Still lower noise radio astronomy (was: low-noise amplifiers for radio astronomy )

wrote:
"John (Liberty) Bell" wrote in message
...
wrote:
John (Liberty) Bell wrote:

snip agreed material

Let antenna noise T = Tant, and amp noise T = Tam. Then

Hold on, what comes out of the antenna is the
signal we are trying to measure.

Precisely, plus the noise sent into the amplifier, from the antenna
plus couplings.

Right but you don't know those when designing a new
product.

Precisely, which is why the specification of Noise Temperature makes
perfect sense in such a situation, but the specification of Noise
Figure does not. The Noise Temperature of the amplifier (at a designed
operating temperature) stays the same whatever the Noise Temperature of
the source you plug into it. In contrast, the Noise Figure will change
with every change in the Noise Temperature of the source. We have
already even agreed on the appropriate formula:

NF = 10log[1 + Tamp/Tsource].

If you design your new and improved amp
you will need to publish these figures and you need
to measure them in a way that the industry recognises.

Quite so. That is typically Noise Temperature in the application areas
of radio astronomy, and of cryogenic amps, for the reasons I have given
above.

It is only in situations where we know that the source will typically
have a noise temperature of ~ 300K, that Noise Figure is meaningfully
specified.

In such situations, there is little point in having a preamp with a
Noise Temperature of 2K. That is confirmed by the NF spec. for such an
amp:

NF = 0.029 dB.

Contrast with a preamp with a Noise Temperature which is a factor of 20
worse:

NF = 0.544 dB

The difference is only half a decibel, using NF.
Nevertheless, the first is better than the state-of-the-art for radio
astronomy, whereas the second would be crippling, because it would then
more than triple the total noise of the system.

The NF is
published for an amplifier (as in the articles
we were discussing) without knowing what
the customer will use it for. The way that is
done is that NF is _specified_ assuming the
input is only a matched resitive load.

Which gives a typical Noise Temperature of the source of ~ 293K,
presumably.

Well, that appears to contradict the NF definition you gave me.
(s/n in [in dB] - s/n out [in dB])

As I said, there are a number of equivalent
definitions. I asked three guys from the RF
department and they gave me three different
answers. One was your analysis above, another
the equation you derive from it below.

That is encouraging.

Nevertheless, I did assume that in posting 73.
Be warned. This gives a noise T for all these bipolar and CMOS LNAs
which is substantially 100K.

Yes, that is expected, but see below.

Noise Figure = 10log [k del f {Tant + Tam} /Tant k del f )]
= 10log [{Tant + Tam} /Tant)]
= 10log [{1 + Tam/Tant)]
2 dB for CMOS
Hence
log [{1 + Tam/Tant)] log 10^ 0.2
1 + Tam/Tant 10^ 0.2
Tam Tant (10^ 0.2 - 1)
Tant (antilog 0.2, - 1)

Do you agree with that conclusion?

The maths is right but replace Tant with Tterm
where Tterm is a dummy matching terminator.

If so, that Noise Figure claim is meaningless, without a known noise
temperature of the source.

You do not need to know the temperature of the source,
what you need to know is the temperature at which the
published NF for the amplifier was measured. For
commercial gear it is "room temperature", usually
assumed to be 290K as stated in the Wiki article you
quoted.

Precisely. So what appears to look good as a published NF turns out to
be terrible when converted to an amplifier noise T of significantly100K.
Despite this, you showed in your example, that such a crappy LNA would
match NASA's (without cooling), if you applied the NF method in that
situation.
You have thus helped me to prove my point.

If you run the amplifier hot, yes it will degrade
performance badly but that is why they cool it.

Hold on now. You are telling me to specify my amp using NF at room
temperature, and then you are going to compare it to every other amp
cooled by a factor of 20, and then complain that it doesn't look like
much of an improvement, under those circumstances? That doesn't sound
like a level playing field to me!

There would be no point in cooling if it didn't
reduce the actual noise power. The sources of
noise produce less power as you cool the amp
just as the input impedance produces less, hence
the ratio tends to remain the same.

No. The input impedance does not produce less noise, because you are
only cooling the amplifier, not the source, when you do this. If you
examine http://www.skatelescope.org/document...8MTTSPaper.pdf
again, you will see that cooling produces a large improvement in Noise
T of the amplifier, but only a small improvement in the Noise T of the
source (attributed to incidental cooling of the coupling).

So an amp suitable
for cooling that has a noise figure of 1.5dB at
room temperature should also be close to that when
cooled. OK, practicalities get in the way and
that won't relly be true but you should follow
the theoretical argument.

I would say the theoretical argument is sufficiently flawed, and the
practicalities that get in the way are so pronounced, in this
application area, that you are better off using something that is
simpler in calculations whilst remaining both accurate and unambiguous.
That something is Noise Temperature.

For a cryogenic amp the temperature might be
stated in the specification or more likely the will quote
the noise temperature of the amp which is just

Tamp = n amp / kB

The latter being simple, rigorous, and not open to misinterpretation,
except to the extent of us possibly confusing noise T with operating T.

Yep, but it will vary with temperature

True. However, variations in abient temperature are typically a small
fraction of that temperature when measured in degrees absolute
(Kelvin). If cooling is used, even domestic refrigerators are
thermostatically controlled. Similarly, the boiling point of liquid
nitrogen varies very little, if that method is used instead.

NF tends to be independent to a first approximation
over the cryogenic range.

I remain to be convinced of that. I refer you again to
http://www.skatelescope.org/document...8MTTSPaper.pdf. If the
noise temperature of the amp changes by a factor of 10 through cooling
by a factor of 10 [1], but the noise temperature of the source changes
very little, it is obvious that Tamp/Tsouce then changes by a factor of
~ 10.

Thus, the NF could only be independent to a first approximation, if
Tamp/Tsouce 1. In which case, there was no point in cooling the amp
in the first place, or , indeed, in using an ultralow noise (even at
room temperature) amp design, such as mine.

