Maker Pro
Maker Pro

Low noise transistors

P

Phil Hobbs

Jan 1, 1970
0
I'm building a test jig to measure the noise floor of my antenna-coupled
tunnel junctions. This is slightly awkward to do by cobbling test
equipment, because they're only a few hundred ohms, and I need to be
able to measure their noise under bias. I'm rooting through my Lista
drawers full of ten-year-old parts for something to slap together.

So I have this nice way-way-sub-poissonian bipolar current source, with
1-Hz voltage feedback so that I don't blow anything up. Works groovy.
Now I'm building a low-noise amplifier to do the noise measurement.
It's a discrete diff amp made from Rohm 2SD786 BJTs, running at a
voltage gain of about 250, with an inverted cascode on each input
transistor to deal with the huge C_cb while bringing the signals out
near ground. There's an LM6361 as the output stage, with overall
AC-coupled feedback to get a gain of 1000, and an auxiliary bias
correction loop dumping a bit of current into one of the collector loads
to keep delta V_BE of the diff stage at zero. Should work fine
sometime tomorrow.

Which brings me to my actual question: Are the 2SD786 and 2SD737 still
the BJT noise champions at low source resistance?

Cheers,

Phil Hobbs
 
P

Pooh Bear

Jan 1, 1970
0
Phil said:
Which brings me to my actual question: Are the 2SD786 and 2SD737 still
the BJT noise champions at low source resistance?

The *2SB737* used to be one of the best devices you could get at a sensible
price for mic preamps. Rohm no longer make it though.

Graham
 
P

Phil Hobbs

Jan 1, 1970
0
Pooh said:
Phil Hobbs wrote:




The *2SB737* used to be one of the best devices you could get at a sensible
price for mic preamps. Rohm no longer make it though.

Graham
Ah, 2SB, right. Used to be able to read things that small, but not any
more. Thanks. So what do you use instead?

Cheers,

Phil Hobbs
 
P

Phil Hobbs

Jan 1, 1970
0
Pooh said:
Phil Hobbs wrote:




The *2SB737* used to be one of the best devices you could get at a sensible
price for mic preamps. Rohm no longer make it though.

Graham
Ah, 2SB, right. Used to be able to read things that small, but not any
more. Thanks. So what do you use instead?

Cheers,

Phil Hobbs
 
P

Pooh Bear

Jan 1, 1970
0
Phil said:
Ah, 2SB, right. Used to be able to read things that small, but not any
more. Thanks. So what do you use instead?

The 2SA1084 ( widely used in this application ) suffices. I think most of the
really nice devices designed for audio ( like moving coil preamps ) are now out
of production. In the distant past we used 2N4403s btw ( larger than usual die
AIUI makes for lower intrinsic resistance and hence lower noise ). It can't match
the more modern parts though. 2SA970 was another we used too. Soundcraft has used
the 2SA1316 and that can still be found on Toshiba's website - and the 970 as
well. Woohoo !

There's a thread in rec.audio.pro where Jim Williams mentions a Renesas (
fromerly Hitachi ) part. 2SC2545. Ah ! PNP - it's the complement of the
aforementioned 2SA1084.

Graham
 
W

Winfield Hill

Jan 1, 1970
0
Phil Hobbs wrote...
I'm building a test jig to measure the noise floor of my antenna-coupled
tunnel junctions. This is slightly awkward to do by cobbling test
equipment, because they're only a few hundred ohms, and I need to be
able to measure their noise under bias. I'm rooting through my Lista
drawers full of ten-year-old parts for something to slap together.

So I have this nice way-way-sub-poissonian bipolar current source, with
1-Hz voltage feedback so that I don't blow anything up. Works groovy.
Now I'm building a low-noise amplifier to do the noise measurement.
It's a discrete diff amp made from Rohm 2SD786 BJTs, running at a
voltage gain of about 250, with an inverted cascode on each input
transistor to deal with the huge C_cb while bringing the signals out
near ground. There's an LM6361 as the output stage, with overall
AC-coupled feedback to get a gain of 1000, and an auxiliary bias
correction loop dumping a bit of current into one of the collector
loads to keep delta V_BE of the diff stage at zero. Should work
fine sometime tomorrow.

