Amplifiers and return loss...

[billcalley]

Hi All!

When I read about a particular discrete amplifier circuit, and
they begin discussing the S11 and the S22 of the amplifier, the return
loss is normally only around -13dB or less. Some are even less than
-9dB! What I don't quite understand is: Why can't designers
*significantly* improve the amplifier's return loss to far better
levels, since the matching circuit (if designed correctly) will "force"
the transistor to see the conjugate of its impedance; as well as also,
I would think, force the outside world to see an almost *perfect* 50
ohm match. In other words, why can't the input/output return losses of
real amplifiers be, let's say, better than -20 or -30dB (within a
certain narrow frequency band), rather than the mediocre -13dB or less
levels? And if this cannot be done for whatever reason, why not just
add a simple L network at the already matched amplifier's ports to
better the return loss to much greater levels?

Thanks,

-Bill

 

[John Larkin]

Hi All!

When I read about a particular discrete amplifier circuit, and
they begin discussing the S11 and the S22 of the amplifier, the return
loss is normally only around -13dB or less. Some are even less than
-9dB! What I don't quite understand is: Why can't designers
*significantly* improve the amplifier's return loss to far better
levels, since the matching circuit (if designed correctly) will "force"
the transistor to see the conjugate of its impedance; as well as also,
I would think, force the outside world to see an almost *perfect* 50
ohm match. In other words, why can't the input/output return losses of
real amplifiers be, let's say, better than -20 or -30dB (within a
certain narrow frequency band), rather than the mediocre -13dB or less
levels? And if this cannot be done for whatever reason, why not just
add a simple L network at the already matched amplifier's ports to
better the return loss to much greater levels?

Thanks,

-Bill

I've noticed that most InGaP MMICS have rotten S11s, typically running
around 30 ohms wideband Zin. I'm guessing they go for better noise
figures at the expense of matching. Some of the SiGe parts are very
close to 50 ohms... maybe that's because their NFs are better there.

John

 

[Tom Bruhns]

I think you need to determine first if you really care about input or
output return loss. There are reasons to not worrying about it, and
other reasons why you would want to worry about it. We make one piece
of equipment that has a guaranteed 46dB input return loss, as I recall.
I remember thinking it's not much fun to have to test that and
guarantee it over some range of temperatures, etc. It certainly IS
possible to design a circuit to present specific input and output
impedances, at least at certain frequencies.

But as John L. pointed out, the best noise figure is generally not for
an impedance matched input in a simple amplifier; and a power amplifier
may be designed to operate to spec into a 50 ohm load, while itself
presenting some much different source impedance. As an extreme
example, consider a hifi audio amplifier that might present an 0.1 ohm
or lower source resistance, but be designed to operate optimally into a
4 to 8 ohm load. Or consider the source impedance of a 120V (or 240V)
household power outlet. You might see a volt or two drop at full load,
say 15 amps, representing a source resistance well below an ohm. But
the load resistance for proper operation probably should not be below 8
ohms or so!

Cheers,
Tom

 

[Wes Stewart]

I think you need to determine first if you really care about input or
output return loss. There are reasons to not worrying about it, and
other reasons why you would want to worry about it. We make one piece
of equipment that has a guaranteed 46dB input return loss, as I recall.
I remember thinking it's not much fun to have to test that and
guarantee it over some range of temperatures, etc. It certainly IS
possible to design a circuit to present specific input and output
impedances, at least at certain frequencies.

But as John L. pointed out, the best noise figure is generally not for
an impedance matched input in a simple amplifier; and a power amplifier
may be designed to operate to spec into a 50 ohm load, while itself
presenting some much different source impedance. As an extreme
example, consider a hifi audio amplifier that might present an 0.1 ohm
or lower source resistance, but be designed to operate optimally into a
4 to 8 ohm load. Or consider the source impedance of a 120V (or 240V)
household power outlet. You might see a volt or two drop at full load,
say 15 amps, representing a source resistance well below an ohm. But
the load resistance for proper operation probably should not be below 8
ohms or so!

Cheers,
Tom

In addition to Tom's excellent points, consider how S-parameters are
defined and measured.

By definition, all ports are perfectly terminated in the system
impedance and the reflection coefficients and transmission
coefficients are measured one at a time.

If for example, s22 is measured and s12 is not zero then s22 is
influenced by the termination on port 1. In the measurement system,
s11 is zero but in actual operation it is something else, thus s22 is
also something else.

So in the usage environment unless all of the external terminations
are perfect, the s-parameters, while not useless, are deceptive. Gain
and stability can be predicted using them, but only if -all- of them
are considered simultaneously.

 

[Mark]

In other words, why can't the input/output return losses of
real amplifiers be, let's say, better than -20 or -30dB (within a
certain narrow frequency band),


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

Bill,

yes it is not hard to design a matching circuit over a narrow range of
frequencies if that were needed.

It is more difficult to match over a broad range of frequencies.

The greater the transformation needed, the more difficult it is to make
it broadband.

Matching at a single frequency is usually easy.

Mark

 

[Joerg]

Hello Mark,

Matching at a single frequency is usually easy.

Not always. Sometimes you either need narrow tolerance caps (meaning
expensive) or hand tuning in production (very much frowned upon). PCB
material can become quite expensive as well if you want a tightly
toleranced dielectric coefficient.

Regards, Joerg

 

[John Larkin]

In other words, why can't the input/output return losses of
real amplifiers be, let's say, better than -20 or -30dB (within a
certain narrow frequency band),


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

Bill,

yes it is not hard to design a matching circuit over a narrow range of
frequencies if that were needed.

It is more difficult to match over a broad range of frequencies.

The greater the transformation needed, the more difficult it is to make
it broadband.

Matching at a single frequency is usually easy.

Mark

Some of the Sirenza SiGe wideband mmic amps are very close to 50 ohms
from DC to several GHz. In fact, you can adjust their operating
current to tweak them to exactly 50, or anywhere from about 40 to 60
ohms, never mind that Sirenza said it wouldn't work. I have some
graphs around here somewhere...

If you can stand a 2-3 dB noise figure, it's hard to beat a 99-cent
MMIC as a gain element.

John