R11/C3 is the pole you want, rolling the gain off above about 22KHz. R14/C7 is compensating for unwanted poles at higher frequencies.
No, we're not trying to compensate individual stages, it's the stability of the amp as a whole that's important. The output stage may have less than unity gain, but it's gain does roll off at some high frequency, and that adds a pole to the response of the amp as a whole. This is the sort of thing that R14/C7 is trying to compensate for. We add a zero in one place to compensate for a pole somewhere else.
Sorry, don't understand the question. Q?
Q's again? I associate Q with the "sharpness" of a tuned circuit or higher order filter, but it doesn't tell you anything that's not on the bode plots.
I build my line stage filters the old fashioned way, from passive components. I use PSpice and other calculations, mostly equations plugged into Excel spreadsheets that I build, to start. I like RF Circuit Design, 2nd edition by Chris Bowick the best for passive filter design equations. You can find a LOT of programs and books on how to design op-amp based filters, but not too many on how to build a 6th order bessel filter from passive components.
The spreadsheet are helpful to make sure the components values you need are available. The teaser for the book is here. You can learn a lot about practical circuit design from this book, that also applies to audio.
Amazon.com: RF Circuit Design, Second Edition (9780750685184): Christopher Bowick, Cheryl Ajluni, John Blyler: Books
For me, PSPice, or similar programs like TINA-TI are great for then simulating how the design works, especially in a complete stage with gain and buffering. TINA-TI is really easy, and easily shows the group delay, which I keep my eye on.
Of course then I build, test it, and listen. Usually the filters test out fine, but need tweaks, as you can't always get exact value components, especially with large through hole inductors.
But I never would use PSPice for THD. Especially class A discrete transistor designs with no global feedback. For that you need to build it, and try a range of active parts. When you do it that way, you better understand what the real world variation will be in trying to build the design.
It's pretty amazing what you can do in a line stage without a ton of gain and then huge loop feedback - if you get to sort parts, and adjust pot values for every copy made. Measures good, and sounds good.
Hi,
Yes. And there again thermal effects also come into play (as much as some here wish to declare thermal distortion to be a pigment of my imagination). Even perfectly matched pairs will not have the same exact temperature with signal...
Past that, I have yet to see "complementary" Transistors or Fet's.
However, as these mismatches do promote mainly even order HD/IMD etc. and I am not a member of the cult of the low THD it does not worry me.
Yes, funny thing. I'm looking at this Tosh pair myself. I had a look at the datasheets and I noticed an over 3dB Difference in Cob, a nearly 8dB difference in early effect driven Rc' and other stuff.
You need to degenerate the heck out of these (which is precisely what I intend to do) to swamp that out...
Ciao T
Yes. And there again thermal effects also come into play (as much as some here wish to declare thermal distortion to be a pigment of my imagination). Even perfectly matched pairs will not have the same exact temperature with signal...
Past that, I have yet to see "complementary" Transistors or Fet's.
However, as these mismatches do promote mainly even order HD/IMD etc. and I am not a member of the cult of the low THD it does not worry me.
Yes, funny thing. I'm looking at this Tosh pair myself. I had a look at the datasheets and I noticed an over 3dB Difference in Cob, a nearly 8dB difference in early effect driven Rc' and other stuff.
You need to degenerate the heck out of these (which is precisely what I intend to do) to swamp that out...
Ciao T
Not the best small signal pair.....
Very high Cob compared to other similar VCE/IC/HFE devices..
Mind you the BC series can be excellent and has been used
as basis for higher voltage japaneses parts.
2SA992/2SC1845 are better but not as good as the 2SA872A/2SC1775A.
Albeit these latter are preferably for low Ic , these are generaly my devices
of choice for input differentials.
The first mentionned have the advantage of a 500mW TDP ,
while it s 300mW for the latters as well as for the ones you did mention.
I've not seen any frequency-dependent difficulties with Spice filter analysis --- perhaps you could provide an example. Although I agree that with relatively simple topologies (Sallen-Key for example) they are easy to design without simulation.
As for CG0/NP0, other than poor volumetric efficiency, and high cost for large values, most have seemed to be quite adequate. I remember reading about one made for the International Space Station. IIRC, 270uF (yes microfarads!), very low inductance, one big mother, about the size of a small loaf of bread. Don't ask the price. It was for a key switching regulator.
I specified an NP0 recently for a non-audio product because of the requirement for a low temperature coefficient of capacitance. The largest one I could readily find in surface mount was 22nF, which worked out to be about right for the application. Unfortunately the tempco of the resonant inductor's loss turned out to be quite peculiar, but I will resist further off-topic excursions
Brad
Not off topic at all boss .......... Continue !!!