Which brings me to my actual question: Are the 2SD786 and 2SD737
still the BJT noise champions at low source resistance?

2sd786 and 2sb737 BJTs are close to the available limit, and folded
cascode circuits are a good choice. But if the source impedance is
near 200 ohms (real), that corresponds to 1.8nV Johnson noise, much
higher than the 0.55nV that 2sd786 are capable of. What about the
base-current noise, is that a problem with your source impedance?
Have you run a network analyzer on your source impedance? You may
want to run your transistors at lower currents to reduce the i_n
base-current shot noise and suffer the higher voltage noise.

BTW, my hp 4470A Transistor Noise Analyzer is a convenient tool to
evaluate low-noise transistors, it's worth looking for one on eBay.
 
P

Phil Hobbs

Jan 1, 1970
0
Winfield said:
Phil Hobbs wrote...



2sd786 and 2sb737 BJTs are close to the available limit, and folded
cascode circuits are a good choice. But if the source impedance is
near 200 ohms (real), that corresponds to 1.8nV Johnson noise, much
higher than the 0.55nV that 2sd786 are capable of. What about the
base-current noise, is that a problem with your source impedance?
Have you run a network analyzer on your source impedance? You may
want to run your transistors at lower currents to reduce the i_n
base-current shot noise and suffer the higher voltage noise.

BTW, my hp 4470A Transistor Noise Analyzer is a convenient tool to
evaluate low-noise transistors, it's worth looking for one on eBay.

Thanks, Win. Glad that I seem to be on the right track. I'll have a
look for a 4470A--though I'll want IBM to buy it, so eBay is usually
out, more's the pity.

The emitters of the diff pair are driven by a low-noise adjustable
current source, and the diff pair collector loads are current sources as
well, driven from the same adjustment pot via massively degenerated
current mirrors. (+-15 V supplies are such a luxury for low-noise
circuits.) The tracking isn't perfect, but the error mainly shows up as
a common-mode shift in the dc level at the output of the inverted
cascodes, so it isn't a worry. The idea is to be able to change the
bias with one pot to optimize the noise for the impedance of the device.

The devices of interest range from about 10 ohms to 500 ohms.
Theoretically they should exhibit about twice full shot noise power,
which will give them a noise temperature of about 300K (PN junctions
used as resistors have a noise temperature of 150K at room temperature,
as I mentioned in another post.) The quieter I can get the amplifier,
the more confident I am in the measurements--my goal is to be able to
calibrate it with a 100-ohm resistor at room temperature and 77K, the
way you do with an RF LNA.

I'm very interested in their popcorn and 1/f performance too, since that
will tell me a fair amount about device reliability and the
contributions of individual atomic asperities and trap states. I don't
anticipate there'll be much. The tunnel barriers in Ni-NiO-Ni are so
low that the tunnel current isn't dominated by the thin spots in the
barrier the way it is in the Al/Al2O3/Al system, for example.

My optical coupling efficiency into the silicon waveguides still stinks,
and I broke my last lensed tapered fibre coupler, so for the next few
days I'm going to be doing noise measurements. This is all to the good,
because I need to know the noise-equivalent power (NEP) as well as the
responsivity before I can claim specific SNR numbers.

I ordered some very low Vce(sat) devices from Zetex, which have f_T of
around 100 MHz at 10 mA and betas of 300-800. Their saturation V_BE and
V_CE suggest that the extrinsic resistances are less than an ohm each,
so they may be decent input transistors. I'll have a whack with them
and see.

Cheers,

Phil Hobbs
 
P

Phil Hobbs

Jan 1, 1970
0
Winfield said:
Phil Hobbs wrote...



2sd786 and 2sb737 BJTs are close to the available limit, and folded
cascode circuits are a good choice. But if the source impedance is
near 200 ohms (real), that corresponds to 1.8nV Johnson noise, much
higher than the 0.55nV that 2sd786 are capable of. What about the
base-current noise, is that a problem with your source impedance?
Have you run a network analyzer on your source impedance? You may
want to run your transistors at lower currents to reduce the i_n
base-current shot noise and suffer the higher voltage noise.

BTW, my hp 4470A Transistor Noise Analyzer is a convenient tool to
evaluate low-noise transistors, it's worth looking for one on eBay.

