I've seen many, many amp schematics with no VAS emitter resistor (although at this mention I will be simulating). What I understand of your argument is that with Cbe creating a direct short to ground at HF, this would introduce bad stability characteristics. My thought is that the ESR of the BE junction capacitance has significance here. Is it below 1 ohm or is it around 7 ohms like decoupling caps perhaps (and is this ESR included in BJT models)?
I'm off to bed.
- keantoken
I'm off to bed.
- keantoken
I was thinking only the log non-linearity of this bare junction.
Capacitance never crossed my mind, perhaps it should have?
Anyways, it seems risky to depend entirely upon a long loop
to fix errors that can be managed locally, and without delay.
Not that this is an especially long loop or anything... Just that
one can't easily or cheaply simulate effects of a wiring layout
at 100KHz or whatever. Thats where relying upon the loop BW
as a universal cure-all may disappoint you.
Capacitance never crossed my mind, perhaps it should have?
Anyways, it seems risky to depend entirely upon a long loop
to fix errors that can be managed locally, and without delay.
Not that this is an especially long loop or anything... Just that
one can't easily or cheaply simulate effects of a wiring layout
at 100KHz or whatever. Thats where relying upon the loop BW
as a universal cure-all may disappoint you.
Last edited:
KenPeter,
Good question. I did the sims on this, and with due alteration to the LTP stage current to ensure LTP balance (and zero DC offset) it revealed the following THD20 distortion at 0dB output (1Vp) in 32R load:
With 22R emitter degeneration on the VAS: H2: -83dB H3: -93dB
With NO degeneration, emitter to rail: H2: -85dB H3: -98dB
Degeneration did improve the width of the constant loop gain shelf from 5Khz or so to almost 10KHz. However, 22R degen dropped loop gain at 1KHz from 49dB to 38dB, a considerable drop, severely reducing the effectiveness of global feedback.
0dB Unity loop gain frequency was unchanged at 1MHz, with -94 degrees phase shift, also unchanged.
Conclusion: VAS emitter degeneration does not improve overall THD of this circuit at 20KHz.
Mechanism: Output of the VAS is high Z (common emitter), making it susceptible to voltage sag under load. Adding emitter degen serves only to further increase Zout and decrease loop gain, making the voltage drive to the output stage even more compromised and the correction less effective. In terms of linearity, this is chemo, worse than the disease itself. Best to aim for highest possible loop gain.
Cheers,
Hugh
Good question. I did the sims on this, and with due alteration to the LTP stage current to ensure LTP balance (and zero DC offset) it revealed the following THD20 distortion at 0dB output (1Vp) in 32R load:
With 22R emitter degeneration on the VAS: H2: -83dB H3: -93dB
With NO degeneration, emitter to rail: H2: -85dB H3: -98dB
Degeneration did improve the width of the constant loop gain shelf from 5Khz or so to almost 10KHz. However, 22R degen dropped loop gain at 1KHz from 49dB to 38dB, a considerable drop, severely reducing the effectiveness of global feedback.
0dB Unity loop gain frequency was unchanged at 1MHz, with -94 degrees phase shift, also unchanged.
Conclusion: VAS emitter degeneration does not improve overall THD of this circuit at 20KHz.
Mechanism: Output of the VAS is high Z (common emitter), making it susceptible to voltage sag under load. Adding emitter degen serves only to further increase Zout and decrease loop gain, making the voltage drive to the output stage even more compromised and the correction less effective. In terms of linearity, this is chemo, worse than the disease itself. Best to aim for highest possible loop gain.
Cheers,
Hugh
<Mechanism: Output of the VAS is high Z (common emitter), making it susceptible to voltage sag under load. Adding emitter degen serves only to further increase Zout and decrease loop gain>
Output impedance of VAS with 22R emitter resistor is 7.65 Ohms
Output impedance with VAS with NO emitter resistor is 20.1Ohms
Negative feedback actually lowers output impedance, the increase in THD is not because of the emitter resistor but because of the change in bias voltage.
Nico
Output impedance of VAS with 22R emitter resistor is 7.65 Ohms
Output impedance with VAS with NO emitter resistor is 20.1Ohms
Negative feedback actually lowers output impedance, the increase in THD is not because of the emitter resistor but because of the change in bias voltage.
