lumanauw said:
I hope you noticed all the fan fare around Gilbert's introduction of:
Ic = Is exp(Vbe/VT)
and
gm = Ic/VT
or at Vbe = 800 mV, Ic = 100 mA, and T = about 27 to 30 deg C:
Ic = 3.6E-18 exp(Vbe/26mv)
and
gm = Ic/26mV
You see him use words like heart and soul of the BJT - yes it is a very good model of a transistor. I saw these equations over, and over, and over again in semiconductor technology and design classes. Even in RF design classes when BJTs were used. If you call a device a bipolar junction transistor it better follow those equations. It is the foundation for how we bias them, and drive them, but device capacitances and secondary effects are ignored. Certainly, much more complexity would be used say when designing a complete OP amp, but Gilbert was just looking at one distortion mechanism.
Technically minded people tend to want exact answers, especially when they don't understand something, but often we would get completely bogged down in the analysis if we seek only an exact answer. Often, we have to break a problem down, keeping in mind the simplifications, gain some understanding, then add complexity as needed to get a refined and better solution.
Thiel of Thiel and Small speaker analysis fame, made a few very reasonable approximations that offered an analogy between a speaker as an electro acoustical filter and electronic filter design. It provided completely new insight into the problem, it was revolutionary.
Some current is required to forward bias a BJT base to emitter junction, this is in contrast to a tube or FET which usually only draw very small leakage current from their control element. Assuming a common emitter configuration with the emitter grounded, your not going to move Vbe very far past that 800 mV because Ic increases exponentially and therefore it is a bit like a virtual ground and in this sense draws current. But the collector current is always controlled by the small changes in Vbe following the above equations. We can raise the input impedance through emitter degeneration (Re) which acts as local feedback. Think of the emitter as the neg input and since Ie=Ic+Ib Re produces an in phase signal feeding this negative input.
Yes it is valid, it is how transistors work.
Pete B.
I'll throw in 2 cents here:
We probably should have backed up here to mention the traditional methods for compensating amplifiers for those who've perhaps not seen them before. When we analyze an amp, or are designing one there are a few models that make the analysis easier.
Dominant pole - we know that most multistage amplifiers have many poles due to different effects, each voltage gain stage usually has it's first pole due to internal device Miller capacitance, but there are others usually at higher frequencies - at very high RF even the component lead inductance becomes a factor. The current gain stages can introduce poles as a result of the input impedance changing the load on the earlier voltage gain stages. The simplest approach is to add Cdom capacitance to the stage with the highest gain, usually the VAS, to increase it's input capacitance and push the pole formed between this capacitance and the diff pair output impedance even lower. Dominant pole compensation is pushed, in unity gain stable OP amps for example, to the point where the forward gain reaches 0 dB before we hit the next pole, it's more than we need for a power amp which has more gain.
There's an interesting effect when dominant pole compensation is implemented through Cdom called pole splitting. This single capacitance has the effect of not only lowering the first stage pole, but also raising the second stage pole since it is high frequency negative feedback around that stage. But remember that it results in a highly capacitive input impedance and is very different than local feedback implemented through emitter degeneration. I've been trying to see why Baxandall suggests Miller/Cdom is the best way to compensate an amplifier and see now that this negative feedback improves the linearity of the VAS. I knew about the capacitance - forgot about the linearity improvement. I'm still not convinced but still thinking about it.
This is not the only way to implement dominant pole compensation, stages can be loaded with capacitors from the output to ground to lower the pole resulting from the output impedance and capacitance loading that stage. There's no improvement in stage linearity with this technique, other than perhaps swamping a non-linear load impedance, and this is probably why Baxandall states that Miller is best. I also took Baxandall's "Miller" compensation to be dominant pole with a large cap, but this is not necessarily the case. I see Baxandall's point but I'm not sure it is always the case. I've not thought it all the way through but I wonder about issues specific to power amps such as rail-to-rail signal swings and the non-linear output stage load. In addition, I don't think that he considered the frequency dependent non-linearity of the diff amp/Cdom integrator which causes distortion of the instantaneous phase of a signal as we discussed earlier in this thread. More analysis/measurement is needed to determine how significant this point is.
