Determining bjt noise parameters for Spice models?
A.Iakovlev
I wanted to predict noise properties of some bjt-based circuits using
Spice. Then I've noticed that the bjt models (almost) never include the
noise parameters (AF, KF ). How can those be determined from the data
sheets?
I've tried the advice found in a Spice manual: set AF=1 and then find
KF such that it gives the requested noise figure; I've tried this for
2SC3329. Ok, I've found something like KF=5e-15 to get the specified
NF at 1 kHz with Ri=20k. Then I've varied the collector current but have
not observed the specified NF changes! I've then looked at the NF at
10Hz. Apparently, AF=1 does not fit this transistor; I've tweaked it
to AF=1.13 to get the noise figures ressembling those in the datasheet,
but without knowing what I am doing!
So:
- would it be possible that someone explains the meaning of AF and KF
(refer to an on-line doc, etc.) and how to calculate them from the
datasheet?
- in general, is the bjt model in Spice capable of precisely enough
modelling the noise behaviour of all the bjts out there (in particular:
2SC/2SA Toshiba/Hitachi low-noise series)?
Thank you.
Regards,
A.Iakovlev
Jim Thompson
"A.Iakovlev" <ade...@worldonline.fr>,
In Newsgroup: sci.electronics.cad,
Article: <3DF6537E...@worldonline.fr>,
Entitled: "Determining bjt noise parameters for Spice models?",
Wrote the following:
AF and KF only affect low frequency (1/f) noise performance.
Bipolar transistor equations for noise.........
Noise is calculated assuming a 1.0-hertz bandwidth, using the
following spectral power densities (per unit bandwidth):
parasitic resistances thermal noise (AC terms on the left of the =
sign, DC terms on the right).....
RC: Ic^2 = 4搔愁/(RC/area)
RB: Ib^2 = 4搔愁/RB
RE: Ie^2 = 4搔愁/(RE/area)
base and collector currents shot and flicker noise
IB: Ib^2 = 2敬弒b + KF弒b^AF / FREQUENCY
IC: Ic^2 = 2敬弒c
...Jim Thompson
--
| James E.Thompson, P.E. | mens |
| Analog Innovations, Inc. | et |
| Analog/Mixed-Signal ASIC's and Discrete Systems | manus |
| Phoenix, Arizona Voice:(480)460-2350 | |
| Jim-T@analog_innovations.com Fax:(480)460-2142 | Brass Rat |
| http://www.analog-innovations.com | 1962 |
For proper E-mail replies SWAP "-" and "_"
I love to cook with wine. Sometimes I even put it in the food.
Kevin Aylward
> Hello,
>
> I wanted to predict noise properties of some bjt-based circuits using
> Spice. Then I've noticed that the bjt models (almost) never include
> the noise parameters (AF, KF ). How can those be determined from the
> data sheets?
>
> I've tried the advice found in a Spice manual: set AF=1 and then find
> KF such that it gives the requested noise figure; I've tried this for
> 2SC3329. Ok, I've found something like KF=5e-15 to get the specified
> NF at 1 kHz with Ri=20k.
Then I've varied the collector current but
> have not observed the specified NF changes! I've then looked at the
> NF at 10Hz. Apparently, AF=1 does not fit this transistor; I've
> tweaked it to AF=1.13 to get the noise figures ressembling those in
> the datasheet, but without knowing what I am doing!
>
This won't really work. KF is the flicker noise co-coefficient, or 1/f
noise contribution. A low noise transistor will have a 1/f corner such
that it won't contribute much noise at 1Khz. Secondly, you should use a
zero source resistance. So, look at the data sheet graph, and set KF to
match the data sheet at a low frequency, say 10 Hz. Do this in voltage
noise terms (nv/sqrthz), NF is pretty much useless.
And... AF is the flicker noise exponent. i.e.
Vn^2 = KF/f^AF in V/sqrthz
This noise is added to the normal white noise of the transistor.
> So:
> - would it be possible that someone explains the meaning of AF and KF
> (refer to an on-line doc, etc.) and how to calculate them from the
> datasheet?
> - in general, is the bjt model in Spice capable of precisely enough
> modelling the noise behaviour of all the bjts out there (in
> particular: 2SC/2SA Toshiba/Hitachi low-noise series)?
>
Just about. You need to make sure the base resistance is set right. This
is the main variable that distinguishes transistor noise form exh other.
Other noise contributions are inherently fixed by Ic and hfe. So, after
the 1/f curve flattens out, set rbb until the flat portion reads
correctly. Do this at a highish collector current, e.g 10ma, also with r
source set to zero.
Kevin Aylward
sa...@anasoft.co.uk
http://www.anasoft.co.uk
SuperSpice, a very affordable Mixed-Mode
Windows Simulator with Schematic Capture,
Waveform Display, FFT's and Filter Design.
A.Iakovlev
Jim Thompson wrote:
> <snipped>
>
> AF and KF only affect low frequency (1/f) noise performance.
>
> Bipolar transistor equations for noise.........
>
> Noise is calculated assuming a 1.0-hertz bandwidth, using the
> following spectral power densities (per unit bandwidth):
>
> parasitic resistances thermal noise (AC terms on the left of the =
> sign, DC terms on the right).....
>
> RC: Ic^2 = 4搔愁/(RC/area)
I know that the termal noise of a resistance is that: Ic^2=4*k*T/Rc, what is
the "area" ?
>
> RB: Ib^2 = 4搔愁/RB
>
> RE: Ie^2 = 4搔愁/(RE/area)
Are RC, RB, RE the same as the RC, RB, RE specified in the bjt Spice model?
Do you mean that those noise currents are "flowing out" of B,C,E of the
following model
RB RC
B ---/\/\/\/----B'----/\/\/\/-----C
|
>
\ RB
<
|
E
?
>
> base and collector currents shot and flicker noise
>
> IB: Ib^2 = 2敬弒b + KF弒b^AF / FREQUENCY
>
> IC: Ic^2 = 2敬弒c
>
To which resistances should these currents be applied? For instance, AoE
talks about the shot current across RE..