Nico
I don't think it's the VAS that's affected most by this change. What we're trying to do is give the LTP a linear load so that the error it does produce is still a reproduction of the original signal.
Adding an emitter resistor adds the amount of voltage swing that goes across R6, increasing how much current the LTP needs to give for a given current swing of the VAS. When the LTP works harder, it produces more distortion. The resistor could, if the LTP was more linear, increase stability and/or decrease distortion.
That's what I see. Zout of the VAS depends solely on Early affect and Cbc.
- keantoken
Adding an emitter resistor adds the amount of voltage swing that goes across R6, increasing how much current the LTP needs to give for a given current swing of the VAS. When the LTP works harder, it produces more distortion. The resistor could, if the LTP was more linear, increase stability and/or decrease distortion.
That's what I see. Zout of the VAS depends solely on Early affect and Cbc.
- keantoken
Last edited:
I cannot follow your logic. What you are saying does not make sense. Could you explain a little better?
When one leg of the LTP draws more current, there is a voltage difference between the two bases of the LTP. In order to power the VAS, the LTP must unbalance itself in this way.
The affect of the degeneration resistor is that as more current flows through the VAS, the voltage across this resistor increases, lwhich increases the Vbe of the VAS as "seen" by the LTP. This larger voltage fluctuation at AC causes more AC current to flow through R6, causing the LTP to be more imbalanced, increasing distortion.
Even with NFB, the LTP has to unbalance a bit in order to control the VAS, which produces say 1mV difference between the bases of the LTP. This error difference is added to the output. Since the LTP has a linear transfer curve, then this error voltage will still be a reproduction of the original signal to a degree. If the load (VAS and R6) on the VAS is nonlinear, then it will distort the error voltage as the LTP tries to correct this error, adding this distortion to the output.
We have two competing mechanisms:
1: distortion caused by VAS nonlinearity.
2: distortion caused by LTP nonlinearity.
IF the VAS is more nonlinear than the LTP, then this resistor would decrease distortion. But AKSA's simulation proves the opposite, which is that the LTP is in fact more nonlinear than the VAS, and so because the emitter resistor increases the workload of the LTP, it simply increases distortion.
I hope this helps,
- keantoken
The affect of the degeneration resistor is that as more current flows through the VAS, the voltage across this resistor increases, lwhich increases the Vbe of the VAS as "seen" by the LTP. This larger voltage fluctuation at AC causes more AC current to flow through R6, causing the LTP to be more imbalanced, increasing distortion.
Even with NFB, the LTP has to unbalance a bit in order to control the VAS, which produces say 1mV difference between the bases of the LTP. This error difference is added to the output. Since the LTP has a linear transfer curve, then this error voltage will still be a reproduction of the original signal to a degree. If the load (VAS and R6) on the VAS is nonlinear, then it will distort the error voltage as the LTP tries to correct this error, adding this distortion to the output.
We have two competing mechanisms:
1: distortion caused by VAS nonlinearity.
2: distortion caused by LTP nonlinearity.
IF the VAS is more nonlinear than the LTP, then this resistor would decrease distortion. But AKSA's simulation proves the opposite, which is that the LTP is in fact more nonlinear than the VAS, and so because the emitter resistor increases the workload of the LTP, it simply increases distortion.
I hope this helps,
- keantoken
Nico, KT,
I follow now. Nico means that if you add emitter degeneration you really need to increase the collector load of the input transistor to maintain the same stage current through LTP. If you do not, and keep this resistor 470R, then you must raise the stage current and this affects the operating point of the input stage.
OTOH, I was not aware that inserting emitter degeneration actually decreased the Zout at the collector. This seems a bit odd to me; it is certainly not true with triodes.
Nico has sent me an amended schematic with most impressive results, I will study it and get back to you.
My preference is to keep it simple and not use VAS degeneration, if only to keep parts count low. However, back to the simulator.....
Cheers,
Hugh
I follow now. Nico means that if you add emitter degeneration you really need to increase the collector load of the input transistor to maintain the same stage current through LTP. If you do not, and keep this resistor 470R, then you must raise the stage current and this affects the operating point of the input stage.
OTOH, I was not aware that inserting emitter degeneration actually decreased the Zout at the collector. This seems a bit odd to me; it is certainly not true with triodes.
Nico has sent me an amended schematic with most impressive results, I will study it and get back to you.