I think that there's also some confusion with terminology, when I hear Miller/Cdom compensation I have always viewed it as a very low frequency dominant pole as with the 741 type OP amps. I certainly knew that a smaller "Miller" cap could be used, but I'd just call that HF compensation. I think that the term Miller capacitance is usually taken to be a large cap on the VAS, others take it to be any C-B capacitance, I prefer it to be the internal device capacitance, but am fully aware that this is not the convention.
The next advancement in stability analysis is two pole, this is where we have two poles below the 0 dB gain crossing that are spaced far enough apart to provide stability. Pole splitting can be used to do this also.
Next we have the two-pole/zero model. Here we start with three poles below the 0 dB gain point, we push one down but add a zero (resistor in series with the cap to ground provides a zero when it's used on the VAS) that cancels one of the higher poles again providing two below the 0 dB crossing.
I was about to mention that I've never seen pole zero compensation (a series RC network) in the VAS Cdom location, then was looking at the Jensen JE-990 and it is the only design where I've seen it implemented. We've also talked about pole cancellation on the VAS emitter degeneration in the Brystons the 990 has an even more complex form that can be seen in the schematics and is discussed in the AES paper. The JE-990 has a diff amp, VAS, and single stage EF output stage. They use compensation that provides zeroes at 240 kHz, 3.3, 25.4, and 8.1 MHz, and poles at 83 kHz and 5.8 MHz. This is complex compensation for a fairly simple topology and they used computer modeling to get it right. It has 125 dB of LF gain and is unity gain stable, with a unity GB of 10 MHz. Imagine trying to get more complex topologies refined to this degree.
Finally to your question, it is generally advised in feedback amplifiers to have only two voltage gain stages since each one will have one main pole, and it fits, with some coaxing, the two pole model that is known to provide stability when properly implemented. These would be the diff amp and VAS in most designs. The current gain stages add more poles but they're usually at higher frequencies due to the low gain. Look at the effort put into the JE-990 to get it right, imagine one with even more gain stages.
I believe it's better in wide bandwidth power amps to have HF feedback, even some broadband feedback, before the output stage (VAS to diff - input) because the output stage is usually not working very well up above 1 MHz, even 1 MHz is high. Output stage performance at HF (above say 500 k or 1 MHz) is highly dependent on signal amplitude making compensation around it even more difficult perhaps even not practical - consider that the output devices are not RF types. Consider that global feedback is lost when the output clips, or sticks, this cannot be a good thing as the front end can overload. Both the Leach and Otalla amps had stability problems when others tried to build them from schematics, and it's no wonder since the output stages were not laid out using good RF practice. The layout, preferably on a single PC board, is part of the design when providing such wide bandwidth, intended or not.
This was a very quick overview, much more can be found through Google.
lumanauw said:Hi, Terry Demol,
Yes, you are right. I never tought of that
It reminds me of a question I've been holding for so long. I saw this double differential in another commercial amp (attached), also in Kaneda amp.
If I need differential gain of 100X, what is the difference between doing it with a single differential 100X gain, compared with 2 series differential, each have 10X gain? The double differential will add more component, but some does it this way. What's the excel?
We probably should have backed up here to mention the traditional methods for compensating amplifiers for those who've perhaps not seen them before. When we analyze an amp, or are designing one there are a few models that make the analysis easier.
Dominant pole - we know that most multistage amplifiers have many poles due to different effects, each voltage gain stage usually has it's first pole due to internal device Miller capacitance, but there are others usually at higher frequencies - at very high RF even the component lead inductance becomes a factor. The current gain stages can introduce poles as a result of the input impedance changing the load on the earlier voltage gain stages. The simplest approach is to add Cdom capacitance to the stage with the highest gain, usually the VAS, to increase it's input capacitance and push the pole formed between this capacitance and the diff pair output impedance even lower. Dominant pole compensation is pushed, in unity gain stable OP amps for example, to the point where the forward gain reaches 0 dB before we hit the next pole, it's more than we need for a power amp which has more gain.
There's an interesting effect when dominant pole compensation is implemented through Cdom called pole splitting. This single capacitance has the effect of not only lowering the first stage pole, but also raising the second stage pole since it is high frequency negative feedback around that stage. But remember that it results in a highly capacitive input impedance and is very different than local feedback implemented through emitter degeneration. I've been trying to see why Baxandall suggests Miller/Cdom is the best way to compensate an amplifier and see now that this negative feedback improves the linearity of the VAS. I knew about the capacitance - forgot about the linearity improvement. I'm still not convinced but still thinking about it.
This is not the only way to implement dominant pole compensation, stages can be loaded with capacitors from the output to ground to lower the pole resulting from the output impedance and capacitance loading that stage. There's no improvement in stage linearity with this technique, other than perhaps swamping a non-linear load impedance, and this is probably why Baxandall states that Miller is best. I also took Baxandall's "Miller" compensation to be dominant pole with a large cap, but this is not necessarily the case. I see Baxandall's point but I'm not sure it is always the case. I've not thought it all the way through but I wonder about issues specific to power amps such as rail-to-rail signal swings and the non-linear output stage load. In addition, I don't think that he considered the frequency dependent non-linearity of the diff amp/Cdom integrator which causes distortion of the instantaneous phase of a signal as we discussed earlier in this thread. More analysis/measurement is needed to determine how significant this point is.
I think that there's also some confusion with terminology, when I hear Miller/Cdom compensation I have always viewed it as a very low frequency dominant pole as with the 741 type OP amps. I certainly knew that a smaller "Miller" cap could be used, but I'd just call that HF compensation. I think that the term Miller capacitance is usually taken to be a large cap on the VAS, others take it to be any C-B capacitance, I prefer it to be the internal device capacitance, but am fully aware that this is not the convention.
The next advancement in stability analysis is two pole, this is where we have two poles below the 0 dB gain crossing that are spaced far enough apart to provide stability. Pole splitting can be used to do this also.
Next we have the two-pole/zero model. Here we start with three poles below the 0 dB gain point, we push one down but add a zero (resistor in series with the cap to ground provides a zero when it's used on the VAS) that cancels one of the higher poles again providing two below the 0 dB crossing.
I was about to mention that I've never seen pole zero compensation (a series RC network) in the VAS Cdom location, then was looking at the Jensen JE-990 and it is the only design where I've seen it implemented. We've also talked about pole cancellation on the VAS emitter degeneration in the Brystons the 990 has an even more complex form that can be seen in the schematics and is discussed in the AES paper. The JE-990 has a diff amp, VAS, and single stage EF output stage. They use compensation that provides zeroes at 240 kHz, 3.3, 25.4, and 8.1 MHz, and poles at 83 kHz and 5.8 MHz. This is complex compensation for a fairly simple topology and they used computer modeling to get it right. It has 125 dB of LF gain and is unity gain stable, with a unity GB of 10 MHz. Imagine trying to get more complex topologies refined to this degree.
Finally to your question, it is generally advised in feedback amplifiers to have only two voltage gain stages since each one will have one main pole, and it fits, with some coaxing, the two pole model that is known to provide stability when properly implemented. These would be the diff amp and VAS in most designs. The current gain stages add more poles but they're usually at higher frequencies due to the low gain. Look at the effort put into the JE-990 to get it right, imagine one with even more gain stages.
I believe it's better in wide bandwidth power amps to have HF feedback, even some broadband feedback, before the output stage (VAS to diff - input) because the output stage is usually not working very well up above 1 MHz, even 1 MHz is high. Output stage performance at HF (above say 500 k or 1 MHz) is highly dependent on signal amplitude making compensation around it even more difficult perhaps even not practical - consider that the output devices are not RF types. Consider that global feedback is lost when the output clips, or sticks, this cannot be a good thing as the front end can overload. Both the Leach and Otalla amps had stability problems when others tried to build them from schematics, and it's no wonder since the output stages were not laid out using good RF practice. The layout, preferably on a single PC board, is part of the design when providing such wide bandwidth, intended or not.
This was a very quick overview, much more can be found through Google.
Throwing in 2 cents again:
I believe that some people see the term EMF (electro motive force) and panic thinking it is some unknown thing. It is a term that's used with motors, specifically ones that obey the law of reciprocity - voltage/current in produces motion/force and force/motion on the secondary produces voltage and current in the primary. Transformers also follow the law of reciprocity. HOWEVER, the acousto-mechanical circuit elements have electrical analogies, and when these elements are reflected into the primary they look exactly like the analogous electrical components. They are non-linear but electrical elements are also. A speaker is simply a reactive load, no big deal if you design for high current.
Can't measure EMF? We measure EMF when we measure the input impedance of a speaker, it is reactive because of EMF. I don't even like the term EMF, reciprocal network is better. EMF is out dated, from when motors were invented. If you don't believe this, clamp the cone so that it can't move (or de-energize the magnet) the input impedance will become Rdc + Lvc the "motional" components will be gone.
Interesting, that we went through bad sounding amps that could not deliver high current, then had a high current phase in the 70s - 90s. I like those amps. I don't like VI limiting or amps that can't drive low impedances. Now we see, designers skimping on output stages and relying on VI limiting again. They'll probably discover high current again some day. Some very reputable designers suggest designing for 4/8 ohm loads with VI limiting, that mistake has already been made, let's not do it again.
If you worry about back EMF, design for high current and clean recovery from clipping, it's that simple.
Pete B.
lumanauw said:Hi, JCX,
I also interested in this Back EMF issue (like Graham Maynard said). I think this is more important than some think to sonic result of audio power amp (global feedback class AB amp). Is it because this is too difficult to measure, that people are pretending it dont exist/dont important? My simple DIY experiment (including/excluding output stage from feedback loop) shows that this is important, at least I can hear the difference.
See..... it is important
I believe that some people see the term EMF (electro motive force) and panic thinking it is some unknown thing. It is a term that's used with motors, specifically ones that obey the law of reciprocity - voltage/current in produces motion/force and force/motion on the secondary produces voltage and current in the primary. Transformers also follow the law of reciprocity. HOWEVER, the acousto-mechanical circuit elements have electrical analogies, and when these elements are reflected into the primary they look exactly like the analogous electrical components. They are non-linear but electrical elements are also. A speaker is simply a reactive load, no big deal if you design for high current.
Can't measure EMF? We measure EMF when we measure the input impedance of a speaker, it is reactive because of EMF. I don't even like the term EMF, reciprocal network is better. EMF is out dated, from when motors were invented. If you don't believe this, clamp the cone so that it can't move (or de-energize the magnet) the input impedance will become Rdc + Lvc the "motional" components will be gone.
Interesting, that we went through bad sounding amps that could not deliver high current, then had a high current phase in the 70s - 90s. I like those amps. I don't like VI limiting or amps that can't drive low impedances. Now we see, designers skimping on output stages and relying on VI limiting again. They'll probably discover high current again some day. Some very reputable designers suggest designing for 4/8 ohm loads with VI limiting, that mistake has already been made, let's not do it again.
If you worry about back EMF, design for high current and clean recovery from clipping, it's that simple.
Pete B.
Hi, PB2,
Another good article (for me, not too difficult to read) comes from Walt Jung, 4 in series, about audio opamp. He wrote that if one opamp has limited bandwith, and the input is not low-passed, the higher frequencies will drive the input differential to saturation, because (like you said, the opamp will do anything to level the input differential), the feedback will give more and more signal to inverting input due to no effect at output at higer/highest frequencies. The opamp tries to get the "target" high frequency signal magnitude by loading the input differential so much, but the output cannot exist, due to limited opamp bandwith.
Again, back to Doug Self book. Sorry, I repeated it. This book is the "not very difficult" to read, due minimal math equations.
Doug Self also write that Miller cap in VAS is the best method to stabilize an amp. But he also pointed out, that other method existed, like putting C from collector of VAS to ground.
Interestingly, in his references there is TAKAHASHI showed up. Takahashi uses this C load to ground, but with "special" cct that makes the VAS gives enormous current due to large Capacitor value (compared to miller cap) in this place. (Anyone has this Takahashi cct?)
So, maybe it is different soniccally in audio amp to use C miller (B-C) or loading Collector VAS with cap to ground. Some designer try not to use Miller cap.
He also said, that loading the VAS collector with R to ground is not good.
But Walt Jung in opamp series, shows that to widen the bandwith, we can put R from VAS to ground. Is loading VAS collector with R to ground good or bad? Many high end amps loads their VAS collector with R to ground. Why don't make the OL gain lower in the first place?
PB2,
I still don't understand the explenation about making 1stage gain 100X and 2 stages series gain 10X and 10X. Is there something else besides 2 poles explenation? Maybe the total OL bandwith is wider with 2 series stages?
Another good article (for me, not too difficult to read) comes from Walt Jung, 4 in series, about audio opamp. He wrote that if one opamp has limited bandwith, and the input is not low-passed, the higher frequencies will drive the input differential to saturation, because (like you said, the opamp will do anything to level the input differential), the feedback will give more and more signal to inverting input due to no effect at output at higer/highest frequencies. The opamp tries to get the "target" high frequency signal magnitude by loading the input differential so much, but the output cannot exist, due to limited opamp bandwith.
Again, back to Doug Self book. Sorry, I repeated it. This book is the "not very difficult" to read, due minimal math equations.
Doug Self also write that Miller cap in VAS is the best method to stabilize an amp. But he also pointed out, that other method existed, like putting C from collector of VAS to ground.
Interestingly, in his references there is TAKAHASHI showed up. Takahashi uses this C load to ground, but with "special" cct that makes the VAS gives enormous current due to large Capacitor value (compared to miller cap) in this place. (Anyone has this Takahashi cct?)
So, maybe it is different soniccally in audio amp to use C miller (B-C) or loading Collector VAS with cap to ground. Some designer try not to use Miller cap.
He also said, that loading the VAS collector with R to ground is not good.
But Walt Jung in opamp series, shows that to widen the bandwith, we can put R from VAS to ground. Is loading VAS collector with R to ground good or bad? Many high end amps loads their VAS collector with R to ground. Why don't make the OL gain lower in the first place?
PB2,
I still don't understand the explenation about making 1stage gain 100X and 2 stages series gain 10X and 10X. Is there something else besides 2 poles explenation? Maybe the total OL bandwith is wider with 2 series stages?
lumanauw said:
The article you're referring to was discussed at length here http://www.diyaudio.com/forums/showthread.php?postid=419336#post419336 Open loop bandwidth by itself is not meaningful. It's the gain-bandwidth product that's important.
lumanauw said:Hi, PB2,
Waw, I'll have to detail read your explenation
So, Doug Self is right? He wrote that BJT is Voltage driven device (like FETs), and Ib is just side effect. Does the HFE=Ic/Ib is derivative equation, not main equation?
Greetings Lumamauw,
When you ask about a bjt being "driven", are you referring to external drive, I vs. V, or internal operating mechanism, or "cause and effect?"
Externally, bjt's are always, as far as I know, current-driven. Driving the base emitter junction directly with a low impedance voltage source will result in a lot of destroyed transistors. Internally, though, many have pondered the question, "What causes Ic, is it Vbe, or Ib?" The problem here is that this is really a chicken/egg type of issue, an endless circular debate. It is impossible to change the value of Ic without changing both Ib and Vbe simultaneously. Don't ask me which is the "cause" and which is the "effect" because I don't know, and anyone who claims to know is misinformed. Likewise, with a FET, a change in Id requires a simultaneous change in BOTH Ig and Vgs. Externally, FET's are voltage driven, but internally its a chicken/egg issue again. In order to modulate the electric field inside the FET, Vgs must be changed. Since the gate to source possesses a substantial (and non-linear) capacitance, Vgs cannot change unless current is inputted to charge/discharge Cgs. All transistors, FET, and BJT require both current and voltage INTERNALLY to operate because you can never change one of them without changing the other. The action is simultaneous and one does not precede the other. Those who have read my posts on this and any other I-V related issue have heard me refer to this property as *mutual inclusion*.
Anyone who claims that base current, base-emitter voltage, gate current, or gate-source voltage are "effects" or "causes" is mistaken. Ib/Vbe, as well as Ig/Vgs are simultaneous and mutually inclusive, with *no pecking order*. The terms "voltage-controlled" and "current-controlled" or "driven" only describe how a device is *externally* driven. They have nothing whatsoever to do with "cause/effect" or internal operation. All BJT and FET devices are charge-operated (and energy-operated) internally. Base current, Ib is NOT an effect. Ib cannot exist without Vbe, but at the same time, Ib is not caused by Vbe, and vice-versa. Am I getting through. Best regards to all.
PB2 said:
I should probably clarify that global feedback is lost with regard to the clipped part of the waveform. Let's say that the output reaches clipping with a 2V peak input and that there's enough gain so that the differential voltage is nearly zero under normal conditions, then if the input increases to 2.2 V peak (only 10% overdrive) the output cannot increase any further so the diff amp sees 2.2 - 2 V or 200 mV this will then be amplified by the front end as if it was operating open loop and with high OL gain will drive the VAS into clipping. The diff amp will also be deep into overload with a simple undegenerated input pair. Of course the output will simply clip with a good design but recovery may not be so fast when the VAS and output stage saturate. The severity of this problem is a function of diff amp linearization, the ratio of OL to CL gain, and any design features that limit device saturation.
Clipping Induced Overload (CIO) even I can invent marketing terms, TM pending.
I believe that systems run into clipping much more often than people realize especially with uncompressed material.
Pete B.
lumanauw said:
It's helpful to look at two extreme cases. The Gilbert paper covers one extreme case where all the distortion of the OL amp is assumed to be in the input stage, ahead of the integrator. The VAS is assumed to be a distortionless integrator and the output stage is assumed to have no distortion. The OL amp only has AM-to-AM distortion (IOW, it takes a bit higher sinusoidal input signal to produce the desired sinusoidal output signal than if the amp were purely linear, but the OL amp has no signal-dependent phase shift). When the loop is closed, the overall distortion is reduced, but it now consists of some AM-to-AM distortion and some AM-to-PM distortion (PIM). So while you could say that the feedback is "making PIM", a more accurate description might be "in the process of reducing the overall distortion, some of the leftover distortion is PIM (AM-to-PM), and some is AM-to-AM, even though the only type of distortion of the OL amp was AM-to-AM".
Now consider another extreme, that of a perfectly linear input stage gm, and an integrator that distorts. As a quick side note, if you calculate the gain-bandwidth product of the Leach amp from the formula gm/C, where C is the 10 pF compensation capacitor, you end up with a number that's quite a bit higher than the measured value of the amp's gain-bandwidth product. That's because the collector-base capacitance of the VAS, which appears in parallel with the 10 pF C, can't be neglected. Now this collector-base capacitance depends on the collector-base voltage. Remember the earlier discussion about the instantaneous value of the gain-bandwidth product gm/C varying with the signal, producing PIM? In the case of the Gilbert paper, gm was varying with the signal and C was constant. Now, with something like a Leach amp, but with a perfect input stage, gm is constant with the signal level and C is varying with the signal level. This produces PIM too. But this PIM is just distortion of the open-loop amplifier. So the feedback reduces it just like any other type of distortion of the OL amp.
In both these cases, the feedback is still reducing the distortion of the OL amplifier. It's just that in the first case (Gilbert), the left-over distortion has some AM-to-PM components even though the OL amp had only AM-to-AM components. That's not the same thing as "making PIM" where no distortion existed before.
