Ultima Thule said:
Could you ellaborate a bit why you think so?
If we change the VAS BJT transistor to a FET, is it still a transimpedance???
Cheers
Yes, think of it as the (-) input of an OpAmp, it is a virtual ground and input current flows then basically trhough Cdom (be BJT of FET inside).
Higher hFE provided by follower improves OL gain. FET probably not good idea because of lower gain.
Rodolfo
Re: Re: Re: Re: Re: Re: Re: Gain output stage
Janneman
One high open loop output impedance amp , when driving a load with a high phase shift (not resistive ) , can have added to the initial 90º phase shift already present in the frequency compensation , the phase shift caused by the load.
Result, the phase of the feedback , can become even positive ( resulting in oscilation) .
No!...I don't mean that.
In this ...I agree 100%!
janneman said:
Janneman
One high open loop output impedance amp , when driving a load with a high phase shift (not resistive ) , can have added to the initial 90º phase shift already present in the frequency compensation , the phase shift caused by the load.
Result, the phase of the feedback , can become even positive ( resulting in oscilation) .
No!...I don't mean that.
In this ...I agree 100%!
Miller or not, 3 stages or not?
There's an old saying in life, never say never or always. I'm very uncomfortable with blanket statements, there are always exceptions, innovative design sometimes does not follow the "rules". I think that Self just follows Baxandall's theory about Miller compensation being best which I discussed in a previous post. I say it depends on the circuit topology and how large of a Miller cap is used, it depends where you want the poles, it depends on many things. I think a small external Miller cap, not device capacitance, can improve the linearity of the VAS, as long as the diff amp has enough current drive to provide high slew rate and linearity to minimize phase distortion. If this pole is above the audio band then it's non-linearity will not phase modulate audio signals and I like this approach even if the effect is very small. I prefer having a reasonable amount of open loop bandwidth.
You don't like the 2 pole explanation? OK, these are just rough guidelines to help new designers get started, no one wants to stop innovation but we also do not want an amplifier that oscillates. It is hard to get a stable amp and high open loop bandwidth, usually the fewer stages the better. Remember that you want to space the poles far appart and remember that transistors also have a gain bandwidth tradeoff. Just as a rule of thumb if you make both gain stages 10X then assuming similar devices with similar internal Miller capacitance they'll have about the same pole position and your on your way to making a very good oscillator. Let me try to say something positive about the Otala design. Let's think, is there a way to add a gain stage and actually improve the stability of the amp? Perhaps we can add a stage that adds a zero to cancel a pole that's giving us trouble, this new stage will also add a pole but perhaps we can place it above the 0 dB crossing where it will do no harm. A highly degenerated stage, a diff amp for example, pushes the pole up higher, then we can put a cap (or RC network for a zero-pole) across the pair's emitters to cancel an HF pole in the design that's causing trouble. The first stage in Otala's design has 1K of emitter degeneration and a cap across the emitters, it appears that this is what he's doing, but it's probably not an easy approach unless you really know what your doing. I'm guessing here, I've not studied Otala's design very closely. Other positive aspects of the design were using an inverting configuration, a very linear diff amp, and having excellent slew rate.
lumanauw said:
There's an old saying in life, never say never or always. I'm very uncomfortable with blanket statements, there are always exceptions, innovative design sometimes does not follow the "rules". I think that Self just follows Baxandall's theory about Miller compensation being best which I discussed in a previous post. I say it depends on the circuit topology and how large of a Miller cap is used, it depends where you want the poles, it depends on many things. I think a small external Miller cap, not device capacitance, can improve the linearity of the VAS, as long as the diff amp has enough current drive to provide high slew rate and linearity to minimize phase distortion. If this pole is above the audio band then it's non-linearity will not phase modulate audio signals and I like this approach even if the effect is very small. I prefer having a reasonable amount of open loop bandwidth.
