Maker Pro
Maker Pro

Bipolar Noise at HF

S

Steve Kavanagh

Jan 1, 1970
0
I want to design some low noise bipolar amplifiers at HF (say 1-30
MHz,typically 5 MHz). But almost no data sheets contain info on
transistor noise performance at these frequencies. I would appreciate
any info on how to estimate/predict HF noise figure. Some specific
questions:

1. I could use low noise UHF/microwave transistors if I can keep them
stable. But usually they have no noise figure characterization below
200 MHz or so and complete noise parameters are usually given only
above 500 MHz or so. Is it practical to estimate HF performance from
this data ?

2. I have run across many data sheets for low noise transistors with
ft of a few GHz where the noise figure flattens out below 300 MHz or
so, which would appear to mean the behaviour is not much affected by
the device capacitances and I should be able to use the sort of
approximations typically used at audio frequencies. But some of these
also show (for a few mA Ic) the NF with optimum and 50 ohm source
impedance to be the same in the flat region, which doesn't seem to
make any sense to me. I would expect lower NF at higher source
impedance. What is going on here ?

3. How much can I trust manufacturers' SPICE models ? I've noticed a
tendency for RB to be a round number like 10 ohms which makes me
suspect that (a) the model is not intended to work well for noise
figure or (b) the unit-to-unit noise variation is too much for the
modellers to worry about getting an exact RB, or some combination of
these.

4. Have you any suggestions for commonly available devices which give
really low noise at HF (better than the venerable 2N5109, for
example), especially at low currents ?

Steve

Note: The email address above may not work...you can easily figure out
what will !
 
R

Robert Baer

Jan 1, 1970
0
Steve said:
I want to design some low noise bipolar amplifiers at HF (say 1-30
MHz,typically 5 MHz). But almost no data sheets contain info on
transistor noise performance at these frequencies. I would appreciate
any info on how to estimate/predict HF noise figure. Some specific
questions:

1. I could use low noise UHF/microwave transistors if I can keep them
stable. But usually they have no noise figure characterization below
200 MHz or so and complete noise parameters are usually given only
above 500 MHz or so. Is it practical to estimate HF performance from
this data ?

2. I have run across many data sheets for low noise transistors with
ft of a few GHz where the noise figure flattens out below 300 MHz or
so, which would appear to mean the behaviour is not much affected by
the device capacitances and I should be able to use the sort of
approximations typically used at audio frequencies. But some of these
also show (for a few mA Ic) the NF with optimum and 50 ohm source
impedance to be the same in the flat region, which doesn't seem to
make any sense to me. I would expect lower NF at higher source
impedance. What is going on here ?

3. How much can I trust manufacturers' SPICE models ? I've noticed a
tendency for RB to be a round number like 10 ohms which makes me
suspect that (a) the model is not intended to work well for noise
figure or (b) the unit-to-unit noise variation is too much for the
modellers to worry about getting an exact RB, or some combination of
these.

4. Have you any suggestions for commonly available devices which give
really low noise at HF (better than the venerable 2N5109, for
example), especially at low currents ?

Steve

Note: The email address above may not work...you can easily figure out
what will !

1) You are talking about what my electronic friends ani call "DC
frequencies", since they are so low.
2) Those transistors are made for (true) high frtequencies, and the NF
is low from audio to the max specified.
50 ohms is the typical impedance for RF work, with 75 ohms being the
second common impedance, and NF at 50 ohmjs can be re-figured for 75
ohms.
3) Perhaps the Rb is specified at a given emitter current?

For low noise at low impedances, the Rbb' or base spreading
resistance, must be as low as possible. As the source impedance goes
higher, this need becomes moot, and Re becomes a major consideration
(actually, it is beta*Re).
The lower the current, to a certain degree, the lower the noise figure
at the higher impedance.
And at a given source impedance, the noise figure as a function of
frequency, when plotted looks like a bathtub; the same is true if one
plots noise figure as a function of source impedance at a given
frequency.
So, ther is an optimum range of frequencies (ie: bandwidth) and source
impedances for minimum noise.