Thanks, Win. Glad that I seem to be on the right track. I'll have a
look for a 4470A--though I'll want IBM to buy it, so eBay is usually
out, more's the pity.

The emitters of the diff pair are driven by a low-noise adjustable
current source, and the diff pair collector loads are current sources as
well, driven from the same adjustment pot via massively degenerated
current mirrors. (+-15 V supplies are such a luxury for low-noise
circuits.) The tracking isn't perfect, but the error mainly shows up as
a common-mode shift in the dc level at the output of the inverted
cascodes, so it isn't a worry. The idea is to be able to change the
bias with one pot to optimize the noise for the impedance of the device.

The devices of interest range from about 10 ohms to 500 ohms.
Theoretically they should exhibit about twice full shot noise power,
which will give them a noise temperature of about 300K (PN junctions
used as resistors have a noise temperature of 150K at room temperature,
as I mentioned in another post.) The quieter I can get the amplifier,
the more confident I am in the measurements--my goal is to be able to
calibrate it with a 100-ohm resistor at room temperature and 77K, the
way you do with an RF LNA.

I'm very interested in their popcorn and 1/f performance too, since that
will tell me a fair amount about device reliability and the
contributions of individual atomic asperities and trap states. I don't
anticipate there'll be much. The tunnel barriers in Ni-NiO-Ni are so
low that the tunnel current isn't dominated by the thin spots in the
barrier the way it is in the Al/Al2O3/Al system, for example.

My optical coupling efficiency into the silicon waveguides still stinks,
and I broke my last lensed tapered fibre coupler, so for the next few
days I'm going to be doing noise measurements. This is all to the good,
because I need to know the noise-equivalent power (NEP) as well as the
responsivity before I can claim specific SNR numbers.

I ordered some very low Vce(sat) devices from Zetex, which have f_T of
around 100 MHz at 10 mA and betas of 300-800. Their saturation V_BE and
V_CE suggest that the extrinsic resistances are less than an ohm each,
so they may be decent input transistors. I'll have a whack with them
and see.

Cheers,

Phil Hobbs
 
B

Ban

Jan 1, 1970
0
Phil said:
I'm building a test jig to measure the noise floor of my
antenna-coupled tunnel junctions. This is slightly awkward to do by
cobbling test equipment, because they're only a few hundred ohms, and
I need to be able to measure their noise under bias. I'm rooting
through my Lista drawers full of ten-year-old parts for something to
slap together.
So I have this nice way-way-sub-poissonian bipolar current source,
with 1-Hz voltage feedback so that I don't blow anything up. Works
groovy. Now I'm building a low-noise amplifier to do the noise
measurement. It's a discrete diff amp made from Rohm 2SD786 BJTs,
running at a voltage gain of about 250, with an inverted cascode on
each input transistor to deal with the huge C_cb while bringing the
signals out near ground. There's an LM6361 as the output stage, with
overall AC-coupled feedback to get a gain of 1000, and an auxiliary
bias correction loop dumping a bit of current into one of the
collector loads to keep delta V_BE of the diff stage at zero. Should work
fine sometime tomorrow.

Which brings me to my actual question: Are the 2SD786 and 2SD737
still the BJT noise champions at low source resistance?

Phil, I drew a basic preamp, which is basically independent of temperature
and allows also very low frequencies to be measured. The gain is also 60dB,
but a following differential amp can make up more. The gain depends only on
passive components, (1+2*4k99/10) and the current can be tailored to meet
the source impedance.
But I used PNP transistors because they have lower noise. Maybe 1nV/sqrtHz
overall in this case.