I've tried to do some simulation in Spice. For a bjt, it calculates some
output square noise voltage values called "RB", "RC", "RE", "IBSN", "IC",
"IBFN" ; I have not yet succeded in matching the calculated values of the
thermal (to RB, RC, RE) and shot noise to those (to ISBN, IC). Can you help?
Thanks.
Best Regards,
A.Iakovlev
>
> ...Jim Thompson
<snipped>
A.Iakovlev
Kevin Aylward wrote:
> <snipped>
>
> This won't really work. KF is the flicker noise co-coefficient, or 1/f
> noise contribution. A low noise transistor will have a 1/f corner such
> that it won't contribute much noise at 1Khz. Secondly, you should use a
> zero source resistance. So, look at the data sheet graph, and set KF to
> match the data sheet at a low frequency, say 10 Hz. Do this in voltage
> noise terms (nv/sqrthz), NF is pretty much useless.
>
> And... AF is the flicker noise exponent. i.e.
>
> Vn^2 = KF/f^AF in V/sqrthz
>
I read the formula given by Jim Thompson as AF being the exponent for Ib,
not the frequency, could you confirm? Is this formula to be applied as is
to obtain the equivalent input voltage noise source in series with base?
>
> This noise is added to the normal white noise of the transistor.
>
> > So:
> > - would it be possible that someone explains the meaning of AF and KF
> > (refer to an on-line doc, etc.) and how to calculate them from the
> > datasheet?
> > - in general, is the bjt model in Spice capable of precisely enough
> > modelling the noise behaviour of all the bjts out there (in
> > particular: 2SC/2SA Toshiba/Hitachi low-noise series)?
> >
>
> Just about. You need to make sure the base resistance is set right. This
> is the main variable that distinguishes transistor noise form exh other.
> Other noise contributions are inherently fixed by Ic and hfe. So, after
> the 1/f curve flattens out, set rbb until the flat portion reads
> correctly. Do this at a highish collector current, e.g 10ma, also with r
> source set to zero.
>
I've tried to do this at 10kHz in the following way:
- common emitter;
- collector connected to a voltage source in series with a resistance;
- base directly connected to a voltage source (VSRC, small AC);
- the base voltage source adjusted to get about 10mA collector current;
- equivalent input noise calculated as V(ONOISE) on the collector divided
by the gain calculated (by Spice)
on the input AC.
Doing that with the standard 2SC3329 model and varying RB from 2 to 10 Ohm
could not reach the 0.6nV/rtHz specified by the datasheet - much lower
levels obtained; the datasheet announces rb about 2Ohm.
Did I do something wrong?
Thank you.
Best Regards,
A.Iakovlev
>
> Kevin Aylward
>
Robert Baer
I have some old but useful references concerning noise as a function
of frequency and source resistance.
Keithly: "electrometer measurements" 1972
Princeton Applied Research Corp : "How to use Noise Figure Contours"
1969
Michael Steffes: "Understanding and Improving Op Amp Noise and
Distortion" date unknown and company unknown; presumed date around 1970.
I may have others; these weer handy at a moments notice.
If you are interested, i could scan at 200 DPI to keep the sizes down
and e-mail them on request.
Start with a differential bipolar transistor pair, emitters connected
together and driven by a current source; bases are inputs and equal
resistors on collectors with signal testing across them.
That is a crude description of the input stage to the uA709; a rather
low noise op amp of those days (ignore the popcorn which was due to
impurities in the process chemicals and environment).
For a given current, one can calculate the noise voltage, and then
measure actual noise voltage; they would agree within measurement
accuracy (in the "flat" band of noise) over a rather wide range.
This appears to work over a fairly wide range of source impedances,
but as the source R increases, that "flat band" decreases materially -
and one has to add in (noise current)*(source R).
I have done this, but did little work to see about hi R noise
correlation (measured VS calculated).
The PAR cover has a graph, log freq horiz VS log Rin at 290 degrees
Kelvin (temperature is *very* important).
Looks somewhat like a topo plot of an oblong valley in the middle;
lowest regions following a downward line (to right).
Every system has its own characteristic curves, which greatly depends
on the devices used, their operating current, and the topology
(circuit).
I do not quite understand what you meant by "..(using) the standard
2SC3329 model..could not reach..much lower levels obtained.."
Sort of sounds like you got better noise levels than the reference.
And that may be correct; were your conditions identical (circuit,
operating current, temperature)?
The Rb on most bipolars is so low that it can be ignored in most
cases; Rbb' (base spreading resistance) is in a similar class.
The Re resulting from the biasing current is the major En contributor,
and this is where one gets very good correlation measured VS calculated
(take the simple single transistor case).
Now when Ie is "large" to make for low Re, then Rb and Rbb' must be
added; most data sheets will not give you any information about these
parameters.
So, measure noise with low Ic and with high Ib (low source R); in both
cases, the noise is very close to the vector sum of Rs, Re, Rb and Rbb'.
However, at the lower current, Re is higher and the base resistances
can be ignored enough to approximate (closely) their combined
(effective) value.
Armed with that info, one can calculate NF at any reasonable frequency
(say 1Khz to near Ft) and reasonable current; there is excellent
correlation in AF and RF measured values.
Sorry about the long-winded mess above; hope is of some help.
Kevin Aylward
> Kevin, Thank you.
>
> Kevin Aylward wrote:
>
>> <snipped>
>
>>
>> This won't really work. KF is the flicker noise co-coefficient, or
>> 1/f noise contribution. A low noise transistor will have a 1/f
>> corner such that it won't contribute much noise at 1Khz. Secondly,
>> you should use a zero source resistance. So, look at the data sheet
>> graph, and set KF to match the data sheet at a low frequency, say 10
>> Hz. Do this in voltage noise terms (nv/sqrthz), NF is pretty much
>> useless.
>>
>> And... AF is the flicker noise exponent. i.e.
>>
>> Vn^2 = KF/f^AF in V/sqrthz
>>
>
> I read the formula given by Jim Thompson as AF being the exponent
> for Ib, not the frequency, could you confirm? Is this formula to be
> applied as is to obtain the equivalent input voltage noise source in
> series with base?
