WaltJ said:
Walt,
Yes of course it is relevant to know it, if only because it tells you that (as an example) you better use emitter resistors on the bjt to get as good linearity as the fet. But once you know and do that, use both with similar or the same result *in the application*, how relevant is then that difference to base your choice on? I'd say it no longer is.
Elso says that he clearly prefers the fet. I don't know his exact implementation, but knowing that you CAN use an implementation, again, with pretty much the same results just by using some emitter degeneration, then what causes the audible difference? Not anymore the linearity difference of the naked device I would say.
It's like two black boxes, one has the fet, the other has the bjt + emitter degeneration. Assuming for the sake of argument that the resulting linearities are the same, there's no way to find out which is which by measuring linearity. Therefore, any perceived or measured difference cannot be caused by differences in linearity.
Jan Didden
andy_c said:
Andy,
Saw your post too late, but you confirm what I perceived with eye-ball acquisition on the curves posted above: the bjt + Re is at least as good if not better then the fet. I understand the noise issue, and there may be more things that come into play, aside from the linearity.
Still, my question remains, and I think it is a valid one. Can we identify why there is an audible preference for fets not based on the difference in linearity. Noise?
Does a bjt have stronger parasitic cap modulation with signal voltage than the fet?
Jan Didden
Re: FET vs. BJT
Elso,
With Andy's post in mind, could it be that your preference is influenced by differences in noise factors? You seem to imply that the lower the signal (prepre) the larger your preference for fets. If it was the linearity it would be the opposite, as smaller signals diminish the advantage of the fet.
Jan Didden
Elso Kwak said:
Elso,
With Andy's post in mind, could it be that your preference is influenced by differences in noise factors? You seem to imply that the lower the signal (prepre) the larger your preference for fets. If it was the linearity it would be the opposite, as smaller signals diminish the advantage of the fet.
Jan Didden
janneman said:
It may be that these differences are based on evaluation of ultra-low noise designs for phono preamp circuits, MIC amps or similar. For optimum noise, the designer uses no degeneration at all with two different designs, one BJT and one FET. The FET circuit has way better linearity in this case. So maybe the designer concludes that FETs are better for all designs. In other words, they might extrapolate that experience to other types of designs where the noise constraint may not apply.
I'm just speculating though.
janneman said:
Jan, Emitter degeneration does not change the character of the sound in my experience. Of course it can also be used with FETs, source degeneration.
Again what are you so eager trying to prove?
😕
janneman said:
Jan:
This comes as a surprise to me though may be you are right. I tend to align with Walt on the basis of the underlying physics.
Anyway, it should be pointless to argue without hard data at hand, something that can be readily be done by simulation (provided device modeling is acurate enough at this detail level).
The point is, BJT's with emitter degeneration dilute the gm nonlinearity by the way of adding a series voltage (emitter drop) to Vbe. This has the effect of bringing down the equivalent gm (looking at the BJT plus emitter resistor).
The comparison should be among a BJT with emitter degeneration and a FET (that could even have source degeneration) but with circuit configuration adjusted so as to provide the same equivalent gm.
On this equal footing, then the DERIVATIVE of gm as a function of Vin should be computed at the operating point and compared.
This could also be extended to vacuum tubes.
To be fair, the check should be made selecting several diferent devices but matching BJT - FET - VT design goals.
Someone (I am absolutely out of time for this now) may derive the results in a more general way analiticaly, and of course publish a paper if this has not already been done.
Rodolfo
re: "What's your reasoning?"
Nobody thinks that this is like discussing what tire technology makes your commute more enjoyable.
Sure, there are technical reasons for one type versus another, and sure everyone is entitled to his or her own opinion--everyone is right, but really is there much difference in the larger picture (electrical signals being an intermediate form of audio anyway...).
JF
Nobody thinks that this is like discussing what tire technology makes your commute more enjoyable.
Sure, there are technical reasons for one type versus another, and sure everyone is entitled to his or her own opinion--everyone is right, but really is there much difference in the larger picture (electrical signals being an intermediate form of audio anyway...).
JF
Leach input stage.....
The text seems to borrow from an AES preprint that he did, mid-80s. Anyone here remember it, and the usual reply from the usual supects?
Garde stated that he saw no reason that emitter resistors need to be more that 10-22 ohms, IIRC. I believe that he allowed that special cases might need 100 ohms.