I don't know the TAKAHASHI circuit. I believe that the "Miller" cap is associated with a low slew rate, 741 level of poor performance, that is perhaps why many do not use it. I believe that an undegenerated diff pair driving a VAS with a large Miller cap is not the way to go in an ultimate amp. It should be used in moderation to position poles, and perhaps a zero if an R is also used. It's a matter of jugling, slew rate, pole position, and stage linearity. It's complicated, it depends, and this is what good design is about.lumanauw said:
It depends, there is another consideration in power amps - when the output load changes, remember it can go from no load to perhaps a 2 ohm load, the VAS load changes and trying to compensate an output stage where the input impedance varies so much can be difficult when a two pole model is used. A parallel resistor to ground lowers the range of this variation, also a linear resistor there lowers the non-linearity seen by the VAS. It makes the open loop amp better, but reduces the forward path gain. It depends. R to ground mainly lowers the LF gain and only in that sense gives wider bandwidth. Think in terms of gain bandwidth.lumanauw said:
lumanauw said:
You don't like the 2 pole explanation? OK, these are just rough guidelines to help new designers get started, no one wants to stop innovation but we also do not want an amplifier that oscillates. It is hard to get a stable amp and high open loop bandwidth, usually the fewer stages the better. Remember that you want to space the poles far appart and remember that transistors also have a gain bandwidth tradeoff. Just as a rule of thumb if you make both gain stages 10X then assuming similar devices with similar internal Miller capacitance they'll have about the same pole position and your on your way to making a very good oscillator. Let me try to say something positive about the Otala design. Let's think, is there a way to add a gain stage and actually improve the stability of the amp? Perhaps we can add a stage that adds a zero to cancel a pole that's giving us trouble, this new stage will also add a pole but perhaps we can place it above the 0 dB crossing where it will do no harm. A highly degenerated stage, a diff amp for example, pushes the pole up higher, then we can put a cap (or RC network for a zero-pole) across the pair's emitters to cancel an HF pole in the design that's causing trouble. The first stage in Otala's design has 1K of emitter degeneration and a cap across the emitters, it appears that this is what he's doing, but it's probably not an easy approach unless you really know what your doing. I'm guessing here, I've not studied Otala's design very closely. Other positive aspects of the design were using an inverting configuration, a very linear diff amp, and having excellent slew rate.
lumanauw said:
Hi lumanauw,
I've read much of Walt Jung's work, not sure if I caught that one. What your describing is slew rate limiting and TIM. You write: "(like you said, the opamp will do anything to level the input differential)" What I said is true under normal reasonably linear operation, not when it's in slew rate limiting. When the input slews the output does not follow since the diff amp doesn't have enough current drive to charge the Miller cap fast enough, it is in a sense open loop during this time. Since the output can't "keep up" and maintain the differential voltage at zero, the input stage not only slews, but now has enough signal to overload it, assuming an undegenerated diff pair. This is what causes SID/TIM.
About the input filter, it's been shown that most audio signals do not have enough rise time to cause slewing, with even typical OP amps and reasonable signal levels, but this assumes only music signals. We have to consider RFI, LP ticks and pops, synthesized music, leakage of bias or pilot tones, lab test signals etc. and therefore I would always include an input filter that keeps the amp out of slew rate limiting. Here's a link that mentions Peter Baxandall's work examining slew rate requirements:
http://www.angelfire.com/ab3/mjramp/golopid/golopid5.html
There was a claim in the early days that the preamp should limit the bandwidth for the power amp, and while this may work it really doesn't make sense since it requires matching pre amps and power amps. The filter belongs at the input of each amplifier.
Some background about diff amps might help here. Undegenerated diff amps are very useful in communications circuits and we say as a rule of thumb that diff amps are fairly linear when the differential signal is below 26 mV, however by about 78 mV the diff amp makes a fairly good limiter or clipping amp. A two stage diff amp makes a good limiter with 1 mV signals at the input. Three stage diff amps are available as limiter ICs. Limiters are used in FM systems where the information is in the frequency or zero crossings of the signal and not the amplitude. Diff amps are also excellent for linear amplifiers when emitter degeneration is used, or the signal is very small.
andy_c said:
Hi Andy,
I follow Self's logic and agree when Cdom closes the loop, but when I first saw the emitter-follower used there many years ago (I think it was in Wireless World) I don't think there was a Cdom cap in the circuit so it's reasonable to view it as an emitter follower "buffering the input stage". The diff amps stage gain will go up, it will see a more linear load, the emitter follower will reduce distortion by driving the non-linear parasitic Miller capacitance from a lower source impedance. Self's claim of others being wrong is only true when his configuration is assumed. I like the emitter follower idea in both cases.