My response to the last part of your posting needs some more thought,
so......To be continued.

John

[1] Note, however, from table 1 of
http://www.skatelescope.org/document...8MTTSPaper.pdf, that this
relationship does not, in fact, appear to be precisely linear.

John (Liberty) Bell wrote:
wrote:
"John (Liberty) Bell" wrote in message
...
wrote:
John (Liberty) Bell wrote:

snip

Let antenna noise T = Tant, and amp noise T = Tam. Then

Hold on, what comes out of the antenna is the
signal we are trying to measure.

Precisely, plus the noise sent into the amplifier, from the antenna
plus couplings.

Right but you don't know those when designing a new
product.

Precisely, which is why the specification of Noise Temperature makes
perfect sense in such a situation, but the specification of Noise
Figure does not. The Noise Temperature of the amplifier (at a designed
operating temperature) stays the same whatever the Noise Temperature of
the source you plug into it. In contrast, the Noise Figure will change
with every change in the Noise Temperature of the source. We have
already even agreed on the appropriate formula:

NF = 10log[1 + Tamp/Tsource].

Again, it is part of the definition of noise figure that
you test it with a terminated input, not a real source.

It is only in situations where we know that the source will typically
have a noise temperature of ~ 300K, that Noise Figure is meaningfully
specified.

That is the case for equipment other than cooled
amps. but where you have cooling involved than it
is less clear what temperature you should use for
the dummy load on the input. Since the other parts
of the system and things like sky noise are also
known as temperatures, giving noise temperature
is convenient.

In such situations, there is little point in having a preamp with a
Noise Temperature of 2K. That is confirmed by the NF spec. for such an
amp:

NF = 0.029 dB.

Contrast with a preamp with a Noise Temperature which is a factor of 20
worse:

NF = 0.544 dB

The difference is only half a decibel, using NF.

And if your source is at ambient, that is a sensible
indicator of the relative improvement.

Nevertheless, the first is better than the state-of-the-art for radio
astronomy, whereas the second would be crippling, because it would then
more than triple the total noise of the system.

The NF is
published for an amplifier (as in the articles
we were discussing) without knowing what
the customer will use it for. The way that is
done is that NF is _specified_ assuming the
input is only a matched resitive load.

Which gives a typical Noise Temperature of the source of ~ 293K,
presumably.

I have seen 290K and 300k used, the difference
is not significant (0.15dB) and probably less than
the change due to the weather over a year.

Precisely. So what appears to look good as a published NF turns out to
be terrible when converted to an amplifier noise T of significantly100K.
Despite this, you showed in your example, that such a crappy LNA would
match NASA's (without cooling), if you applied the NF method in that
situation.
You have thus helped me to prove my point.

If you run the amplifier hot, yes it will degrade
performance badly but that is why they cool it.

Hold on now. You are telling me to specify my amp using NF at room
temperature, and then you are going to compare it to every other amp
cooled by a factor of 20, and then complain that it doesn't look like
much of an improvement, under those circumstances? That doesn't sound
like a level playing field to me!

I think we are at cross purposes, I was saying that
you cannot compare a cooled amp with an uncooled
amp on the assumption that the noise of the uncooled
amp would remain constant if it was cooled.

There would be no point in cooling if it didn't
reduce the actual noise power. The sources of
noise produce less power as you cool the amp
just as the input impedance produces less, hence
the ratio tends to remain the same.

No. The input impedance does not produce less noise, because you are
only cooling the amplifier, not the source, when you do this.

Not true, that's the point. Noise figure is specified
into a perfect matched input at the same ambient
as the amp, it has nothing to do with the actual
source temperature or noise level.

If you
examine http://www.skatelescope.org/document...8MTTSPaper.pdf
again, you will see that cooling produces a large improvement in Noise
T of the amplifier, but only a small improvement in the Noise T of the
source (attributed to incidental cooling of the coupling).

Right, but the noise from the source is not relevant
to the way NF is specified. It is a measure of amplifier
performance alone and has to be divorced from the
system so that manufacturers can measure it and
publish the spec.

So an amp suitable
for cooling that has a noise figure of 1.5dB at
room temperature should also be close to that when
cooled. OK, practicalities get in the way and
that won't relly be true but you should follow
the theoretical argument.

I would say the theoretical argument is sufficiently flawed, and the
practicalities that get in the way are so pronounced, in this
application area, that you are better off using something that is
simpler in calculations whilst remaining both accurate and unambiguous.
That something is Noise Temperature.

In the cryogenic field I agree but the rest of the
industry tends to use noise figure as being more
convenient.

For a cryogenic amp the temperature might be
stated in the specification or more likely the will quote
the noise temperature of the amp which is just

Tamp = n amp / kB

The latter being simple, rigorous, and not open to misinterpretation,
except to the extent of us possibly confusing noise T with operating T.

Yep, but it will vary with temperature

True. However, variations in abient temperature are typically a small
fraction of that temperature when measured in degrees absolute
(Kelvin). If cooling is used, even domestic refrigerators are
thermostatically controlled. Similarly, the boiling point of liquid
nitrogen varies very little, if that method is used instead.

Exactly, hence a published NF is useful.

NF tends to be independent to a first approximation
over the cryogenic range.

I remain to be convinced of that. I refer you again to
http://www.skatelescope.org/document...8MTTSPaper.pdf. If the
noise temperature of the amp changes by a factor of 10 through cooling
by a factor of 10 [1], but the noise temperature of the source changes
very little, it is obvious that Tamp/Tsouce then changes by a factor of
~ 10.

Again, the source is not relevant. From that paper

Ambient Noise curve A

300K 23K
80K 4.7K
20K 1.5K

The ratio is close to 13:1 over more than a decade of
temperature despite the matching being optimised for
each and one using GaAs pHEMT while the other two
used InP.

Thus, the NF could only be independent to a first approximation, if
Tamp/Tsouce 1. In which case, there was no point in cooling the amp
in the first place, or , indeed, in using an ultralow noise (even at
room temperature) amp design, such as mine.