AF is the exponent, which is what I wrote. f^AF means f to the power
(^)of AF
>>
>> This noise is added to the normal white noise of the transistor.
>>
>>> So:
>>> - would it be possible that someone explains the meaning of AF and
>>> KF (refer to an on-line doc, etc.) and how to calculate them from
>>> the datasheet?
>>> - in general, is the bjt model in Spice capable of precisely enough
>>> modelling the noise behaviour of all the bjts out there (in
>>> particular: 2SC/2SA Toshiba/Hitachi low-noise series)?
>>>
>>
>> Just about. You need to make sure the base resistance is set right.
>> This is the main variable that distinguishes transistor noise form
>> exh other. Other noise contributions are inherently fixed by Ic and
>> hfe. So, after the 1/f curve flattens out, set rbb until the flat
>> portion reads correctly. Do this at a highish collector current, e.g
>> 10ma, also with r source set to zero.
>>
>
> I've tried to do this at 10kHz in the following way:
KF needs to be set at a low frequency, say 10 Hz.
> - common emitter;
> - collector connected to a voltage source in series with a resistance;
> - base directly connected to a voltage source (VSRC, small AC);
> - the base voltage source adjusted to get about 10mA collector
> current;
> - equivalent input noise calculated as V(ONOISE) on the collector
> divided by the gain calculated (by Spice)
> on the input AC.
>
> Doing that with the standard 2SC3329 model and varying RB from 2 to
> 10 Ohm could not reach the 0.6nV/rtHz specified by the datasheet -
> much lower levels obtained; the datasheet announces rb about 2Ohm.
>
I dont understand whatvyou say here. Is the spice value too high or too
low?
A.Iakovlev
> A.Iakovlev wrote:
> > Kevin, Thank you.
> >
> > Kevin Aylward wrote:
> >
> <snipped>
> >
> >>
> >> Just about. You need to make sure the base resistance is set right.
> >> This is the main variable that distinguishes transistor noise form
> >> exh other. Other noise contributions are inherently fixed by Ic and
> >> hfe. So, after the 1/f curve flattens out, set rbb until the flat
> >> portion reads correctly. Do this at a highish collector current, e.g
> >> 10ma, also with r source set to zero.
> >>
> >
> > I've tried to do this at 10kHz in the following way:
>
> KF needs to be set at a low frequency, say 10 Hz.
>
I've started with the thermal noise, estimating that the flicker noise
vanishes at 10kHz. Was it a wrong approach?
>
> > - common emitter;
> > - collector connected to a voltage source in series with a resistance;
> > - base directly connected to a voltage source (VSRC, small AC);
> > - the base voltage source adjusted to get about 10mA collector
> > current;
> > - equivalent input noise calculated as V(ONOISE) on the collector
> > divided by the gain calculated (by Spice)
> > on the input AC.
> >
> > Doing that with the standard 2SC3329 model and varying RB from 2 to
> > 10 Ohm could not reach the 0.6nV/rtHz specified by the datasheet -
> > much lower levels obtained; the datasheet announces rb about 2Ohm.
> >
>
> I dont understand whatvyou say here. Is the spice value too high or too
> low?
>
Sorry. The value calculated (not directly by Spice but by me, from the
V(ONOISE) provided by Spice) was much lower than 0.6 nV/rtHz.
Thank you.
Regards,
A.Iakovlev
Kevin Aylward
> Kevin Aylward wrote:
>
>> A.Iakovlev wrote:
>>> Kevin, Thank you.
>>>
>>> Kevin Aylward wrote:
>>>
>> <snipped>
>>>
>>>>
>>>> Just about. You need to make sure the base resistance is set right.
>>>> This is the main variable that distinguishes transistor noise form
>>>> exh other. Other noise contributions are inherently fixed by Ic and
>>>> hfe. So, after the 1/f curve flattens out, set rbb until the flat
>>>> portion reads correctly. Do this at a highish collector current,
>>>> e.g 10ma, also with r source set to zero.
>>>>
>>>
>>> I've tried to do this at 10kHz in the following way:
>>
>> KF needs to be set at a low frequency, say 10 Hz.
>>
>
> I've started with the thermal noise, estimating that the flicker noise
> vanishes at 10kHz. Was it a wrong approach?
That's ok for rbb noise and shot noise. I as stating that you cannot
determine the 1/f co-efficient at high frequencies.
>
>>
>>> - common emitter;
>>> - collector connected to a voltage source in series with a
>>> resistance;
>>> - base directly connected to a voltage source (VSRC, small AC);
>>> - the base voltage source adjusted to get about 10mA collector
>>> current;
>>> - equivalent input noise calculated as V(ONOISE) on the collector
>>> divided by the gain calculated (by Spice)
>>> on the input AC.
>>>
>>> Doing that with the standard 2SC3329 model and varying RB from 2 to
>>> 10 Ohm could not reach the 0.6nV/rtHz specified by the datasheet -
>>> much lower levels obtained; the datasheet announces rb about 2Ohm.
>>>
>>
>> I dont understand whatvyou say here. Is the spice value too high or
>> too low?
>>
>
> Sorry. The value calculated (not directly by Spice but by me, from
> the V(ONOISE) provided by Spice) was much lower than 0.6 nV/rtHz.
>
At 290 deg, 20 ohms gives 0.57 nv/rthz. At 10ma the collector noise =
0.141 nv, which is getting close to 0.6nv when you sum the squares.
What does the spice V(INOISE) give?
fred bartoli
A.Iakovlev <ade...@worldonline.fr> a écrit dans le message :
3DF98E5B...@worldonline.fr...
> Kevin Aylward wrote:
>
> > A.Iakovlev wrote:
> > > Kevin, Thank you.
> > >
<SNIP>
> > > - common emitter;
> > > - collector connected to a voltage source in series with a resistance;
> > > - base directly connected to a voltage source (VSRC, small AC);
> > > - the base voltage source adjusted to get about 10mA collector
> > > current;
> > > - equivalent input noise calculated as V(ONOISE) on the collector
> > > divided by the gain calculated (by Spice)
> > > on the input AC.