I have various references from PAR and Analog printed in the 60's and
70's that may be useful.
 
S

Steve Kavanagh

Jan 1, 1970
0
Robert Baer said:
1) You are talking about what my electronic friends ani call "DC
frequencies", since they are so low.
2) Those transistors are made for (true) high frtequencies, and the NF
is low from audio to the max specified.

That is my assumption and why I am considering them (i.e. 1/f corner
frequency is well below the range I am concerned with). There may be
lower ft devices which are just as good at HF but there are no NF
specs for them so it is hard to identify them. I'm not sure if the
1/f corner frequency might be too high in GaAsFETs.
50 ohms is the typical impedance for RF work, with 75 ohms being the
second common impedance, and NF at 50 ohmjs can be re-figured for 75
ohms.

The applications I have are at various source impedances in the
10-1000 ohm range. Right now I am looking at a 5 MHz, 120 ohm source
impedance example, where I hope that a device may have a bit better
noise figure than at 200 MHz and 50 ohms, but am not sure how to
determine how much better.
3) Perhaps the Rb is specified at a given emitter current?

For low noise at low impedances, the Rbb' or base spreading
resistance, must be as low as possible. As the source impedance goes
higher, this need becomes moot, and Re becomes a major consideration
(actually, it is beta*Re).
The lower the current, to a certain degree, the lower the noise figure
at the higher impedance.
And at a given source impedance, the noise figure as a function of
frequency, when plotted looks like a bathtub; the same is true if one
plots noise figure as a function of source impedance at a given
frequency.
So, ther is an optimum range of frequencies (ie: bandwidth) and source
impedances for minimum noise.

I am familiar with most of this, but if you care to expand on the Re
impact I would appreciate it.
I have various references from PAR and Analog printed in the 60's and
70's that may be useful.

My currently preferred reference on (low freq.) noise is P.J.Fish,
Electronic Noise and Low Noise Design, McGraw-Hill, 1994, which uses
the common hybrid-pi model with no explicit Re with the result that it
doesn't enter into the equations. Guess I need to think a bit about
feedback configurations (or common base with Re added to source
resistance).

Steve
 
J

John S. Dyson

Jan 1, 1970
0
That is my assumption and why I am considering them (i.e. 1/f corner
frequency is well below the range I am concerned with). There may be
lower ft devices which are just as good at HF but there are no NF
specs for them so it is hard to identify them. I'm not sure if the
1/f corner frequency might be too high in GaAsFETs.
For the best 1/f corner for a UHF transistor, you might look at
the SiGe components. Keeping them stable might be difficult, so
you might need to avoid 'tuned circuits' per se. The high beta
of the SiGe will tend to help the current noise also (give you
a wider range of low noise operation.)

Also, some of the (normal UHF BJT) NEC transistors are specified
(through their spice models) for middle frequency (HF/VHF) behavior.
Some KF/AF parameters are modeled, and some of their components are
specified for low noise oscillators. Take a look at www.cel.com
or somesuch.
The applications I have are at various source impedances in the
10-1000 ohm range. Right now I am looking at a 5 MHz, 120 ohm source
impedance example, where I hope that a device may have a bit better
noise figure than at 200 MHz and 50 ohms, but am not sure how to
determine how much better.
It really sounds like you'd do well to investigate SiGe BJTs. The
biggest disadvantage are their very low operating voltage. They have
both high beta and high fT (the high fT helps to minimize the increase
in current noise at high frequencies. The high Beta helps with current
noise at middle frequencies... The 1/f is fairly well controlled on
SiGe. Their rbb type parameters are also good. )

Your layout would have to be appropriate for UHF+ components, but
with some components that 'spoil' the UHF behavior, then the region
where UHF+ design is needed could be limited.

