Hi Rafael,
It is in Chapter 13.8. It is true that many amplifiers still use the capacitor in the shunt leg of the feedback network instead of a DC servo, so this issue remains important. A radial 100uF, 100V NP electrolytic is not too much of a space-eater.
A DC servo is actually competitive in space taken up. This is especially true if the +/-15V or so to run the servo op amp is already available.
Cheers,
Bob
Hi Waly,
Good point about the Zener current. I should have set it higher, but basically just started with the simple feedback current mirror in coming up with the drawing. Thanks for pointing this out.
The bottom line here is that we must run the same current into the feedback transistor as we would have had to run through the Zener diode in the Zener-referenced current source if we want the same precision.
It is notable that a 2-terminal IC voltage reference can be used in place of the Zener diode. The 2.5V LM336 voltage reference, for example, is rated to operate at bias currents as low as 400uA. Using a 2.5V reference also does not force us to use up as much voltage headroom.
Guess I should have used the LM336 in my schematic instead
.Cheers,
Bob
This is an interesting point about H2 being produced in the VAS and possibly adding to, or interacting with, any H2 produced in the LTP. Your observation that the signal current in the LTP will be 90 degrees out of phase with the VAS output voltage in an ordinary Miller-compensated amplifier is an astute one.
I think, ultimately, the question is to evaluate/discuss the H2 distortion of the LTP in isolation (no NFB) and go from there. And of course this should be done with and without degeneration. For me, of course, the case with degeneration is the only one I care about, since I never use a BJT LTP without degeneration. My favorite number for amount of degeneration is 10:1.
Cheers,
Bob
Cherry, Edward M "Estimates of Nonlinear Distortion in Feedback Amplifiers" JAES Volume 48 Issue 4 pp. 299-313; April 2000
of course deep nulling of 2nd only applies within the diff pair - I keep pointing the Cherry article for those who want a possible tool to look into amplifier stage by stage distortion contribution
of course deep nulling of 2nd only applies within the diff pair - I keep pointing the Cherry article for those who want a possible tool to look into amplifier stage by stage distortion contribution
Good paper.
Of course, in the context here, "deep nulling" is not necessarily what we want to get a handle on. We only need the LTP balance to be good enough to not matter over the reasonable operating range of the LTP. For example, we would like the balance to be sufficient that the H2 is, say, less than 1/4 the H3 of the LTP for all operating signal amplitudes from 1/10 to full amplitude of the LTP. Something like that. Just thinking out loud.
A really deep null where the H2 is, say, 1/100 of H3 is not necessary.
Cheers,
Bob
In my post before that I talked about the effects of just nonlinear Vbe on the LTP harmonics, which applies both with an without degeneration, and also how it applies in the context of an amplifier. That was a start on the analysis you describe, but it does miss out on Early effect, Hfe nonlinearity, and thermal effects.
Like in the case of the Wilson mirror, we can greatly diminish Early effect using a tracking cascode. One of my favorite illustrations of this is the OPA627 internal schematic, but the usual tracking cascode should do as well. As for Ib, using the BC560C/550C and having a low feedback and input impedance, which some say is a necessity for noise anyway, helps. So my question is, when all these are heeded, is a advantageous H2 null still possible?
As in, can we get the H2 of the LTP contribution itself to be <H3, AND can this possibly be a significant improvement in total output distortion?
What is being discussed seems suggest that the LTP needs to be deliberately imbalanced in order to cancel the H2 contribution of secondary effects like Early effect, Ib nonlinearity and so on. Unless every one of these factors is meticulously taken into account and SOMEHOW controlled within component tolerances and relative phase, environmental conditions, and dynamic loading, you will only get a deep null within a small region of the audio band. We don't know if this would even sound good, and why strive for a distortion null at ONLY 1.85KHz? Perhaps if the design is basically finished and you want to tune the sound - but for better or worse?
We can only define a crude margin for LTP balance, as the point where we know for sure that the H2 of the LTP will dominate output distortion.
