Methods to to effective broadband matching for amplifier

[sze wen]

I have done load pull simulations and found different Zopt for the 2GHz and 3 GHz respectively.
The Zopt will give me the maximum power transfer at the load side at different frequency. Hence needs 2 different matching network it seems to me.
Is T or Pi network good enough for broadband matching 2 GHz to 3 GHz?
How do we ensure that the output power and efficiency are still good enough?

 

[sze wen]

What matching methodology is recommended.

 

[BradtheRad]

Two different topics: 1) bandpass filters and 2) impedance matching...

It becomes difficult to combine both purposes into one filter. Furthermore input resistance might be unknown and unseen coming from your preceding circuitry.

We have the formula that tells what center frequency is produced by a given LC combination. LC ratio makes a difference in rolloff curve. To keep things simple I'm creating a basic simulation. Two LC configurations create bandpass. 47 Ω at both input and output.
This is merely an example. It's not likely to be optimum for your purposes.

 

[sze wen]

Two different topics: 1) bandpass filters and 2) impedance matching...

It becomes difficult to combine both purposes into one filter. Furthermore input resistance might be unknown and unseen coming from your preceding circuitry.

We have the formula that tells what center frequency is produced by a given LC combination. LC ratio makes a difference in rolloff curve. To keep things simple I'm creating a basic simulation. Two LC configurations create bandpass. 47 Ω at both input and output.
This is merely an example. It's not likely to be optimum for your purposes.

View attachment 199228

Any expert in this forum?

 

[vfone]

For this wide bandwidth, theoretically a 3 pole Pi LC network should do the job, but 5 pole Pi network would be much better.

 

[sze wen]

For this wide bandwidth, theoretically a 3 pole Pi LC network should do the job, but 5 pole Pi network would be much better.

Any technique that could be used to improve the bandwidth after matching is variable matching etc?

 

[FvM]

The question can't be well discussed without knowing amplifier impedance characteristic. If you want to match a moderately varying impedance without distinct resonances to constant load impedance (e.g. 50 ohms), a LC network should work. Real source and load impedance components help to avoid new resonances introduced by the network. Starting with a useful network topology, the solution can be found by numerical optimization methods. Required network complexity depends on wanted matching performance.

If source impedance has already distinct resonances, it's usually not possible to cancel it by a lossless network.

 

[sze wen]

The question can't be well discussed without knowing amplifier impedance characteristic. If you want to match a moderately varying impedance without distinct resonances to constant load impedance (e.g. 50 ohms), a LC network should work. Real source and load impedance components help to avoid new resonances introduced by the network. Starting with a useful network topology, the solution can be found by numerical optimization methods. Required network complexity depends on wanted matching performance.

If source impedance has already distinct resonances, it's usually not possible to cancel it by a lossless network.

Let's say the impedance seen by the power amplifier is 8. How do I transform them to 50 ohm?

 

[FvM]

For wideband, most simply by a transformer.

 

[BradtheRad]

Let's say the impedance seen by the power amplifier is 8. How do I transform them to 50 ohm?

The goal of impedance matching is to preserve power transfer (Watts=V x A). Therefore raw math says that when you have any amount of Watts through 8 Ω, you can convey identical power (theoretically) through 50 Ω...by multiplying V x 6.25 ... and reduction of current to 16%.

In addition to transformers as one method...
With sine waves you can achieve this by assembling one or more networks of inductors and capacitors, in correct values custom designed to perform with a particular frequency or range of frequencies.

 

[sze wen]

The goal of impedance matching is to preserve power transfer (Watts=V x A). Therefore raw math says that when you have any amount of Watts through 8 Ω, you can convey identical power (theoretically) through 50 Ω...by multiplying V x 6.25 ... and reduction of current to 16%.

In addition to transformers as one method...
With sine waves you can achieve this by assembling one or more networks of inductors and capacitors, in correct values custom designed to perform with a particular frequency or range of frequencies.

For wideband, most simply by a transformer.

How to do it in Cadence.

--- Updated Apr 28, 2025 ---

The goal of impedance matching is to preserve power transfer (Watts=V x A). Therefore raw math says that when you have any amount of Watts through 8 Ω, you can convey identical power (theoretically) through 50 Ω...by multiplying V x 6.25 ... and reduction of current to 16%.

In addition to transformers as one method...
With sine waves you can achieve this by assembling one or more networks of inductors and capacitors, in correct values custom designed to perform with a particular frequency or range of frequencies.

Should we look at efficiency + output power instead of maximum power transfer.

 

[BradtheRad]

Should we look at efficiency + output power instead of maximum power transfer.

Example clarifying which is which...
Cheap transistor radio running on 9V battery, with class A amplifier providing audio signal which is a few volts amplitude at tiny current. In this form iit would fall flat if directly connected to 8Ω speaker. It's a mismatch of impedance.

Now install an audio transformer in the collector leg. Step down volts. Step up current. As a side benefit it eliminates the DC component and instead provides AC to the speaker. Impedance is matched. Power level is preserved (Watts=V * A). Yet now signal is in a form such that speaker receives enough current to drive it efficiently.

 

[sze wen]

Example clarifying which is which...
Cheap transistor radio running on 9V battery, with class A amplifier providing audio signal which is a few volts amplitude at tiny current. In this form iit would fall flat if directly connected to 8Ω speaker. It's a mismatch of impedance.

Now install an audio transformer in the collector leg. Step down volts. Step up current. As a side benefit it eliminates the DC component and instead provides AC to the speaker. Impedance is matched. Power level is preserved (Watts=V * A). Yet now signal is in a form such that speaker receives enough current to drive it efficiently.

is it recommended to use PI network

 

[BradtheRad]

is it recommended to use PI network

Perhaps two or more networks, and maybe one or more of another type also. Your operating frequencies are in a range that requires particular know-how. Fast-responding components. They may need to be arranged to avoid certain hurdles.

And chosen for compatibility with your previous stage or amplifier.

 

[FvM]

A straightforward method is using a LC bandpass filter of sufficient order. Parameters can be determined by numerical optimization. Mathworks is describing it in an application note https://mathworks.com/help/rf/ug/designing-broadband-matching-networks-part-1-antenna.html

 

[sze wen]

I use Cadence. Any similar methods for it?

A straightforward method is using a LC bandpass filter of sufficient order. Parameters can be determined by numerical optimization. Mathworks is describing it in an application note https://mathworks.com/help/rf/ug/designing-broadband-matching-networks-part-1-antenna.html

 

[sze wen]

I notice my output power dropped from 30dBm to 21 dBm after the output matching network. Is it because the input network is not designed yet? The input that I have is 50 ohm resistor only.

 

[sze wen]

The S22 ran way from 2 GH to 1 GHz.

 

[mtwieg]

I notice my output power dropped from 30dBm to 21 dBm after the output matching network. Is it because the input network is not designed yet? The input that I have is 50 ohm resistor only.

Probably one or more of the following is true:
1. The Zopt you calculated is wrong
2. The output impedance matching network is incorrectly designed, and it not presenting Zopt to the PA output
3. The PA has poor reverse isolation (S12), and the output match is affecting the input match significantly.

 

[sze wen]

Probably one or more of the following is true:
1. The Zopt you calculated is wrong
2. The output impedance matching network is incorrectly designed, and it not presenting Zopt to the PA output
3. The PA has poor reverse isolation (S12), and the output match is affecting the input match significantly.

How to design the output impedance matching given that we know there are parasitics cap?