Interesting question. When transistors are used for switching they are typically switched between collector current cut-off and saturation, passing through the linear/active region on the way (the goal is to pass through the active region as quickly as possible because this is where the device dissipates most power). There are charge storage effects that mean the Ic does not respond immediately to change in Ib outside the active region. From cut-off a certain amount of charge needs to be injected into the base before the device becomes active. Same with a FET.
As for audio output stages it depends on the design. Typically, output transistors are not going to become saturated (maybe in clipping in some designs) but in a class B they will alternately enter the cut-off region. In a push-pull with adequate bias one of the devices is always in the linear region so there is no delay to the output. But as one transistor enters cut-off its gain drops off rapidly so you get the usual gain non-linearity issues. And in an emitter follower, Ie is the sum of Ib and Ic.
Last edited:
I don't think that is right. Transit time is not a pure delay (unless you only have one electron to play with). It relates to average drift velocity of gazillions of carriers. It is called a "time" because its units are time. Much like the units of RC are time.
In general, it does really matter whether a NFB loop contains a delay as opposed to an inertial phase shift. This is why I caution not to confuse the two.
Hi Bob,
What I noticed was a decrease in Slew-Rate, but the effect decreases with resitor. Also noticed that you use in amp MOSFET Error Correction.
http://www.cordellaudio.com/papers/MOSFET_Power_Amp.pdf
I was surprised of you obtain 300V/us using the filter RC in resistor NFB.
In my project, have the formation of a second pole to -20dB and phase margin ~80° to first pole (dominant), but I not have good result with the filter RC in resistor NFB. What I have used is a filter RC in input amplifier, it improves the stability of capacitive loads and of according to information what have, it is needed to avoid RFI of external environment. I do not know if what I did, was the best option.
BTW: Recently improved the stability of capacitive loads with increasing current in VAS (my compensation is in VAS, I not use as integrator in output of the LTP). I think that stability is much more complicated than just having the phase margin.
Jan,
Short answer: no.
Long answer: Everything below is a first order phenomena discussion.
The main issue with BJT switching is pulling the device from saturation. The excess charge (that is, any charge that was added to the device base beyond what's strictly required to saturate the device) needs to be eliminated from the base. There are essentially two ways to do this:
a) Wait until the charge recombines - this may take an awful amount of time, in particular for device not designed for fast switching as audio power BJT. Note: short recombination times (aka fast switching devices) are always associated with low beta.
b) Help the charge recombine, by injecting carriers of opposite sign - this requires forcing a base current of opposite sign, usually by reverse biasing the BE junction using a cap.
The other option for avoiding the storage time effect is to avoid strong saturation (that is, adding only so much base current to bring the device into what is called "incipient saturation"). This is usually accomplished by a clamp, the most familiar being the Baker breed.
Fortunately, an emitter follower stage never really saturates! That's obvious if you consider the transistor as a two diode device. Even if the diodes have wildly different Is, because the collector is connected at the supply voltage, the BC is never forward biased, so VCE never reaches the large IS BC junction forward drop. This interesting property of an emitter follower was used since the late 60's in the ECL devices.
One to another, in an EF stage, the storage time of the device is not that critical. Again, it is important to distinguish between the large signal parameters (as switching/stored charges) and small signal parameters (like Cob, defining the AC behaviour of the devices). There is no direct relationship between stored charge and e.g. Cob!
This would also help understanding why the output stage "switching" or not has little to nothing to do with the crossover distortions. The real problem with the crossover distortion is the gm variation with Ic. We want to keep the "off" branch biased only to avoid a significant beta drop around cut-off.
finally, an insightfull post from syn08...
just a question : since the devices whose gm is varying are included in the
feedback loop, shouldn t this latter compensate for the varying gm ?...
it seems to me that unless one of the device switch off completly, there can t be any significant high frequency subproducts...
so it appears as when one of the device came from a non conducting state
to enter conduction, this latter is abrupt, thus creating the harmonics
commonly measured in class ab push pulls...
For clarity, that should read:
so VCE never reaches the large IS BC junction low forward drop
This can be quasi-gradual the on device can be "on" at only a few micro-amps. This "what is class A really" has to be one of the more misunderstood concepts in audio. A diamond with no degeneration has theoretically a totally smooth transfer function with neither device turning "off" but plenty of high order harmonics when weakly biased.
Right Scott.
The thing is, wahab, an abrupt gain change is not needed to produce high harmonics, a smooth exponential will do. So even if you have an ideal push-pull of exponential BJTs that never completely turn off, you'll still get high harmonics.
The thing is, wahab, an abrupt gain change is not needed to produce high harmonics, a smooth exponential will do. So even if you have an ideal push-pull of exponential BJTs that never completely turn off, you'll still get high harmonics.