I've always thought it would be a good way to lower the HF distortion in the Leach. Stability would have to be considered and it would raise the open loop gain when the intention was low open loop gain. Have you considered this?
Do you still have your Leach amps?
PB2 said:
Hi Pete,
I realize that you were originally talking about literally "buffering the VAS" with the EF outside the Miller loop. I didn't intend for the stuff I pasted in from Self's site to refer to your earlier post, but rather just to describe the function of the EF inside the Miller loop in that particular configuration. Sorry if it may have come across otherwise.
But having the EF outside the Miller loop, what is the gain-bandwidth product of that config? Seems like the gain would be about the same or higher, and the bandwidth would be a lot higher than a conventional configuration. I'd have to think about that one. Could the GBW be controlled easily in a similar way to the conventional gm/C?
But having the EF inside the Miller loop might be a good way to improve the Leach if VAS distortion is significant. To tell the truth though, I haven't gone through the design to try to figure out whether the VAS has a significant effect on the overall distortion or not, or if it's mostly crossover distortion that's not covered up by the feedback.
In the late '70s, Leach basically abandoned the low OL gain, wide OL bandwidth approach. The original article in Audio used that approach http://users.ece.gatech.edu/~mleach/papers/lowtim/feb76feb77articles.pdf with the help of R20 and R21 (figure 2). But if you look at the latest version http://users.ece.gatech.edu/~mleach/lowtim/graphics/ckt.pdf, the equivalent of R20 and R21 aren't there. Here's a quote from his site that clarifies the decision a bit:
"Many of the critics of TIM disputed the design criteria that the open-loop bandwidth must be greater than the signal bandwidth to prevent TIM. After my article was published, I came to realize that this is true, provided the open-loop gain and bandwidth are varied in such a way that the product of the two remains constant. If this is done correctly, static distortions such as THD and IM can be reduced while not affecting the stability of the amplifier or its susceptibility to TIM."
He made this change around '78 or '79 I think, and talked about it in class. The latest version has much higher DC OL gain and narrower OL bandwidth than the original.
Nelson Pass never gives comment on this discussion, but his work on SuSy amps (I look at half of it) fits nicely in the requirements. Input stage that can accept high voltage, yes SuSy does, it uses MOSFET. Low gain with high bandwith, yes SuSy does, it uses passive folded cascode, very low OL gain with very wide OL bandwith.
Re: Re: Re: Re: Re: Re: Re: Re: Gain output stage
I think you are talking about the situation before the loop is closed, and then I agree. But, as I see it, when you close the loop, the feedback voltage will be in phase with the input voltage. So the output voltage is in phase with the input voltage, but the output current need not be in phase with the output voltage.
Jan Didden
Tube_Dude said:
I think you are talking about the situation before the loop is closed, and then I agree. But, as I see it, when you close the loop, the feedback voltage will be in phase with the input voltage. So the output voltage is in phase with the input voltage, but the output current need not be in phase with the output voltage.
Jan Didden
As I see it the C miller cap can be used for etither linearizing the VAS or just to set a C dom that will limit the bandwith to make the amplifier stable if that's the issue, but an amplifier which is suffering from unstability could be compensated elsewhere.
Whetere the VAS is linearized further by an additional miller cap we will have either lower bandwtih but also lower gain at higher frequency and thereby a lower part of the unlinearities is taken care by the diff stage so it's a trade how we want to deal with distortion and gain bandwith, and how we want to distribute and take care of the issues.
Anyhow the amount of miller cap added to linearize has it's bandwith dictated by the input impedance established by the input stages collector resistor and the current bias in cct.
So how can a so called "buffer" help up the situation if an additional miller cap is connected from VAS collector to the base of the additional "buffer" transistor which is still giving the very same bandwith.
I guess what we are doing here is actually that we are ISOLATING the very unlinear input capasitor in the VAS transistor and left is only a pure capacitive load loading the diffstages output impedance, and thereby we are getting further linearisation gains.
The buffer transistor don't have such an unlinear capacitive loading on the input stage since it's collector is tied up to a constant voltage point.