SiGe BJTS make VERY GOOD low noise UHF oscillators, because the lower
frequency noise is fairly small. This helps to keep the noise
sidebands minimized, but this is also an indicator of working well
at lower frequencies. Also, NEC does specify some of their components
for your apparent needs.

John
 
R

Robert Baer

Jan 1, 1970
0
Steve said:
That is my assumption and why I am considering them (i.e. 1/f corner
frequency is well below the range I am concerned with). There may be
lower ft devices which are just as good at HF but there are no NF
specs for them so it is hard to identify them. I'm not sure if the
1/f corner frequency might be too high in GaAsFETs.


The applications I have are at various source impedances in the
10-1000 ohm range. Right now I am looking at a 5 MHz, 120 ohm source
impedance example, where I hope that a device may have a bit better
noise figure than at 200 MHz and 50 ohms, but am not sure how to
determine how much better.


I am familiar with most of this, but if you care to expand on the Re
impact I would appreciate it.

My currently preferred reference on (low freq.) noise is P.J.Fish,
Electronic Noise and Low Noise Design, McGraw-Hill, 1994, which uses
the common hybrid-pi model with no explicit Re with the result that it
doesn't enter into the equations. Guess I need to think a bit about
feedback configurations (or common base with Re added to source
resistance).

Steve

You failed to mention 1/f and/or flicker noise in your "high
frequency" ampliier.
Usually, RF amplifiers have a low frequency rolloff far above 1KHz,
which makes the contribution irrelevant.
JFETs and MOSFETs have a more serious 1/f problem than bipolars.
1/f may become a problem for DC, and audio amplifiers with HF cutoff
above 1KHz; that certainly is not what you have.
You may safely ignore 1/f.
 
R

Robert Baer

Jan 1, 1970
0
John S. Dyson said:
For the best 1/f corner for a UHF transistor, you might look at
the SiGe components. Keeping them stable might be difficult, so
you might need to avoid 'tuned circuits' per se. The high beta
of the SiGe will tend to help the current noise also (give you
a wider range of low noise operation.)

Also, some of the (normal UHF BJT) NEC transistors are specified
(through their spice models) for middle frequency (HF/VHF) behavior.
Some KF/AF parameters are modeled, and some of their components are
specified for low noise oscillators. Take a look at www.cel.com
or somesuch.

It really sounds like you'd do well to investigate SiGe BJTs. The
biggest disadvantage are their very low operating voltage. They have
both high beta and high fT (the high fT helps to minimize the increase
in current noise at high frequencies. The high Beta helps with current
noise at middle frequencies... The 1/f is fairly well controlled on
SiGe. Their rbb type parameters are also good. )

Your layout would have to be appropriate for UHF+ components, but
with some components that 'spoil' the UHF behavior, then the region
where UHF+ design is needed could be limited.

SiGe BJTS make VERY GOOD low noise UHF oscillators, because the lower
frequency noise is fairly small. This helps to keep the noise
sidebands minimized, but this is also an indicator of working well
at lower frequencies. Also, NEC does specify some of their components
for your apparent needs.

John

He is working with what i call "DC frequencies", and has no need to
use UHF (exotic) devices.
Many common silicon transistors are adequate for his needs.
 
S

Steve Kavanagh

Jan 1, 1970
0
[email protected] (John S. Dyson) wrote in message
Here's a break from the politics, John...
Also, some of the (normal UHF BJT) NEC transistors are specified
(through their spice models) for middle frequency (HF/VHF) behavior.
Some KF/AF parameters are modeled, and some of their components are
specified for low noise oscillators. Take a look at www.cel.com
or somesuch.

Thanks for the reminder about the CEL site. I have some NE85637's
lying around that could be good for initial experiments. Obsolete
package with no SPICE package parameters but the package parasitics
shouldn't matter at 5 MHz anyway.

Lowest frequency I saw there was 50 MHz, but I doubt if much could go
wrong with the models between there and a few kHz, given the ft's in
the GHz range, could it ?