Again, NF is specified independent of the actual
source!

My response to the last part of your posting needs some more thought,
so......To be continued.

John

[1] Note, however, from table 1 of
http://www.skatelescope.org/document...8MTTSPaper.pdf, that this
relationship does not, in fact, appear to be precisely linear.

It is closer to linear than constant, hence the ratio
of amp noise to the thermal noise is a matched
resistor tends to have a fixed value. That is the
(or one) definition of NF.

George

wrote:
"John (Liberty) Bell" wrote in message
...

.....Continued, starting with continuation of my response to:

NF tends to be independent to a first approximation
over the cryogenic range.

I suggest that statement would only be true, to a first approximation,
if you use the ground plane, fed via a 'matching' series resistor (of
typically 50 ohms), into the input of the amplifier, as the dummy
source for initial and continued testing. In that case, yes, the
decrease in noise temperature of that matching resistor should then
approximately match the decrease in noise temp of the amp.

[We do seem to be in agreement that that is the method used in
practice, for specifying NF at room T, when the source is unknown. ]

Hiowever, on the basis of those above statements, I would now like to
bring up a more interesting question (that I am surprised you have not
brought up yourself).

Given the above, and the fact that, in practical applications, the LNA
will typically also need a 50 ohm matching resistor from signal to
ground for transmission line impedance matching purposes, prior to
connection to the actual transistor gate;- why does the source Noise T
not reflect that matching resistor's noise T (which is ambient T),
within the determination of total noise?

My answer to that question would be empirical. The evidence shows that
it doesn't. I would give the same answer to the corresponding
question:- why does the antenna noise not reflect the ambient T of the
antenna? Because the evidence demonstrates otherwise.

If we look again at
http://www.skatelescope.org/document...8MTTSPaper.pdf, figure 2
clearly shows an antenna matching impedance of 50 ohms, connected
between earth and the beginning of the amplifier stage. Nevertheless,
table 1 equally clearly shows that the total noise of the amplifier
stage at an ambient T of 300K is 23K not 300K. Why would that be? One
possibility is that this resistor is now connected in parallel with a
real source, not in series, and so its noise energy has three likely
exit routes. One is direct to ground via a zero resistance path. The
next is to ground via the antenna's 50 ohm impedance, with the least
likely route being into the higher impedance of the transistor gate.

I suggest in consequence that we stick to a discussion of noise
temperatures in future, for precisely this reason.
(If we come across an amp with a quoted NF, we convert that to noise
temperature first, and then proceed more rigorously and unambiguously,
using that data.)

Agreed?

I think we need to understand both definitions
and use them as appropriate.

Agreed.

They would

(I assume an unwritten 'not' here)

have survived in the industry if they didn't have
their place. Engineers are pragmatic that way
as I'm sure you know.

The bottom line on this, given your clarification
of your claim, is that a cryogenic amp with a noise
figure of 1.5dB at 10K has a noise temperature of
around 4K.

ONLY if the souce has a noise T whuich is then equal to the ambient T
of the amp.

Reducing internal noise _only_ would
have the effect of reducing that to 0.04K

In that case, I have made an arithmetic mistake here, if I quoted a 20
dB reduction (which I probably did).

In reality, I calculated a 10 fold reduction in noise power for a
relaxed practical application of the design concept, which really means
10 dB. It would only mean 20 dB if that was a 10 fold reduction in
noise voltage. Consequently, that only means, in practice, a noise T
reduction from 4K to 0.4K or, equivalently, from 40K to 4K, which is
precisely what I claimed originally, before incorrectly converting back
to dB for your benefit, thus managing to commit the most common blunder
that the dB scale is notorious for.

(Can you see now why, having had to understand the use of noise T in
radio astronomy at the start of this thread, I now appreciate its
superior elegance and simplicity for unambiguous and easy specification
of noise contributions, in this application?)

but the
majority of the noise in the DSN systems would then
be from other influences so there is some benefit.

Certainly, after factor of 10 further improvement in amplifier noise T,
an additional factor of 10 improvement would have negligible benefit,
unless, of course, we are then talking about a factor of 10 improvement
over cryogenically controlled amplifier performance, using a room
temperature device.

However, if you run the LNA at room temperature,
say 300K, then its noise temperature would be about
120K. Reduce the noise by 20dB and you get 1.2K so
cryogenic performance without the cooling. That is
the big cost driver. It means your concerns about
the failure modes of devices at low temperature
aren't a worry.


In fact, I get ~ 2.3 to 2.5 K at an ambient T of 300K for a~ a 1.5 GHz
amplifier
based on data in the repeatedly quoted references. (This would
correspond to a NF of ~ 0.035 dB, using your described method of room
testing for NF.)

There are 2 ways to go with this:
1) Cryogenic performance without the cryogenics
2) Now reducing the amplifier noise contribution in eg the DSN from 25%
to 2.5%. That is equivalent to increasing the range by 10%

I am not really sure about the current distance ranges in radio
astronomy (except for the CMB which is over a fairly narrow band of
frequencies [and essentially omnidirectional]). Consequently I can only
currently express this by analogy with optical astronomy. In that area,
I understand we already have evidence of galaxies at about 1.5 billion
light years from the big bang. A better than 10% improvement in optical
range would, in principle, then seem to take us all the way to the
presumed beginning of time.

John Bell
(Change John to Liberty to bypass anti-spam email filter)

wrote:

Again, it is part of the definition of noise figure that
you test it with a terminated input, not a real source.

OK, if you say so, it must be, and that does then make a bit more sense
of NF. But that was certainly not made clear to me from your
explanatory pointer to http://en.wikipedia.org/wiki/Noise_figure, and
was specifically not made clear to me via their given mathematical
definition:
NF = SNRin - SNRout
where everything is in dB.

On second reading of wiki however, what you say appears to be
compatible with what they say, but this is another example of where I
still think the wiki explanation was nowhere near as clear and
unambiguous as I would have liked it to be.

