Member
Joined 2009
Paid Member
if matching is unimportant the I'm wondering why structure it as an LTP instead of running the emitters of both devices independently to the bottom rail ?
I'm still struggling with this.. I may be too stuoopid to get there quickly. The top LTP still forms an error amplifier, it has signal input and feedback signal input - I thought an LTP requires good balance for low distortion when working like this ?
Schottky diode law (something close to square law) AB + 150mA.
I am doubtful the reality follows perfect "law", but not important...
Whats important is that the crossing glitch is smooth and gentle
enough for the main loop's gain bandwith to finish correcting the
problem. Without adding a mess of high order harmonics and IMD.
Also eases symmetry requirements for the LTP that will otherwise
have to sense these crossing errors.
You want linear class A, simply replace sense diodes with resistors.
Still be far less quiescent than old JLH with a blind current source.
I take it all the way to AB with diodes, only to show how the SRPP
current control is versatile enough to make any shape inbetween.
Any offset of the unmatched LTP is copied at only unity gain to
the output. A capacitor in the feedback loop sabotages gain for
the lowest of low frequencies. That DC offset and rumble is not
otherwise amplified by the full closed loop gain. All non-inverting
amplifiers can use this trick...
I may not have drawn that capacitor into every version of my
schematic, but you can assume it needs to be there...
Last edited:
Member
Joined 2009
Paid Member
still digesting...
Kenpeter - I tried simulation of the output, without gnf, with and without schottky and although the curves changed the overall distortion wasn't necessarily better ?
Elvee - sorry if you said already, but can you explain more about what is the function of R14 attached to Q12 in your schematic (I didn't understand your earlier explanation) ? what would have to be adjusted to maintain proper operation if it were replaced with zero ohm ?
Kenpeter - I tried simulation of the output, without gnf, with and without schottky and although the curves changed the overall distortion wasn't necessarily better ?
Elvee - sorry if you said already, but can you explain more about what is the function of R14 attached to Q12 in your schematic (I didn't understand your earlier explanation) ? what would have to be adjusted to maintain proper operation if it were replaced with zero ohm ?
Last edited:
Member
Joined 2009
Paid Member
Hi Kenpeter, sorry, my mistake what I should have siad is that I was simulating a complementary EF output and I swapped out the emitter resistors for schottky diodes. I did see the smooth cross over curves you predicted when looking at the current flow through the outputs but I haven't tried this with the JLH.
It could also work like that, but there are reasons in favour of the LTP.
Here are the most important:
-Common mode. Not exactly the usual one like CMRR, but the fact it doesn't convert common-mode (vertical path, ancillary control loop for current servo) into differential-mode (horizontal path, noble signal loop), but, and this is important, it transmits the common-mode to control the quiescent current.
-Symmetry: the current from one emitter has to come from the other one, and this forces a symmetry in the output, even if the devices are different.
The current flowing in the input transistors equals 1.2/R6 (or R7).
Mismatches in lower LTP will only add or subtract millivolts to the 1.2V potential, and mismatches in the input LTP will only force an output offset sufficient to rebalance the currents in the input transistors.
QED: mismatches in input or VAS transistors have no first order effect on linearity.
It is probably simpler to look at what would happen without R14.
If the supply voltage changes, the current in R21 changes accordingly.
Connected as a simple diode, like Q7, transistor Q12 has a dynamic resistance ~0.026/Ie, and this means that some of the supply voltage variations will be fed to Q13, the first transistor of the current servo amplifier.
This is undesirable, as it will give a poor PSRR, and create parasitic modulation effects.
By adding a B-C resistor having the same value as the dynamic resistance of the transistor, this effect cancelled for a certain range of supply voltage.
An equivalent result could be achieved by using a CCS instead of R21, and it would also make constant the load current of Q2, but it is more complicated and unnecessary:
Variations at the collector of Q2 are simply perturbations inside the current-servo loop, and are mostly rejected thanks to the loop gain.
