JFETs vs. BJTs
May I join the fun?
While I don't have the knowledge or experience to have any
strong personal opinion on the JFET vs. BJT matter myself, I
get the impression that National had some favourable things
to say about JFET inputs already 30 years ago, but what do
they know?
http://www.national.com/an/AN/AN-A.pdf p. 10
May I join the fun?
While I don't have the knowledge or experience to have any
strong personal opinion on the JFET vs. BJT matter myself, I
get the impression that National had some favourable things
to say about JFET inputs already 30 years ago, but what do
they know?
http://www.national.com/an/AN/AN-A.pdf p. 10
To John
---quote
You are incorrect. Fets are now, and have been for the last 25 years, as quiet as bipolar devices in almost every application, if you use Toshiba input fets. Bipolar transistors, because they have somewhat higher Gm, can have slightly lower short circuit noise, but at realistic impedance levels and currents, they are usually just as noisy or noisier than discrete fet input stages.
That is easy, a 2SK146 will do the trick. It will also keep low noise at all reasonable impedances, up to 1 meg ohm or so. Of course, these devices are not easily available, so you could parallel a pair of 2SK389's and get the same performance. I work at 0.4nV/rt hz with complementary paralleled fets in my Vendetta Research design. I have achieved a 10 ohm overall equivalent noise level with this circuit for the last 20 years. Before that, I used bipolar transistors to achieve the same result, approximately 30 years ago, with the introduction of the Mark Levinson JC-1 pre-preamp.
The problem with transistors is their NOISE CURRENT which is very high when the NOISE VOLTAGE is low. It is virtually impossible to get both very low noise voltage and noise current with a bipolar device with an inductive source, such as a tape head or a moving magnet phono cartridge, and they can be very noisy with source impedances above 1K ohm or so. Fets can be operated at higher currents for low voltage noise, and still have almost unmeasurable current noise.
---quote
As I remember, Walker in his RIAA design (published in WW in the early 70th) showed that the lowest noise could be reached with MM cartidge assuming RIAA frequency curve with input p-n-p long tail pair with current of 110 uA through each BJT. Skritek in IEEE Trans. published the same result. My personal experience (may by I'd used unproper JFETs) also the same. You say - inductive source - but shunted by 47 kOhm and ~200 pF.
---quote
You are incorrect. Fets are now, and have been for the last 25 years, as quiet as bipolar devices in almost every application, if you use Toshiba input fets. Bipolar transistors, because they have somewhat higher Gm, can have slightly lower short circuit noise, but at realistic impedance levels and currents, they are usually just as noisy or noisier than discrete fet input stages.
That is easy, a 2SK146 will do the trick. It will also keep low noise at all reasonable impedances, up to 1 meg ohm or so. Of course, these devices are not easily available, so you could parallel a pair of 2SK389's and get the same performance. I work at 0.4nV/rt hz with complementary paralleled fets in my Vendetta Research design. I have achieved a 10 ohm overall equivalent noise level with this circuit for the last 20 years. Before that, I used bipolar transistors to achieve the same result, approximately 30 years ago, with the introduction of the Mark Levinson JC-1 pre-preamp.
The problem with transistors is their NOISE CURRENT which is very high when the NOISE VOLTAGE is low. It is virtually impossible to get both very low noise voltage and noise current with a bipolar device with an inductive source, such as a tape head or a moving magnet phono cartridge, and they can be very noisy with source impedances above 1K ohm or so. Fets can be operated at higher currents for low voltage noise, and still have almost unmeasurable current noise.
---quote
As I remember, Walker in his RIAA design (published in WW in the early 70th) showed that the lowest noise could be reached with MM cartidge assuming RIAA frequency curve with input p-n-p long tail pair with current of 110 uA through each BJT. Skritek in IEEE Trans. published the same result. My personal experience (may by I'd used unproper JFETs) also the same. You say - inductive source - but shunted by 47 kOhm and ~200 pF.
bipolar disorder
Mike, this is the "book" answer, and I will leave it there in the interest of academic conservatism (i.e., you can't be wrong if you quote the book). This leaves for another day the more interesting question of why bench measurements so often fail to predict amplifier sound quality.mikek said:
Next time your wife thinks that you are strange......