I played a bit with a SPICE model of an NE85619 CE amp (Ic about 4 mA)
which turned out to have an optimum source impedance not too far from
50 ohms (i.e. NF was more-or-less constant and minimum from about
50-200 ohms). Just like the data sheets that were puzzling me. I
guess I just wasn't accustomed to the low rbb' in these devices.
SiGe BJTS make VERY GOOD low noise UHF oscillators, because the lower
frequency noise is fairly small. This helps to keep the noise
sidebands minimized, but this is also an indicator of working well
at lower frequencies. Also, NEC does specify some of their components
for your apparent needs.

An interesting observation. I had noticed that they were being touted
for low noise oscillators and had pondered the thought of them making
good low noise audio amps.

Steve
 
J

John S. Dyson

Jan 1, 1970
0
You failed to mention 1/f and/or flicker noise in your "high
frequency" ampliier.
Usually, RF amplifiers have a low frequency rolloff far above 1KHz,
which makes the contribution irrelevant.
Typically, high frequency transistors have a much higher 1/f
cutoff than AF transistors. The SiGe type components
(no longer exotic) have both the high frequency performance
and good 1/f performance. The key to low noise is BOTH
the rbb type contribution and the fT/Beta contribution. Note
that AFAIR the input impedance range wasn't just the 50ohm
figure, where higher impedances do make the current noise
more of an issue.

John
 
J

John S. Dyson

Jan 1, 1970
0
He is working with what i call "DC frequencies", and has no need to
use UHF (exotic) devices.
Remember, SiGe devices are NO LONGER EXOTIC, and aren't really very
expensive anymore. Also, note that the input current noise is related
to the fT/Beta figure, and the high frequency desired is 30MHz. Therefore
300MHz fT components aren't really a good choice for the lowest
noise. There are lots of applications that aren't just connecting a
transistor to an antenna, but this MIGHT be a measurement application.
Many common silicon transistors are adequate for his needs.
'Common' transistors don't just include the 2n3904/2n2222/MPSH10
types anymore. SiGe and PHEMT are often less than $1/each now...

Considering a component with a Beta of 50 and fT of 300MHz isn't
really going to give ideal noise performance at 5MHz through
30MHz (esp over a wide impedance range.) For vanilla buffers,
the $0.05each parts will work okay, though.

Choosing a traditional VHF transistor that minimizes the high
frequency current noise isn't necessarily going to be a
device with a 1kHz 1/f noise corner, but might be much much
higher than that. The cool thing about the SiGe components
is that you can have good 1/f and fantastic current noise
performance also.

For oscillators, SiGe is a fantastic choice, providing both
VHF/UHF performance for osciallation, and very low 1/f noise.

The newer technologies provide for a MUCH MUCH less critical
optimum source impedance for low noise. The old components
tended to be less pleasant to match (for noise.)

For a good time (easy to match, relatively low cost, near
theoretical performance, good noise match at 50ohms) the
new HP PHEMTs are incredible components. (In this application,
the PHEMTs aren't likely the best choice, however.)

John
 
J

John S. Dyson

Jan 1, 1970
0
[email protected] (John S. Dyson) wrote in message
Here's a break from the politics, John...

Thanks for the reminder about the CEL site. I have some NE85637's
lying around that could be good for initial experiments. Obsolete
package with no SPICE package parameters but the package parasitics
shouldn't matter at 5 MHz anyway.
You should have NO problems with package parasitics at 5MHz (assuming
reasonable layouts.)
Lowest frequency I saw there was 50 MHz, but I doubt if much could go
wrong with the models between there and a few kHz, given the ft's in
the GHz range, could it ?
The spice models should be very good at low frequencies. I'll check
the models for your part (the 856 series) and see if I have info
about the KF/AF noise behavior. For your application, the Sparams
aren't going to be as helpful as the spice parameters. I know
that I have a fairly large library of CEL and NEC (direct from
NEC) models -- but my archives are very large and very disorganized.