My preference is to keep it simple and not use VAS degeneration, if only to keep parts count low. However, back to the simulator.....
Cheers,
Hugh
Hugh, you have it right there. Often we change a single component value and look at the results without considering the effect it has on other components in the chain. It is nice to have an "optimise" feature in a simulator which tests other conditions and displays a range of result from which you can choose from.
Nico
Nico
I am simulating right now. I'm adding different degen values and then re-balancing the LTP. Currently there is a slight distortion decrease at 10 ohms VAS degen. At 2.5V pk-pk output, there is a decrease from .025%THD to .024%. I'll wait for Hugh's next post.
I somewhat agree to making changes but I don't think there's much that can stop us from trying.
- keantoken
I somewhat agree to making changes but I don't think there's much that can stop us from trying.
- keantoken
except that relying on V+ stability after tail R trim to match VAS Vbe bias and keep the ltp currents balanced is asking too much - it not a "stable" (bias) design over anticipated environmental variations
a current mirror load fixes the basic ltp balancing issue, improves loop gain and reduces input cm distortion
current mirror diff pair load may effect "sonics" but I assure you that improves accuracy and "buildablity"
Not at all KT, but consider now that we are pulling teeth, we are looking at fractions of a percent better linearity - where does this lead.
If you stumbled onto something that improves the results by an order of magnitude or more, I think Hugh will put down his pencil immediately and rethink his design.
We are at optimum with what is presented here, not much more one can do with this design. Even adding complexity does not make that much difference at all.
I say we build it.
Nico
If you stumbled onto something that improves the results by an order of magnitude or more, I think Hugh will put down his pencil immediately and rethink his design.
We are at optimum with what is presented here, not much more one can do with this design. Even adding complexity does not make that much difference at all.
I say we build it.
Nico
Hi Hugh,
first thanks for your previous reply to my post!
I was going to ask the same thing as KenPeter...
My question now is did you got the higher order harmonics beyond H2 and H3 from the THD20 analyse you did?
About the THD20 went up when T3 was degenerated, I would have thought the gain-bandwidth product should have stayed fairly the same, no? I would have expected the gain to be fairly the same for any reasonable frequency abvoe 10 kHz independently of the T3 is degenerated or not, 22 Ohm is not a huge value though, could you confirm about that?
Thinking here for a while one of Mr Pass recent white paper "Audio, Distortion and Feedback" (direct link below) comes to my mind where he deals around the single gain stage and how high gain versus low gain but more linearising around each gain stage affect the outcome when everything is wrapped around by the GFB, BTW it's a wonderful piece of paper composed by a great gentleman and I would warmly recommend anyone read this paper.
http://www.passlabs.com/pdf/articles/distortion_and_feedback.pdf
Check out also Mr.Pass other paper "The Sweet Spot"!
Hugh, if you look at that paper and take a look at Fig.10 (by John Linsley-Hood) and Fig.11, these are the reason I am wondering about higher order harmonics, and as the paper deals with linearising every each single gain stage, yes the gain goes down (below the coner frequency for sure!) but the single gain stage also gets more linear and we have less to correct.
My take in general have been for the past years we should linearise each gain stage as much as suitable before applying GFB and am particular happy to see M.r Pass paper which is in congruence to 100% with my philosophy in audio amplifier designing... I mentioned for couple of years ago here that (even if we have very high bandwidth in our amplifier) we have to remember every gain stage is a DELAY (think bas/gate charge/discharge), and what I meant by that is when our audio signal have to go through a lot of gain stages it is "too late" to correct it, it's better to take care of it locally, eg. around each gain stage.
Would you agree here Hugh?
(A bit of topic and speaking a bit from my point of view it's kind of life-philosophy in general but also in audio ampifier designing, when we get older we realise simplicity is the beauty.)
Getting back to your headphone amp I honestly don't think it's much of a problem as we have such a short "audio signal path" (few gain stages), and your simulations indicates we have a good bandwidth and I'm sure we have a pretty darn nice design that will perform very well pleasing many DIY:ers!
Cheers Michael
Last edited:
Hi jcx,
You are absolutely correct in your statements. A CCS would improve the LPT performance and a double current mirror would make it even better, and we might as well go single ended because 180 mA bias is not that much. We could also throw out the bootstrap and replace it with a CCS and yes, the improvements will be easily detected, but then it is no more a simple headphone amplifier and not what Hugh set-out to do.