I think we can define such a margin based on the THD figure of the amp. Amps with low THD, provided low Nfb impedance, will benefit the least from LTP balance. The higher the THD, the more benefit can be had with LTP matching, I think. For instance the Optimos in my simulations couldn't be helped by more than 5% matching. An amp with 10 times the THD would probably want .5% matching. So here's my idea. .005/THDmax=min%Icmatching optimal for performance.
So if your amp has .01% max THD, you want at least .5% matching. But if your amp has .001% THD, you need no more than 5% matching. Maybe you guys can open up simulations of multiple amps and if the relationship applies, we can converge on the most reasonable proportion?
Like in the case of the Wilson mirror, we can greatly diminish Early effect using a tracking cascode. One of my favorite illustrations of this is the OPA627 internal schematic, but the usual tracking cascode should do as well. As for Ib, using the BC560C/550C and having a low feedback and input impedance, which some say is a necessity for noise anyway, helps. So my question is, when all these are heeded, is a advantageous H2 null still possible?
As in, can we get the H2 of the LTP contribution itself to be <H3, AND can this possibly be a significant improvement in total output distortion?
What is being discussed seems suggest that the LTP needs to be deliberately imbalanced in order to cancel the H2 contribution of secondary effects like Early effect, Ib nonlinearity and so on. Unless every one of these factors is meticulously taken into account and SOMEHOW controlled within component tolerances and relative phase, environmental conditions, and dynamic loading, you will only get a deep null within a small region of the audio band. We don't know if this would even sound good, and why strive for a distortion null at ONLY 1.85KHz? Perhaps if the design is basically finished and you want to tune the sound - but for better or worse?
We can only define a crude margin for LTP balance, as the point where we know for sure that the H2 of the LTP will dominate output distortion.
I think we can define such a margin based on the THD figure of the amp. Amps with low THD, provided low Nfb impedance, will benefit the least from LTP balance. The higher the THD, the more benefit can be had with LTP matching, I think. For instance the Optimos in my simulations couldn't be helped by more than 5% matching. An amp with 10 times the THD would probably want .5% matching. So here's my idea. .005/THDmax=min%Icmatching optimal for performance.
So if your amp has .01% max THD, you want at least .5% matching. But if your amp has .001% THD, you need no more than 5% matching. Maybe you guys can open up simulations of multiple amps and if the relationship applies, we can converge on the most reasonable proportion?
Last edited:
Let me clarify my reasoning for this "rule of thumb".
The signal at the VAS input, with its own fundamental and harmonics, is a reflection of the THD of the output stage. Because output stages generally give little in the way of diminishing distortion, so too is the signal at the VAS input going to have a similar THD regardless of amp topology. How well this signal is attenuated, by buffering throughout the amp, or rejected by I/V conversion, then will more accurately predict output THD. Degeneration is considered anti-buffering. Because the signal to the LTP will regardless of topology have a similar THD, differing only in total signal magnitude, the linearity of the amp itself and the loading of the LTP will be inversely proportional. For this reason, the optimal minimum LTP matching, being proportional to LTP loading, will be roughly inversely proportional to THD. In this proportion we use max THD, because max THD is always relative to output power, and max output power is inversely proportional to output stage distortion, although the proportion changes somewhat between OPS types.
The signal at the VAS input, with its own fundamental and harmonics, is a reflection of the THD of the output stage. Because output stages generally give little in the way of diminishing distortion, so too is the signal at the VAS input going to have a similar THD regardless of amp topology. How well this signal is attenuated, by buffering throughout the amp, or rejected by I/V conversion, then will more accurately predict output THD. Degeneration is considered anti-buffering. Because the signal to the LTP will regardless of topology have a similar THD, differing only in total signal magnitude, the linearity of the amp itself and the loading of the LTP will be inversely proportional. For this reason, the optimal minimum LTP matching, being proportional to LTP loading, will be roughly inversely proportional to THD. In this proportion we use max THD, because max THD is always relative to output power, and max output power is inversely proportional to output stage distortion, although the proportion changes somewhat between OPS types.