The amount of negative feedback you can apply is limited by delay, which is the non-minimum phase component of the system response, and by the margin that has to be allowed for the variability of both the non-minimum and the minimum phase components with load and operating point variation
Minimum phase response of the usually limiting output stage, for example single pole RC roll-offs, can in principle be equalized out (or compensated by tailoring the feedback transfer function) to the extent they are stable, repeatable
Distributed R-C lossey transmission lines and LC ladder networks formed from parasitic packaging and device R,L,C or geometric layout limitations can add limiting non-minimum phase (~= delay) even when fundamental device physics would allow more speed – power MOSFET polySi gate resistance may be an example
From my readings I still believe that high voltage audio power output BJT have minority carrier base transit delay that limits their phase response – although the fastest types may be approaching packaging and internal distributed RC limits as well
as for Class AB stages I believe "shoot though" increase in output stage current at high frequencies is well documented in inadequate designs
Minimum phase response of the usually limiting output stage, for example single pole RC roll-offs, can in principle be equalized out (or compensated by tailoring the feedback transfer function) to the extent they are stable, repeatable
Distributed R-C lossey transmission lines and LC ladder networks formed from parasitic packaging and device R,L,C or geometric layout limitations can add limiting non-minimum phase (~= delay) even when fundamental device physics would allow more speed – power MOSFET polySi gate resistance may be an example
From my readings I still believe that high voltage audio power output BJT have minority carrier base transit delay that limits their phase response – although the fastest types may be approaching packaging and internal distributed RC limits as well
as for Class AB stages I believe "shoot though" increase in output stage current at high frequencies is well documented in inadequate designs
Last edited:
That reminds me. Another way of telling a time delay from a phase shift is that one cannot equalize out a delay. In theory, a filter can recover a phase shift completely and even create a phase lead. But a time delay is irreversible.
Brian, I am not confusing the two. However, especially if the delay is on the order of more than, say, one-eighth period of the gain crossover frequency, it certainly does matter whether the cause is inertial phase shift or pure delay.
Let me also emphasize that I am not defending those who make wrong arguments about how feedback's operation and effectiveness are affected by delay around the loop. Those who make arguments like "feedback can't work because it is attempting to correct an error after it has happened" are completely wrong.
Feedback generally works as advertized as long as it is adequately stable. Understanding the origins of, and behavior of, excess phase in the loop is helpful in achieving that needed stability. Those who assume that the excess phase originates from essentially one additional pole are fooling themselves.
Cheers,
Bob
The high slew rate in the MOSFET power amplifier results because I am not using conventional Miller compensation. I am using Miller Input Compensation (MIC), where the Miller integrator feedback is returned to the input stage rather than just to the input of the VAS. This breaks the usual rules that govern and limit slew rate. It also increases the dynamic range of the input stage at high frequencies by enclosing it in a local feedback loop. The price paid for this kind of compensation is the need to compensate the Miller compensation loop, but that loop typically can have a much higher gain crossover frequency than the global feedback loop because it does not enclose the output stage.
Cheers,
Bob
I would really like to agree with everything you post but your middle paragraph doesn't read quite right.
If there is a time delay in the loop then indeed NFB cannot correct it because it literally is trying to correct an error after it has happened. You'd need feedforward to fix a delay issue. Phase shift is a different matter because the correction speed is unlimited - output response is theoretically instantaneous even though the phase shift may be very large.
thanks for the insights...
i m not from engineering background, at least not in
electronics ,so it takes me some time so catch some subtilities...
regards, wahab
I disagree. Negative feedback, loop gain, etc... are only concepts and tools that EEs use to analyze realizable circuits. There is absolutely no trouble to integrate a time delay in the base amp transfer function (or feedback network transfer function) as f(t-T)=1/s*exp(-Ts) where f is the Heaviside unit step function, and then use the feedback stuff to determine the closed loop response. The effect of the time delay has nothing to do with negative feedback itself, it exists even if you call the negative feedback circuit topology "reverse signal feeding".
Of course the closed loop response is different to a regular phase shift, and the pole/zero only convenient analysis method will probably fail, but you can still draw trajectories in the phase space, analyze the asymptotic properties and use the stability criteria.
This being said, time delays as above are, for the purpose of audio design, totally negligible. Typical transit times in modern semiconductors are in the pS range. You can estimate yourself an order of magnitude by dividing a typical discrete device geometry (base thickness, channel length, etc...) to the EM field velocity in silicon, c/SQRT(Er). You'll get something like 6pS.
I think you are disagreeing with my use of the word "fix" since I agree with your post. Yes, delays can be accommodated in NFB loop design just like hysteresis and deadbands and so on but the maths gets complicated.
I disagree with the above statement, from at least two perspective.
a) It's not NFB that can't 100% correct a time delay, there's nothing you can do about, as it would violate the causality principle and any synthesized circuit won't be realizable.
b) Though, NFB can correct time delays, but only in a statistical sense, and to a certain limit. These strategies are usually called "predictive algorithms" and they can be pretty effective. They all come at a price and, as usual in EE, there are a number of trades to make.
Last edited: