Yes you are right, for dipper investigation we should do that for all loops. I tried to simulate inner TMC loop but somehow I couldn't get it satisfatory, could be that I don't know how to do that. Now it's time to learn more.
I simulated TMC whitin global loop only, but you should pay attention were to put the probe. If the probe is between output and global feedback resistor then result is similar to normal Miller comp, but if the TMC resistor is on the left side of the probe(in my shematic) then result is similar to the TPC simulation with second order slope. With that method I simulated my TT amp(http://www.diyaudio.com/forums/solid-state/182554-thermaltrak-tmc-amp-3.html#post2643498) and the amp was stable when built.
That is what I am trying to get answer from others if that way to simulate TMC LG is good enough.
BR Damir
Yes he can. Similarly we can forget to buy the book. I made that decision round about the time Doug admitted that an entire chapter was so bad it needs to be completely rewritten, but he doesn't have the time now so it will remain unchanged in the new edition.
I've noticed that his responses to suggestions tend to fall broadly into one of three categories:
- You'll be glad to know that's already in the next edition.
- That would be a valuable addition. Unfortunately it's now too late to include it in the next edition.
- I don't want to cover that subject in my book.
That makes me wonder about the accuracy of the thread title and the first post. Does he really want our opinions, or does he just want to generate interest in the new edition, in the hope that we'll buy it?
Perhaps this thread should be moved to the commercial sector.
p.s. Come to think of it, I remember feeling much the same way about his previous thread advertising the (then) upcoming book on crossovers.
Last edited:
godfrey
I think there is not only a single reason for this thread.
I think it is a survey what people want to read about.
But also of course a way to generate interest about the coming book.
I see no problem to have this thread here in Solid State.
I am also sure Douglas Self will cover some topics initiated here.
That way he will get readers of his book.
Regards
I think there is not only a single reason for this thread.
I think it is a survey what people want to read about.
But also of course a way to generate interest about the coming book.
I see no problem to have this thread here in Solid State.
I am also sure Douglas Self will cover some topics initiated here.
That way he will get readers of his book.
Regards
I think I explained why an audio power amplifier has to be unconditionally stable. Op amps do not have the same failure mechanisms if they transiently burst into oscillations. As an example of possible failure mechanisms, think the Zobel resistor. This resistor is usually a 2-3W device, that may fail after repeatedly surviving short power surges of hundreds of watts. If the Zobel resistor fails, then the whole stability in difficult loads may be compromised.
Unfortunately, I don't think it's possible to design a TMC+TPC amp that is unconditionally stable (pretty easy to understand why), so this entire simulation exercise is, at best, of theoretical importance only. Your personal experience with conditional stable amps doesn't make it look better. I haven't looked closely, but if you purposely built a TMC amp that is only conditionally stable, then you were pushing your luck, to the point that one may think it's an incompetent design.
Maybe I am wrong, but I dislike to use any input device without a base or grid stopper. My prefered value is around 100 Ohm. This is a lesson I retained from some old amps which, having their input directly AC connected to the cursor of a volume pot, suffered from annoying little auto-oscillations when the cursor was at the ground potential. This does not happen with my Blameless which has no base stopper.
In the fifth edition, I noticed input RC low-pass filtering at very high frequencies, around 16 MHz :
R = 10 Ohm, C = 15 pF to 1 nF (typo "INF", trimodal circuit, page 319, Figure 10.19)
R = 10 Ohm, C = 1 nF (Load-invariant circuit, page 169, Figure 6.31; class G circuit; page 356, Figure 12.12)
R = 100 Ohm, C = 100 pF (balanced input circuit using 5532s, page 539, Figure 20.8)
In the February and March 1994 issues of Electronics World + Wireless World, the first published Blamelesses had a feedback network from the output stage made of 10 kOhm in parallel with an RC series of 330 Ohm + 15 pF.
More recent Blamelesses do not include this RC, but the 10 kOhm has been lowered to 2.2 kOhm as well as the input biasing resistor (a bootstrap connexion can be provided to enhance the AC input impedance).
Last edited:
If the data I have on the differences between the 4*** and other NJL transistors is still accurate then R13 will have to be reduced to about zero, or perhaps lower.
Multiple, parallel ThermalTrak diodes to share the current could be part of the solution if the bias is still excessive after R13 is shorted.
If you actually build this then please post updates on the thermal stability.
Best wishes
David
Last edited:
Hi Damir
I think it is safest to use a Tian probe rather than the simple Vsource loop probe that you have used in this case. See Loopgain2.ASC in the LTSpice examples.
The Tian probe is insensitive to circuit impedance and so can be placed where ever you want to measure, not just at "special" places.
Best wishes
David
Could you quote GNFB commercial power amp to be stable with the gain set to one???
Bode Plots, AD797 etc
But I was actually referring to the section on BYPASSING CONSIDERATIONS and other pearls of wisdom scattered in the datasheet. I'll add that small Aluminium Electrolytics have almost the perfect amount of ESR to replace his evil Tantalum + carbon resistor. I like very simple circuits with tight PCB layouts which allow 100u decoupling close to the outputs & driver stages. One caveat is that it puts electrolytics close to hot heatsinks but commercial examples of my sins seem to have survived at least 2 decades of this.
To those horrified by his evil shunt compensation to an even more evil negative Vss, I'll point out that the 'distortion cancelling capacitor' (Cn in figs 28 & 29 and C2 in fig 41) is actually a 'pure Cherry'.
_________________________
Perhaps the emphasis on Bode plots of Open Loop gain is misguided. I look at them to see how much loop gain I have at audio frequencies and to see if a circuit is impossibly unstable. Very dangerous to do this in real life cos the cost might be the release of Holy Smoke.
