Ultra WideBand Low Noise Amplifier Design

[Mitch]

I'm currently working on a project for school about UWB LNA's. There
seems to be several different opinions about circuits that work best
in UWB applications. However, there doesn't seem to be any concrete
proof and due to my lack of experience, i'm finding it hard to choose
a circuit topology.

There are four main ideas I'm looking at:

1. Single cmos transistor amplifier (or possibly make it a darlington
pair)
- lots of material
- BW may not be large enough ???
- gain flatness

2. Differential pair
- BW large enough ??
- low noise

3. New Filter configuration
- leading researchers propose LC filters with gain or chebyshev
filter with high gain in pass band
- difficult to find material

4. Distributed Amplifiers
- power consumption issue
- need transmission lines
- practicaly any BW achievable

Choosing a good circuit can be a little tricky because UWB requires
about a bandwidth of over 7 GHz. At the moment Im leaning toward
number 4, the distributed amplifiers because they can also be
positioned to remove the 5.1 GHz to 5.3 GHz range which is occupied by
another wireless technology.

Does anyone have any experience in UWB or wideband LNAs ? I'd
appreciate any help to choose the best topology.

thanks,
Mitch

 

[Rene Tschaggelar]

I'm currently working on a project for school about UWB LNA's. There
seems to be several different opinions about circuits that work best
in UWB applications. However, there doesn't seem to be any concrete
proof and due to my lack of experience, i'm finding it hard to choose
a circuit topology.

There are four main ideas I'm looking at:

[ snip ]

Choosing a good circuit can be a little tricky because UWB requires
about a bandwidth of over 7 GHz. At the moment Im leaning toward
number 4, the distributed amplifiers because they can also be
positioned to remove the 5.1 GHz to 5.3 GHz range which is occupied by
another wireless technology.

Does anyone have any experience in UWB or wideband LNAs ? I'd
appreciate any help to choose the best topology.

Mitch,
You could help by giving some information about the intended
amplifiers. Just knowing the bandwidth, ahem 7GHz, is not sufficient.

Where is the centerfrequency ? Or is it DC to 7GHz ?
Gain ? Noise figure ? Output power ?

Rene

 

[Mitch]

I'm currently working on a project for school about UWB LNA's. There
seems to be several different opinions about circuits that work best
in UWB applications. However, there doesn't seem to be any concrete
proof and due to my lack of experience, i'm finding it hard to choose
a circuit topology.

There are four main ideas I'm looking at:

[ snip ]

Choosing a good circuit can be a little tricky because UWB requires
about a bandwidth of over 7 GHz. At the moment Im leaning toward
number 4, the distributed amplifiers because they can also be
positioned to remove the 5.1 GHz to 5.3 GHz range which is occupied by
another wireless technology.

Does anyone have any experience in UWB or wideband LNAs ? I'd
appreciate any help to choose the best topology.

Mitch,
You could help by giving some information about the intended
amplifiers. Just knowing the bandwidth, ahem 7GHz, is not sufficient.

Where is the centerfrequency ? Or is it DC to 7GHz ?
Gain ? Noise figure ? Output power ?

Rene

haha ok, I figured that these specs would be very obvious. But I guess
that is not the case. They were never really defined for me so I did a
little reading a choose to aim for these:

- Gain: as much as possible ( minimum 5 or 6 dB and flat over 7
GHz )
- NF: 3 to 4 dB
- Bandwidth: 3.1 GHz to 10.6 GHz
- Center Frequency: none in UWB ( but I guess you could
do 10.6 - 3.1 = 7.5 GHz )
- Ouput power: not sure about this one ( I assume this means
enough power to drive the next stage in the IC. We are working in cmos
0.18um technology )

The whole entire transceiver is going to be integrated into 0.18um
cmos technology and I assume I am matching to 50 ohms input and
output. We are not implementing a preselect filter or the antenna
which would be before the LNA so I am assuming I receive an
appropriate signal as input.

For the output, the LNA must drive the correlator which is imediately
after the LNA. The correlator is composed of a mixer followed by an
integrator. So, all this to say, the LNA must drive the mixer. I dont
know if this helps to determine the required output power or if it
even applies here. I haven't seen anything about this in my research.

I found an interesting summary of a paper on UWB development. There is
a going to be a paper presented at the next IEEE meeting on Chebyshev
Filter used as LNAs for UWB. The abstract of the paper described a
system capable of a gain of 9.? dB and NF of 4 dB over the 3.1 to 10.6
GHz range. I dont expect to go as far as an experienced researcher,
however, this seems like the best-to-date topology and could be
promising. Im looking into this topic at the moment.