Earlier in the thread I mentioned that one starts with a specification and design toward achieving it. Here we started with a simple design and worked toward optimising it without adding complexity.
Unlike popular belief there is only one way to skin a cat.
Nico
You are absolutely correct in your statements. A CCS would improve the LPT performance and a double current mirror would make it even better, and we might as well go single ended because 180 mA bias is not that much. We could also throw out the bootstrap and replace it with a CCS and yes, the improvements will be easily detected, but then it is no more a simple headphone amplifier and not what Hugh set-out to do.
Earlier in the thread I mentioned that one starts with a specification and design toward achieving it. Here we started with a simple design and worked toward optimising it without adding complexity.
Unlike popular belief there is only one way to skin a cat.
Nico
Hello Michael
I have read the first Mr.Pass excellent article, and I will read Mr.Pass other paper "The Sweet Spot".
Here is the direct link to "The Sweet Spot" article;
http://www.passlabs.com/pdf/articles/sweet_spot.pdf
Thank for mentioning them.
Bye
Gaetan
Last edited:
I understand Hugh's preference for the R tail source which wasn't the point of my post in this thread
you seem to miss my point on the current mirror diff pair collector load
as it is with the collector R you have a very hard to meet trim condition to balance the diff pair current as a function of VAS Vbe and V+ to signal gnd V supply differences
diff pair current balance = necessary for low distortion given the diff pair tanh curve and low (I would say ineffective emitter degen resistors - but they are handy to measure diff pair current balance if you use 1% R) degeneration
V+ to signal gnd variations cause a large variation in diff pair bias balance as feedback holds VAS bias constant any tail current change is totally steered to one side of diff input pair
I have seen other "simple " Lin schematics with diff pair collector R that you could calculate >2:1 current imbalance in the input Q - at least Hugh provides a trim pot which lets you get it right once - but not necessarily over expected thermal and operating point variations
for the combination of diff pair R tail source and R collector load the V+ to signal gnd V should be a function of VAS Vbe and ib - and the VAS is not operating at constant power
I think this is a real design defect rather than simply a free design choice with little real world consequence – of course amps work with unbalanced diff pair current – but why choose the diff pair and then risk severe imbalance/pwr supply interaction?
re the "linearize each stage 1st" comments - as much as I respect Nelson Pass as a circuit designer I can't agree with some of his feedback paper's conclusions - try reading Cherry, especially Cherry’s “ESTIMATES OF NONLINEAR DISTORTION IN FEEDBACK AMPLIFIERS” JAES V48#4 2000 p299-313 - clearly demonstrates that loop gain linearizes earlier stages' thru reducing the signal level at that stage - a much more powerful linearizing effect than local feedback
also, by establishing and maintaining diff pair current balance the current mirror collector load is linearizing the diff pair locally
Last edited:
The 100u decoupling cap presumably cuts down any PSU interaction in the manner you describe except at DC.
Unfortunately it is hard to identify whether wandering balance or PSU drift are issues because no one has said whether or not this affects the sound negatively.
AFAIK, 2nd harmonics is part of the "AKSA" sound, and so perhaps a perfectly balanced LTP is not so important since it contributes 3rd harmonics.
I agree with Nico, let's just build it and go from there.
- keantoken
Unfortunately it is hard to identify whether wandering balance or PSU drift are issues because no one has said whether or not this affects the sound negatively.
AFAIK, 2nd harmonics is part of the "AKSA" sound, and so perhaps a perfectly balanced LTP is not so important since it contributes 3rd harmonics.
I agree with Nico, let's just build it and go from there.
- keantoken
This would apply to having multiple fast low-gain stages to "build up" gain, rather than having a few slow high-gain stages which are inherently less linear and will, regardless of OLG, take the whole power swing, if I understand correctly. Since one of our design goals is simplicity and another is low parts count, we don't want to add unnecessary complexity.
The problem here is that in order to keep high OLG under control, you have to implement incredible stability measures which cut out feedback at high frequencies. Another design goal is to have as little high-order harmonics as possible, and because of the stability measures these go uncorrected and even amplified. So NFB is a double-sided coin and much of our focus is on increasing OLG while still making our subcircuits inherently linear enough to not produce high order harmonics.
- keantoken