Since much of your further response was a reaffirmation of your above
point, I have now snipped further repetitions, for both of us.

John (Liberty) Bell wrote:
It is only in situations where we know that the source will typically
have a noise temperature of ~ 300K, that Noise Figure is meaningfully
specified.

That is the case for equipment other than cooled
amps.

It is not the case in radio astronomy even when the amp is not cooled.
(See eg Tamp/Tsys for 300K , in Table 1 of mutually repeated
reference).

((That is, in fact, the application area we are supposed to be
discussing.))

snip

I think we are at cross purposes, I was saying that
you cannot compare a cooled amp with an uncooled
amp on the assumption that the noise of the uncooled
amp would remain constant if it was cooled.

And I was saying that you cannot derive better than an order of
magnitude improvement in the internal noise of that uncooled amp,
merely by replacing a room temperature resistor at the input, with a
radio telescope. That was what you appeared to be claiming in your
given example, given the wiki maths definition of NF.

We seem to be in agreement on all important points elsewhere.

I will be interested to see your response to my continuation.

John

I have answered both your posts here.

"John (Liberty) Bell" wrote in message
...
wrote:
"John (Liberty) Bell" wrote in message
...

....Continued, starting with continuation of my response to:

NF tends to be independent to a first approximation
over the cryogenic range.

I suggest that statement would only be true, to a first approximation,
if you use the ground plane, fed via a 'matching' series resistor (of
typically 50 ohms), into the input of the amplifier, as the dummy
source for initial and continued testing. In that case, yes, the
decrease in noise temperature of that matching resistor should then
approximately match the decrease in noise temp of the amp.

[We do seem to be in agreement that that is the method used in
practice, for specifying NF at room T, when the source is unknown. ]

Picture yourself in the manufacturer's test lab being
asked to give the value the sales department should
put in the glossy brochure to get maximum orders but
without getting sued.

Hiowever, on the basis of those above statements, I would now like to
bring up a more interesting question (that I am surprised you have not
brought up yourself).

Given the above, and the fact that, in practical applications, the LNA
will typically also need a 50 ohm matching resistor from signal to
ground for transmission line impedance matching purposes, prior to
connection to the actual transistor gate;- why does the source Noise T
not reflect that matching resistor's noise T (which is ambient T),
within the determination of total noise?

Go back to the definition of noise figure as the ratio
of the internal noise to that in the input matching
termination expressed in dB.

The thermal noise power in a resistor is kTB which is
independent of resistance. Placing one resistor in
parallel with another therefore does not change the
total noise power. In practice, some of the time the
noise out of one resistor will happen to be of the
opposite polarity to that from the other since they
are uncorrelated.

Hence placing an extra resistor inside the amp in
parallel with the input doesn't change the input
noise power and when multiplied by the power gain
that gives the same reference value of output noise
power. The noise temperature of the amp is a measure
of how much the amp noise output exceeds that
reference value so is also unaffected.

At least I think that's the way it would work out.

My answer to that question would be empirical. The evidence shows that
it doesn't. I would give the same answer to the corresponding
question:- why does the antenna noise not reflect the ambient T of the
antenna? Because the evidence demonstrates otherwise.

The answer that is that it partially does. Consider a
mirror which reflects heat in the form of infra-red. If
it emitted well it could lose energy by radiation without
absorbing it so it could cool itself without using energy.
That cold mirror could then run a Stirling engine hence
you violate conservation of energy - a perpetual motion
machine. Good reflectors must be poor emitters.

The DSN dish is an RF mirror which reflects virtually all
the incoming radiation to the receiver, so it must be a
very poor emitter and very little thermal radiation comes
from the antenna compared to what there would be if it
was painted matt black.

If we look again at
http://www.skatelescope.org/document...8MTTSPaper.pdf, figure 2
clearly shows an antenna matching impedance of 50 ohms, connected
between earth and the beginning of the amplifier stage. Nevertheless,
table 1 equally clearly shows that the total noise of the amplifier
stage at an ambient T of 300K is 23K not 300K. Why would that be?

Good question and it raises another, but let me first
answer yours. As above, the noise temperature of 23K
is a measure of the amount by which the output exceeds
the gain times the thermal power in the input resistance.
The amp is only adding about 7.7% to the amount that
would be there due to the gain times the noise from the
input resistance alone.

Now the question that raises in my mind is this - I
thought HEMT amplifiers were particularly good because
the electrons were able to cool by emitting into space.
With a dummy load in place, they can only cool until
they reach equilibrium which must be the same
temperature as the source termination on thermodynamic
grounds. Surely then the performance would depend on
the temperature of the terminator. I haven't figured
that one out yet.

One
possibility is that this resistor is now connected in parallel with a
real source, not in series, and so its noise energy has three likely
exit routes. One is direct to ground via a zero resistance path. The
next is to ground via the antenna's 50 ohm impedance, with the least
likely route being into the higher impedance of the transistor gate.

That would allow the transistors to cool but the matching
would be screwed (real source or terminator but not both)
and I doubt the guys had access to a DSN dish to try this
out. I'm sure they would be running this in the university
lab on the bench, at least for the 300K design.

I suggest in consequence that we stick to a discussion of noise
temperatures in future, for precisely this reason.
(If we come across an amp with a quoted NF, we convert that to noise
temperature first, and then proceed more rigorously and unambiguously,
using that data.)

Agreed?

I think we need to understand both definitions
and use them as appropriate.

Agreed.

They would

(I assume an unwritten 'not' here)

You are correct.

have survived in the industry if they didn't have
their place. Engineers are pragmatic that way
as I'm sure you know.

The bottom line on this, given your clarification
of your claim, is that a cryogenic amp with a noise
figure of 1.5dB at 10K has a noise temperature of
around 4K.

ONLY if the souce has a noise T whuich is then equal to the ambient T
of the amp.

That I believe is the definition.