Jocko
The text seems to borrow from an AES preprint that he did, mid-80s. Anyone here remember it, and the usual reply from the usual supects?
Garde stated that he saw no reason that emitter resistors need to be more that 10-22 ohms, IIRC. I believe that he allowed that special cases might need 100 ohms.
Jocko
Re: Leach input stage.....
I read it just recently. I could dig it out if anybody's interested. There was another reply to another letter too, where Leach seemed frustrated with all the terms TIM, PIM, SID, etc. He basically said "It's all the same thing: input stage overload." Or as Czerwinski said "It's not the velocity of the rock that hurts you. It's the rock itself, though the velocity can be measured".
Jocko Homo said:
I read it just recently. I could dig it out if anybody's interested. There was another reply to another letter too, where Leach seemed frustrated with all the terms TIM, PIM, SID, etc. He basically said "It's all the same thing: input stage overload." Or as Czerwinski said "It's not the velocity of the rock that hurts you. It's the rock itself, though the velocity can be measured".
The wheels are still turning...
To bring back the past (and to continue--unintentionally--to annoy others), I put my money on an all JFET design like this...
My reasoning...lowest noise... After six months, I still enoy listening to it (with Sennheiser headphones...). Thinking about paralleling a bunch of JFETs to drive speakers (again for super low noise...a main advantage of discrete over integrated...understand that I'll need use low impedance to deal with accumulation of CISS). I find that I don't need a lot of voltage gain...so I prefer the follower... Sorry, if I'm off track...unlikely that I'll continue long...
JF
To bring back the past (and to continue--unintentionally--to annoy others), I put my money on an all JFET design like this...
An externally hosted image should be here but it was not working when we last tested it.
My reasoning...lowest noise... After six months, I still enoy listening to it (with Sennheiser headphones...). Thinking about paralleling a bunch of JFETs to drive speakers (again for super low noise...a main advantage of discrete over integrated...understand that I'll need use low impedance to deal with accumulation of CISS). I find that I don't need a lot of voltage gain...so I prefer the follower... Sorry, if I'm off track...unlikely that I'll continue long...
JF
Elso, not at the core of this discussion but anyway here is a short answer. If you do a Taylor series expansion of the exponential function you end up with a polynomial with all odd exponents. In the frequency domain that translates to odd (3rd, 5th, 7th etc) harmonics for a BJT.
Whereas the distortion for the (ideal) FET will be purely second harmonic which you probably know is much more pleasant for the ear.
I have a whitepaper somewhere where feedback for a single stage (BJT, FET and tube) is examined. While it reduces distortion for all those gain devices it makes the distortion messier i.e. with lots of higher order harmonics instead of purely second order for the FET and vacuum tube. Non GFB-lovers rejoice!
/Magnus
Was pain/Is pain with 36db of Overhead
Is pain: replace eight 9V batteries after every 24 hours of play time. (Batteries being a low noise power source. A 20 minute operation and ~$12 a set...)
Was pain (construction): copper heatsinks on all JFETs, sheet copper power planes, silver wiring, silver epoxy, double steel enclosure with air vents yet no line-of-sight penetration. Super matched eight sets of JFETs. Everything that might minimize noise.
Gain for me is not a problem. With my Sennheisers at my normal listening levels, I still have 36db of overhead (on the step attenuators)... hehehe...I say voltage gain is overrated...
JF
Is pain: replace eight 9V batteries after every 24 hours of play time. (Batteries being a low noise power source. A 20 minute operation and ~$12 a set...)
Was pain (construction): copper heatsinks on all JFETs, sheet copper power planes, silver wiring, silver epoxy, double steel enclosure with air vents yet no line-of-sight penetration. Super matched eight sets of JFETs. Everything that might minimize noise.
Gain for me is not a problem. With my Sennheisers at my normal listening levels, I still have 36db of overhead (on the step attenuators)... hehehe...I say voltage gain is overrated...
JF
Hi, PMA,
Yes you are right. I dont know why people "think" NE5532 is standard for any audio application.
While I dont know what's inside every opamp (I just experiment with exchanging opamps in preamp) I like the sound of LF353.
The OPA2134 gives less detail than the LF353. Producing kind of "dull" sound. The LM833 is very harsh, the TL072 is too light sound. 4558 gives no special sound.