+-----+--------------+--o+15V
| | | /|
| | | OP177 /+|-GND
| | +------)-----+-< |
.-. .-. | .-. | \-|-+-+-+
| |3k3| | .-. | | | ||\| | | |
2k| | | | | | | |2k +-||---+ | |
'-' '-' 6k8| | '-' 0u47|| .-.-.
| | '-' | | | |
| | | |< 1M| | |
|---+-------+----| MAT03 '-'-'
/| ___ |\ | |
GND +--------|___|-------)--------------)-+
| | 4k99 | ___ | |
.-. +---+ +---+---|___|--+ |
1k2| | |22p| ___ |22p| 4k99 | |
| | --- +-----+--|___|--+----+ --- | |
'-' --- | | 10R | | --- | |
| | |< |< >| >| | | |
o--+---+-------| THAT120A |------+ | |
|\ |\ /| /| | OP227 | |
| | | | | |\ | |
+-----+ +----+---)-----|-\ | |
| | | | >-+-)---o
o--+-------|--------------------)---+ +-|+/ |
| | | | |/ |
.-. | | | |\ |
1k2| | +--------------------)-------)-|-\ |
| | | | | | >---+---o
'-' .-. .-. +-|+/
| | | -15V | | | |/
=== | |2k4 o 2k4| | |
GND '-' | '-' o
| _V_ | -10
+--------|___|-------+
200R
(created by AACircuit v1.28 beta 10/06/04 www.tech-chat.de)
 
W

Winfield Hill

Jan 1, 1970
0
Phil Hobbs wrote...
Thanks, Win. Glad that I seem to be on the right track. I'll have a
look for a 4470A --though I'll want IBM to buy it, so eBay is usually
out, more's the pity.

http://www.tucker.com/java/jsp/product_partno4470A_invid8898_condR.htm
Tucker has one you can order with a P.O., but of course they charge
a lot more than a typical eBay seller will get from we cheap bidders.
However 4470A don't come up very often on ebay, for example none are
available now, or in the last 30 days either. A Google search will
show a few other used-instrument sources that might have one.
 
P

Phil Hobbs

Jan 1, 1970
0
Ban said:
Phil Hobbs wrote:

Phil, I drew a basic preamp, which is basically independent of temperature
and allows also very low frequencies to be measured. The gain is also 60dB,
but a following differential amp can make up more. The gain depends only on
passive components, (1+2*4k99/10) and the current can be tailored to meet
the source impedance.
But I used PNP transistors because they have lower noise. Maybe 1nV/sqrtHz
overall in this case.


Thanks, Ban. My circuit has more transistors and fewer op amps than
yours, but operates not so differently.

Interim report:

It's running about 1.1 nV/sqrt(Hz) in 1-11 kHz, with some line harmonics
lower down that I have to fix--probably with an on-card regulator, since
my cap multipliers don't have enough rejection down there. Adding a
50-ohm source resistance (300K) brings it back up to 1.5 nV, which is
about right.

There's something broken about the inverted cascodes, because ripping
them out dropped the noise by about 3 dB, which it shouldn't. The clue
was that bypassing their bases either to ground or to the supply
*raised* the noise by about 10 dB! Their emitters are supposed to stand
still, so this is quite odd. There's probably an optimal V_CE for the
2SD786s, too--interelectrode capacitance and all that.

I'll try using 2N4250s instead of 2N3906s for the cascodes--I probably
damaged one of them during torture testing. The feedback network has a
10k : 10 ohm divider and two 1000 uF, 16V aluminums back to back, giving
unity gain at dc, with a low-frequency gain 3 dB point of about 32 Hz,
which is fine for my purposes, but it would be nice to get rid of the
capacitors. I may re-do the bias loop to eliminate them, but that will
take a bit of work.

Cheers,

Phil Hobbs
 
J

Joerg

Jan 1, 1970
0
Hello Win,
http://www.tucker.com/java/jsp/product_partno4470A_invid8898_condR.htm
Tucker has one you can order with a P.O., but of course they charge
a lot more than a typical eBay seller will get from we cheap bidders.
However 4470A don't come up very often on ebay, for example none are
available now, or in the last 30 days either. A Google search will
show a few other used-instrument sources that might have one.

If you buy through dealers on occasion what is your favorite used
equipment dealer? Tucker? I bought a lot of my stuff at auctions when
companies closed down. But lately that has dried up, mostly because
companies often just don't have anything other than computer stuff
anymore. IOW they didn't have any staff that could operate an analyzer.

Regards, Joerg
 
W

Winfield Hill

Jan 1, 1970
0
Joerg wrote...
Hello Win,

If you buy through dealers on occasion what is your favorite used
equipment dealer? Tucker? I bought a lot of my stuff at auctions when
companies closed down. But lately that has dried up, mostly because
companies often just don't have anything other than computer stuff
anymore. IOW they didn't have any staff that could operate an analyzer.

I've bought plenty of stuff through Tucker (mostly their eBay stuff),
but I prefer some of the other Internet used-equipment vendors. Most
of them also have lots of eBay activity. The best let you return the
equipment, no questions asked, and use Purchase-Order purchases, just
like all of the other true industrial-instrument sellers.
 
B

Ban

Jan 1, 1970
0
Phil said:
Thanks, Ban. My circuit has more transistors and fewer op amps than
yours, but operates not so differently.

Interim report:

It's running about 1.1 nV/sqrt(Hz) in 1-11 kHz, with some line
harmonics lower down that I have to fix--probably with an on-card
regulator, since my cap multipliers don't have enough rejection down
there. Adding a 50-ohm source resistance (300K) brings it back up to
1.5 nV, which is about right.

There's something broken about the inverted cascodes, because ripping
them out dropped the noise by about 3 dB, which it shouldn't. The
clue was that bypassing their bases either to ground or to the supply
*raised* the noise by about 10 dB! Their emitters are supposed to
stand still, so this is quite odd. There's probably an optimal V_CE
for the 2SD786s, too--interelectrode capacitance and all that.

I'll try using 2N4250s instead of 2N3906s for the cascodes--I
probably damaged one of them during torture testing. The feedback
network has a 10k : 10 ohm divider and two 1000 uF, 16V aluminums
back to back, giving unity gain at dc, with a low-frequency gain 3 dB
point of about 32 Hz, which is fine for my purposes, but it would be
nice to get rid of the capacitors. I may re-do the bias loop to
eliminate them, but that will take a bit of work.

Phil,
me think what you desribe is a design flaw as a differential amp has quite a
different and non balanced impedance for common mode signals. So it might be
the gain goes up and down and not the noise. It will also induce a lot of
distortion. Maybe you should also put a ferrite bead on the base of those
3906 to get rid of the negative input impedance there, seem to oscillate
with the lowish cap impedance.
Can you post a schematic, because I do not quite understand where your
feedback goes. Another thing I do not understand is this gain of 250 what
you mentioned before. Do you really have 2.5k resistors in front of the
differential amp or what?
When you want to try the circuit I posted, do not forget to put 2 diodes
4148 from emitter to base, as the opamp can pull the leg quite high when
overdriven and you do not want to damage those expensive matched Quads. I
also have some MAT04 here which could be taken as well. If you want to
reduce gain at DC, put just a 3300uF or bigger cap in series with the 10R,
since the differential voltage swing will be very small there, no need for
bipolar Cs. The input impedance is somehow higher, so a pair of 100u is
enough there. I also omitted some small(15p) caps from output to inv.IN of
those opamps.
 
P

Phil Hobbs

Jan 1, 1970
0
Phil said:
Interim report:

It's running about 1.1 nV/sqrt(Hz) in 1-11 kHz, with some line harmonics
lower down that I have to fix--probably with an on-card regulator, since
my cap multipliers don't have enough rejection down there. Adding a
50-ohm source resistance (300K) brings it back up to 1.5 nV, which is
about right.

There's something broken about the inverted cascodes, because ripping
them out dropped the noise by about 3 dB, which it shouldn't. The clue
was that bypassing their bases either to ground or to the supply
*raised* the noise by about 10 dB! Their emitters are supposed to stand
still, so this is quite odd. There's probably an optimal V_CE for the
2SD786s, too--interelectrode capacitance and all that.

I'll try using 2N4250s instead of 2N3906s for the cascodes--I probably
damaged one of them during torture testing. The feedback network has a
10k : 10 ohm divider and two 1000 uF, 16V aluminums back to back, giving
unity gain at dc, with a low-frequency gain 3 dB point of about 32 Hz,
which is fine for my purposes, but it would be nice to get rid of the
capacitors. I may re-do the bias loop to eliminate them, but that will
take a bit of work.


Okay, it's done, and all measured up. Replacing the inverted cascodes
with PN4250s fixed the excess noise problem. The measurement problem
was that my HP dynamic signal analyzer lies through its teeth about its
measurement bandwidth per point! When measuring power spectral
density--the real bandwidth is almost twice what they quote (Grr.).

I plotted the curve, and whadda ya know, it was nearly perfect but ~3 dB
higher than I calculated, everywhere. I made up a stupid little RC
lowpass/highpass box and ran it into my trusty HP 3400 true-RMS AC
voltmeter, and whadda ya know, the numbers came out 2.84 dB lower, which
is right on the money.

With two 2SD786s running at 3 mA each, beta = 750 or so on the input
device (4 uA base current), I get an astonishingly good fit to the
parameters:

eN=0.635 nV/sqrt(Hz), iN = 1.7 pA/sqrt(Hz).

(That's R_N = 373 ohms, T_N = 19.5 K. Pretty amazing for a BJT--those
Rohm guys knew what they were doing.)

The measured noise is well within 0.2 dB of that calculated from these
values everywhere from 0 to 10k source resistance. These correspond to
the shot noise of the two collector currents divided by the
transconductances, plus the Johnson noise of 7.3 ohms, or 3.6 ohms per
transistor, which is about right.

The only discrepancy from the prediction of fundamental physics is that
the noise current should be 1.6 pA/sqrt(Hz) instead of 1.7, so that's
probably caused by the base current cancellation circuit.

Amazing, this theory stuff--it really works. Off to measure some junctions!

Cheers,

Phil Hobbs
 
J

Joerg

Jan 1, 1970
0
Hello Win,
I've bought plenty of stuff through Tucker (mostly their eBay stuff),
but I prefer some of the other Internet used-equipment vendors. Most
of them also have lots of eBay activity. The best let you return the
equipment, no questions asked, and use Purchase-Order purchases, just
like all of the other true industrial-instrument sellers.

Thanks for sharing your opinion. Buying used equipment is sometimes like
buying used cars. I have done both but one has to be careful and mind
the experience of others.

Regards, Joerg
 
M

Mikko Kiviranta

Jan 1, 1970
0
Which brings me to my actual question: Are the 2SD786 and 2SD737 still
the BJT noise champions at low source resistance?

I have measured ~ 0.3 nV/rtHz white noise for a discrete
medium-power SiGe bipolar transistor at room temperature. Wrote
about it into Superconductor Science and Technology. Even if
they are intended for GHz frequencies, you can operate them
at low frequencies, too.

1/f corner is pretty high (~100 kHz) at standard bias powers
but I just found out that the corner goes down when the bias
power is reduced. Exactly how low I don't know yet (intend to
measure it), in particular at room temperature. There are some
noise spectra measured at 4.2K in a paper I just submitted to
the WOLTE-7 workshop.

Regards,
Mikko
 
W

Winfield Hill

Jan 1, 1970
0
Mikko Kiviranta wrote...
I have measured ~ 0.3 nV/rtHz white noise for a discrete
medium-power SiGe bipolar transistor at room temperature.
Wrote about it into Superconductor Science and Technology.
Even if they are intended for GHz frequencies, you can
operate them at low frequencies, too.

What are some part numbers you took measurements on? BTW,
at how low a temperature can you use these transistors?
1/f corner is pretty high (~100 kHz) at standard bias powers
but I just found out that the corner goes down when the bias
power is reduced. Exactly how low I don't know yet (intend to
measure it), in particular at room temperature. There are some
noise spectra measured at 4.2K in a paper I just submitted to
the WOLTE-7 workshop.

Mikko, what's the easiest way to read your papers or view
your presentations?
 
M

Mikko Kiviranta

Jan 1, 1970
0
What are some part numbers you took measurements on? BTW,
at how low a temperature can you use these transistors?

The Infineon BFP650 appears to be the best performer for
my purposes. The measured noise is consistent with 1.5 ohm
base-spreading resistance, whereas the Infineon Spice parameters
specify 1.03 ohm. I suppose such a small Rbb is a combination
of the geometry for a medium-power transistor and the high
doping level in the base region typical for SiGe. At least those
were my clues when I started looking for a low u_N.

I use them in liquid He ( 4.2 K ), there is no point in
going lower because of the heat load to the refrigerator.
Mikko, what's the easiest way to read your papers or view
your presentations?

You can find preprints in the XEUS web page
http://www.iki.fi/msk/xeus/
which exists (non-updated) for the moment in my personal web space
due to some ridiculous web publishing policies my institution
brought in effect recently.

Regards,
Mikko
 
Top