Radiofrequency amplifier.

[[email protected]]

Hi,

I am trying to build an amp capable of driving +/- 12 volts into a
50 ohm load over a range of frequencies from DC - ~100 mhz. As if that
wasn't hard enough, it needs to be about 2 cm^2 in size. Additionally
the output must be able to go into a high impedance state which
presents very little capacitance to the bus. I've gone through a lot
of different design ideas and have settled on this kind of weird
architecture. The schematic is here:

http://prymfg.com/schema.jpg

The output transistors are 2sk1403 and 2sc3597. The opamps are
actually ad8045's. Even with only one half of the push-pull, this
design isn't stable. At least according to spice it oscillates in the
5-65GHZ region. I'm not really sure if spice can be trusted at all.
How would a circuit like this be analyzed? Will it work or how can I
make it. I've tried things like using transconductance amplifiers,
using a mosfet as a second stage and adding feedback filters.

Thanks in advance.

Jon Pry

 

[[email protected]]

Hi,

    I am trying to build an amp capable of driving +/- 12 volts into a
50 ohm load over a range of frequencies from DC - ~100 mhz. As if that
wasn't hard enough, it needs to be about 2 cm^2 in size. Additionally
the output must be able to go into a high impedance state which
presents very little capacitance to the bus. I've gone through a lot
of different design ideas and have settled on this kind of weird
architecture. The schematic is here:

http://prymfg.com/schema.jpg

The output transistors are 2sk1403 and 2sc3597. The opamps are
actually ad8045's. Even with only one half of the push-pull, this
design isn't stable. At least according to spice it oscillates in the
5-65GHZ region. I'm not really sure if spice can be trusted at all.
How would a circuit like this be analyzed? Will it work or how can I
make it. I've tried things like using transconductance amplifiers,
using a mosfet as a second stage and adding feedback filters.

Thanks in advance.

Jon Pry

Jon, Perhaps using the Wilson Current Mirror for the output stage is a
solution. It has good linearity if you use matched transistors like
the lm394, etc.. Just parallel enough of them to drive 50 ohms at
12v. It's not rocket science, you should get a flat response out to
100k without feedback. You cant use feedback , it will lower your
output impedance.

Good luck and let me know what happens.

 

[Vladimir Vassilevsky]

Hi,

I am trying to build an amp capable of driving +/- 12 volts into a
50 ohm load over a range of frequencies from DC - ~100 mhz. As if that
wasn't hard enough, it needs to be about 2 cm^2 in size.

What are the requirements to linearity and gain flatness?

Additionally
the output must be able to go into a high impedance state which
presents very little capacitance to the bus. I've gone through a lot
of different design ideas and have settled on this kind of weird
architecture. The schematic is here:

http://prymfg.com/schema.jpg

The output transistors are 2sk1403 and 2sc3597. The opamps are
actually ad8045's. Even with only one half of the push-pull, this
design isn't stable. At least according to spice it oscillates in the
5-65GHZ region.

You betcha. This schematics has severe problems with stability. Especially
if you are planning on driving 50 Ohm cables.

I'm not really sure if spice can be trusted at all.

Only if the parasitics is accounted for.

How would a circuit like this be analyzed? Will it work or how can I
make it.

No it won't.

I've tried things like using transconductance amplifiers,
using a mosfet as a second stage and adding feedback filters.

Avoid general feedback.
I would use a common gate FET cascade as the output stage.


Vladimir Vassilevsky
DSP and Mixed Signal Consultant
www.abvolt.com

 

[MooseFET]

Hi,

    I am trying to build an amp capable of driving +/- 12 volts into a
50 ohm load over a range of frequencies from DC - ~100 mhz.

Is this 50 Ohm load always there during normal operation?

As if that
wasn't hard enough, it needs to be about 2 cm^2 in size. Additionally
the output must be able to go into a high impedance state which
presents very little capacitance to the bus.

If you always have the 50 Ohm loading, then the output impedance can
be always above 50 Ohms and reflections won't be a problem. You can
drive the load directly from the collector / drain with no need for
feedback from that node.

Is this 2 cm^2 connected to a big heatsink?

Can you have any supply voltages you want?

How big of an input signal can you get?

Can you stand distortion?


Consider something like the LT1939 doing this:


!+\
! >---!<---
!-/ !
+---
!+\ !
! >--->!---
!-/

The diodes can provide good isolation when the two drivers are shifted
to back bias both diodes.

I've gone through a lot
of different design ideas and have settled on this kind of weird
architecture. The schematic is here:

http://prymfg.com/schema.jpg

That looks like an oscillator to me.