The other method that shows alot of promise is to use a distributed
topology. Here we can split up the bandwidth into 3 or 4 chunks like
so:
(view in fixed witdh font)


_____ _____ _____
/ \/ \/ \
| | | |
-|--------------------------------
0 3 10 GHz


This (in my view) seems like the most appropriate way to achieve a
good LNA. The other aspect that I appreciate about this topology is
that it will allow for band blocking. This is especially important
around the 5.1 to 5.3 GHz range where another wireless technolody
exists. The only catch with distributed amplifier system is that its
going to required alot of power, relative to the other topologies. But
i dont expect that to be a problem.

The other tricky part will be to implement the inductors and
capacitors in the distributed topology as transmission line instead of
lumped components. For space-saving and quality my supervisor wont
allow inductors. There is apparently active circuits which will
simulate inductors but I have yet to understand them.

The problem i have here is that there are many circuits available for
this application but I have very little time to implement them. Id try
each one out individually if i had time but unfortunately i dont. So
id im hoping you, or someone can help me get started on the most
appropriate one. As you can see, my choice would be the distributed
amplifier topology. What do you think?

 

[John Woodgate]

I read in sci.electronics.design that Mitch
osting.google.com>) about 'Ultra WideBand Low Noise Amplifier Design',

The other tricky part will be to implement the inductors and
capacitors in the distributed topology as transmission line instead of
lumped components. For space-saving and quality my supervisor wont
allow inductors.

OMG! At your frequencies, EVERYTHING is an inductor. Furthermore,
nothing that looks like a conventional inductor is relevant for your
frequency range.

There is apparently active circuits which will
simulate inductors but I have yet to understand them.

Don't bother; they are irrelevant at your frequencies.

The problem i have here is that there are many circuits available for
this application but I have very little time to implement them. Id try
each one out individually if i had time but unfortunately i dont. So
id im hoping you, or someone can help me get started on the most
appropriate one. As you can see, my choice would be the distributed
amplifier topology. What do you think?

I think you should seek another post, with intelligent management that
does not impose unrealistic time-scales.

 

[Mike]

I read in sci.electronics.design that Mitch
osting.google.com>) about 'Ultra WideBand Low Noise Amplifier Design',


OMG! At your frequencies, EVERYTHING is an inductor. Furthermore,
nothing that looks like a conventional inductor is relevant for your
frequency range.

Hold on a second, John. He's implementing this in a 0.18u CMOS process, so
dimensions scale down quite a bit. I've built 2.5GHz SONET chips in a
similar (if not the same) process, and we didn't worry about on-chip wire
inductance at all. Package, bond-wire, and board inductance are another
matter, of course, but on the chip, capacitance is still a much larger
problem than inductance.

-- Mike --

 

[John Larkin]

The other method that shows alot of promise is to use a distributed
topology. Here we can split up the bandwidth into 3 or 4 chunks like
so:
(view in fixed witdh font)


_____ _____ _____
/ \/ \/ \
| | | |
-|--------------------------------
0 3 10 GHz


This (in my view) seems like the most appropriate way to achieve a
good LNA. The other aspect that I appreciate about this topology is
that it will allow for band blocking. This is especially important
around the 5.1 to 5.3 GHz range where another wireless technolody
exists. The only catch with distributed amplifier system is that its
going to required alot of power, relative to the other topologies. But
i dont expect that to be a problem.

Too bad it must be fully integrated; a 75 cent InGaAs or SiGe MMIC
would do this with ease.

The classic "distributed amplifier" is a number of fets "in parallel"
spaced along gate (input) and drain (output) transmission lines. Here,
the Gm's add up but the parasitic capacitances don't. This integrates
well with serpentine transmission lines, and the GBW and NF are both
better than you can get with a single device. Basicly you can split
one large FET into three or four small ones along the line. It's flat
wideband, usually, so won't give you frequency notches.

You might take a peek at Wong's book, Fundamentals of Distributed
Amplification. It's OK.

Tektronix considered the frequency-band splitting idea for the 7104 1
GHz analog scope, but did it another way in the end. See Jim Williams'
first book.


