Q: EXTREMELY LOW NOISE TRANSISTORS and INPUT TRANSFORMERS....
j.h.van...@student.utwente.nl
The following concerns only those people expert on low noise electronic design:
Question is put quite verbally to short-out any misconceptions.....
I am a student working on a front-end amplifier operating at room temp. in the
10Hz-100KHz range from an almost purely resistive (about) 0.1 Ohm signal
source (=detector {also room temp.}). Input signals go down to 100 Nanovolt
(about 18Db above detector's noisefloor). It is better to use bipolars at
this low source impedance level than Jfets allthough the latter can be cooled
to say liquid Nitrogen. Amplifier input current shot noise contribution is
generally neglegible with this 0.1 Ohm source. Rbb' however does matter. With
some looking around it is not too difficult to find commercial bipolars (even
single chip paired) with about 30 Ohms Rbb' and an acceptable 1/f corner in
E-noise-input. I have to work at room temperatures and the detector's signal
is non periodic so I cannot average using a computer, nor do any other
algorithms help. Using a good input transformer might be the solution.... see
"second question" below.
I however first am looking for bipolars (probably PNP) that have a Rbb'
considerably lower than the earlier mentioned 30 Ohms: Hence my first
question considering the 0.1 Ohm source:
--> Looking for (commercial?) Bipolar transistors on following criteria:
-Rbb' in the 1 Ohm range or below (Ree' of course low enough)
-Low 1/f corner in E-noise-input (preferrably below 10 Hz, but 100Hz
might do).
-DC forward current transfer ratio (HFE) better than say 50.
-Current gain transit freq. (Ft) > 1 Mhz.
Of course I have looked at paralleling devices to achieve goal....
Some solutions (prematurely) failed:
-Transmitter bipolars (lots of parallel devices inside)(f.i. MRF xxx types):
1/f corner in E-noise-input is too high; current gain is lowish.
-Most super low noise UHF/VHF transistors (e.g. BFT66) have the same problems.
-Power audio frequency (i.e. large geometry) transistors mostly fail on 1/f
and/or HFE considerations here.
-How about those old Germanium bipolars? Too much leakage, too much 1/f
effects? Guess Rbb' must be much lower than with Silicon however.
-GaAs is too noisy compared to silicon regarding 1/f effects in the
10Hz-100KHz range?
Any expert constructive comments on the last above points?
Second question:
I know that (noise) impedance matching using a transformer preceding the
amplifier could be a good solution; I however have not been able to find any
suitable commercial transformers; I have even tried a big power toroid ((=low
winding resistances, low barkhausen core noise, high winding inductances; but
bandwidth, loss inductance, parasitic capacitances etc. really messed up
things)); So hence my second question:
--> Who knows any (commercial?) transformers that might come close to:
-Undistorted (=phase, amplitude and intermodulation) transfer bandwidth of
at least 100KHz.
-Minimal parasitic inductances and -capacitances.
-Large core --> Low Barkhausen magnetic core noise modulation effects?
-Perfect shielding (all modes of -).
-I guess I may have to build one myself......
Constructive ideas are welcomed.
Thank you ahead for your expert bothering.
It would be MOST convenient if answers could be E-mailed to
j.h.van...@student.utwente.nl
Kindest regards,
j.h.van...@student.utwente.nl
----Yes, The universe is Analog; It's background noise is too. But still we
----can learn from it. How much Kelvin did you say?
Janos Szamosfalvi
: I am a student working on a front-end amplifier operating at room temp. in the
: 10Hz-100KHz range from an almost purely resistive (about) 0.1 Ohm signal
Wow!! 0.1 Ohm!! Can't you get a little higher impedance detector?
: Second question:
: I know that (noise) impedance matching using a transformer preceding the
: amplifier could be a good solution; I however have not been able to find any
: suitable commercial transformers; I have even tried a big power toroid ((=low
: winding resistances, low barkhausen core noise, high winding inductances; but
: bandwidth, loss inductance, parasitic capacitances etc. really messed up
: things)); So hence my second question:
: --> Who knows any (commercial?) transformers that might come close to:
: -Undistorted (=phase, amplitude and intermodulation) transfer bandwidth of
: at least 100KHz.
