Yep, you are corect.....
There is a word for that type of problem, but I can't remember what it is called. All my buddies in the dics cutting business have retired, so I can't get the answer from them. Maybe someone else will come up with it.
Back to the original topic:
I agree with JC. Discrete circuits with JFET inputs always sound better, and they don't need GFB. Yet, they don't measure as well as op-amps with tons of GFB.
I tried to ge the AES to examine this about 15 years ago, and was told "Our experts (Cherry/Garde/Cordell/etc.) all prove you wrong, so there. End of discussion." At least we have some folks here who aren't so sure of themselves.
Still waiting for someone to come up with "The Reason" before I retire.
Jocko
There is a word for that type of problem, but I can't remember what it is called. All my buddies in the dics cutting business have retired, so I can't get the answer from them. Maybe someone else will come up with it.
Back to the original topic:
I agree with JC. Discrete circuits with JFET inputs always sound better, and they don't need GFB. Yet, they don't measure as well as op-amps with tons of GFB.
I tried to ge the AES to examine this about 15 years ago, and was told "Our experts (Cherry/Garde/Cordell/etc.) all prove you wrong, so there. End of discussion." At least we have some folks here who aren't so sure of themselves.
Still waiting for someone to come up with "The Reason" before I retire.
Jocko
Maybe the reason is because jFets have a more similar transfercurve
to tubes than bjts ? My experience with jfet-input was, that jfets
sounds sweeter/softer, somehow like they have less trouble with treble.
And additional to nicer sounding, they don't produce that much
input-DC-offset. That enables a noncomplementary diffamp-design
without big caps for getting rid of this annoying dc-offset. Matching
the 2 jfets is enough to get DC below 10mv. (With matched bjts easily
>2v)
Mike
to tubes than bjts ? My experience with jfet-input was, that jfets
sounds sweeter/softer, somehow like they have less trouble with treble.
And additional to nicer sounding, they don't produce that much
input-DC-offset. That enables a noncomplementary diffamp-design
without big caps for getting rid of this annoying dc-offset. Matching
the 2 jfets is enough to get DC below 10mv. (With matched bjts easily
>2v)
Mike
janneman said:
Yea, the context here is that I swear it used to be buried in the noise and now I can hear it on (I think) even some Sheffield Labs recordings. I do have some records where it is audible on an ordinary preamp too. The low inductance high output Grados let you get noise figures of a few tenths of a dB with somewhat ordinary JFET inputs. It gets harder with 10 Ohm moving coils. This was mainly an exercise in "how little noise can I get out of my electronics". YMMV on the whole openloop, passive equalization aspect of things. At mV levels an openloop JFET pair has a LOT less distortion than the needle tracking the groove.
BJT / FET input stage
Check Walt Jung's EDN November/98 third article BJT / JFET relative transconductance diagram.
Rodolfo
MikeB said:
Check Walt Jung's EDN November/98 third article BJT / JFET relative transconductance diagram.
Rodolfo
Rodolfo,
I guess you specifically refer to this figure (Walt, I trust you allow me to reproduce it here, since it is accessible to anyone anyway).
I have been thinking about that for some time. The BJT has higher gain by some 5 or 10 times, but lower linearity. What if we use that higher gain to apply 5 or 10 times more feedback. Would that not bring the situation back to a level field, same gain overall and same (non)linearity?
I guess what I am saying is that these plots as such don't tell the final story, but it is how you implement them in the total circuit topology. Possibly there is a topology that gives OVERALL better linearity with BJT's than with FETs? Even a simple emitter resistor for the BJT to bring its gain down to FET level would give it a much better linearity.
Jan Didden
I guess you specifically refer to this figure (Walt, I trust you allow me to reproduce it here, since it is accessible to anyone anyway).
I have been thinking about that for some time. The BJT has higher gain by some 5 or 10 times, but lower linearity. What if we use that higher gain to apply 5 or 10 times more feedback. Would that not bring the situation back to a level field, same gain overall and same (non)linearity?
I guess what I am saying is that these plots as such don't tell the final story, but it is how you implement them in the total circuit topology. Possibly there is a topology that gives OVERALL better linearity with BJT's than with FETs? Even a simple emitter resistor for the BJT to bring its gain down to FET level would give it a much better linearity.
Jan Didden
Attachments
SY said:
In some cases maybe, but what I hear is mostly after something loud everything has so much more silence to decay into. The spookiest thing I ever heard was the subway in old shaded dogs recorded in London (not my system). The guy had all the noise in the lows down so far (unfortunately this required a Goldmund Reference table and appropriate isolation). You could hear it rumble in and then leave a few seconds later.