Regarding Leach:
Intersting note is that he is increasing the diff stage degeneration a lot with every new amplifier version:
TIM 1 100 Ohm
TIM 2 220 Ohm
Tim 3 270 Ohm
TIM 4 300 Ohm
the diffstages bandwith is decreasing but the diff signal overload capacity is growing more than the bandwith is decreasing I guess and thereby the diff stage is getting less prone to produce TIM anyhow.
This was just a quick noting since I have not extensively studied his articles at the URL given below so I can be wrong in my asumption: http://users.ece.gatech.edu/~mleach/papers/lowtim/feb76feb77articles.pdf
Michael
Whetere the VAS is linearized further by an additional miller cap we will have either lower bandwtih but also lower gain at higher frequency and thereby a lower part of the unlinearities is taken care by the diff stage so it's a trade how we want to deal with distortion and gain bandwith, and how we want to distribute and take care of the issues.
Anyhow the amount of miller cap added to linearize has it's bandwith dictated by the input impedance established by the input stages collector resistor and the current bias in cct.
So how can a so called "buffer" help up the situation if an additional miller cap is connected from VAS collector to the base of the additional "buffer" transistor which is still giving the very same bandwith.
I guess what we are doing here is actually that we are ISOLATING the very unlinear input capasitor in the VAS transistor and left is only a pure capacitive load loading the diffstages output impedance, and thereby we are getting further linearisation gains.
The buffer transistor don't have such an unlinear capacitive loading on the input stage since it's collector is tied up to a constant voltage point.
Regarding Leach:
Intersting note is that he is increasing the diff stage degeneration a lot with every new amplifier version:
TIM 1 100 Ohm
TIM 2 220 Ohm
Tim 3 270 Ohm
TIM 4 300 Ohm
the diffstages bandwith is decreasing but the diff signal overload capacity is growing more than the bandwith is decreasing I guess and thereby the diff stage is getting less prone to produce TIM anyhow.
This was just a quick noting since I have not extensively studied his articles at the URL given below so I can be wrong in my asumption: http://users.ece.gatech.edu/~mleach/papers/lowtim/feb76feb77articles.pdf
Michael
andy_c said:
Hi Andy,
I didn't see it directed at me, just saw the conversation getting back to the topic and wondered about your position. My feeling without detailed analysis is that configurations such as the Leach and Citation 12 could be improved by adding an EF in the Miller loop to raise the open loop gain and up the diff pair degeneration perhaps to 1K. I like the simplicity of the old reliable Citation 12 topology and this simple mod might offer a bit more performance.
I found the article, "Spot-frequency distortion meter" "Measures very low (0.00001%) levels of harmonic distortion." by J. L. Linsley Hood, Wireless World July 1979. I'm not bringing this up to argue, I find the history interesting and it offers food for thought. He uses an OP amp configuration for the oscillator with an undegenerated diff amp, single Darlington device (MPSA-64) VAS, and single Darlington (MPSA-14) emitter follower. It is very simple but does use a few FET current sources.
His simplified block diagram shows no "Miller" cap. He states: "This has a low frequency open-loop gain of the order 200,000 or greater, which allows a substantial measure of loop feedback to be applied and avoids the pitfall demonstrated by Baxandall that low levels of negative feedback may exchange a small measure of non-linearity for a whole host of higher-order distortions."
Baxandall "Audio power amplifier design" Wireless World, December 1978
Anyone have this paper scanned?
Hood also states: "Loop stabilisation is achieved by adding a dominant lag capacitor between the collector and base of Tr4. (Note that it is actually a pole-zero series RC, 470 ohms, 47p) The values shown have proved to prevent squegging in three experimental models of this oscillator, but in two of the three cases a 3 pF capacitor was quite adequate, with consequent improvement in the h.f. open-loop gain and rather lower t.h.d. figures at 10 kHz than those shown in Fig. 10."
Observations such as this are probably what give the impression that less Miller compensation is better. I believe it is true because often more HF open loop gain is needed to linearize sections of the design that are assumed linear in the simplified models used in many of the papers.