: -Minimal parasitic inductances and -capacitances.
: -Large core --> Low Barkhausen magnetic core noise modulation effects?
: -Perfect shielding (all modes of -).
: -I guess I may have to build one myself......
I'm not sure the exact impedance of moving coil cartridges, but they're
very low and most of them are used with special transformators. Those
trafos made for expensive HI-FI gears, so they must meet all your criteria,
with the possible exception of 100 khz where you may experience some drop.
Dave Dilatush
>I am a student working on a front-end amplifier operating at room temp. in the
>10Hz-100KHz range from an almost purely resistive (about) 0.1 Ohm signal
>source (=detector {also room temp.}). Input signals go down to 100 Nanovolt
>(about 18Db above detector's noisefloor).
Assuming you want an amplifier with an input-referred noise level of
100 nV peak-to-peak over a 100 KHz bandwidth, this would mean in input
noise voltage density of about 0.05 nV/root-Hz rms. This is AWFULLY
low, probably at or beyond the absolute limits of what you can achieve
at room temperature.
>It is better to use bipolars at
>this low source impedance level than Jfets allthough the latter can be cooled
>to say liquid Nitrogen. Amplifier input current shot noise contribution is
>generally neglegible with this 0.1 Ohm source. Rbb' however does matter.
>With
>some looking around it is not too difficult to find commercial bipolars (even
>single chip paired) with about 30 Ohms Rbb' and an acceptable 1/f corner in
>E-noise-input.
The only ones I'm familiar with are National Semiconductor's LM394
matched dual superbeta NPNs; these have about 40 ohms Rbb', Ree about
0.4 ohm, HFE about 500, and low 1/f corner frequency.
>I have to work at room temperatures and the detector's signal
>is non periodic so I cannot average using a computer, nor do any other
>algorithms help. Using a good input transformer might be the solution.... see
>"second question" below.
>I however first am looking for bipolars (probably PNP) that have a Rbb'
>considerably lower than the earlier mentioned 30 Ohms: Hence my first
>question considering the 0.1 Ohm source:
>--> Looking for (commercial?) Bipolar transistors on following criteria:
> -Rbb' in the 1 Ohm range or below (Ree' of course low enough)
> -Low 1/f corner in E-noise-input (preferrably below 10 Hz, but 100Hz
> might do).
> -DC forward current transfer ratio (HFE) better than say 50.
> -Current gain transit freq. (Ft) > 1 Mhz.
>Of course I have looked at paralleling devices to achieve goal....
If you absolutely cannot transform your source impedance upward, the
only thing I could think of would be to parallel a number of LM394's
at about 2 mA Ie each, perhaps in a circuit similar to that of Figure
4 of National's LM394 application note AN-222 (in the Linear
Applications Handbook). You might be able to get down in the 0.1
nanovolt/root-Hz range this way.
>Some solutions (prematurely) failed:
>-Transmitter bipolars (lots of parallel devices inside)(f.i. MRF xxx types):
> 1/f corner in E-noise-input is too high; current gain is lowish.
>-Most super low noise UHF/VHF transistors (e.g. BFT66) have the same problems.
>-Power audio frequency (i.e. large geometry) transistors mostly fail on 1/f
> and/or HFE considerations here.
>-How about those old Germanium bipolars? Too much leakage, too much 1/f
> effects? Guess Rbb' must be much lower than with Silicon however.
>-GaAs is too noisy compared to silicon regarding 1/f effects in the
> 10Hz-100KHz range?
>Any expert constructive comments on the last above points?
>Second question:
>I know that (noise) impedance matching using a transformer preceding the
>amplifier could be a good solution; I however have not been able to find any
>suitable commercial transformers; I have even tried a big power toroid ((=low
>winding resistances, low barkhausen core noise, high winding inductances; but
>bandwidth, loss inductance, parasitic capacitances etc. really messed up
>things)); So hence my second question:
>--> Who knows any (commercial?) transformers that might come close to:
> -Undistorted (=phase, amplitude and intermodulation) transfer bandwidth of
> at least 100KHz.