> > >
> > > Doing that with the standard 2SC3329 model and varying RB from 2 to
> > > 10 Ohm could not reach the 0.6nV/rtHz specified by the datasheet -
> > > much lower levels obtained; the datasheet announces rb about 2Ohm.
> > >
> >
> > I dont understand whatvyou say here. Is the spice value too high or too
> > low?
> >
>
> Sorry. The value calculated (not directly by Spice but by me, from the
> V(ONOISE) provided by Spice) was much lower than 0.6 nV/rtHz.
>
Just in case : beware that usually Spices won't give you the result in terms
of V/rtHz but in terms of V^2/Hz. You have to square root the result you
have.
Fred.
Kevin Aylward
But not in SuperSpice. Its read out is in v/rthz:-)
A.Iakovlev
>
<snipped>
>
> >>
> >>> - common emitter;
> >>> - collector connected to a voltage source in series with a
> >>> resistance;
> >>> - base directly connected to a voltage source (VSRC, small AC);
> >>> - the base voltage source adjusted to get about 10mA collector
> >>> current;
> >>> - equivalent input noise calculated as V(ONOISE) on the collector
> >>> divided by the gain calculated (by Spice)
> >>> on the input AC.
> >>>
> >>> Doing that with the standard 2SC3329 model and varying RB from 2 to
> >>> 10 Ohm could not reach the 0.6nV/rtHz specified by the datasheet -
> >>> much lower levels obtained; the datasheet announces rb about 2Ohm.
> >>>
> >>
> >> I dont understand whatvyou say here. Is the spice value too high or
> >> too low?
> >>
> >
> > Sorry. The value calculated (not directly by Spice but by me, from
> > the V(ONOISE) provided by Spice) was much lower than 0.6 nV/rtHz.
> >
>
> At 290 deg, 20 ohms gives 0.57 nv/rthz. At 10ma the collector noise =
> 0.141 nv, which is getting close to 0.6nv when you sum the squares.
>
> What does the spice V(INOISE) give?
A) the initial, stock model: has RC=5.9, RB and RE missing (so =0); with
collector current of about 10mA, collector voltage about 9V, collector load
of 1kOhm, V(INOISE) is about 0.14 nV/rtHz.
In the noise analysis printout, Spice gives "RC=1,800e-22
IC=2,938e-15".
B) RB=10 added to the model, in the same conditions, V(INOISE) is about
0.43 nV/rtHz.
BTW, the datasheet is located here:
http://www.semicon.toshiba.co.jp/en/bucat_8/bucat_3/bucat_2/td_63/TD.pdf
It says, in particular, "low base spreading resistance: rbb ' = 2.0 Ohm
(typ.)".
Kevin Aylward
I don't believe it. It can't possible be 2 ohms. The NF graphs show
around 4 db at 10ma at ~12 ohms. This means rbb' has to be greater then
12 ohms.
Just set rbb' in the model to give the 0.65nv.
A.Iakovlev
> A.Iakovlev wrote:
> > Kevin Aylward wrote:
> >
<snipped>
>
> >> What does the spice V(INOISE) give?
> >
> > A) the initial, stock model: has RC=5.9, RB and RE missing (so =0);
> > with collector current of about 10mA, collector voltage about 9V,
> > collector load of 1kOhm, V(INOISE) is about 0.14 nV/rtHz.
> > In the noise analysis printout, Spice gives "RC=1,800e-22
> > IC=2,938e-15".
> >
> > B) RB=10 added to the model, in the same conditions, V(INOISE) is
> > about
> > 0.43 nV/rtHz.
> >
> > BTW, the datasheet is located here:
> >
> http://www.semicon.toshiba.co.jp/en/bucat_8/bucat_3/bucat_2/td_63/TD.pdf
> > It says, in particular, "low base spreading resistance: rbb ' = 2.0
> > Ohm (typ.)".
>
> I don't believe it. It can't possible be 2 ohms. The NF graphs show
> around 4 db at 10ma at ~12 ohms. This means rbb' has to be greater then
> 12 ohms.
>
> Just set rbb' in the model to give the 0.65nv.
>
I am yet far from mastering the Spice bjt model, but I suspect that the RB
parameter is not exactly rbb'....
But I've adjusted RB, then KF and checked for several Ic, Ri - the updated
model 2SC3329 showed a behaviour ressemling that in the datasheet! I needed
to set RB=24, KF=0.38e-16 (AF=1).
I've then done the same thing for 2SC1775 ; obtained RB=200, KF=1e-17.
But the third stock model, for 2SA876, behaved differently: even without RB
in the model, it exhibited the thermal noise slightly exceeding the
datasheet specs... In order to set up the flicker noise I had to add some
RB=6, and large KF=1e-10.
Thanks!!
Regards,
A.Iakovlev
>
> Kevin Aylward
A.Iakovlev
> I am yet far from mastering the Spice bjt model, but I suspect that the RB
> parameter is not exactly rbb'....
>
> But I've adjusted RB, then KF and checked for several Ic, Ri - the updated
> model 2SC3329 showed a behaviour ressemling that in the datasheet! I needed
> to set RB=24, KF=0.38e-16 (AF=1).
>
> I've then done the same thing for 2SC1775 ; obtained RB=200, KF=1e-17.
>
> But the third stock model, for 2SA876, behaved differently: even without RB
> in the model, it exhibited the thermal noise slightly exceeding the
> datasheet specs... In order to set up the flicker noise I had to add some
> RB=6, and large KF=1e-10.
>
Well, I've then simulated my circuit (a voltage follower) and have obtained
huge noise figures (about 0.2 uV/sqHz @ 5kHz)! Something looks wrong with the
model. I have then set RB and KF for 2SA876 the same as for 2SC1775 (its
complementary), and the circuit noise figure became reasonable... But
I believe that the model could still be improved.
Thanks.
A.Iakovlev
Don Pearce
<ade...@worldonline.fr> wrote:
>Well, I've then simulated my circuit (a voltage follower) and have obtained
>huge noise figures (about 0.2 uV/sqHz @ 5kHz)! Something looks wrong with the
>model. I have then set RB and KF for 2SA876 the same as for 2SC1775 (its
>complementary), and the circuit noise figure became reasonable... But
>I believe that the model could still be improved.
>
>Thanks.
>
>A.Iakovlev
If you do the calculation again, but using uV/rootHz, what do you see?
d
_____________________________
Telecommunications consultant
http://www.pearce.uk.com
A.Iakovlev
> On Sat, 14 Dec 2002 10:43:52 +0100, "A.Iakovlev"
> <ade...@worldonline.fr> wrote:
>
> >Well, I've then simulated my circuit (a voltage follower) and have obtained
> >huge noise figures (about 0.2 uV/sqHz @ 5kHz)! Something looks wrong with the
> >model. I have then set RB and KF for 2SA876 the same as for 2SC1775 (its
> >complementary), and the circuit noise figure became reasonable... But
> >I believe that the model could still be improved.
> >
> >Thanks.
> >
> >A.Iakovlev
>
> If you do the calculation again, but using uV/rootHz, what do you see?
>
Yes, I meant sqHz = square root of Hertz. The circuit noise density (input or
output) became about 3.25 nV / Sqrt(Hz) in the 100Hz-10kHz band. I've made a
typo, it's 2SA872.
A.I.
A.Iakovlev
"A.Iakovlev" wrote:
> Jim, Thank you. I have a few follow-up questions...
>
> Jim Thompson wrote:
>
> > <snipped>
> >
> > AF and KF only affect low frequency (1/f) noise performance.
> >
> > Bipolar transistor equations for noise.........
> >
> > Noise is calculated assuming a 1.0-hertz bandwidth, using the
> > following spectral power densities (per unit bandwidth):
> >
> > parasitic resistances thermal noise (AC terms on the left of the =
> > sign, DC terms on the right).....
> >
> > RC: Ic^2 = 4搔愁/(RC/area)
> >
> > RB: Ib^2 = 4搔愁/RB
> >
> > RE: Ie^2 = 4搔愁/(RE/area)
>
> Are RC, RB, RE the same as the RC, RB, RE specified in the bjt Spice model?
>
>
> RB RC
> B ---/\/\/\/----B'----/\/\/\/-----C
> |
> >
> \ RE (corrected)
> <
> |
> E
>
> >
> > base and collector currents shot and flicker noise
> >
> > IB: Ib^2 = 2敬弒b + KF弒b^AF / FREQUENCY
> >
> > IC: Ic^2 = 2敬弒c
> >
>
>
My current idea is the following; the noise currents (AC terms) should be
applied:
"RC", "IC" : across RC, RE, Rload (between Collector and ground);
"RB", "IB" : across RB, RE, Rinput (between Base and ground);
"RE" : across RE, RC and Rload - multiplied by Alpha, RB and Rinput- mutiplied
by (1-Alpha);
Is this correct?
Thanks.
A.I.
> >
> > ...Jim Thompson
>
> <snipped>
A.Iakovlev
> "A.Iakovlev" wrote:
> >
> <snipped>
>
> I have some old but useful references concerning noise as a function
> of frequency and source resistance.
> Keithly: "electrometer measurements" 1972
> Princeton Applied Research Corp : "How to use Noise Figure Contours"
> 1969
> Michael Steffes: "Understanding and Improving Op Amp Noise and
> Distortion" date unknown and company unknown; presumed date around 1970.
> I may have others; these weer handy at a moments notice.
> If you are interested, i could scan at 200 DPI to keep the sizes down
> and e-mail them on request.
>
Thank you. At this time, I would like just to understand the basic things about
the bjts' noise.
>
> Start with a differential bipolar transistor pair, emitters connected
> together and driven by a current source; bases are inputs and equal
> resistors on collectors with signal testing across them.
> That is a crude description of the input stage to the uA709; a rather
> low noise op amp of those days (ignore the popcorn which was due to
> impurities in the process chemicals and environment).
> For a given current, one can calculate the noise voltage, and then
> measure actual noise voltage; they would agree within measurement
> accuracy (in the "flat" band of noise) over a rather wide range.
> This appears to work over a fairly wide range of source impedances,
> but as the source R increases, that "flat band" decreases materially -
> and one has to add in (noise current)*(source R).
> I have done this, but did little work to see about hi R noise
> correlation (measured VS calculated).
Do you mean the correlation between the different noise sources? The thermal
noise of RB,RC,RE, shot noise of Ic and Ib?
>
> The PAR cover has a graph, log freq horiz VS log Rin at 290 degrees
> Kelvin (temperature is *very* important).
> Looks somewhat like a topo plot of an oblong valley in the middle;
> lowest regions following a downward line (to right).
> Every system has its own characteristic curves, which greatly depends
> on the devices used, their operating current, and the topology
> (circuit).
>
> I do not quite understand what you meant by "..(using) the standard
> 2SC3329 model..could not reach..much lower levels obtained.."
> Sort of sounds like you got better noise levels than the reference.
>
Yes, obtained lower noise levels than what said the datasheet.
> And that may be correct; were your conditions identical (circuit,
> operating current, temperature)?
Not sure about the circuit, but the datasheet does not specify much. I used
common emitter configuration, a resistance as collector load, the Vce, Ice,
Temperature as specified in the ds.
>
> The Rb on most bipolars is so low that it can be ignored in most
> cases; Rbb' (base spreading resistance) is in a similar class.
>
> The Re resulting from the biasing current is the major En contributor,
> and this is where one gets very good correlation measured VS calculated
> (take the simple single transistor case).
> Now when Ie is "large" to make for low Re, then Rb and Rbb' must be
> added; most data sheets will not give you any information about these
> parameters.
> So, measure noise with low Ic and with high Ib (low source R); in both
> cases, the noise is very close to the vector sum of Rs, Re, Rb and Rbb'.
> However, at the lower current, Re is higher and the base resistances
> can be ignored enough to approximate (closely) their combined
> (effective) value.
> Armed with that info, one can calculate NF at any reasonable frequency
> (say 1Khz to near Ft) and reasonable current; there is excellent
> correlation in AF and RF measured values.