Just remind her of all your DIY pals. Make sure that I am on top of the list, since I helped you sneak parts into your house.
Anyway........
Yep, he knows it all......our DIY pal........good ol' MikeK.
Maybe this will just encourage him to do more.
"Who, Jocko? MikeK, or your buddy Phred?"
Well, both actually. This is more fun that listenting to Frank Zappa going on about the enchilada with pickle sauce..........
See......even more Italian references......Sinatra, Zappa......and not one mention of those other things Italians like so much.......you know.......the ones that can be seen in "Texas".
"Hilts, 10 days."
"That's Capt. Hilts."
"20 days."
"Oh, you still be here when I get out, won't you?"
Jocko
Just remind her of all your DIY pals. Make sure that I am on top of the list, since I helped you sneak parts into your house.
Anyway........
Yep, he knows it all......our DIY pal........good ol' MikeK.
Maybe this will just encourage him to do more.
"Who, Jocko? MikeK, or your buddy Phred?"
Well, both actually. This is more fun that listenting to Frank Zappa going on about the enchilada with pickle sauce..........
See......even more Italian references......Sinatra, Zappa......and not one mention of those other things Italians like so much.......you know.......the ones that can be seen in "Texas".
"Hilts, 10 days."
"That's Capt. Hilts."
"20 days."
"Oh, you still be here when I get out, won't you?"
Jocko
going by the book
The FET (JFET or MOSFET) has a considerably lower transconductance than a bipolar device operating at the same current. While this is normally considered a drawback of the FET, we note that this "poor"' behavior is in fact highly desirable in applications to fast amplifiers. … [A ] JFET-input op amp is about forty times faster than a simple bipolar input. Further, if JFET's are properly substituted for the slow p-n-p's in a monolithic design, bandwidth improvements by at least a factor of ten are obtainable. JFET-input op amps, therefore, offer slew rate improvements by better than two orders of magnitude when compared with the conventional IC op amp. (Similar improvements are possible with MOSFET-input amplifiers.) This characteristic, coupled with picoamp input currents and reasonable offset and drift, make the JFET-input op amp a very desirable alternative to conventional bipolar designs.
Granted, this thinking is now almost 30 years old, but I doubt that much has happened in the interim to invalidate it.
You may not yet have read the National Semiconductor app note (Dec 1974) to which you have a link in your thread on Self amp oscillation. From page 10 of that app note:mikek said:
The FET (JFET or MOSFET) has a considerably lower transconductance than a bipolar device operating at the same current. While this is normally considered a drawback of the FET, we note that this "poor"' behavior is in fact highly desirable in applications to fast amplifiers. … [A ] JFET-input op amp is about forty times faster than a simple bipolar input. Further, if JFET's are properly substituted for the slow p-n-p's in a monolithic design, bandwidth improvements by at least a factor of ten are obtainable. JFET-input op amps, therefore, offer slew rate improvements by better than two orders of magnitude when compared with the conventional IC op amp. (Similar improvements are possible with MOSFET-input amplifiers.) This characteristic, coupled with picoamp input currents and reasonable offset and drift, make the JFET-input op amp a very desirable alternative to conventional bipolar designs.
Granted, this thinking is now almost 30 years old, but I doubt that much has happened in the interim to invalidate it.
Well folks, I don't know what has happened in the past, so I can't easily comment on it, however: As far as noise is concerned, in this case, fets can easily keep up with bipolars.
Some examples shown here, prove it so.