> -Minimal parasitic inductances and -capacitances.
> -Large core --> Low Barkhausen magnetic core noise modulation effects?
> -Perfect shielding (all modes of -).
> -I guess I may have to build one myself......
>
>Constructive ideas are welcomed.
I really think that a proper input transformer will do more to solve
this problem than any amount of amplifier design magic: this is a
REALLY tough problem.
Have you considered making your own transformer? It isn't terribly
difficult. I prefer ferrite cup cores to toroids because they're so
much easier to wind; if you start out with a large cup core (say 200
grams of ferrite or more) you might be able to make one with the
necessary low leakage inductance and low distributed capacitance.
You don't need a very high ratio of impedance transformation here, and
I think a 2:1 winding ratio (4:1 impedance step-up) would suffice to
make your amplifier design task a lot easier.
Whatever this project is, you've picked a real challenge! If I may
ask, just what is this signal you're trying to amplify? You've
excited my curiosity.
Best of luck,
Dave
_ _ _ _ _ MARK W. LUND _ _ _ _ _
> Dear reader,
>
> The following concerns only those people expert on low noise electronic design:
> Question is put quite verbally to short-out any misconceptions.....
>
> 10Hz-100KHz range from an almost purely resistive (about) 0.1 Ohm signal
> source (=detector {also room temp.}). Input signals go down to 100 Nanovolt
> to say liquid Nitrogen. Amplifier input current shot noise contribution is
> generally neglegible with this 0.1 Ohm source. Rbb' however does matter. With
> some looking around it is not too difficult to find commercial bipolars (even
> single chip paired) with about 30 Ohms Rbb' and an acceptable 1/f corner in
> algorithms help. Using a good input transformer might be the solution.... see
> "second question" below.
>
> I however first am looking for bipolars (probably PNP) that have a Rbb'
> considerably lower than the earlier mentioned 30 Ohms: Hence my first
> question considering the 0.1 Ohm source:
> --> Looking for (commercial?) Bipolar transistors on following criteria:
> -Rbb' in the 1 Ohm range or below (Ree' of course low enough)
> -Low 1/f corner in E-noise-input (preferrably below 10 Hz, but 100Hz
> might do).
> -DC forward current transfer ratio (HFE) better than say 50.
> -Current gain transit freq. (Ft) > 1 Mhz.
>
> Of course I have looked at paralleling devices to achieve goal....
> It would be MOST convenient if answers could be E-mailed to
> j.h.van...@student.utwente.nl
>
> Kindest regards,
> j.h.van...@student.utwente.nl
>
> ----Yes, The universe is Analog; It's background noise is too. But still we
> ----can learn from it. How much Kelvin did you say?
JH,
JFETs are the lowest noise devices available in your frequency range.
MOXTEK makes the lowest noise JFETs in the world, but priced in the
$300 range. They are low capacitance devices with no g-r noise
primarily used in x-ray detectors. If you don't need low capacitance
the noise reduces as the capacitance increases. Our JFETs are 1 and
3 pF (approximately). A good 4 pF device is the 2N4416. The 2SK152
is also used, I don't remember its capacitance. InterFET makes a
range of low noise transistors. With any of the non-MOXTEK devices
you will want to sort for noise, since g-r peaks appear in most devices
in every batch. Some of these peaks will overlap into your temperature
range. Also if you are interested in the very lowest noise you have
to repackage the fets into a low noise package, since epoxy and glass
give dielectric noise.
A good reference for standard low noise JFET work is "Low noise
electronic system design" by CD Motchenbacher and JA Connely.
best regards
mark
Mark W. Lund, PhD
Director
MOXTEK, Inc.