Reducing internal noise _only_ would
have the effect of reducing that to 0.04K

In that case, I have made an arithmetic mistake here, if I quoted a 20
dB reduction (which I probably did).

You did, that's a factor of 100 in power and power
is proportional to temperature.

In reality, I calculated a 10 fold reduction in noise power for a
relaxed practical application of the design concept, which really means
10 dB. It would only mean 20 dB if that was a 10 fold reduction in
noise voltage. Consequently, that only means, in practice, a noise T
reduction from 4K to 0.4K or, equivalently, from 40K to 4K, which is
precisely what I claimed originally, before incorrectly converting back
to dB for your benefit, thus managing to commit the most common blunder
that the dB scale is notorious for.

OK, _very_ easily done.

(Can you see now why, having had to understand the use of noise T in
radio astronomy at the start of this thread, I now appreciate its
superior elegance and simplicity for unambiguous and easy specification
of noise contributions, in this application?)

but the
majority of the noise in the DSN systems would then
be from other influences so there is some benefit.

Certainly, after factor of 10 further improvement in amplifier noise T,
an additional factor of 10 improvement would have negligible benefit,

That was the point really, reducing 4K to 0.4K is nice
but when the upstream plumbing adds 11K anyway, it
doesn't give a factor of 10 increase in range to the
craft. Cleaning up the plumbing has more scope.

unless, of course, we are then talking about a factor of 10 improvement
over cryogenically controlled amplifier performance, using a room
temperature device.

That was my point really, your idea still has merit
but is better applied to non-cooled amps.

However, if you run the LNA at room temperature,
say 300K, then its noise temperature would be about
120K. Reduce the noise by 20dB and you get 1.2K so
cryogenic performance without the cooling. That is
the big cost driver. It means your concerns about
the failure modes of devices at low temperature
aren't a worry.


In fact, I get ~ 2.3 to 2.5 K at an ambient T of 300K for a~ a 1.5 GHz
amplifier
based on data in the repeatedly quoted references. (This would
correspond to a NF of ~ 0.035 dB, using your described method of room
testing for NF.)

The way I did it, a noise figure of 1.5dB is a noise
factor of sqrt(2) = 1.414 so the amp noise is 0.414
times the reference. Temperature is proportional so
0.414 * 300K ~ 120K. Reduce that by your 20dB and
you get 1.2K or a NF of 0.018 dB. For an improvement
of 10dB, the temperature goes to 12K and the NF would
improve from 1.50 dB to 0.179 dB.

There are 2 ways to go with this:
1) Cryogenic performance without the cryogenics
2) Now reducing the amplifier noise contribution in eg the DSN from 25%
to 2.5%. That is equivalent to increasing the range by 10%

I am not really sure about the current distance ranges in radio
astronomy (except for the CMB which is over a fairly narrow band of
frequencies [and essentially omnidirectional]).

Yes but that uses differential radiometers, a very
different technique.

Consequently I can only
currently express this by analogy with optical astronomy. In that area,
I understand we already have evidence of galaxies at about 1.5 billion
light years from the big bang.

The put out a lot more power though.

A better than 10% improvement in optical
range would, in principle, then seem to take us all the way to the
presumed beginning of time.

No, the big problem there is red shift. To go any
farther back than we have, we need to move to
deeper infra-red, hence the James Webb.

To assess the improvement just bear in mind the
received signal falls as the inverse square of the
range so a 10% reduction in noise for the same SNR
means range increases by sqrt(1.1) = 1.0488 or an
improvement of 4.88%. At 70AU, it would increase
the range by about 3.4 AU. Pioneer covers 1AU every
138 days so a 10% would have allowed contact to be
maintained for an extra 470 days after ~ 30 years.

================================================== =

"John (Liberty) Bell" wrote in message
...
wrote:

Again, it is part of the definition of noise figure that
you test it with a terminated input, not a real source.

OK, if you say so, it must be, and that does then make a bit more sense
of NF. But that was certainly not made clear to me from your
explanatory pointer to http://en.wikipedia.org/wiki/Noise_figure, and
was specifically not made clear to me via their given mathematical
definition:
NF = SNRin - SNRout
where everything is in dB.

On second reading of wiki however, what you say appears to be
compatible with what they say, but this is another example of where I
still think the wiki explanation was nowhere near as clear and
unambiguous as I would have liked it to be.

I think the first definition is about as clearly written
as it could be:

"Noise Figure is the ratio of the output noise power
of a device to the portion thereof attributable to
thermal noise in the input termination at standard
noise temperature (usually 290 K)."

The words "attributable to thermal noise in the
input termination" say it all for me. However,
Wiki is a collaborative effort so if you can
think of better wording, register and edit it :-)


Since much of your further response was a reaffirmation of your above
point, I have now snipped further repetitions, for both of us.

Cool.

John (Liberty) Bell wrote:
It is only in situations where we know that the source will typically
have a noise temperature of ~ 300K, that Noise Figure is meaningfully
specified.

That is the case for equipment other than cooled
amps.

It is not the case in radio astronomy even when the amp is not cooled.
(See eg Tamp/Tsys for 300K , in Table 1 of mutually repeated
reference).

((That is, in fact, the application area we are supposed to be
discussing.))

Indeed but we seemed to have picked up the general
use of "noise figure" as an aside. The RF gear we
make is used for ship-to-shore comms, air traffic
control, coastguards and the like and noise figure
is used pretty universally. It is horses for courses
really.

snip

I think we are at cross purposes, I was saying that
you cannot compare a cooled amp with an uncooled
amp on the assumption that the noise of the uncooled
amp would remain constant if it was cooled.

And I was saying that you cannot derive better than an order of
magnitude improvement in the internal noise of that uncooled amp,
merely by replacing a room temperature resistor at the input, with a
radio telescope. That was what you appeared to be claiming in your
given example, given the wiki maths definition of NF.