John

 

[Mike]

The other tricky part will be to implement the inductors and
capacitors in the distributed topology as transmission line instead of
lumped components. For space-saving and quality my supervisor wont
allow inductors. There is apparently active circuits which will
simulate inductors but I have yet to understand them.

In general, "active inductor" and "low noise" are mutually exclusive. Make
sure you understand the noise contribution before you use them...

Inductors do take space, and for a wideband amplifier they might not be
much good, since the most common use is in tank circuits. On the other
hand, you're talking about implementing a distributed amplifier or a sum of
parallel bandbass sections - both of which take space and power; if you had
to make a choice, it seems to me that inductors might not be such a bad
thing.

There are several professors there at Carleton with some RF CMOS
experience: Rogers and Plett have published a book, Kwasniewski has done a
fair bit of work in GHz CMOS. Have you talked to them?

-- Mike --

 

[John Larkin]

Hold on a second, John. He's implementing this in a 0.18u CMOS process, so
dimensions scale down quite a bit. I've built 2.5GHz SONET chips in a
similar (if not the same) process, and we didn't worry about on-chip wire
inductance at all. Package, bond-wire, and board inductance are another
matter, of course, but on the chip, capacitance is still a much larger
problem than inductance.

-- Mike --

I've seen GaAs integrated distributed amps that use on-chip spiral
inductors; at low-GHz frequencies, they take less chip area than
serpentine transmission lines. But unless you really want inductance,
ICs don't have much.

Some cute tricks have been done with wire bonds, though.

John

 

[John Woodgate]

<[email protected]>) about 'Ultra WideBand Low
Noise Amplifier Design', on Sat, 3 Jan 2004:

Hold on a second, John. He's implementing this in a 0.18u CMOS process, so
dimensions scale down quite a bit.

I didn't see that; it was in his follow-up post. So now I don't
understand why his supervisor won't let him use inductors. Surely the
process practically rules that out anyway?

I've built 2.5GHz SONET chips in a
similar (if not the same) process, and we didn't worry about on-chip wire
inductance at all. Package, bond-wire, and board inductance are another
matter, of course, but on the chip, capacitance is still a much larger
problem than inductance.

Of course I agree with both of those points.

 

[Mike]

I didn't see that; it was in his follow-up post. So now I don't
understand why his supervisor won't let him use inductors. Surely the
process practically rules that out anyway?

At GHz frequencies, integrated inductors are pretty common, even if the Q's
are limited (the upper bound is typically 10 or less). In 2.5GHz SONET, I'd
guess that half the parts in production use integrated inductors, and half
don't. In RF, virtually every part uses integrated inductors.

-- Mike --

 

[Robert Baer]

I'm currently working on a project for school about UWB LNA's. There
seems to be several different opinions about circuits that work best
in UWB applications. However, there doesn't seem to be any concrete
proof and due to my lack of experience, i'm finding it hard to choose
a circuit topology.

There are four main ideas I'm looking at:

[ snip ]

Choosing a good circuit can be a little tricky because UWB requires
about a bandwidth of over 7 GHz. At the moment Im leaning toward
number 4, the distributed amplifiers because they can also be
positioned to remove the 5.1 GHz to 5.3 GHz range which is occupied by
another wireless technology.

Does anyone have any experience in UWB or wideband LNAs ? I'd
appreciate any help to choose the best topology.

Mitch,
You could help by giving some information about the intended
amplifiers. Just knowing the bandwidth, ahem 7GHz, is not sufficient.

Where is the centerfrequency ? Or is it DC to 7GHz ?
Gain ? Noise figure ? Output power ?

Rene

....maybe it is ultra narrow-band because the center frequency is "UV"
(light).

 

[Robert Baer]

I'm currently working on a project for school about UWB LNA's. There
seems to be several different opinions about circuits that work best
in UWB applications. However, there doesn't seem to be any concrete
proof and due to my lack of experience, i'm finding it hard to choose
a circuit topology.

There are four main ideas I'm looking at:

[ snip ]

Choosing a good circuit can be a little tricky because UWB requires
about a bandwidth of over 7 GHz. At the moment Im leaning toward
number 4, the distributed amplifiers because they can also be
positioned to remove the 5.1 GHz to 5.3 GHz range which is occupied by
another wireless technology.

Does anyone have any experience in UWB or wideband LNAs ? I'd
appreciate any help to choose the best topology.

Mitch,
You could help by giving some information about the intended
amplifiers. Just knowing the bandwidth, ahem 7GHz, is not sufficient.