Orem UT 84057
801-225-0950
FAX 801-221-1121
lu...@xray.byu.edu
Luis Palafox
I wouldn't call myself an expert, but it might be interesting to takle
a look at the following paper: Neri B., Pellegrini B. and Saletti R.,
ultra Low-Noise Preamplifier for Low-Frequency noise Measurements in
Electron Devices, IEEE Tr. Instr and Meas. Vol 40. No.1 Feb 91 p.2
They use FETs in parallel (2SK146, formerly from Toshiba, now
available from InterFET in the states) and an op27 and an op17. Gain
is 80 dB +_ 0.6 dB between 20 mHz and 100 kHz. Corner frequency is 10
Hz.
They published another paper more recently, reporting further
developments. I think it was also in the IEEE Tr IM, but I'm not sure.
If your bibliography search is unsuccessful, let me know and I'll try
to find it for you.
I matched my own 2SK147s, and paired them for a differential input
stage. Managed to get 13 nV/sqrt(Hz) using two 9 Volt batteries, but
the bandwidth was a lot smaller than yours. I still suspect the noise
contribution of the OP27 used in the second stage as
non-negligible. The data sheet only speaks about +-15V supplies and we
suspected that noise performance degrades significantly at +-9V.
[snip, snip]
Sorry, I can't help you with transformer coupling.
:>Constructive ideas are welcomed.
:>Thank you ahead for your expert bothering.
:>It would be MOST convenient if answers could be E-mailed to
:> j.h.van...@student.utwente.nl
:>
:>Kindest regards,
:> j.h.van...@student.utwente.nl
:>
:>----Yes, The universe is Analog; It's background noise is too. But still we
:>----can learn from it. How much Kelvin did you say?
BTW, your units are all wrongly spelled. It's dB and kHz, not Db and
KHz. It would matter in the specs of the preamplifier presented in the
paper, a mHz is not the same as a MHz! But I've always been known to
be picky. Don't take this as a flame, please.
Hope this helps!
Luis
Luis Palafox pal...@goofy.mpi-hd.mpg.de
Max-Planck-Institut fuer Kernphysik
Postfach 103980
D-69029 Heidelberg
------------------------------------------------------------------------------
Maybe I'm wrong. It wouldn't be the first time, it won't be the last!!
--
Luis Palafox pal...@goofy.mpi-hd.mpg.de
Max-Planck-Institut fuer Kernphysik
Postfach 103980
D-69029 Heidelberg
------------------------------------------------------------------------------
Maybe I'm wrong. It wouldn't be the first time, it won't be the last!!
John 015
(Luis Palafox) said:
>I matched my own 2SK147s, and paired them for a differential input
>stage. Managed to get 13 nV/sqrt(Hz) using two 9 Volt batteries, but
>the bandwidth was a lot smaller than yours. I still suspect the noise
>contribution of the OP27 used in the second stage as
>non-negligible. The data sheet only speaks about +-15V supplies and we
>suspected that noise performance degrades significantly at +-9V.
I am curious: the theoretical limit for the 2SK147s is much lower
than 13nV/sqrt(Hz). What brought the noise up to 13nV ?
john
Vladimir Koifman
>I am a student working on a front-end amplifier operating at room temp. in the
>10Hz-100KHz range from an almost purely resistive (about) 0.1 Ohm signal
>source (=detector {also room temp.}). Input signals go down to 100 Nanovolt
>(about 18Db above detector's noisefloor). It is better to use bipolars at
>this low source impedance level than Jfets allthough the latter can be cooled
>to say liquid Nitrogen. Amplifier input current shot noise contribution is
You can cool bipolars as well. In that particular application when the
source is negligible, the low beta is not a problem. NPNs should work
better than PNPs when cooled, so try them first.
>generally neglegible with this 0.1 Ohm source. Rbb' however does matter. With
>some looking around it is not too difficult to find commercial bipolars (even
>single chip paired) with about 30 Ohms Rbb' and an acceptable 1/f corner in
>E-noise-input. I have to work at room temperatures and the detector's signal
>is non periodic so I cannot average using a computer, nor do any other
>algorithms help.
An algorithmic solution can help in the case if you do not need to
know the exact signal shape, but just its spectrum, or correlation
function, etc. If this is the case, you can do following:
Connect two amps to your detector. Then the two amp outputs would give
you signal component + their internal noise components. Their noises
are not correlated, while the signal comonent is. If you compute a
cross-spectrum between these two outputs, it will contain just your
signal spectrum and not these two amp noises. Obviously you have to
have enough time to average your cross-spectrum. All modern
two-channel FFT analysers have cross-spectrum and cross-correlation
capability.