>
The question: how to calculate re, rb, rbb', rc?
>
> Sorry about the long-winded mess above; hope is of some help.
>
> > Doing that with the standard 2SC3329 model and varying RB from 2 to 10 Ohm
> > could not reach the 0.6nV/rtHz specified by the datasheet - much lower
> > levels obtained; the datasheet announces rb about 2Ohm.
Thanks.
A.I.
Kevin Aylward
> I do not quite understand what you meant by "..(using) the standard
> 2SC3329 model..could not reach..much lower levels obtained.."
> Sort of sounds like you got better noise levels than the reference.
> And that may be correct; were your conditions identical (circuit,
> operating current, temperature)?
> The Rb on most bipolars is so low that it can be ignored in most
> cases; Rbb' (base spreading resistance) is in a similar class.
I disagree that rbb should be ignored. BC109s might be 400 ohms, 2n4401s
might be 12 ohms, others might vary from 10 to 500 ohms. For low
resistance sources, rbb *is* the *dominate* noise source in an optimally
designed amplifier. e.g. as I noted, shot noise might be only 0.14nv at
10 ma, verses a 50 ohm resistor at 0.89nv.
> The Re resulting from the biasing current is the major En
> contributor,
Only at low collector currents, which are only used for high source
resistances (well excepting low power, not noise, as a strong
reqiement).
> and this is where one gets very good correlation
> measured VS calculated (take the simple single transistor case).
> Now when Ie is "large" to make for low Re, then Rb and Rbb' must be
> added; most data sheets will not give you any information about these
> parameters.
Re is a bit confusing, this usually refers to an external emitter
resistor. re, imo, is usually clearer for the dynamic transistor emitter
resistance.
> So, measure noise with low Ic and with high Ib (low source R); in
> both cases, the noise is very close to the vector sum of Rs, Re, Rb
> and Rbb'. However, at the lower current, Re is higher and the base
> resistances can be ignored enough to approximate (closely) their
> combined (effective) value.
> Armed with that info, one can calculate NF at any reasonable
> frequency (say 1Khz to near Ft) and reasonable current; there is
> excellent correlation in AF and RF measured values.
>
> Sorry about the long-winded mess above; hope is of some help.
But there is not much point in doing tests at low currents at all. *All*
transistors have the *same* white noise at any current, except for rbb
generated noise and the assumtion that typically hfes are 100-400. All
you want to detrmine is what rbb is.
e.g. noise due to collector current accross re is:
V1noise=re.sqrt(2qIc) = sqrt(2q/40^2IC)
noise due to base current accross Rs is:
v2noise=Rs.sqrt(2q.IC/hfe)
A rough optimum being to set re=Rs/sqrt(hfe)
Noting from this that the noise variation due to spreads of hfe is
rather low, so hfe can usually be ignored.
It is the fact that transistors have a large spread in rbb' from device
to device that actually differentiates low noise devices from any othe
device. Its inherent that if one tries to chose a low noise device, then
one is selecting principly for *low* rbb. Nothing else matters much.
Summary:
A low noise transistor is essentailly a transistor with low rbb, and low
1/f noise. End of story.
John S. Dyson
"Kevin Aylward" <ke...@anasoft.co.uk> wrote in message news:MZOK9.1335$TY1....@newsfep1-gui.server.ntli.net...
>
> It is the fact that transistors have a large spread in rbb' from device
> to device that actually differentiates low noise devices from any othe
> device. Its inherent that if one tries to chose a low noise device, then
> one is selecting principly for *low* rbb. Nothing else matters much.
>
> Summary:
>
> A low noise transistor is essentailly a transistor with low rbb, and low
> 1/f noise. End of story.
>
low/middle freqs, and also hfe(Beta) does count for high source resistance
applications, especially in cases where the Ic might be chosen to be high
for bandwidth or other reasons. There are cases where an rbb decrease
from 50 down to 10 will cause essentially nil difference in noise, but a doubling
of beta will make a significant difference. There are also cases vice/versa.
What you say for maybe a 100 ohm source is generally true, but that isn't
always the case.
John
Kevin Aylward
> "Kevin Aylward" <ke...@anasoft.co.uk> wrote in message
> news:MZOK9.1335$TY1....@newsfep1-gui.server.ntli.net...
>>
>> It is the fact that transistors have a large spread in rbb' from
>> device to device that actually differentiates low noise devices from
>> any othe device. Its inherent that if one tries to chose a low noise
>> device, then one is selecting principly for *low* rbb. Nothing else
>> matters much.
>>
>> Summary:
>>
>> low 1/f noise. End of story.
>>
> I'd agree with you given the following qualification: you are
> speaking of low/middle freqs, and also hfe(Beta) does count for high
> source resistance applications, especially in cases where the Ic
> might be chosen to be high for bandwidth or other reasons.
Ahmmm. Did you read my post? I did address the hfe issue, to wit:
*********
A rough optimum being to set re=Rs/sqrt(hfe)
Noting from this that the noise variation due to spreads of hfe is
rather low, so hfe can usually be ignored.
*********
It is why I said "essentially" and "usually". I always cover my arse:-)
For instance, if the source is highly inductive, this can make the
effect of base current noise much more significant.
> There
> are cases where an rbb decrease from 50 down to 10 will cause
> essentially nil difference in noise,
>but a doubling of beta will make
> a significant difference. There are also cases vice/versa.
>
Sure if the source is >> rbb.
But that does not really change the statement that a low noise
transistor is essentially a transistor with low rbb. If you chose *any*
transistor *type*, it will usually have a hfe from 100 to 400 say. Its
not often you really want to use a superbeta transistor because of ease
of its availability. So, given that you set the circuit up for say a hfe
of 200, then slap in different transistors, the noise won't really
change much. The idea here is that in reality, you don't have much
choice in actually selecting a transistor for hfe in that most are all
in the same ballpark, however, rbb might vary from 10 to 500 ohms from
device to device. In fact, 2n3055s have been used for MC inputs as their
rbbs are around 2 ohms, with their hfes being rather low.