At the level of Q12/Q13, the input signal itself is corrupted, and that's outside the loop.
For your unity gain buffer here are already two possibilities: first Unigabuf:
http://www.diyaudio.com/forums/solid-state/185501-unigabuf-follower-cut-out-leader.html
Note that a prototype has actually been built and tested, it isn't pure simulation.
And here is another idea, for a symmetrical buffer:
Attachments
Member
Joined 2009
Paid Member
Hi Elvee,
thanks for all the details of the UFO ! - I will need to study further.
Back to the circlophone - my current and incomplete understanding of this topology is a little different from your explanation of it and if correct I'm afraid that you are getting IM products. It makes me wonder if this is what you were seeing in the simulations when you commented that you don't think the sims are working properly at higher frequencies. I can't confirm this at home (iMac doesn't run LTSpice), I will have to wait til I have access to a PC during the week. What I would expect to see is 3rd harmonic and above being slightly accenuated, depending on the effectiveness of the nfb, and the sound being slightly coloured,.
thanks for all the details of the UFO ! - I will need to study further.
Back to the circlophone - my current and incomplete understanding of this topology is a little different from your explanation of it and if correct I'm afraid that you are getting IM products. It makes me wonder if this is what you were seeing in the simulations when you commented that you don't think the sims are working properly at higher frequencies. I can't confirm this at home (iMac doesn't run LTSpice), I will have to wait til I have access to a PC during the week. What I would expect to see is 3rd harmonic and above being slightly accenuated, depending on the effectiveness of the nfb, and the sound being slightly coloured,.
Last edited:
No, what I mean is that there are big discrepancies between simulated and actual, measured parameters, like power bandwidth.
IM can be produced by "simple" non-linearity mechanisms, just like harmonic distortion, and by more complex, dynamic interactions.
In the case of the Circlophone, the THD figure is good, nothing to see here, and the transient behaviour is spotless, as the oscillograms given at the beginning demonstrate.
Can you explain exactly why you think the topology could generate excess IM?
Since this is a balanced, symmetrical design, odd order harmonics are predominant.
In an ideal world, there would only be odd harmonics, nothing wrong there.
Now, whether you like it or not is another story: some prefer SE amplifiers for that very reason, while others go in the opposite direction with super-symmetrical amplifiers.
But all the harmonics even or odd, have their amplitudes nicely decreasing with increasing order, thus quite healthy and normal.
Member
Joined 2009
Paid Member
The LTP has a good CMRR, but only to signals presented at the differential inputs. An LTP doesn't have great PSRR and so benefits from a CCS. Note the first "C" in CCS.
You are modulating the current flow through the LTP. This turns it into a simple analogue multiplier.
The sub-circuit that measures current flow through the output devices produces a signal that is predominantly twice the frequency of the input signal and so you are modulating the LTP with a current that varies at twice the signal frequency. The differential input will be modulated by it to produce the common mode output that does what you want in terms of the output stage current, but it also outputs sum and difference side-bands. The upper side band will be at three times the signal frequency, hence 3rd harmonic.
For complex music there are many different frequencies and the modulated LTP will turn them into multiple IM products. Now you may have only low levels of IM products being generated and the gnf network will reduce them but I believe they will be there nonetheless.
It is a shunt regulated current source. And not modulated at random.
The modulation inherently includes correction for CM and PSRR. 2nd
order harmonics so created are complimentary, and cancel at output.
A highly regulated complimentariness, with a high cancellation....
In the output, I see 2nd and 3rd down -70db, with hardly anything of
high order splatter....
---------
I have seen quadrature modulators do some amazing things for the
cancellation of undesired sidebands and/or even the fundamental...
I am not certain what was fed to I and Q to make that happen, but
was controlled by a feedback of some sort... I don't think we have
even one full Gilbert cell going on here, and you would need at least
two to play those tricks. So maybe what I've seen was irrelevant.