>--> Looking for (commercial?) Bipolar transistors on following criteria:
> -Rbb' in the 1 Ohm range or below (Ree' of course low enough)
> -Low 1/f corner in E-noise-input (preferrably below 10 Hz, but 100Hz
> might do).
> -DC forward current transfer ratio (HFE) better than say 50.
> -Current gain transit freq. (Ft) > 1 Mhz.
I don't think you really need such a high HFE or Ft. Your source
resistance is so small, that current noise component is not so high.
Even more, there is much easier to manufacture low-Rbb transistor with
low gain, then a high-gain one.
If you happen to have a good friend in your university
microelectronics lab., he can easily make a bipolar with Rbb=1ohm and
Hfe~1.
There is no problem to do it, just a lack of market demand prevents
semiconductor companies from making such creatures.
Your case is really something not common.
>Some solutions (prematurely) failed:
>-Transmitter bipolars (lots of parallel devices inside)(f.i. MRF xxx types):
> 1/f corner in E-noise-input is too high; current gain is lowish.
>-Most super low noise UHF/VHF transistors (e.g. BFT66) have the same problems.
>-Power audio frequency (i.e. large geometry) transistors mostly fail on 1/f
> and/or HFE considerations here.
>-How about those old Germanium bipolars? Too much leakage, too much 1/f
> effects? Guess Rbb' must be much lower than with Silicon however.
>-GaAs is too noisy compared to silicon regarding 1/f effects in the
> 10Hz-100KHz range?
I would recommend you to take a look on static induction transistors
(SIT). Potentially they are very low-noise devices. Yet I suspect they
have high 1/f noise. Anyway, worth looking. As I recall some Japanese
company is producing them (NEC?).
>Second question:
>I know that (noise) impedance matching using a transformer preceding the
>amplifier could be a good solution; I however have not been able to find any
Personally I do not believe in such a solution because you have too
wide bandwidth - 4 decades.
In order do not loose signal at low frequencies you have to provide
reactance > 0.1Ohm at 10Hz. Then at 100kHz the reactance would become
> 1kOhm. Then if your coil have Q~100, your magnetic looses will
generate noise like 10Ohm resistor, which is too high.
Vladimir Koifman (vl...@NetVision.net.il)
Engineering Work Cycle: Think, Do, Regret.
Ken Smith
Dave Dilatush <ddi...@ix.netcom.com> wrote:
>Applications Handbook). You might be able to get down in the 0.1
>nanovolt/root-Hz range this way.
Interfet has a jfet with about 0.4nV/root Hz noise. You may have better
luck with these.
>Have you considered making your own transformer? It isn't terribly
>difficult. I prefer ferrite cup cores to toroids because they're so
>much easier to wind; if you start out with a large cup core (say 200
>grams of ferrite or more) you might be able to make one with the
>necessary low leakage inductance and low distributed capacitance.
- If you dont need the isolation make it an autotransformer
- Shock mount it
- You may need a mu-metal house around it
--
--
kens...@rahul.net forging knowledge
john 015
C.D.H.W...@exeter.ac.uk (Dr Charles DH Williams) said:
>2SB737 (Rohm) PNP Rbb'=2ohm, enoise=0.55V/sqrt(Hz) at 10Hz, Hfe>120,
>Ft=100MHz
Question. Is the 2SB737 that say MCM sells the 'right' 2SB737 or
must one be careful to make sure one gets the right one ?
john
Dr Charles DH Williams
> Dear reader,
Hello
> The following concerns only those people expert on low noise electronic
> design:
Well I'm not sure what you mean by expert but...
> --> Looking for (commercial?) Bipolar transistors on following criteria:
> -Rbb' in the 1 Ohm range or below (Ree' of course low enough)
> -Low 1/f corner in E-noise-input (preferrably below 10 Hz, but 100Hz
> might do).