Again, its a *practical* selection issue, most common transistors have
about the same hfe, so hfe don't make much of a selection guide. Sure,
there are a few down at 30, and some at 800, but overall, hfe is not a
significant selection criteria imo, for most low noise design.
There is some other history to this. It is/was often quoted that the
BC109C is a low noise device because it has a relatively high hfe of
500. However, its rbb is around 400 ohms, making it quite a noisy device
for low source resistances.
> What you say for maybe a 100 ohm source is generally true, but that
> isn't always the case.
>
Well, for high impedance sources, one is usually much better using a
jfet, its certainly what I recommend for say, a guitar preamp.
JD
"Kevin Aylward" <ke...@anasoft.co.uk> wrote in message news:E3XK9.162$cy2.14496@newsfep2-gui...
> John S. Dyson wrote:
> > "Kevin Aylward" <ke...@anasoft.co.uk> wrote in message
> > news:MZOK9.1335$TY1....@newsfep1-gui.server.ntli.net...
> >>
> >> It is the fact that transistors have a large spread in rbb' from
> >> device to device that actually differentiates low noise devices from
> >> any othe device. Its inherent that if one tries to chose a low noise
> >> device, then one is selecting principly for *low* rbb. Nothing else
> >> matters much.
> >>
> >> Summary:
> >>
> >> A low noise transistor is essentially a transistor with low rbb, and
> >> low 1/f noise. End of story.
> >>
> > I'd agree with you given the following qualification: you are
> > speaking of low/middle freqs, and also hfe(Beta) does count for high
> > source resistance applications, especially in cases where the Ic
> > might be chosen to be high for bandwidth or other reasons.
>
>
> Ahmmm. Did you read my post? I did address the hfe issue, to wit:
>
low 1/f noise) as being the key... Remember your 'end of story'
statement?
> *********
> A rough optimum being to set re=Rs/sqrt(hfe)
>
> Noting from this that the noise variation due to spreads of hfe is
> rather low, so hfe can usually be ignored.
>
Maybe, where you cannot use high beta parts, you can ignore it, but
you aren't always limited to medium beta parts.
>
> For instance, if the source is highly inductive, this can make the
> effect of base current noise much more significant.
>
This makes low rbb insufficient in many cases, doesn't it? Remember,
your 'end of story' comment?
>
> Sure if the source is >> rbb.
>
It only has to be >rbb (not >>rbb) in cases where doubling the Beta
is useful.
>
> But that does not really change the statement that a low noise
> transistor is essentially a transistor with low rbb. If you chose *any*
> transistor *type*, it will usually have a hfe from 100 to 400 say. Its
> not often you really want to use a superbeta transistor because of ease
> of its availability.
>
Your permise about availability is wrong. Transistors with high beta
(not really superbeta) are easily available (I buy them by the 1000's easily,
at very low price, typical of LF bipolars today.) It makes almost no
sense to purchase BJTs from suppliers anymore in small quantity
(at 2-8 cents apiece), because of UPS/FedEX charges.
There are MANY cases of an rbb (the effective rbb for noise) being 50
instead of 10 is better if the beta is 600 instead of 200. rbb of 50
isn't all that uncommon, nor is a minimum beta of 400. Super
low rbb of 5-10 doesn't buy that much noise performance, even with
common low-z audio source of 150-200 ohms. For the normal cases
of super low noise applications, you'll often do better with paralleled
sets of higher beta transistors. Some transistors do have excessively
high (effective) rbb, but I am not speaking of those either -- just as
I am not speaking of BJTs with Beta of 50.
Super-beta is a term that applies to parts that generally have very
high beta, but serious tradeoffs in behavior are also made (e.g. the
ZTX689 might be marginally considered a discrete superbeta), but a very
common part like an MPSA18 isn't really a horrid set of tradeoffs. It isn't
perfect, and I wouldn't necessarily use it on a low-z mic preamp
(without a transformer, or significant parallleling), but a device with
even with an rbb of 10 and Beta of 100-400 isn't going to be able to
compete where the MPSA18 type part does well. (Assuming
good 1/f performance.)
There are indeed some especially 'nice' parts out there specified
for low impedance noise, but claiming that many are readily available
is rather specious.
For availability, I buy the MPSA18 by the bag, just like the 2n3904. There
are some parts that are much less available (not using normal suppliers, or
huge quantities) that have interesting low noise characteristics for
high current/low impedances, but those are yet another class of low noise
that is optimized for different applications.
John
Kevin Aylward
> "Kevin Aylward" <ke...@anasoft.co.uk> wrote in message
> news:E3XK9.162$cy2.14496@newsfep2-gui...
>> John S. Dyson wrote:
>>> "Kevin Aylward" <ke...@anasoft.co.uk> wrote in message
>>> news:MZOK9.1335$TY1....@newsfep1-gui.server.ntli.net...
>>>>
>>>> It is the fact that transistors have a large spread in rbb' from
>>>> device to device that actually differentiates low noise devices
>>>> from any othe device. Its inherent that if one tries to chose a
>>>> low noise device, then one is selecting principly for *low* rbb.
>>>> Nothing else matters much.
>>>>
>>>> Summary:
>>>>
>>>> A low noise transistor is essentially a transistor with low rbb,
>>>> and low 1/f noise. End of story.
>>>>
>>> I'd agree with you given the following qualification: you are
>>> speaking of low/middle freqs, and also hfe(Beta) does count for high
>>> source resistance applications, especially in cases where the Ic
>>> might be chosen to be high for bandwidth or other reasons.
>>
>>
>> Ahmmm. Did you read my post? I did address the hfe issue, to wit:
>>
> I did, but you said in an unqualified way that rbb (and the obvious
> low 1/f noise) as being the key... Remember your 'end of story'
> statement?
But there was an epilogue to the story.
John S. Dyson
"Kevin Aylward" <ke...@anasoft.co.uk> wrote in message news:7V3L9.1619$me3....@newsfep1-gui.server.ntli.net...
> JD wrote:
> > "Kevin Aylward" <ke...@anasoft.co.uk> wrote in message
> > news:E3XK9.162$cy2.14496@newsfep2-gui...