No that wasn't what I was saying at all. Let me try
again. The paper you quoted gave examples of a 300K
amp with 23K noise and an amp cooled to 20K with 1.5K
noise - which is the better design? Obviously the
latter has better noise temperature but the noise
figure is about 0.32 dB for each. What that tells
me is that if the 300K design were cooled to 20K and
survived, its performance would be similar to that of
the InP design. That is a more level playing field to
borrow your expression and shows the usefulness of NF
even in this application. Of course it wouldn't scale
exactly but you get the idea.

We seem to be in agreement on all important points elsewhere.

I will be interested to see your response to my continuation.

All in one now though not in any specific order.

George

There is one further angle I would like to cover while we are on the
subject of level of performance improvement that can theoretically be
achieved with this LNA design concept. As George doubtless knows, all
electronic designs involve a tradeoff. At microwave frequencies, a
variety of effects come into play which can affect this tradeoff, thus
limiting the performance improvement that can be achieved in practice,
using my technique. Consequently, at such frequencies I think it is
still desirable to use the lowest noise (and highest cost) transistor
available, at the input stage. However, at the lower end of the
frequency range of radio astronomy, it does currently seem that the
technique could possibly be applied more ambitiously, within a single
chip CMOS design, to additionally overcome the inherently higher noise
generated by such CMOS transistors (at the cost of increased power
consumption). Whether that is worth doing in practice, would presumably
depend on the quantity required, as well as other engineering factors.

John Bell
http://accelerators.co.uk
(Change John to Liberty to bypass anti-spam email filter)

I am just going to concentrate primarily on the astronomy side of this
response right now, as I would like a bit of a break from discussing
engineering details, (as I guess other readers might too).

wrote:

"John (Liberty) Bell" wrote in message
...

That was the point really, reducing 4K to 0.4K is nice
but when the upstream plumbing adds 11K anyway, it
doesn't give a factor of 10 increase in range to the
craft.

I don't think anyone ever claimed it did. Taking your figures, and
assuming no additional noise sources, a reduction from 15K to 11.4K is
an improvement in s/n ratio of 24%. Since signal strength decreases as
r squared, this represents an increase in range of 11%

Cleaning up the plumbing has more scope.

Sure, which helps to explain why that was done first, thus leaving a
greater proportionate scope for improvement in the LNA noise.

I am not really sure about the current distance ranges in radio
astronomy (except for the CMB which is over a fairly narrow band of
frequencies [and essentially omnidirectional]).

Yes but that uses differential radiometers, a very
different technique.

Which helps to justify me ignoring that particular knowledge

Consequently I can only
currently express this by analogy with optical astronomy. In that area,
I understand we already have evidence of galaxies at about 1.5 billion
light years from the big bang.

The put out a lot more power though.

Not necessarily. Such observations tend to take advantage of
gravitational lensing and/or extremely long exposures, and suggest the
existence of a wealth of different types of galaxies similar to those
we observe locally. (We are not talking about quasars here)

A better than 10% improvement in optical
range would, in principle, then seem to take us all the way to the
presumed beginning of time.

No, the big problem there is red shift. To go any
farther back than we have, we need to move to
deeper infra-red, hence the James Webb.

I think you may be missing something here. The EM spectrum is a
continuum.
Optical - infrared - microwave - lower RF, so, in principle there is
no problem going beyond, say, Z = 1000, if there iwas anything there
to see.

To assess the improvement just bear in mind the
received signal falls as the inverse square of the
range so a 10% reduction in noise for the same SNR
means range increases by sqrt(1.1) = 1.0488 or an
improvement of 4.88%.

You have failed to notice that ! have already borne that in mind.
Look at the Deep Space Network noise contributions again. (Subsequent
to plumbing improvements). From memory, the LNA now contributes ~ 27%.
Reducing that by 90% thus increases the range by ~ 12 %

When this level of improvement is applied by analogy to the current
state-of-the-art in optical/radio astronomy (ie EM spectrum astronomy),
this does, in fact, take us to the presumed time of the big bang.

John Bell
http://accelerators.co.uk
(Change John to Liberty to bypass anti-spam email filter)

I have already partially answered this using Chalky's computer, (which
also connects to our server). [However, this will probably mean that I
will now be listed as Chalky (from chalkie's cookie), when those
response(s) get published.]

wrote:

"John (Liberty) Bell" wrote in message
...

Picture yourself in the manufacturer's test lab being
asked to give the value the sales department should
put in the glossy brochure to get maximum orders but
without getting sued.

I would have been content to specify noise temperature.
Now that you have more clearly explained noise figure, and how it is
measured in practice, I would be happy to specify that too, if helpful.

snip

Hiowever, on the basis of those above statements, I would now like to
bring up a more interesting question (that I am surprised you have not
brought up yourself).

Given the above, and the fact that, in practical applications, the LNA
will typically also need a 50 ohm matching resistor from signal to
ground for transmission line impedance matching purposes, prior to
connection to the actual transistor gate;- why does the source Noise T
not reflect that matching resistor's noise T (which is ambient T),
within the determination of total noise?

Go back to the definition of noise figure as the ratio
of the internal noise to that in the input matching
termination expressed in dB.

The thermal noise power in a resistor is kTB which is
independent of resistance. Placing one resistor in
parallel with another therefore does not change the
total noise power. In practice, some of the time the
noise out of one resistor will happen to be of the
opposite polarity to that from the other since they
are uncorrelated.

Hence placing an extra resistor inside the amp in
parallel with the input doesn't change the input
noise power and when multiplied by the power gain
that gives the same reference value of output noise
power. The noise temperature of the amp is a measure
of how much the amp noise output exceeds that
reference value so is also unaffected.

This, however, gets trickier to explain when the first 'resistor'
actually has a noise temperature of 11K (since it is, in, fact, the
source impedance of the radio telescope), whereas the second resistor
has a noise T of 300K.. (I am talking here about actual noise generated
in a practical application, not laboratory testing of noise figure,
which is something else.)