 

[[email protected]]

Jon, Perhaps using the Wilson Current Mirror for the output stage is a
solution. It has good linearity if you use matched transistors like
the lm394, etc..  Just parallel enough of them to drive 50 ohms at
12v. It's not rocket science, you should get a flat response out to
100k without feedback. You cant use feedback , it will lower your
output impedance.

Good luck and let me know what happens.

I don't understand how a current mirror is going to work. I need the
common emitter outputs because they allow me to make the output
impedance go high. Also, a push-pull amp is necessary for that reason
and heat reduction.

 

[[email protected]]

What are the requirements to linearity and gain flatness?


You betcha. This schematics has severe problems with stability. Especially
if you are planning on driving 50 Ohm cables.


Only if the parasitics is accounted for.


No it won't.


Avoid general feedback.
I would use a common gate FET cascade as the output stage.

I need the common emitter BJT's so that the amp has low enough output
capacitance when off.

 

[[email protected]]

Is this 50 Ohm load always there during normal operation?

There can be loads of different impedance attached to the device. I
plan on using some kind of programmable back termination. Probably
using a 10 ohm resistor and then a variable amount of positive
feedback with a digital pot. I've left this out of the circuit as it
is an extra complication.

If you always have the 50 Ohm loading, then the output impedance can
be always above 50 Ohms and reflections won't be a problem.  You can
drive the load directly from the collector / drain with no need for
feedback from that node.

Is this 2 cm^2 connected to a big heatsink?

Yes, but there are going to be 16 of these amplifiers attached to it.

Can you have any supply voltages you want?

+/-12 are available. Other voltages could be made with converters or
regulators but they take away from my 2 cm^2

How big of an input signal can you get?

as big as an opamp can make. signal is coming from an adc.

Can you stand distortion?

i would like to make pretty clean triangle waves as 2 mhz and a sqaure
wave at 30mhz.

 

[[email protected]]

I'd say skip the entire "designing your own" thing and just specify an
opamp with output disable function.
Start here
http://www.maxim-ic.com/quick_view2.cfm/qv_pk/1655

You probably won't find one that easily meets your drive spec though.

 

[MooseFET]

There can be loads of different impedance attached to the device. I
plan on using some kind of programmable back termination. Probably
using a 10 ohm resistor and then a variable amount of positive
feedback with a digital pot. I've left this out of the circuit as it
is an extra complication.

At 100MHz, you are likely to have a lot of trouble making such a
circuit be mostly real impedance. The reactive component will
reflect.

With a 50 Ohm constant load I was going to suggest making the driver a
constant current source.


[...]

Yes, but there are going to be 16 of these amplifiers attached to it.

Water cooling or do we not count the fins?

5V /50 Ohms = 100mA

100mA * (12-5) = 0.7W

0.7W * 16 = 11.2W

This is with a class B output stage. With class A the situation will
be worse.

+/-12 are available. Other voltages could be made with converters or
regulators but they take away from my 2 cm^2

Can the DC-DC be shared?

 

[[email protected]]

Real impedance...

Since my signals are really only 30mhz. The power at the harmonics
should be kind of low. As long as I can absorb a decent amount of
reflected power, all is well.

Water cooling or do we not count the fins?

5V /50 Ohms = 100mA

100mA * (12-5) = 0.7W

0.7W * 16 = 11.2W

This is with a class B output stage.  With class A the situation will
be worse.

Some sort of class C or class AB is going to be the only option. There
will end up being some limit on the number of outputs that are
simultaneously driving some amount of current because of thermal
limits. Keep Q low is going to be very important.

Can the DC-DC be shared?

Yes

 

[JosephKK]

There can be loads of different impedance attached to the device. I
plan on using some kind of programmable back termination. Probably
using a 10 ohm resistor and then a variable amount of positive
feedback with a digital pot. I've left this out of the circuit as it
is an extra complication.


Yes, but there are going to be 16 of these amplifiers attached to it.

+/-12 are available. Other voltages could be made with converters or
regulators but they take away from my 2 cm^2


as big as an opamp can make. signal is coming from an adc.


i would like to make pretty clean triangle waves as 2 mhz and a sqaure
wave at 30mhz.

Sounds like fun. Some 25 years ago i was building a pulse amplifier
that would deliver 10 v/v with 50 v into 50 ohms with 3 ns edges and
flat tops. The proof of design unit proved the design, but the
project got killed. I only needed another couple of weeks and a lot
more not cheap (about $3 each) transistors. No "high z mode" possible
in the design, but open circuit and short circuit safe (at least for a
few seconds).