Hi,
The 2SA872 has halve the Cob of the 2SA970, which is not a huge difference and both are lower than the parallel Philips/NXP Parts.
But from the curves it looks like the 2SA970 has better open loop linearity (beta, early effect etc.).
The 2SA992 is at 3/4 of the Cob of the 2SA970 and linearity again looks not so horrorshow, but as you say, it has a trifle more Pd, which can help.
Looking at these choices I see no really compelling reason to use these over the 2SA970/2SC2240 pair.
Different strokes for different blokes I guess. I prefer device level linearity, you focus on low Cob.
Ciao T
The 2SA872 has halve the Cob of the 2SA970, which is not a huge difference and both are lower than the parallel Philips/NXP Parts.
But from the curves it looks like the 2SA970 has better open loop linearity (beta, early effect etc.).
The 2SA992 is at 3/4 of the Cob of the 2SA970 and linearity again looks not so horrorshow, but as you say, it has a trifle more Pd, which can help.
Looking at these choices I see no really compelling reason to use these over the 2SA970/2SC2240 pair.
Different strokes for different blokes I guess. I prefer device level linearity, you focus on low Cob.
Ciao T
If you look at carefully you ll see that the 2SA872 HFE/IC curve
start at 10uA while the 2SA970 start at 100uA and in this respect
linearity of the former (low current bjt) is as good from 10uA to 2mA
that the latter from 100uA to 20mA , that is a 200 amplitude in each case..
Second , the 2SA872 cob at 10V has collapsed to 2pF while the 2SA970
has smoother slope and is still at 4.5pF at the same VCE , this potentialy
reduce drasticaly the device Ft..
2SA992 has also good hfe linearity in the same range as 2SA872
but with higher cob at 10V VCE.
In short the 2SA970 is to be used for currents up to 20mA while
the two others do better at currents ten times lower and below.
I am probably over thinking. C3/R11 is a problem in the first place I now understand after the very strong hints most helpfully provided. The f3 being very low, high load on the VAS etc. Between making C7 somewhere around 220p and correcting the gate resistors, I should be able to be stable removing them. I am not quite sure how I would do Miller compensation with all the protection and bias diodes around Q6 and 7. More reading. I am still working the details on C5,6 and 8.
What I was looking at is Q11 sees this pole through the bias spreader, Q10 does not. Seems to me Q11 would see a lower Q. This might effect the symmetry. All filters have Q, so this can effect the overshoot or damping of any load. Should not this be considered within an amplifier?
What I was looking at is Q11 sees this pole through the bias spreader, Q10 does not. Seems to me Q11 would see a lower Q. This might effect the symmetry. All filters have Q, so this can effect the overshoot or damping of any load. Should not this be considered within an amplifier?
Hi,
C3/R11 do two things.
C3 provides likely the dominant pole for the compensation (at 22KHz, which incidentally is very low at al) and R11 limits the DC gain of the circuit.
If you remove R11 all that happens is that you get more feedback at frequencies below 22KHz. This may or may not be desirable.
If you remove C3 you will likely have the whole amplifier oscillating. You can experiment with C3 Values IF you have a 100MHz 'scope and experience tracing oscillation.
If you increase the degeneration of the Input and VAS stage you can scale up R11 to restore the DC Gain, while you can reduce C3 by the same factor. This would for example retain a stable amplifier but increase the slew rate.
There are many subtleties to amplifier compensation.
Also, the bias spreader is normally bypassed with a capacitor, however, what will be the impedance of the bias spreader (guesstimate - 20 to 30 ohm)? I don't think Q per se comes into this.
If you want to be more extreme, connect 43K and 150pF from each side of the bias spreader instead of the single RC.
Ciao T
C3/R11 do two things.
C3 provides likely the dominant pole for the compensation (at 22KHz, which incidentally is very low at al) and R11 limits the DC gain of the circuit.
If you remove R11 all that happens is that you get more feedback at frequencies below 22KHz. This may or may not be desirable.
If you remove C3 you will likely have the whole amplifier oscillating. You can experiment with C3 Values IF you have a 100MHz 'scope and experience tracing oscillation.
If you increase the degeneration of the Input and VAS stage you can scale up R11 to restore the DC Gain, while you can reduce C3 by the same factor. This would for example retain a stable amplifier but increase the slew rate.
There are many subtleties to amplifier compensation.
Also, the bias spreader is normally bypassed with a capacitor, however, what will be the impedance of the bias spreader (guesstimate - 20 to 30 ohm)? I don't think Q per se comes into this.
If you want to be more extreme, connect 43K and 150pF from each side of the bias spreader instead of the single RC.