Yes, it may be accessible already, and all you needed to do was *link* to it, and say "Look at http://home.comcast.net/~walt-jung/wsb/PDFs/WTnT_Op_Amp_Audio_3.pdf
page 2 figure." This takes care of any rights issues.
But to your technical points:
Of course this plot doesn't say it all. It says that for these conditions, the BJT pair is much less linear, and has more gain than the JFET pair.
Consider the attached plot, which adds a 6SL7 diff pair to the mix. Now which pair is more linear? You'll need to look carefully, but the info is there. Tube lovers, rejoice!
Emitter (or source, or cathode) degeneration is another dimension that could be added, and is most interesting. I leave that as an excercise. It has been a long day here.
Walt Jung
Attachments
janneman said:
Yes, that figure.
They should be freely interchangeable if their respective differential nonlinearities were equivalent. I do not remember right now how differential gm nonlinearities compare among BJT and FET (be JFET or MOS).
I suspect the underlying physical phenomena make for in principle different behaviour. BJT's are current controled devices where forward biased junction voltage is log related to current. FETs on the other hand are essentially voltage controlled devices, channel depletion or enhancement is a function of gate induced field which in turn depends of geometry and border fringe effects.
With respect to GNFB, it was noted in previous posts the dynamic problem arising from the shift of operating point with increasing signal departure from quiescent condition.
Perhaps it should be noteworty to consider this viewpoint:
The output from an amplifier is nothing more than the input (diferential) signal times the amplifier transfer function. As such, given a nonlinear transfer, for the output to be as good as possible, the differential input must be predistorted to be strightened in turn by the nonlinear transfer. The predistorted differential input in time is the result of combining the "perfect" input with the "as good as it gets" output.
So, if someone is still reading this, it should be clear first and foremost the transfer itself should be the best possible in order to leave for GNFB the task of only cleaning up the "lint" not in charge of strighten everything.
Rodolfo
WaltJ said:
Good point.
Could it be this is basically unmasking the underlying physical nature of the different devices?
The comparison among BJT and FET I suggested in my previous post.
May be the FET - Tube difference is related to the field dependent dielectric constant of solid state (channel region) with respect to vacuum?
Rodolfo
WaltJ said:
Only me who can't open that PDF file!??
About emitter degenerstion, I just wonder how would a JFET compare to a BJT that has such an emitter degeneration which match the JFET's Gm curve?
The Walt J. Gm curve picture Jan D. first showed here leaded me to think about Leach's worthwhile words about emitter degeneration which can be seen
here
Cheers
Terry, you have my direction. I built what I previously described, as an open loop line preamp, the CTC. 350KHz open loop bandwidth anyone? Yes, I shamelessly copied Charles Hansen of Ayre, who did it first. I was chicken to do it, before I saw that Charles had gotten away with it. Why, because it is difficult to get GREAT specs with ANY open loop circuit. But what the heck! It REALLY WORKS!. I spec the distortion at .01%, IM or harmonic, at 3V balanced out. The distortion is, more or less, within this spec. With feedback, I could have gotten the same or better spec at 10V out, BUT it would not sound the same. How do I know? Because the JC-80 is a great example of a quality feedback line preamp design, using essentially the same parts.
Scott, you are far more sensitive than me regarding pre-'print-through' caused by records themselves. I have heard it before, myself, but I usually ignore it. Good to hear from you.
Scott, you are far more sensitive than me regarding pre-'print-through' caused by records themselves. I have heard it before, myself, but I usually ignore it. Good to hear from you.
Here's the 4558 as promised...
John, you asked about the 4558. I've attached the plots. The TI data sheet said unity gain bandwidth of 3 MHz typical, so I used that as the gain-bandwidth product. As before, this is for a closed loop gain of 10, freq = 20 kHz per the graph title. Performance at 20 kHz is a lot like the 797 at 1 MHz. Interesting pattern repeating itself. The 0.5 percent 3rd harmonic for a phase error of 1 degree once again shows up. At 2 Volts peak, the data files generated by MathCad show -.063 degrees phase error with a distortion of 68.77 dBc or 0.036 percent.
Donning my objectivist lab coat, three-foot-thick glasses, clipboard, and bald head, I'd say that the third harmonic is a more sensitive measure of the input stage nonlinearity than the phase error is.
John, you asked about the 4558. I've attached the plots. The TI data sheet said unity gain bandwidth of 3 MHz typical, so I used that as the gain-bandwidth product. As before, this is for a closed loop gain of 10, freq = 20 kHz per the graph title. Performance at 20 kHz is a lot like the 797 at 1 MHz. Interesting pattern repeating itself. The 0.5 percent 3rd harmonic for a phase error of 1 degree once again shows up. At 2 Volts peak, the data files generated by MathCad show -.063 degrees phase error with a distortion of 68.77 dBc or 0.036 percent.