But I do think it is meaningful that any modulation may occur in
the LTP and/or SRPP is at 90 degree angle to the fundamental.
Not mixing random things at random angles, and with both eyes
always on the result.
--------
I don't care much for EF + Schottky, as the emitter drops are still
dominant in the equation, and a blind thermal variable. If you are
using Schottky drops as an active feedback to the spreader, now
EF or anything else is fine because the hot devices are no longer
blindly operating upon their own curve.
Last edited:
Member
Joined 2009
Paid Member
It matters not whether there is a much of a correlation between the input signal and the signal doing the modulating - it'll still mix 'em. I realize we're not looking at a full Gilbert cell here, just a simple dual-quadrature mixer but the math tells you what the outcome should be. Gilbert Cells
Single sine waves are one way to probe the workings of such mixers but a music signal will result in a much more complex interaction.
The output passes on only the differential to the load so the common-mode modulation frequency isn't there - but the sidebands will be there as far as I can tell. Anyhow, it may not matter, the IM products may be low enough at low power and under the control of the NFB that it's a non-issue or it may even enhance the sound. There may be a sweet spot in terms of how it's set up regarding idle currents.
Single sine waves are one way to probe the workings of such mixers but a music signal will result in a much more complex interaction.
The output passes on only the differential to the load so the common-mode modulation frequency isn't there - but the sidebands will be there as far as I can tell. Anyhow, it may not matter, the IM products may be low enough at low power and under the control of the NFB that it's a non-issue or it may even enhance the sound. There may be a sweet spot in terms of how it's set up regarding idle currents.
You are right from a general perspective, but here, the tail current modulation is essential to the performances of the Circlophone.
It works in multiple and subtle ways.
If you look at the performances of the "raw" amplifier, you can expect the main distortion product, H3 to be 70dB down.
Normal for a three stage amplifier without gain enhancement or fancy techniques, using crappy output transistors.
When the tail modulation is present (~6%pp at the full power output), it will add a further -71dB of H3.
Thus, will the resulting distortion be -64 to -65dB H3?
The modulation current is in fact an error signal from the output. It tells the previous stages to increase the conductance of the output stage.
This is the closed-loop part.
But this signal can also serve as a "helper" in an open-loop, feedforward fashion, and it has the right phase to compensate, at least partially the compression, gain droop etc of the "noble" path, including the small and medium-signal part: the transfer of a differential pair is not a linear function, but a tan h function.
In the end, the result of the compensation will be a composite H3 level of ~-90dB.
Other harmonic and IM products will follow the same trend, because the compensation doesn't work in the frequency domain, it globally straightens the transfer function of the amplifier in the amplitude domain.
The compensation is not as accurate and as deterministic than in the Unigabuf f.e., but it does provide valuable improvements.
Another important aspect to understand is the way the modulation signal affects the functionning of the amplifier, in particular its class.
Let us establish the convention that in order to let +1A of current flow through the load, the upper device must provide +1A, and for -1A output, the lower device has to deliver +1A.
In a class B push-pull, when the current to the load is +1A, the current is provided by the upper transistor, and the lower one is off.
In fact, it is "more" than off: it provides a virtual current of -1A. Or at least, it is biased that way by the driver and the preceding stages. But because it is a unidirectionnal device, it is simply off.
Not let us see what happens in the Circlophone under the same conditions.
The real time modulation of the current will make sure the lower device passes a minimum of 100mA; the upper transistor will have to provide 1.1A, and the resulting current to the load is again 1A.
But the instantaneous quiescent current is now 0.6A.
This is class A. That is why I said the circuit is a quasi-class A: it is an adaptative class A, in fact it never works in class B or class AB, it just gives that impression.
But it has the attributes of class A, in particular the "grip" both output devices have at all times on the load: it never works in a pure source or sink, open collector mode.