Ciao T
Should not LF feedback be set by the R14/R13 ratio? R11 just loading the VAS output. I don't quite see how increasing C7 to 330 would not have the same effect. (or more like 220) Is that not the traditional approach? Should C6 come out at that point?
More than subtle for sure.
I was holding off on bypassing the spreader. One step at a time. All the later designs do have a .1 across them. Same with adding emitter resistors to the CM.
Off to work.
More than subtle for sure.
I was holding off on bypassing the spreader. One step at a time. All the later designs do have a .1 across them. Same with adding emitter resistors to the CM.
Off to work.
Hi,
The closed loop DC gain is determined by C4 (unity plus leakage) and the closed loop AF gain is determined by R14/R13.
The open loop gain at DC is determined by the bipolar transistors Re (appx. 26 / Ic in mA) and degeneration resistors R3, R4, R16 and R22 as load.
The open loop bandwidth depends on a lot of complex stuff.
There is a pole/zero combo on the collectors of the input transistors, the zero is at around 3.4MHz, the pole is around 1MHz if I calculate right. Plus there is another zero in the emitter circuit at 7.7MHz and an (invisible) pole at around 6.2MHz.
There is another zero in the VAS emitter degeneration at around 6Khz but combined by an (invisible) pole 16KHz, I suspect this is just fine tuning the phase...
And there is a pole in R11/C3 at 22KHz, which would seem to define the open loop gain bandwidth.
Finally there is another pole in the feedback network at around 2MHz, another round of fine tuning phase I suspect.
Don't take that as the gospel of mark BTW (or any Gospel), Errors and omissions reserved.
I was just amusing myself with looking casually, do perform a more rigorous analysis...
Hard to say. If you want to seriosuly mess with this, have the 100MHz + Bandwidth scope handy.
Try 10 or even 100Khz squarewaves at low levels into small load capacitaces. This tends to reveal lot's of interesting stuff.
Ciao T
The closed loop DC gain is determined by C4 (unity plus leakage) and the closed loop AF gain is determined by R14/R13.
The open loop gain at DC is determined by the bipolar transistors Re (appx. 26 / Ic in mA) and degeneration resistors R3, R4, R16 and R22 as load.
The open loop bandwidth depends on a lot of complex stuff.
There is a pole/zero combo on the collectors of the input transistors, the zero is at around 3.4MHz, the pole is around 1MHz if I calculate right. Plus there is another zero in the emitter circuit at 7.7MHz and an (invisible) pole at around 6.2MHz.
There is another zero in the VAS emitter degeneration at around 6Khz but combined by an (invisible) pole 16KHz, I suspect this is just fine tuning the phase...
And there is a pole in R11/C3 at 22KHz, which would seem to define the open loop gain bandwidth.
Finally there is another pole in the feedback network at around 2MHz, another round of fine tuning phase I suspect.
Don't take that as the gospel of mark BTW (or any Gospel), Errors and omissions reserved.
I was just amusing myself with looking casually, do perform a more rigorous analysis...
Hard to say. If you want to seriosuly mess with this, have the 100MHz + Bandwidth scope handy.
Try 10 or even 100Khz squarewaves at low levels into small load capacitaces. This tends to reveal lot's of interesting stuff.
Ciao T
One also needs to know semiconductor limitations: wide spread of parameters for a device, large temp dependance (heating effects), and parasitic effects.(not to mention parasitics from layout and wiring) All of these can be accounted for in sims (monte carlo, for one), but its not easy.
Too many people these days use sims to learn electronics and skip essential knowledge like semiconductor physics and electromagnetics.
Some other advantages of sims: you can see what perfect devices will do, you can run 1000 amps/volts thru a circuit if you want. It "replaces" thousands of dollars of test equipment. You can try exotic (expensive) devices with out cost. You can easily show open loop gain vs freq. And you never let the magic smoke out of anything. Im sure theres more.
I dont believe sims replace building the circuit, but they can help a lot. And they dont replace knowledge, one needs to know a few things about semis.
+1 to #1657
But it is little different to those who believe their DVM but do not know how it behaves with non DC signals, or believe their diugital scope without understand its aliasing behaviour - or even how to calibrate a probe!
Hopefully, the onset of grey hairs coincides with a healthy scepticism of all "tools" - particularly if they give the right (=expected) answer first time.
But it is little different to those who believe their DVM but do not know how it behaves with non DC signals, or believe their diugital scope without understand its aliasing behaviour - or even how to calibrate a probe!
Hopefully, the onset of grey hairs coincides with a healthy scepticism of all "tools" - particularly if they give the right (=expected) answer first time.
- Status
- Not open for further replies.