In sound reproduction, I prefer OPA2604 than OPA2134 (I also dont understand this. Why for me the 2604 sounds better, while the 2134 is advertized "for audio")
But NE5532 have an interesting property. Once I series 8 of these opamps in buffer mode. The final result, the array of these 8 buffers with NE5532 gives no (little) difference with the real input sound. With other opamps, the difference is bigger, so other opamps are"making" their own sound, instead of reproducing what is fed to them.
What do you think of LF353? Is it good enough for "average"/low price opamp? Is it better than NE5532?
Yes you are right. I dont know why people "think" NE5532 is standard for any audio application.
While I dont know what's inside every opamp (I just experiment with exchanging opamps in preamp) I like the sound of LF353.
The OPA2134 gives less detail than the LF353. Producing kind of "dull" sound. The LM833 is very harsh, the TL072 is too light sound. 4558 gives no special sound.
In sound reproduction, I prefer OPA2604 than OPA2134 (I also dont understand this. Why for me the 2604 sounds better, while the 2134 is advertized "for audio")
But NE5532 have an interesting property. Once I series 8 of these opamps in buffer mode. The final result, the array of these 8 buffers with NE5532 gives no (little) difference with the real input sound. With other opamps, the difference is bigger, so other opamps are"making" their own sound, instead of reproducing what is fed to them.
What do you think of LF353? Is it good enough for "average"/low price opamp? Is it better than NE5532?
"The LM833 is very harsh........"
How can you say that? National specifically designed it for audio applications. And Philips probably put millions of them into Magnavox CD players.
Snicker.........
Jocko
How can you say that? National specifically designed it for audio applications. And Philips probably put millions of them into Magnavox CD players.
Snicker.........
Jocko
Re: "The LM833 is very harsh........"
I had one of those. I think it was a CDP-650 uhhh... or something. It had an incredibly cheap PC board (phenolic?) with resistors that were mounted vertically with big long leads that looped around. Great for minimizing clock pickup!
You're bringing back great memories of "the golden age of CD" Jocko. Pure perfect sound forever! Or at least until next year when your machine breaks. And the next one you buy, the latest and greatest, sounds just as bad as the previous one. Oh well, that situation only lasted for about 15 years or so.
But in all seriousness, things are looking up now on the digital side of things.
Jocko Homo said:
I had one of those. I think it was a CDP-650 uhhh... or something. It had an incredibly cheap PC board (phenolic?) with resistors that were mounted vertically with big long leads that looped around. Great for minimizing clock pickup!
You're bringing back great memories of "the golden age of CD" Jocko. Pure perfect sound forever! Or at least until next year when your machine breaks. And the next one you buy, the latest and greatest, sounds just as bad as the previous one. Oh well, that situation only lasted for about 15 years or so.But in all seriousness, things are looking up now on the digital side of things.
Easy. Make an array of 8 opamps in series. Each is a buffer. Then you decide what the sound's like compared to the original input.
LM833 kind of having "loudness" button on all frequencies. Using 1 of this, maybe gives slightly image of better sounding opamp. All frequencies are obvious.
But putting in this 8 array can tell you what a certain opamp "contribute" to the original signal.
lumanauw said:
Jocko was just joking. "Snicker" is a quiet laugh. I don't think he likes the LM833 either and I think he agrees with you about it.
I really like your test idea of 8 op-amps. It shows that you are thinking hard about op-amp sound and coming up with solutions of how to figure out what's best. I think your solution is very clever. I'd love to hear more about your results.
Pavel, This is interesting:
Empirical data must be contained in the sound if you can hear a difference with diifferent values of C1.
I realize you say even spectral analysis in audio band says nothing about sonic differences, but have you plotted SPL against the amplifier's frequency response with changes in value of C1, keeping all other things the same?
The original Fletcher-Munson curves and power spectrum masking model logically require me to think that perceived quality might change drastically even if, for example, with different instantiations of the component, SPLs systematically vary even 1 or 2 dB through the range 1 to 5kHz .
Empirical data must be contained in the sound if you can hear a difference with diifferent values of C1.
I realize you say even spectral analysis in audio band says nothing about sonic differences, but have you plotted SPL against the amplifier's frequency response with changes in value of C1, keeping all other things the same?
The original Fletcher-Munson curves and power spectrum masking model logically require me to think that perceived quality might change drastically even if, for example, with different instantiations of the component, SPLs systematically vary even 1 or 2 dB through the range 1 to 5kHz .
An externally hosted image should be here but it was not working when we last tested it.
- Status
- Not open for further replies.