Just to add to the confusion, one might say that in the second case above, the phase shift of the open-loop amplifier doesn't depend on C either, since it's still an integrator which has a phase shift of -90 deg at all frequencies, even if the integration time constant depends on signal level. But in order for the OL amp to work as a real amp to perform the tests, Cordell had to put collector resistors in the VAS to control the gain to the desired value. After doing this it's no longer an integrator of course, and the OL phase shift really does depend on the signal level (and frequency). One might argue that mathematically this is a bit questionable, but in a purely practical sense he's doing exactly what a designer must do to create a workable amplifier without global feedback. IOW it actually reflects the improvement in PIM measurements that could be achieved by taking an amplifier without feedback and making it into a feedback amplifier.
I'm glad this thread has gotten back on track again. There really is a lot that can be said about these papers - more than meets the eye. I hadn't even thought about the issues in the previous paragraph until I tried just now to explain this.
Hi, AndyC,
In your first case, what makes AM to PM from AM to AM when the loop is closed (if the assumption there is no capacitor at all in the whole cct)?
Is the existance of capacitor (wheter interinsic in devices or miller cap) that makes PIM in closed loop amp? I begin to see why JC tries to eliminate capacitor (at anything) inside audio power amp.
Miller cap seems to be able to stabilize amp, but in other side, it makes PIM in closed loop amp. Maybe this is why some designers do not like Miller cap in VAS.
The main problem of distortions is the fluctuation of GM, especially in differential input.
Is there any "trick" to make GM constant, not varying due to voltage or signal voltage or current? Does cascoding that input device (in always constant VCE) helps in anything or not?
I remember you've given one "easy to understand" how PIM happens. Where is that? I read it once, but I forgot where it is.
In your first case, what makes AM to PM from AM to AM when the loop is closed (if the assumption there is no capacitor at all in the whole cct)?
Is the existance of capacitor (wheter interinsic in devices or miller cap) that makes PIM in closed loop amp? I begin to see why JC tries to eliminate capacitor (at anything) inside audio power amp.
Miller cap seems to be able to stabilize amp, but in other side, it makes PIM in closed loop amp. Maybe this is why some designers do not like Miller cap in VAS.
The main problem of distortions is the fluctuation of GM, especially in differential input.
Is there any "trick" to make GM constant, not varying due to voltage or signal voltage or current? Does cascoding that input device (in always constant VCE) helps in anything or not?
I remember you've given one "easy to understand" how PIM happens. Where is that? I read it once, but I forgot where it is.
lumanauw said:
If there is no capacitance anywhere in the circuit (even stray capacitance in the collector-base junction of the VAS) then the circuit has infinite bandwidth. Then the gain-bandwidth product woudl be infinite also. It's not a meaningful condition.
What makes PIM in the closed-loop amp can be thought of in a simplified way as a variation of gain-bandwidth product (gm/Ccomp) with signal level. This can happen if gm varies with signal level (Gilbert case) or if Ccomp varies with signal level. Ccomp can vary with signal level if Ccomp consists only of the stray capacitance from VAS collector to base (as happens in many exotic "audiophile" designs) or if the external compensation cap is of the same order of magnitude as the stray collector-base capacitance (such as the Leach amp). If Ccomp is dominated by a fixed external capacitor and gm is totally constant with signal level (heavy emitter degeneration in the input diff amp) then PIM will be negligible (assuming the output stage is perfect, which it's not). You can take the stray collector-base capacitance of the VAS out of the picture by cascoding it.
No. The main problem with distortion depends on the design being considered. It may be as you say but often it is not.
Let's talk about power amps. Linearizing the input stage is easily done by increasing the emitter degeneration. But if this is done, one must also reduce the compensation cap accordingly, to keep the gain-bandwidth product constant. For a power amp, this keeps the stability margin constant as gm is varied, which is what you want. For a preamp, you have more leeway to make the gain-bandwidth product as large as you want, because you can put a series resistor at the output to make the stability insensitive to capacitive loads.
Look up posts by me with the word "varactor" in them.
About the Leach Amp,
IMHO, the Miller Cap has 2 roles:
-making the amp stable (Nyquist)
-linearizing the VAS. Like Andy_C said, the stray capacitance of the VAS transistor is not linear at all. So, we add in parallel, an external cap, linear and of much higher value. If Ccomp>Ccb , Ccb can be negliged and you end up with a linear compensation cpacitor
In the Leach Amp, the 2nd criteria isn't achieved, is there any reason for that?
