force" the outputs into acting like we want them to. Sometimes we can do things differently with success, though.

Good luck,
- keantoken

 

[Bigun]

Originally posted by keantoken I hope I'm not tiring your eyes by writing so much. Tell me if I'm getting exhausting.

I'd rather have input, suggestions, corrections and ideas than silence or I won't learn. Whether or not I can keep up the pace is another question, my kids already complain I spend too much time on the keyboard...

I thought this too. The slave device keeps the current through the master very constant, but it does nothing to keep the voltage in check - because there is always about .69V across the Vbe junction. The Vcb of the master device will simply be the output voltage minus .69V. The extra gain only keeps the master operating within a more linear current range, but the voltage range is untouched. You could use a cascoded CFP for the VAS. That covers the master device on all levels. There are surely sweet benefits from doing this!

...so there is no free lunch. I remember now that the 'memory effect theory' uses both a Cascode and a CFP so that both current and voltage are pinned and hence power is constant. And that applies to the VAS too.

I also remember that the usual approach is to Cascode both LTP devices. How about instead of that we Cascode the whole LTP. Allowing that the slave devices will keep the current fixed for the master devices then we don't need a CCS to feed the LTP. So we put a Cascode/voltage regulator at the top instead (attached) ?

I will make no statements regarding subjectivity vs. objectivity

now that's not very brave of you
I wear both hats, they are not in conflict with each other in my mind. If that sounds too simple, try: http://design.caltech.edu/Misc/pirsig.html

Thank you for the info. Usually, the method for output stages is to use drivers to "force" the outputs into acting like we want them to. Sometimes we can do things differently with success, though.

Well I don't think Blomley is going to make it into TGM2 at this time, but with some further thought and some further investigations it may end up in TGM4....

 

[keantoken]

I'd rather have input, suggestions, corrections and ideas than silence or I won't learn. Whether or not I can keep up the pace is another question, my kids already complain I spend too much time on the keyboard...

Don't worry. Everything I have said is stored in your mind. If you keep playing on the simulator, you will eventually come to understand all of it. Time is not much of an object here.

...so there is no free lunch. I remember now that the 'memory effect theory' uses both a Cascode and a CFP so that both current and voltage are pinned and hence power is constant. And that applies to the VAS too.

I also remember that the usual approach is to Cascode both LTP devices. How about instead of that we Cascode the whole LTP. Allowing that the slave devices will keep the current fixed for the master devices then we don't need a CCS to feed the LTP. So we put a Cascode/voltage regulator at the top instead (attached) ?

I've tried some crazy things like this and for me they had bad stability. What you will have is a feedback loop keeping the LTP even, instead of the devices using their own internal feedback. Generally this means larger Cdom and worse HF performance (not that you won't possibly be able to get a pleasing sound). Go ahead and try out your idea, though. Nothing is set in stone.

now that's not very brave of you
I wear both hats, they are not in conflict with each other in my mind. If that sounds too simple, try: http://design.caltech.edu/Misc/pirsig.html

I wear both hats too. I don't despise either. But for my engineering mind I believe in some pretty crazy things... I find that to seriously live a meaningful life, I had to search for something that made sense and that didn't end in paradox, as most things seemingly do... It turned out that in order to find this, I had to look in some "unorthodox" places. I am fine with pure data. But when you use anything else than data, it is subject to viewpoints, and my experience with viewpoints is that they're only good if you're looking for unique ideas (things you would have never thought of on your own), or if you want to tell someone your own spin on something and win their unconditional trust... IE entrapment (good for marketing, as it turns out). I prefer to build up from a objective base and try to see the patterns in what we like to call magic or subjectivity (these terms are interchangeable in my experience). However, I cannot simply shun "Magic" without losing my objectivity. So, I do my best to try and see into both worlds... This really helps with real life and with understanding other people.

I would be happy to converse with you about this, but not on DIYAudio. This forum is for Audio, and this thread is for TGM2. Let's keep it that way. If you're interested, PM me.

As far as Zen and the Art of Motorcycle Maintenance goes, I think it's a good book, but I haven't finished it. I will have to do so. I was certainly impressed by it's Wikiquote page.

"brave and manly" is a gender thing. No need to be brave here, everything is easily handled.

- keantoken

 

[MJL21193]

I wasn't quite right on that, I haven't yet purchased the complementary device for the VAS, although I have plenty of the BD139's already. I'll have to get some BD140's.

I'm with keantoken on the device choices. Instead of BD139/140 for VAS get some
2SA1381 (PNP) and 2SC3503 (NPN). Available at Digikey, these are excellent devices - low Cob, high gain, high voltage and cheap.

BC550/BC560 are low noise, high gain. I can't comment on fets - not my cup of tea.

 

[Bigun]

OS,

I have followed up on your suggestion about current sharing in the LTP and found that the distortion profile does improve if I allow the LTP master devices a bit more juice. I didn't get time to explore a range of values using .step today but the results were positive. I reduced the input device collector resistors from 1k2 to 820R, and used 33R for emitter degeneration.

With a 10nF load in parallel witht he speakers, phase margin evaporates quickly. A choke in the output is now on the agenda.

I also followed up on the suggestion of less capacitance between the VAS and feedback node on stability concerns; I reduced C6 from 15pF to 4.7pF.

The addition of the CFP to the LTP remains a clear winner from a distortion perspective. But I have not convinced myself that the additional improvements from the CFP in the VAS are needed unless I had to drive a more challenging output stage (a non-switching driver stage will be a topic of my next project).

Keantoken,

I haven't tried putting the VAS Cascode into the sim yet (maybe tomorrow).


MJL21193,

Good suggestion for improved VAS devices. However, I do have a pile of BD139's....so maybe will have to wait for my next Digikey order.