My answer to that question would be empirical. The evidence shows that
it doesn't. I would give the same answer to the corresponding
question:- why does the antenna noise not reflect the ambient T of the
antenna? Because the evidence demonstrates otherwise.

The answer that is that it partially does. Consider a
mirror which reflects heat in the form of infra-red. If
it emitted well it could lose energy by radiation without
absorbing it so it could cool itself without using energy.
That cold mirror could then run a Stirling engine hence
you violate conservation of energy - a perpetual motion
machine. Good reflectors must be poor emitters.

The DSN dish is an RF mirror which reflects virtually all
the incoming radiation to the receiver, so it must be a
very poor emitter and very little thermal radiation comes
from the antenna compared to what there would be if it
was painted matt black.

If we look again at
http://www.skatelescope.org/document...8MTTSPaper.pdf, figure 2
clearly shows an antenna matching impedance of 50 ohms, connected
between earth and the beginning of the amplifier stage. Nevertheless,
table 1 equally clearly shows that the total noise of the amplifier
stage at an ambient T of 300K is 23K not 300K. Why would that be?

Good question and it raises another, but let me first
answer yours. As above, the noise temperature of 23K
is a measure of the amount by which the output exceeds
the gain times the thermal power in the input resistance.

No, I don't think so. You seem to be relying on the definition of noise
figure again.

The noise temperature is simply a measure of the total noise generated
by the amplifier, divided by the amplifier's gain. That will, by
definition, include those contributions from circuitry between the
physical input connection and the first transistor gate.

The amp is only adding about 7.7% to the amount that
would be there due to the gain times the noise from the
input resistance alone.

No. The total noise T of the amp (including that resistor) is only
about 7.7% of what one might expect from that resistor alone (if
connected in series, as in NF tests). Otherwise Tsyst would still be
300K

Now the question that raises in my mind is this - I
thought HEMT amplifiers were particularly good because
the electrons were able to cool by emitting into space.

I don't think so. I believe this comes about from the name High
Electron Mobility Transistor. High electron mobility means the
electrons can move more in the desired directions without banging into
obstructions (which is what generates the noise).

With a dummy load in place, they can only cool until
they reach equilibrium which must be the same
temperature as the source termination on thermodynamic
grounds. Surely then the performance would depend on
the temperature of the terminator. I haven't figured
that one out yet.

One
possibility is that this resistor is now connected in parallel with a
real source, not in series, and so its noise energy has three likely
exit routes. One is direct to ground via a zero resistance path. The
next is to ground via the antenna's 50 ohm impedance, with the least
likely route being into the higher impedance of the transistor gate.

That would allow the transistors to cool

I don't think this has anything to do with the transistor cooling. In
order for this resistor to degrade the noise performance of the amp,
its noise power must be pushed through that first transistor gate. This
you can do in your NF test, by placing it effectively in series with
the signal source (which is ground, in practice) . This you cannot do
in a real application, because we are then amplifying a real signal,
not earth potential, and the resistor is then in parallel with that
source, not in series.

but the matching
would be screwed (real source or terminator but not both)

Why? I understood that if the real source is connected to the amp via a
transmission line, such a terminating resistor would, in fact, be
needed.

I can, however, see how we might be able to dispense with that
reasistor if we dispense with the transmission line. That is something
I argued for much earlier, but you argued against.

and I doubt the guys had access to a DSN dish to try this
out. I'm sure they would be running this in the university
lab on the bench, at least for the 300K design.

OK, so in practice, you would use a noisier source, and subtract that
noise (x gain) from the measured output noise of the amp. This
consideration is academic here, however, because the analysis of that
paper was theoretical. The amplifiers were not actually built or tested
on a bench.

John (Liberty) Bell wrote:
It is only in situations where we know that the source will typically
have a noise temperature of ~ 300K, that Noise Figure is meaningfully
specified.

That is the case for equipment other than cooled
amps.

It is not the case in radio astronomy even when the amp is not cooled.
(See eg Tamp/Tsys for 300K , in Table 1 of mutually repeated
reference).

((That is, in fact, the application area we are supposed to be
discussing.))

Indeed but we seemed to have picked up the general
use of "noise figure" as an aside.

In fact, the term "noise figure" was introduced by you, resulting in
considerable confusion, in both directions. Hopefully, that confusion
is now over.

The RF gear we
make is used for ship-to-shore comms, air traffic
control, coastguards and the like and noise figure
is used pretty universally. It is horses for courses
really.

Sure. Nevertheless, noise temperature specification is the convention
in the arena of radio astronomy and, since I also find it far easier to
use, I would prefer us to keep to that convention from now on, at
sci.astro.research.

John Bell
(Change John to Liberty to bypass anti-spam email filter)

Chalky wrote:
I am just going to concentrate primarily on the astronomy side of this
response right now, as I would like a bit of a break from discussing
engineering details, (as I guess other readers might too).

Sounds good to me.

wrote:
"John (Liberty) Bell" wrote in message
...

snip - pretty much agreed

I am not really sure about the current distance ranges in radio
astronomy (except for the CMB which is over a fairly narrow band of
frequencies [and essentially omnidirectional]).

Yes but that uses differential radiometers, a very
different technique.

Which helps to justify me ignoring that particular knowledge

Consequently I can only
currently express this by analogy with optical astronomy. In that area,
I understand we already have evidence of galaxies at about 1.5 billion
light years from the big bang.

The put out a lot more power though.

Not necessarily. Such observations tend to take advantage of
gravitational lensing and/or extremely long exposures, and suggest the
existence of a wealth of different types of galaxies similar to those
we observe locally.

http://www.arxiv.org/abs/astro-ph/0503116
http://www.arxiv.org/abs/astro-ph/0510697


http://uk.news.yahoo.com/13092006/32...-universe.html

"They found hundreds of galaxies at redshifts around
900 million years after the Big Bang.

But when they looked at higher redshifts, at about
700 million years after the Big Bang, they found
unconfirmed evidence for only one galaxy, when
they had expected to find many more.