Where is the centerfrequency ? Or is it DC to 7GHz ?
Gain ? Noise figure ? Output power ?

Rene

haha ok, I figured that these specs would be very obvious. But I guess
that is not the case. They were never really defined for me so I did a
little reading a choose to aim for these:

- Gain: as much as possible ( minimum 5 or 6 dB and flat over 7
GHz )
- NF: 3 to 4 dB
- Bandwidth: 3.1 GHz to 10.6 GHz
- Center Frequency: none in UWB ( but I guess you could
do 10.6 - 3.1 = 7.5 GHz )
- Ouput power: not sure about this one ( I assume this means
enough power to drive the next stage in the IC. We are working in cmos
0.18um technology )

The whole entire transceiver is going to be integrated into 0.18um
cmos technology and I assume I am matching to 50 ohms input and
output. We are not implementing a preselect filter or the antenna
which would be before the LNA so I am assuming I receive an
appropriate signal as input.

For the output, the LNA must drive the correlator which is imediately
after the LNA. The correlator is composed of a mixer followed by an
integrator. So, all this to say, the LNA must drive the mixer. I dont
know if this helps to determine the required output power or if it
even applies here. I haven't seen anything about this in my research.

I found an interesting summary of a paper on UWB development. There is
a going to be a paper presented at the next IEEE meeting on Chebyshev
Filter used as LNAs for UWB. The abstract of the paper described a
system capable of a gain of 9.? dB and NF of 4 dB over the 3.1 to 10.6
GHz range. I dont expect to go as far as an experienced researcher,
however, this seems like the best-to-date topology and could be
promising. Im looking into this topic at the moment.

The other method that shows alot of promise is to use a distributed
topology. Here we can split up the bandwidth into 3 or 4 chunks like
so:
(view in fixed witdh font)

_____ _____ _____
/ \/ \/ \
| | | |
-|--------------------------------
0 3 10 GHz

This (in my view) seems like the most appropriate way to achieve a
good LNA. The other aspect that I appreciate about this topology is
that it will allow for band blocking. This is especially important
around the 5.1 to 5.3 GHz range where another wireless technolody
exists. The only catch with distributed amplifier system is that its
going to required alot of power, relative to the other topologies. But
i dont expect that to be a problem.

The other tricky part will be to implement the inductors and
capacitors in the distributed topology as transmission line instead of
lumped components. For space-saving and quality my supervisor wont
allow inductors. There is apparently active circuits which will
simulate inductors but I have yet to understand them.

The problem i have here is that there are many circuits available for
this application but I have very little time to implement them. Id try
each one out individually if i had time but unfortunately i dont. So
id im hoping you, or someone can help me get started on the most
appropriate one. As you can see, my choice would be the distributed
amplifier topology. What do you think?

Active circuits that simulate inductors or capacitors give rather
lossy (low Q) "components".

 

[Mitch]

At GHz frequencies, integrated inductors are pretty common, even if the Q's
are limited (the upper bound is typically 10 or less). In 2.5GHz SONET, I'd
guess that half the parts in production use integrated inductors, and half
don't. In RF, virtually every part uses integrated inductors.

-- Mike --

Update!!!

Just updating you guys on the LNA.... at the moment its producing 5.6
dB gain at the low end(3 GHz) and 3.6 dB and the high end(10 GHz). Of
course, this is no where near flat gain but there is still alot of
work to be done. This is still using a single transistor
configuration. I Still haven't tried implementing an Ft doubler or
cascode circuit. Some people have suggested making it differential
once everything is nicely working.

Mitch

Ps. This is just to rub it in to the master student that said it
couldn't be done with a single transistor.

 

[Mike]

Update!!!

Just updating you guys on the LNA.... at the moment its producing 5.6
dB gain at the low end(3 GHz) and 3.6 dB and the high end(10 GHz). Of
course, this is no where near flat gain but there is still alot of
work to be done. This is still using a single transistor
configuration. I Still haven't tried implementing an Ft doubler or
cascode circuit. Some people have suggested making it differential
once everything is nicely working.

Mitch

Ps. This is just to rub it in to the master student that said it
couldn't be done with a single transistor.

If you haven't already, check out some of the recent ESSCIRC conference
papers - there have been several on CMOS LNAs (esscirc.org).

If you haven't yet included parasitics, you're in for a nasty surprise (in
which case, I hope you didn't rub it in too much, since last I checked,
humble pie is none too tasty).

-- Mike --