> >> John S. Dyson wrote:
> >>> "Kevin Aylward" <ke...@anasoft.co.uk> wrote in message
> >>> news:MZOK9.1335$TY1....@newsfep1-gui.server.ntli.net...
> >>>>
> >>>> It is the fact that transistors have a large spread in rbb' from
> >>>> device to device that actually differentiates low noise devices
> >>>> from any othe device. Its inherent that if one tries to chose a
> >>>> low noise device, then one is selecting principly for *low* rbb.
> >>>> Nothing else matters much.
> >>>>
> >>>> Summary:
> >>>>
> >>>> A low noise transistor is essentially a transistor with low rbb,
> >>>> and low 1/f noise. End of story.
> >>>>
> >>> I'd agree with you given the following qualification: you are
> >>> speaking of low/middle freqs, and also hfe(Beta) does count for high
> >>> source resistance applications, especially in cases where the Ic
> >>> might be chosen to be high for bandwidth or other reasons.
> >>
> >>
> >> Ahmmm. Did you read my post? I did address the hfe issue, to wit:
> >>
> > I did, but you said in an unqualified way that rbb (and the obvious
> > low 1/f noise) as being the key... Remember your 'end of story'
> > statement?
>
> But there was an epilogue to the story.
>
prejudicially pronounce 'rbb and 1/f' as being the key.
Rather, I tend to prefer these ideas for midband:
en^2 = 4 * k * T * rbb(eff) + 2 * q * Ic * re^2
in^2 = 2 * q * Ib
The input noise voltage is then limited by the
en > re * sqrt(2*q*Ic) and en > sqrt(4*k*T*rbb(eff)) terms.
The input noise current is purely defined (assuming no parasitic
noise sources) by the operating current and Beta. (Again,
ignoring 1/f.)
So, the input noise power is inversely related to the sqrt(B)
and contributions between rbb(eff) and the operating point
physical limitations are of the same kind (often of the same
order in discrete parts, at normal 100ua to 10ma bias.)
NF is kind of 'bogus', but if properly designed, it is useful for
getting a qualitative view of 'low noise'. The noise factor
for an ideal transistor is about:
nfactor (ideal rsource) = 1 + sqrt(1 + 2 * rbb / re) / sqrt (B).
Once the rbb drops much below re, then the only thing that you
can do is to increase beta so as to decrease total noise. rbb
is of diminishing returns.
When using rbb as a criteria for selecting a low noise device,
even for low impedances where the noise is still significant, choosing
increased Beta can still decrease the noise. This is still important
when trying to decrease the en by running at high Ic for a low
rbb device. Increased beta will help to counter the effect of
high Ic on the base noise current.
I don't mean to claim that low rbb components aren't often desirable.
John
A.Iakovlev
|snipped|
> It is probably best to consider both the en and in type effects, and not
> prejudicially pronounce 'rbb and 1/f' as being the key.
>
> Rather, I tend to prefer these ideas for midband:
>
> en^2 = 4 * k * T * rbb(eff) + 2 * q * Ic * re^2
> in^2 = 2 * q * Ib
>
Of what order should the Spice RE be expected? It is also very often missing from the models...
If comparable to RB, it would contribute comparable thermal noise (voltage term) as well...
>
> The input noise voltage is then limited by the
> en > re * sqrt(2*q*Ic) and en > sqrt(4*k*T*rbb(eff)) terms.
>
> The input noise current is purely defined (assuming no parasitic
> noise sources) by the operating current and Beta. (Again,
> ignoring 1/f.)
>
> So, the input noise power is inversely related to the sqrt(B)
> and contributions between rbb(eff) and the operating point
> physical limitations are of the same kind (often of the same
> order in discrete parts, at normal 100ua to 10ma bias.)
>
> NF is kind of 'bogus', but if properly designed, it is useful for
> getting a qualitative view of 'low noise'. The noise factor
> for an ideal transistor is about:
>
> nfactor (ideal rsource) = 1 + sqrt(1 + 2 * rbb / re) / sqrt (B).
>
> Once the rbb drops much below re, then the only thing that you
> can do is to increase beta so as to decrease total noise. rbb
> is of diminishing returns.
>
> When using rbb as a criteria for selecting a low noise device,
And low RE?
>
> even for low impedances where the noise is still significant, choosing
> increased Beta can still decrease the noise. This is still important
> when trying to decrease the en by running at high Ic for a low
> rbb device. Increased beta will help to counter the effect of
> high Ic on the base noise current.
>
> I don't mean to claim that low rbb components aren't often desirable.
> John
Thank you.
A.I.
Kevin Aylward
If by Re you mean the parasitic emitter resistance, then this is usually
less than 1 ohm, so can be ignored noise wise.
However, re is the dynamic resistance of the emitter circuit and is
given by re=1/40Ic at room temperature. re does not generate thermal
noise but it does generate a noise voltage due to emitter shot noise,
v(re) =re.sqrt(2.q.Ie).
So, for approximate optimal noise set re=Rs/sqrt(hfe), then Ie=1/40re
JD
"A.Iakovlev" <ade...@worldonline.fr> wrote in message news:3DFE3F2C...@worldonline.fr...
> "John S. Dyson" wrote:
>
> |snipped|
>
> > It is probably best to consider both the en and in type effects, and not
> > prejudicially pronounce 'rbb and 1/f' as being the key.
> >
> > Rather, I tend to prefer these ideas for midband:
> >
> > en^2 = 4 * k * T * rbb(eff) + 2 * q * Ic * re^2
> > in^2 = 2 * q * Ib
> >
>
> Of what order should the Spice RE be expected? It is also very often missing from the models...
> If comparable to RB, it would contribute comparable thermal noise (voltage term) as well...
>
made up from the 0.025/Ic (at room temperature.) This isn't a physical (or
parasitic) resistor, but is of the dynamic sort. There is some physical resistance,
but that shouldn't be much of an issue at currents that are used for low noise.
>
> >
> > When using rbb as a criteria for selecting a low noise device,
>
> And low RE?
>
You certainly want a low physical RE (as a parasitic or contact resistance),
but the re value is predominantly the dynamic resistance.
John