This backs theories about a "hierarchical" formation
of big galaxies -- that these huge clusters were built
up over time as smaller galaxies collided and merged,
they believe.

"The bigger, more luminous galaxies were just not in
place at 700 million years after the Big Bang," said
Illingworth.

"Yet 200 million years later, there were many more of
them, so there must have been a lot of merging of
smaller galaxies during that time."

Try this calulator to see how the redshift changes from
700Ma to 900Ma and beyond.

http://www.astro.ucla.edu/~wright/CosmoCalc.html

(We are not talking about quasars here)

Indeed. A more relevant point is that such observations
are limited by the finite number of photons received and
CCD quantum efficiency is more important. I'm not even
sure where you idea would fit into optical observations.

A better than 10% improvement in optical
range would, in principle, then seem to take us all the way to the
presumed beginning of time.

No, the big problem there is red shift. To go any
farther back than we have, we need to move to
deeper infra-red, hence the James Webb.

I think you may be missing something here. The EM spectrum is a
continuum.
Optical - infrared - microwave - lower RF, so, in principle there is
no problem going beyond, say, Z = 1000, if there iwas anything there
to see.

Agreed but there may be little to see by then. This
is the record holder at z = 6.96 and the chances are
the first stars formed at less than z=14.

http://arxiv.org/abs/astro-ph/0609393

You won't see stars or galaxies redshifted to RF
frequencies.

To assess the improvement just bear in mind the
received signal falls as the inverse square of the
range so a 10% reduction in noise for the same SNR
means range increases by sqrt(1.1) = 1.0488 or an
improvement of 4.88%.

You have failed to notice that ! have already borne that in mind.

Possibly, anyway we both understand how it works.

Look at the Deep Space Network noise contributions again. (Subsequent
to plumbing improvements). From memory, the LNA now contributes ~ 27%.
Reducing that by 90% thus increases the range by ~ 12 %

When this level of improvement is applied by analogy to the current
state-of-the-art in optical/radio astronomy (ie EM spectrum astronomy),
this does, in fact, take us to the presumed time of the big bang.

Getting from t=700 million years to 378000 years goes
from z=7.6 to z=1090 and the Tolman test says the
brightness goes as (1+z)^4 so requires an increase in
sensitivity of about 260 million times. That takes you
from 13 billion years ago to 13.69622 billion assuming
an age of exactly 13.7 billion. Since a photon's energy
is proportional to its frequency, redshift has a very
significant impact.

The CMBR peaks at around 160 GHz and stars at z~11
would be redshifted by a factor of 100 less, and their
surfaces are perhaps another order of magnitude hotter.

There is one further angle I would like to cover while we are on the
subject of level of performance improvement that can theoretically be
achieved with this LNA design concept. As George doubtless knows, all
electronic designs involve a tradeoff. At microwave frequencies, a
variety of effects come into play which can affect this tradeoff, thus
limiting the performance improvement that can be achieved in practice,
using my technique. Consequently, at such frequencies I think it is
still desirable to use the lowest noise (and highest cost) transistor
available, at the input stage. However, at the lower end of the
frequency range of radio astronomy, it does currently seem that the
technique could possibly be applied more ambitiously, within a single
chip CMOS design, to additionally overcome the inherently higher noise
generated by such CMOS transistors (at the cost of increased power
consumption). Whether that is worth doing in practice, would presumably
depend on the quantity required, as well as other engineering factors.

To build your prototypes I would suggest single transistors
and if you can get that to work then it may persuade a
backer to fund a chip design.

George

wrote:

The DSN dish is an RF mirror which reflects virtually all
the incoming radiation to the receiver, so it must be a
very poor emitter and very little thermal radiation comes
from the antenna compared to what there would be if it
was painted matt black.

snip

Now the question that raises in my mind is this - I
thought HEMT amplifiers were particularly good because
the electrons were able to cool by emitting into space.
With a dummy load in place, they can only cool until
they reach equilibrium which must be the same
temperature as the source termination on thermodynamic
grounds. Surely then the performance would depend on
the temperature of the terminator. I haven't figured
that one out yet.


I have given these matters some further thought since my last response,
and it seems to me that we may be approaching this noise question from
opposing starting points.

Your starting point seems to be the assumption that basic thermal noise
formulae are infallible, which leads you to propose thermodynamic
explanations for why they sometimes seem to give erroneous results.

My starting point is the assumption that they are not infallible. They
predict the thermal agitation of charge carriers in a conductor, on the
assumption that the degree of coupling between free charge carriers and
the thermal agitation of atoms and molecules is independent of
temperature and frequency (the temperature dependence coming from the
degree of agitation, rather than the degree of coupling).

However, we know that this is not universally true. In superconductors,
these atoms and molecules provide no resistance whatsoever, so, by
implication, they should not impart electrical noise. Similarly in
HEMTs, they appear to create less interaction, and so, again impart
less noise.

Finally, we can look again at radio astronomy antennae. Here, you may
have been right in the sense that those atoms could have less thermal
agitation at radio frequencies than at other frequencies, and so impart
less noise at those frequencies. However, it would be misleading to
conclude from this that they have a substantially lower temperature,
since temperature is an unambiguous measure of overall thermal
agitation of atoms and molecules. (I am sure that if I put my hand on a
dish in the tropics, it would not immediately be frozen to that dish by
the cold of deep space.).

Consequently, the reason why these have such low noise temperatures is
probably the opposite of what you have just suggested. These devices
are designed to be particularly good detectors of radio frequencies. By
the corollary of your argument, they should also thus be particularly
good radiators at those frequencies. Consequently, the thermal
agitation of atoms at those frequencies is radiated into space, thus
leaving less that can couple to the desired signal.

Now, whether that behaviour also pumps thermal energy out of a final
impedance matching resistor at such frequencies, is an interesting
question over which I will let you contemplate the answer.


John Bell
http://accelerators.co.uk
(Change John to Liberty to bypass anti-spam email filter)