IMHO, the Miller Cap has 2 roles:
-making the amp stable (Nyquist)
-linearizing the VAS. Like Andy_C said, the stray capacitance of the VAS transistor is not linear at all. So, we add in parallel, an external cap, linear and of much higher value. If Ccomp>Ccb , Ccb can be negliged and you end up with a linear compensation cpacitor
In the Leach Amp, the 2nd criteria isn't achieved, is there any reason for that?
Bricolo said:
I think he was just trying to get the slew rate of the amp to be as large as he could. He wanted the amp to clip before it slewed with a square wave input, so I guess he couldn't achieve that with a larger capacitor. It seems like cascoding the VAS would fix that, as then the nonlinear Ccb becomes irrelevant and he'd also need a larger fixed cap for the same gain-bandwidth product. But that makes tha amp a lot more complicated.
Also, at the time the Leach amp was designed in the early '70s, I don't think the ideas about VAS linearization had been discovered yet, or at least hadn't been made public. I read about them in the Self book, and I think he references some articles by Baxandall in Wireless World. I'd be interested to know who originally discovered this concept.
Looks like the original article in Audio was Feb '76. http://users.ece.gatech.edu/~mleach/papers/lowtim/feb76feb77articles.pdf
Another possibility is an emitter follower inside the VAS integrator like Self does. Then the compensation cap would go from EF base to CE collector and wouldn't be in parallel with the Ccb of the CE amp anymore.
Another possibility is an emitter follower inside the VAS integrator like Self does. Then the compensation cap would go from EF base to CE collector and wouldn't be in parallel with the Ccb of the CE amp anymore.
76... so newer BJTs certainly could imrpove linearity
an EF before the VAS? I can't see the advantage of this topology. The non linear intrinsec cap is still here, the bandwidth and phase margin don't change much... The only difference I see if that the diff pair is isolated from the VAS's input impedance
an EF before the VAS? I can't see the advantage of this topology. The non linear intrinsec cap is still here, the bandwidth and phase margin don't change much... The only difference I see if that the diff pair is isolated from the VAS's input impedance
Just so we're talking about the same topology, there's a schematic of this at http://www.dself.dsl.pipex.com/ampins/dipa/dipa.htm. It's figure C, section 5.2.2 about a third of the way down the page, where it says "increase in local NFB by adding emitter-follower".
It's not immediately obvious why this works, but if you consider the VAS as a local feedback amplifier and compute its loop gain, the loop gain of the VAS goes up by a lot by having the EF inside the loop. He does measurements that show the improvement in distortion performance also. Simulation shows the distortion reduction too. And Cdom does not appear in parallel with the Ccb of either transistor.
He says the following about the configuration "The function of such an emitter-follower is sometimes described as "buffering the input stage from the VAS" but this is quite wrong; its true function is VAS linearisation by enhancing local NFB through Cdom."
It's not immediately obvious why this works, but if you consider the VAS as a local feedback amplifier and compute its loop gain, the loop gain of the VAS goes up by a lot by having the EF inside the loop. He does measurements that show the improvement in distortion performance also. Simulation shows the distortion reduction too. And Cdom does not appear in parallel with the Ccb of either transistor.
He says the following about the configuration "The function of such an emitter-follower is sometimes described as "buffering the input stage from the VAS" but this is quite wrong; its true function is VAS linearisation by enhancing local NFB through Cdom."
Upupa Epops said:Gentlemen, I see here many theory, but any practical examples -not schematics, but real work of your hands. Exist it ?
Not yet. You're way beyond me in that department. I've built Leach amps, but not my own design yet. I basically quit audio for about 10 years until a few years ago. But I think the theory is fun to think about and discuss. As you can see I'm susceptible to "analysis paralysis"
My actual designs are in military radar receivers. These need low distortion too, since distortion is a "false target" in a radar.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.