> -DC forward current transfer ratio (HFE) better than say 50.
> -Current gain transit freq. (Ft) > 1 Mhz.
2SB737 (Rohm) PNP Rbb'=2ohm, enoise=0.55V/sqrt(Hz) at 10Hz, Hfe>120, Ft=100MHz
> Second question:
> I know that (noise) impedance matching using a transformer preceding the
> amplifier could be a good solution; I however have not been able to find any
> suitable commercial transformers; I have even tried a big power toroid ((=low
EG&G model 1900 has voltage gains of 1:100:1000 and is flat from 0.1Hz to
2kHz and is intended to match sources from 0.05ohm to 500ohm. Or there is
the 1900A which is flat from 1Hz to 2kHz with 1:100 ratio. Winding this
sort of transformer is a special task don't bother trying it as a
garden-shed exercise.
Depending on the circumstances you may want to design a hybrid amplifier
with the transformer used in a low-pass amp, summed with something
optimised for 1kHz to 100kHz as the high-pass fill-in and then some
overall feedback to get frequency-phase response right at cross-over. The
design of this type of amp so that it works really well is not entirely
trivial, even if you know what you are doing.
You will may find interference is as significant as noise at the higher
frequencies. If you are using any normal sort of differential amplifier
configuration its CMRR will probably be rather poor at 100kHz.
In article <1995Jul17....@physc2.byu.edu>, lu...@physc2.byu.edu (_
_ _ _ _ MARK W. LUND _ _ _ _ _) wrote:
> JFETs are the lowest noise devices available in your frequency range.
> MOXTEK makes the lowest noise JFETs in the world, but priced in the
> $300 range. They are low capacitance devices with no g-r noise
[...]
> Mark W. Lund, PhD
> Director
> MOXTEK, Inc.
2SB737s are about 50cents each. In fact a SQUID is the lowest noise
device available in his frequency range.
Good luck
Charles
Dr Charles DH Williams
015) wrote:
> In article <C.D.H.Williams-...@144.173.2.204>,
> C.D.H.W...@exeter.ac.uk (Dr Charles DH Williams) said:
>
>
> >2SB737 (Rohm) PNP Rbb'=2ohm, enoise=0.55V/sqrt(Hz) at 10Hz, Hfe>120,
> >Ft=100MHz
>
> Question. Is the 2SB737 that say MCM sells the 'right' 2SB737 or
> must one be careful to make sure one gets the right one ?
It should have been enoise=0.55nV/sqrt(Hz) of course...
The numbers are for Rohm's device because (a) I have the data-sheet [which
is in Japanese!], and (b) I have confirmed the above specs myself.
These are not expensive devices, about $0=30, in moderate quantities. I
don't think there any "magic" about the processing, they just go for a
nice low Rbb'. The noise is exactly what you would expect given Rbb', Re,
Ic, etc. In fact the above figure includes some 1/f noise and drops to
enoise~0.4nV/sqrt(Hz) by 100Hz. The downside is the junction capacitances
are a bit on the high side for some applications and the high collector
current...
Charles
Dr Charles DH Williams
C.D.H.W...@exeter.ac.uk (Dr Charles DH Williams) wrote:
> 2SB737 (Rohm) PNP Rbb'=2ohm, enoise=0.55V/sqrt(Hz) at 10Hz, Hfe>120, Ft=100MHz
Oops, that should have been enoise=0.55nV/sqrt(Hz)
Craig Buckingham
LM394 or PMI's MAT-02.These are matched pairs with very low rbb of about
6 ohm I think. Parallel them for best impedance match with source.
You may not need to use a transformer with this setup.
This will simplify the design and reduce another noise source. Of course
the second stage will have to be low noise if the first stage does not
have much gain. I dont know how much gain you need but Analog Devices makes
a very low voltage noise opamp the AD797. It has a voltage noise of 0.7nV/Hz
and a 1/f noise corner of about 10Hz (or maybe less I dont have the data sheet
in front of me). The data sheets for the above devices have some circuits
which would be relevant for your application. Hope this is of some help.
Regards.