A low noise bipolar transistor has 3 significant noise contributions. One of these r(bb'), which is the intrinsic base resistance, can vary from 2-400 ohms, depending on the device. Of course, the MAT02, which is composed of 10 or more paralleled devices, has a low r(bb'), but interestingly enough, not the lowest in the industry. This would go to a PNP device from Rohm, Hitachi, or even Fairchild.
The second noise contribution, that is collector current dependent, is the second stage shot noise. It ALWAYS measures at the noise equivalent of 1/2 r(e). r(e) is always 1/gm which is: 26 ohms at one ma. 260 ohms at 100ua, 2.6 ohms at 10 ma, etc.
Therefore at 1 ma: [second stage shot noise] is 13ohms equivalent noise.
The third component is base shot noise and is more complicated to calculate, because it is sensitive to source impedance and any resistance in the emitter leg as well. It is also sensitive to the base current, so it is inversely proportional to the BETA of the device. The MAT02 has a beta of 250 or so, which is respectable, but still low enough to insure some current noise in a practical circuit, although still fairly low with a source below 100 ohms or so.
For the moment, we can ignore the input shot noise.
Well the best that can be done at 1 ma is: 13 +r(bb')[which at best is 2 ohms] so we could, in theory get 15 ohms equivalent noise or .3nV/rt Hz. Pretty good, but the MAT02 shows 1 nV/rt Hz, so it must have an r(bb') of approximately 45 ohms in order to make the results match the theory. Perhaps it is somewhat better than that, I know that some matched bipolar pairs are, but they degraded the performance on the spec sheet.
However, a 2SK389 monolithic fet pair can easily match this performance, but you would have to run it at 10 ma or so to get best results. A 2SK146 matched pair can easily beat this by another 3dB.
The important thing about FET's is that they have almost no input stage shot noise, so they do not get noisy with input impedances over 100 ohms or so, like bipolars do, when running at 1 ma. This is a great advantage in most cases. Now folks, what happens if we run the bipolar transistor at 100ua, instead of 1 ma?
Some examples shown here, prove it so.
A low noise bipolar transistor has 3 significant noise contributions. One of these r(bb'), which is the intrinsic base resistance, can vary from 2-400 ohms, depending on the device. Of course, the MAT02, which is composed of 10 or more paralleled devices, has a low r(bb'), but interestingly enough, not the lowest in the industry. This would go to a PNP device from Rohm, Hitachi, or even Fairchild.
The second noise contribution, that is collector current dependent, is the second stage shot noise. It ALWAYS measures at the noise equivalent of 1/2 r(e). r(e) is always 1/gm which is: 26 ohms at one ma. 260 ohms at 100ua, 2.6 ohms at 10 ma, etc.
Therefore at 1 ma: [second stage shot noise] is 13ohms equivalent noise.
The third component is base shot noise and is more complicated to calculate, because it is sensitive to source impedance and any resistance in the emitter leg as well. It is also sensitive to the base current, so it is inversely proportional to the BETA of the device. The MAT02 has a beta of 250 or so, which is respectable, but still low enough to insure some current noise in a practical circuit, although still fairly low with a source below 100 ohms or so.
For the moment, we can ignore the input shot noise.
Well the best that can be done at 1 ma is: 13 +r(bb')[which at best is 2 ohms] so we could, in theory get 15 ohms equivalent noise or .3nV/rt Hz. Pretty good, but the MAT02 shows 1 nV/rt Hz, so it must have an r(bb') of approximately 45 ohms in order to make the results match the theory. Perhaps it is somewhat better than that, I know that some matched bipolar pairs are, but they degraded the performance on the spec sheet.
However, a 2SK389 monolithic fet pair can easily match this performance, but you would have to run it at 10 ma or so to get best results. A 2SK146 matched pair can easily beat this by another 3dB.
The important thing about FET's is that they have almost no input stage shot noise, so they do not get noisy with input impedances over 100 ohms or so, like bipolars do, when running at 1 ma. This is a great advantage in most cases. Now folks, what happens if we run the bipolar transistor at 100ua, instead of 1 ma?
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.