There is one more side benefit to this mode of operation: the excursion seen by the amplification chain is not comprised between A- and A+, total 2Ap.p. (clipped at the output), but |0 to A|, amplitude Ap.p. with the supplement provided "parametrically" by the modulation (pump) current that dynamically shifts the operating point.
The raw linearity improvement brought by the modulation current is ~20dB, but there is more to it than a mere improvement in figures.
Oops, I just realized odysseyB had made the conversion.
Here they are again:
The silkscreen file is too large for the forum:
http://dl.dropbox.com/u/5430178/cir...ix_3 - Bottom Copper_Drill_Top Silkscreen.pdf
Also, remember to use the schematic for the board, as the location #'s are different.
Member
Joined 2009
Paid Member
I understand that it's integral to the workings of the amp - and I see it as a very elegant method to achieve a Sliding Bias arrangement. I wasn't sure that the consequence of the arrangement had been considered or to what extent it might add some distortion and I'm still learning how your circuit works (I can be a bit slow!) - so the conclusion on IM seems that this effect may be benign within the expected performance range of this amplifier. Good !
I don't understand this bit and it sounds rather interesting - you are saying that the common mode loop alleviates gain droop - is this because it maintains an always-on conduction path and so maintains a more stable transconductance ? or are you saying that the common mode feedback contains some distortion produced by the tan h LTP function which it is able to partially compensate for in some way ??
I can see that current requirements are different, although usually higher current means more linearity - but I don't quite understand why the raw linearity is improved by 20dB in operating with sliding bias instead of full class A ? and what is 'more to it' ?
Hope you don't mind my curiosity !
To state the things in a simpler way:
The current servo has to react when the current drive is insufficient to keep both output devices simultaneoulsy active. This is partly caused by the transconductance of the (relatively) inactive device being minimal, gain compression in the active device, and tan h distortion in the driver LTP.
This means that the current-servo error signal is heavily correlated to gain errors in the amplification chain.
That's why it is used to create an open-loop correction, that acts by modulating the gain, increasing it for both positive and negative excursion.
As with most open loop servo's, there is a limit to the level of correction that can be achieved: there are accuracy issues, and the shaping is not optimal, but as long as the polarity is correct and the amplitude is more or less in the right range, the improvement in linearity is valuable.
The only serious drawback of the scheme is the possible leaking of the correction signal into the noble path, due to the bias current of the input transistors combined with unequal impedances seen by the bases.
That's why I warned the impedance at the input has to be under control.
No I don't say linearity is improved compared to pure class A.
I say it is improved compared to standard class AB:
If the quiescent current is fixed at some optimal value, say 100mA for example (thus not class A), with the dynamic modulation completely turned off, the distortion shoots up to ~0.03%, which is a normal value for such an amplifier.
I didn't make any tests, but I suppose that if the amplifier was operated in full, static class A, the distortion would be comparable to or lower than that of the Circlophone, but of course, you would then need much larger heatsinks...
Certainly not
Member
Joined 2009
Paid Member
I tried running the simulation file, but it's looking for device models I don't have. This isn't a big issue as I can substitute with other models, but if you were to re-post it with the device models included that would allow people to reproduce your results. The way I do this is copy the text from the device model onto the 'clipboard', switch to spice, hit 'S' and past it in so that the model is directly in the schematic.
Member
Joined 2009
Paid Member
Elvee:
I looked for it in all these posts (179, I almost went blind), but could not find it:
The Circlophone produces about 20+ watts as shown in the current schematic. Changing the output power requires what changes to the circuit? I believe 25/50/100 watt versions would be of interest to the Circlophonites. Could you post suggestions to the parts value changes?
Thanks, E
I looked for it in all these posts (179, I almost went blind), but could not find it:
The Circlophone produces about 20+ watts as shown in the current schematic. Changing the output power requires what changes to the circuit? I believe 25/50/100 watt versions would be of interest to the Circlophonites. Could you post suggestions to the parts value changes?
Thanks, E