But if I really want to look at stability in SPICE world using Bode plots, I just close the loop and look at Close Loop response. This is a far better guide [*] to instability than trying to translate an open loop Bode plot to a Nyquist plot (which is the real test of stability. Jerald Graeme's "Feedback Plots define Op Amp AC Performance", sboa015.pdf from the TI library is a good treatment of this.
Even better is to do .tran instead of .ac
I sorta 'believe' .tran rather more than .ac (open or close loop). Look at the Transient plot as it starts up too to see if there is oscillation.
* Of course open loop Bode plots are useful when you are tweaking stuff for initial stability.
kgrlee said:
Guru Wurcer's AD797 is indeed an evil VFA with even more evil shunt compensation.Yes but the AD797 is a single gain stage VFA with shunt compensation to ground (well virtually ground, Vcc/Vee) like a CFA so it has to be inferior.
But I was actually referring to the section on BYPASSING CONSIDERATIONS and other pearls of wisdom scattered in the datasheet. I'll add that small Aluminium Electrolytics have almost the perfect amount of ESR to replace his evil Tantalum + carbon resistor. I like very simple circuits with tight PCB layouts which allow 100u decoupling close to the outputs & driver stages. One caveat is that it puts electrolytics close to hot heatsinks but commercial examples of my sins seem to have survived at least 2 decades of this.
To those horrified by his evil shunt compensation to an even more evil negative Vss, I'll point out that the 'distortion cancelling capacitor' (Cn in figs 28 & 29 and C2 in fig 41) is actually a 'pure Cherry'.
_________________________
Perhaps the emphasis on Bode plots of Open Loop gain is misguided. I look at them to see how much loop gain I have at audio frequencies and to see if a circuit is impossibly unstable. Very dangerous to do this in real life cos the cost might be the release of Holy Smoke.
But if I really want to look at stability in SPICE world using Bode plots, I just close the loop and look at Close Loop response. This is a far better guide [*] to instability than trying to translate an open loop Bode plot to a Nyquist plot (which is the real test of stability. Jerald Graeme's "Feedback Plots define Op Amp AC Performance", sboa015.pdf from the TI library is a good treatment of this.
Even better is to do .tran instead of .ac
I sorta 'believe' .tran rather more than .ac (open or close loop). Look at the Transient plot as it starts up too to see if there is oscillation.
* Of course open loop Bode plots are useful when you are tweaking stuff for initial stability.
Last edited:
Yes.
Each and every amp that is Miller compensated (which makes prolly about 99% of the current market). Conditional stability is strictly related to compensation methods of the 2nd (or larger) order.
I am though not aware of any commercial amp that is using a 3rd (or larger) order compensation network.
I'll point out that the 'distortion cancelling capacitor' (Cn in figs 28 & 29 and C2 in fig 41) is actually a 'pure Cherry'.
Since the forward path is the same with or without it (except for small second order effects at high frequencies), I don't see the arguement for this. It affects neither the crossover nor phase response at crossover in more than a trivial way. It certainly has absolutely nothing to do with anything in the 5534.
Last edited:
You forget that there is output stage, specialy triple EF to spoil first order Miller compensation after some frequency, and if we dont want triples then we are back at stone age of VFA. And back to quite high THD. Please don't start of non importance of THD for audio.
Huh? What are you talking about? Do you understand the "dominant pole" concept?
I just simulated a 3 pole case, no TMC involved.
It could keep 90dB feedback at 10KHz. As long as you keep 60 degree phase margin, I think you should be fine.
Yes, but you have to set dominant pole at quite low frequences and then distortion increase at higher frequences. Please, try to simulate (by the way I never sow enything of the kind from you), don't just put all others work to rubbish.
I dont build power amp to sell, but for my enjoyment and I don't need unconditionaly stable power amp. By the way I designed power supply with all needed protection, and if I overdo the protection trigers and disconnect the power. Never happened, I don't do disco parties.
That is correct. Myself, I would never trade an unconditional stable amp for a conditional stable with a little less distortions. And BTW, 2nd order systems like TMC and TPC can be made unconditional stable, if designed properly.
You may not enjoy my VHDL-AMS simulations
.
Last edited:
which "condition"?
there are at least 2 that are popular for defining "unconditional stability"
Bode/Nyquist stability as gain devices warm up, their gain sweeping form 0 to normal operating value - important with tubes - not so justifiable with SS
http://www.diyaudio.com/forums/soli...terview-negative-feedback-66.html#post1162866
and stability to "any" load - usually arbitrary passive load with big, high Q C being a problem for many audio amps
some make a fetish of arbitrary C load stability without series L - I don't see the point - use a uH inductor and you're done
there are at least 2 that are popular for defining "unconditional stability"
Bode/Nyquist stability as gain devices warm up, their gain sweeping form 0 to normal operating value - important with tubes - not so justifiable with SS
http://www.diyaudio.com/forums/soli...terview-negative-feedback-66.html#post1162866
and stability to "any" load - usually arbitrary passive load with big, high Q C being a problem for many audio amps
some make a fetish of arbitrary C load stability without series L - I don't see the point - use a uH inductor and you're done
Last edited:
That is correct. Myself, I would never trade an unconditional stable amp for a conditional stable with a little less distortions. And BTW, 2nd order systems like TMC and TPC can be made unconditional stable, if designed properly.
You may not enjoy my VHDL-AMS simulations
Were I can enjoy it?
The fragment below defines a resistor using the ohm law and the thermal noise current:
entity Resistor is
generic (r: REAL);
port (terminal a, b: electrical);
end entity Resistor;
architecture Noise of Resistor is
quantity v across i through a to b;
quantity tinoise : REAL noise 4.0*ambient_temp*boltzmann/r;
begin
assert r /= 0.0;
i == v / r + tinoise;
end architecture Noise;
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.