Donning my objectivist lab coat, three-foot-thick glasses, clipboard, and bald head, I'd say that the third harmonic is a more sensitive measure of the input stage nonlinearity than the phase error is.
Attachments
Mike Gergen said:
Just to give credit where credit is due, the math involved was taken straight from Barrie Gilbert's article "Are op-amps really linear?" which was emailed to me by another member (and for which I'm grateful). All I did was transcribe Gilbert's results to a MathCad worksheet which allows me to vary the ratio of the test frequency to the gain-bandwidth product of the op-amp, and sweep the output voltage amplitude. Also, only input stage distortion is taken into account. The rest of the op-amp is assumed to be a distortionless integrator. Last assumption is that the input stage is a bipolar diff amp with no emitter degeneration. So this math doesn't represent the 797 exactly. Rather, it represents an idealized op-amp meeting the above requirements, and having a gain-bandwidth product of 100 MHz. That idealized circuit has negligible third harmonic distortion, mainly because the ratio of the gain-bandwidth product to the operating frequency is 5000. I didn't mean to say that the actual 797 would have these low figures in real life.
Gilbert's equations aren't valid for op-amps with FET input stages such as the OPA627 or AD8610. However, for a given gain-bandwidth product, op-amps with FET input stages should theoretically be better than Gilbert's equations predict, because the input stage will be more linear than a bipolar input stage.
Sorry, I don't have a clue about that one! That's actually a much more complex question than just circuit design because it relates to human perception. I'm just an engineer.
Despite what I said above, it's difficult for me to believe that 160 dB down would be audible for 3rd harmonic at 2 Volts peak out at 20 kHz. And certainly phase modulation of 10-6 deg ought to be inaudible by common sense. So, given a large enough gain-bandwidth product (say 100 MHz like the 797), the mechanisms we've been discussing should be irrelevant sound-wise, if only because they're swamped out by other nonlinearities. I'm not going to claim there's no audible difference between feedback amps and non-feedback amps, but if there is, I'd be hard pressed to say it had anything to do with distortion at the level that the idealized 797 shows at 20 kHz, that's for sure.
Once again, I don't have a clue, because that's a human perception issue and not a circuit design issue. To me, common sense seems to indicate that there must be some "fuhgettaboutit" level for distortion
. But objectivists tend to obsess about frequency domain stuff like distortion and often neglect time domain stuff. I'd be inclined to believe that time domain anomalies might be more audible than very low levels of distortion, even if there is no associated non-linearity per se. But that's just my gut feel and I have no rational justification for that at all.Putting my objectivist egghead hat on, when someone asks "Why is X true?", my first reaction is to ask "How was the truth of X established in the first place?". For instance, Cordell's article about PIM deals with the effect of feedback on PIM. He has a test amplifier that can be used both with and without feedback. He draws a schematic of the amplifier, and clearly shows which elements are removed and which added as feedback is added or removed. It gives the reader something to hang his hat on, as it were. A point of reference. But of all the claims about feedback supposedly sounding bad, I've not seen one bit of explanation as to how the experiment was conducted to establish this. Maybe I've just missed it. If you know of a web site that has such results, I'd be interested in seeing them.
ingrast said:
Rodolfo,
Yes, I get your point. What I was getting at with the 5 - 10 times increased feedback around the loop for the BJT, it would mean that the actual input signal at the BJT base, would be about 5 - 10 times smaller than at the FET gate (all other things being equal). In other words, the BJT is linear only over a much smaller range, but with the lower input level only that lower range is traversed, so the two situation could become more or less equivalent.
There is of course the matter of the effect of higher fb, but, as Barrie Gilbert said in his paper, "all of those requirements allmost all translate to a requirement for high slew rate" or words to that effect. So, personally I am not too worried about PIM, TIM, AM>FM and what have you, provided the amp is fast enough, they will not dominate over the usual distortion sources.
Bottom line (for me): one can built great amps with either BJTs or FETs, and the proof is out there on the streets.
Jan Didden
john curl said:
It is not a question of THD at all. I have made buffered link stages that measure below 0.001% THD and sound different. I assume we need a kind of dynamic measuring method (probably AM modulated multitone?), but it is very difficult to evaluate that results.
For the buffered stage as attached the sound depends greatly on C1 capacitor value. This also changes the resulting frequency range of the citcuit. Bandwith limited to some 140kHz gives much better sonic result compared to bandwith in MHz. The THD in audio band remains the same. Even spectral analysis across audio band is unable to say anything about sonic differences. I should add that in any tested case the circuit was unconditionally stable.
Attachments
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.