Not only in simulations but also in real life.
We can easily extrapolate from the usual zeners curves
envelloppe evolution that 0.5mA would still decently work.
http://www.vishay.com/docs/85608/85608.pdf
Isn't "Nonlinear Distortion" a tautology or a pleonasm?
Hi Mike,
Actually, sometimes people refer to linear distortion, so I guess that is why nonlinear distortion may not be redundant. For example, some may refer to phase distortion as a linear distortion. I always avoid using the term distortion in connection with a linear process.
Cheers,
Bob
But a nonlinear transfer curve is precisely what gives rise to distortion; a perfectly linear transfer characteristic does not. So in my view "nonlinear distortion" is a tautology.
"linear distortion" essentially means a deviation from flat frequency response - termed a "distortion" because if the replay chain does not have flat frequency response, the program material has been altered from what was intended. In other words, the envelope of the signal in the frequency domain has changed shape - i.e. has been distorted.
Hi Mike,
The issue is that some people refer to linear distortion. This is a distortion that does not involve a nonlinearity. For example, some refer to phase distortion in an otherwise linear system. I personally think the use of the word distortion in connection with linear systems is dangerously confusing, and I do not endorse it.
BTW, the phase distortion example I cite is not to be confused with "Phase Intermodulation Distortion" (PIM) which indeed does involve a classic nonlinearity and is a nonlinear distortion.
Cheers,
Bob
When you mention the tracking cascode, are you referring to what I called the driven cascode in my book in Chapter 7.5 and Figure 7.17?
My experience has been that this does indeed make a difference. When I applied it to my MOSFET power amplifier with error correction I did see some distortion reduction, even though the distortion had been well under 0.001% at 20kHz.
My favorite way to implement the driven cascode is to create a replica of the feedback signal and apply that to the cascode bases. This avoids the need to mess around with the emitter circuit of the LTP.
Cheers,
Bob
7.17 is the idea of a tracking cascode. However I would reference the cascode to ground instead of the rail, since one of the purposes of an LTP cascode is to improve PSRR. I would bias an LED or zener above ground, then reference the bootstrap to the voltage reference.
WRT the emitter junction of the LTP, there seems to be a general dislike of a resistor in series with the CCS. I want to point out that this improves RF rejection of the CCS, and it also improves the CCS overload behavior, and if the CCS slew limits, it will prevent dramatic contortions of LTP emitter current. Furthermore, it allows to use a smaller CCS transistor, which also results in better performance.
WRT the emitter junction of the LTP, there seems to be a general dislike of a resistor in series with the CCS. I want to point out that this improves RF rejection of the CCS, and it also improves the CCS overload behavior, and if the CCS slew limits, it will prevent dramatic contortions of LTP emitter current. Furthermore, it allows to use a smaller CCS transistor, which also results in better performance.
Hi keantoken,
That figure is a bit simplified. The low-voltage power supply to which the cascode bases are referenced is assumed to be quiet and firmly referenced to ground, and is basically the same supply that would be used if the cascode was not driven. The feedback resistor of the replica feedback circuit is much larger than the shunt resistor of the replica circuit (just like the relationships in the real feedback network), so the direct coupling merely drops the quiescent voltage on the cascode bases a small amount.
I have not heard of a dislike for the resistor in series with the LTP tail, but I always use one.
Cheers,
Bob
Well, I have encountered a dislike of that resistor a few times, and saw no defendants, but maybe it's just bad timing on my part. It can make a big difference in startup behavior, since the LTP current will track the rail voltage until the CCS has adequate collector voltage - so someone might come to anomalous conclusions based on how it affects their specific design.
Hi,
Resistor in series with the LTP tail :
http://www.diyaudio.com/forums/soli...-self-wants-your-opinions-13.html#post3418023
Resistor in series with the LTP tail :
http://www.diyaudio.com/forums/soli...-self-wants-your-opinions-13.html#post3418023