I just fully read your message right before going to bed, so
will look up some of the info tomorrow morning.
I played a bit with a SPICE model of an NE85619 CE amp (Ic about 4 mA)
which turned out to have an optimum source impedance not too far from
50 ohms (i.e. NF was more-or-less constant and minimum from about
50-200 ohms). Just like the data sheets that were puzzling me. I
guess I just wasn't accustomed to the low rbb' in these devices.
Hmmm... In ideal (lf) situations, with a Beta of 100 and Ic
of 4ma, assuming rbb of 0, we should see an Ropt of approx 60ohms.
So, that looks pretty good. If you want to see lower noise,
you certainly don't want to choose a component with a larger
rbb (even though the ideal rs would be higher), but if you want
Ic=4ma, then you need higher Beta (assuming everything else
is the same.) The current noise will be the predominant noise
mechanism above 60-100ohms. This will be worse at higher
frequencies (fT/B). Assuming the 'ideal' of using SiGe with
a Beta of 300 -- it will also have a low rbb, but the ideal
rs will only be sqrt(3) higher (perhaps 100ohms.)

With SiGe, where the fT would be maintained at lower currents
(e.g. 1ma), and the Beta is also maintained at lower currents,
it seems like an ideal source of 300-500ohms is attainable at
5-100MHz... (I mean, truly low noise, not just 1dB NF :)).`

Assuming you have adequate bandwidth at lower currents, you might
find that using Ic=1ma will give you several times higher Ropt.

The worst case would be to use a component with a normal fT (e.g. 300MHz),
where the effective Beta might be significantly decreased. This
would push the ideal source impedance downwards (by increasing the
effective current noise.) You might get reasonably low noise
performance at 5MHz (the current noise is just increasing, thereby
starting to decrease Ropt), but at 50MHz, you'll quickly start seeing
increases of current noise... If you cannot use impedance matching
because of other constraints,
and really want vanishing noise with a 500ohm source (again, assuming
1dB NF isn't good enough), then SiGe or well chosen JFET might be
possible alternatives???

For 'normal' applications, usually circuit configurations and
requirements allow the use of 'normal' fT 500MHz+- BJTs (or even
slower) for a 50MHz application. Special applications can make
the choices more complicated.

So, for a wideband small signal instrumentation low noise amplifier,
you might have enough constraints to make the choice of a
newer technology component advantageous. In a narrow-band application,
choosing the 'newer technology' might just be a waste of money
or complicate the layout because of excess instability.
An interesting observation. I had noticed that they were being touted
for low noise oscillators and had pondered the thought of them making
good low noise audio amps.

My guess is that they'd be more trouble than they'd be worth for
Audio :). For normal audio applications (I mean, non-hypersonic),
some of the large geometry semi-superbeta components from Zetex
or some of the low noise transistors from Japan (low rbb, reasonable
Beta) seem to be good choices. The diminishing returns thing is
something easily achievable in audio.

For RF, even with some of the fancy new (but becoming common) components,
can help to address the large number of constraints often dealt
with in RF (Impedance matching, and trying to optimize the noise for a
given source impedance need to be simultaneously considered.)

I will check into the 856 component series, and respond again if
I learn anything tomorrow.

John
 
S

Steve Kavanagh

Jan 1, 1970
0
Robert Baer said:
He is working with what i call "DC frequencies", and has no need to
use UHF (exotic) devices.
Many common silicon transistors are adequate for his needs.

Robert...have you any specific device suggestions ?

I haven't been able to find any obvious candidates so far.
Manufacturers tend to characterize transistors for NF only when
intended for audio (ft<300 MHz) and for above 500 MHz (ft>4 GHz).

I think you are probably right to a degree (but so is John to a
degree, too). Using the CEL SPICE model for an NE85619 I get the
following typical 5 MHz performance, which are definitely in the
ballpark that I am aiming for

CE, no feedback, 4mA Ic: 1.0 dB NF (awful match to 50-120 ohms
source!)
CE, resistive (Collector-Base) FB to match 50 or 120 ohms: a bit under
2 dB NF
CB, 500/220 uA Ic for 50/120 ohms input matched to source: 2.7 dB NF.