It seems to me that slew rate, VAS linearization, and pole position are linked when the Miller loop includes the EF, but not entirely since there are other factors. This will probably often work without a problem when all these requirements can be met simultaneously. However, it seems to me that a Miller RC (not including the EF) should be used to position the pole/zero, emitter degeneration to set VAS local feedback, and a cap from the EF base to ground to set slew rate and make it less dependent on device variation when independent control of these parameters is required. The Miller RC could include the EF in cases where the slew rate requirements can be met. I've never seen the EF not included and perhaps it would be difficult to position the diff amp, EF, and VAS poles.
I remembered what I thought was probably the best or better configuration, in my mind, after reading that article but clearly the Miller loop included the EF.
I believe that the difference that we see in practice is due to that fact that more open loop gain is needed at HF to linearize the portions of the circuit that are assumed linear in the simplified models. We know that an ideal integrator cannot be realized and that output stages are not perfectly linear in practice. It would be interesting to see a model that includes more of these considerations.
jcx suggested looking at Cherry's work in another thread. I've read some of them in the past but will take another look.
Has Cherry's NDFL design been discussed?
PB2 said:
Are you referring to preamp-type circuits or power amps? Should there be a difference in approach? I would argue "yes", for the following reasons. Let's look at load stability and its role in the process.
For preamps, one can put a reasonable series resistor on the output, and it's not likely to have to drive more than a few hundred pF or so (assuming the user has his power amps right next to his speakers and runs long cables from the preamp to the power amp). Let's say we're designing a discrete op-amp circuit for it. Under these conditions, it makes sense for the op-amp to have the highest GBW possible, to get the most feedback at high frequencies for linearization, just as the articles you quoted said. The limit of how high a GBW we can get will probably come from the location of the non-dominant poles inside the op-amp itself, rather than loading considerations. If we can get a GBW of 100 MHz or more, why not go for it? The more the merrier.
But suppose you're designing a power amp, and for best transient response you'd like to eliminate the output inductor. Further, it might be expected to drive electrostatics, with a load capacitance of 2 uF or so. You'll find that stability with the capacitive load will be a big issue. To cure the potential stability problem, there's simply no substitute for reducing the unity loop gain frequency. In fact, having a unity loop gain frequency of a few hundred kHz or so in this case is about all you can do and still have good stability with a good transient response. Since the gain will be fixed at about 20 or so, this essentially fixes the maximum GBW. So for that case, one can look at the problem as "hey, I've got this fixed GBW constraint now, so how can I minimize the distortion given this constraint?". That's a different sort of problem than the one above.
I think we probably agree on these issues much more so than it appears, and it's just a matter of clarifying which type of problem it is we're looking at.
andy_c said:
I'm generally talking about power amps because I find the problem more challenging. Discussion of both is fine. I prefer to have a comfortable margin of GB and I agree with you.
andy_c said:
I came to the same conclusion that a few hundred kilohertz is about right also but more based on output transistor capability. I've also wanted to eliminate the output inductor, there are several examples out there where it's been done.
Yes I do think we generally agree, and believe that it's good that we're looking at both theoretical and practical designs.
Thanks also for the update on the Leach designs in your previous post. I followed Leach's early work more closely but not his later theories.
Upupa Epops said:Gentlemen, I see here many theory, but any practical examples -not schematics, but real work of your hands. Exist it ?
Hi,
Yes, I like working with my hands and also talking about the theory. I have built several amps and speakers over the years, here's what's left of my first, built in 1967 - I got started young:
http://members.aol.com/basconsultants/lil_tiger_tp.jpg
I built several nicer amps as the years passed both kits and completely from scratch. The best amp had a minor failure, long story, but I ended up at college without a power amp. I built this junk amp at my desk, mostly during the first 3 days of orientation at school, my friends found the chassis outside the school Physics lab. The first channel worked right away, second needed some debugging. Here's my ugly 50W/ch amp, it never failed:
http://members.aol.com/basconsultants/WPI_FRNT_S.jpg
It was not practical to bring my 6' tall TL speakers to school so we built 4' X 2' X 16" TLs during the first few weeks of school and they evolved into a 4 way system.
Any pictures of your work?
Just thought I'd add that a speaker's reciprocal nature can also be seen by the fact that they also work as microphones, the voltage produced is the same as back EMF.PB2 said:
PB2 said:
Also wanted to mention that there are clear advantages involving high impedance current source drive for speakers intended for such use. The point above is in the case of driving speakers designed for use with voltage source amplifiers.
Pete B.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.