RB in this model is about 4 ohms.

The initial application is for an IF amp which must properly terminate
the passive mixer/IF filter combination ahead of it, so the latter two
are candidates. I'd rather avoid the complication of noiseless
feedback configurations for this one.

In the CB case, the collector load must be quite high to achieve a
useful gain and this, combined with the low current and resulting
lowish ft (maybe 1 GHz ?) means that gain drops off noticeably at 30
MHz. But this one is for 5 MHz so there should be some room to play
with lower ft - a rated ft around 1-2 GHz, rather than 7 GHz (at a few
mA) is likely OK. CE cases with fairly low collector load impedances
will also need pretty good ft to ensure enough hfe to get useful gain
but I think there is probably more margin here, partly since they can
be run at higher current (though not a lot higher in the present case
since I am trying to minimize current drain).
You failed to mention 1/f and/or flicker noise in your "high
frequency" ampliier.
Usually, RF amplifiers have a low frequency rolloff far above 1KHz,
which makes the contribution irrelevant.
JFETs and MOSFETs have a more serious 1/f problem than bipolars.
1/f may become a problem for DC, and audio amplifiers with HF cutoff
above 1KHz; that certainly is not what you have.
You may safely ignore 1/f.

That was my assumption - that bipolars were safe as far as 1/f noise
was concerned but that was not necessarily true for UHF/microwave
FETS. A good VHF JFET (like J310) might be reasonable for NF but
likely only at higher current than I am willing to accept, and with
additional matching networks not needed with bipolars. Thanks for the
vote in agreement !

Steve
 
D

ddwyer

Jan 1, 1970
0
Briefly the application was not stated, if for an HF front end then
dynamic range is more important particularly as background and
atmospheric noise exceeds most transistor noise.
It maybe that a low noise op amp with < ,1nV/root Hz is good enough now
commonly available to 1GHz these are specified for voltage and current
noise down to 100kHz and below.
 
S

Steve Kavanagh

Jan 1, 1970
0
A few comments...

I think some of John's explanations are starting to get through my
skull - I've definitely got a better understanding of the low rbb
regime now - thanks !

A comment was made about dynamic range being more important in HF
receivers. The same issues can also arise in some instrumentation
scenarios - spectrum analyzers are a good example. While this is
true, the lower the noise figure in each stage the less gain is needed
to meet whatever noise figure spec one has (and hence signal levels
are lower), so low NF amplifiers can be a useful tool toward better
IP3 (or the same IP3 at lower DC power).

Just for fun I tried a couple of the Zetex super-beta power
transistors in a broadband resistive feedback RF amp circuit (in
SPICE) - they do very nicely for NF but only good up to a MHz or so
with ft of 150-250 MHz. I suppose one might get a bit better
behaviour in a tuned amp.

Also tried an NEC SiGe device (NESG2030) - it didn't do much for me at
120 ohms and 5 MHz, but it is pretty impressive up to a few hundred
MHz in a 500 ohm characteristic impedance circuit at about 0.5 mA Ic.
It would appear, not surprisingly, it is most useful for such high-Z
VHF low-power cases (other than the microwave applications it is
marketed for). Now to think of a good application for such an
amplifier! Not much of a compression point, it seems, with Vce of
1.5V or so.

I did quite a lot of searching to see if I could find transistors with
low rbb (<5 ohms), high beta (>150) and moderately high nominal ft
(350 MHz-2 GHz) that might be both less noisy and less prone to
oscillations than microwave transistors. It seems Zetex used to make
some up to about 350 MHz ft but no longer. They have dropped almost
all their small-signal products. No luck anywhere else so far - I've
been through 8 websites today. Any suggestions would be welcome !

Steve
 
Top