Generally output of VCO or I/Q modulator is differential.
So driving PA(Power Amplifier) by differential signal is natural.
Then output signal of PA is also differential.
But TX load is generally single ended unless we do use loop antenna.
So we have to interface differeitial output to single-ended output.

Here I pick up the following important characteristics of PA.
  (1) Availble Maximum Output Power Level
  (2) Power Added Efficiency
  (3) Second Order Harmonic Level

If we can use fairly good balun which is broad band enough to suppress Second Order Harmonic,
Differential output PA is superior than Single-Ended PA regarding the above three characteristics.

However if outputs of PA are open drain of differential pair
and we can only use very poor balun which is composed from discrete L and C,
I think there is no advantage in Differential output PA compared to Single-Ended output PA.

For example, we often use a balun of Figure-3 in the following.
http://www.zen118213.zen.co.uk/RFMicrowave_Circuits_Files/Balun%20Design.pdf

This balun is not broad band, so Second Order Harmonic can not be suppressed at all.
And power can not be doubled even if we double cosumption current.

My opinions are:
If we can't use good balun, Differential Output PA is
 - Less efficiency regarding available maximum power level and consumption current
 - Second Order Harmonic Suppression is not good at all

So if provided same consumption current, I think Single-Ended output PA is superior than Differential one regarding available maximum power level.
Here consumption currents include not only final stage of PA but also driver stages which might have differential input/output.

I know differential LNA inputs is preferable for RX sensitivity.
But I don't think differential output can be valuable for good performance for TX.

How do you think ?
Is there any advantage in Differential Output PA ?

But I have to share TX pins with RX differential LNA inputs for low cost transceiver.
I'm always in very trouble to get best solution to accomodate best performance for both RX and TX.

Thanks for reply.

aaron_do wrote on Apr 20^th, 2009, 6:41pm:
What do you mean by "I understand the point about the bonding pads" ?

http://v3.espacenet.com/publicationDetails/biblio?adjacent=true&KC=A1&date=20030...

It seems that an inventor of this patent is people belonging to Avago Technologies.
This patent was applied to the Patent Office of our country too.
But an application was finally withdrawn by an applicant and has not been formed as a patent in our country.

This patent assumes a differential antenna, here balun is not required at all.
So there is no problem even though outputs of PA are differential.
If we can use a differential antenna like this patent, we can utilize full ability of differential output PA without any obstacle.

The followings are extractions from this patent.
Quote:

These are all very common knowledges.

If we can't use good balun, Differential Output PA is still advantageous compared to Single-Ended Output PA ?

aaron_do wrote on Apr 21^st, 2009, 8:24am:

Maybe it could be possible if we can realize very sharp resonance at 2*f0 and enough low impedance for f0.

aaron_do wrote on Apr 21^st, 2009, 6:43pm:
I was thinking Single-Ended PA.

If PA is differential, impedance of common node is virtual ground for f0. So impedance of common node is very low for f0.

I think the real benefits of pseudo differential versus single ended PA are

1. class B push pull operation may be more efficient than class AB for the same linearity requirements

2. differential operation divides the output power requirement across two devices instead of one (this is helpful for low voltage applications)

You must have some way to combine the push-pull signals at the single-ended load (I am not considering differential loads), so some sort of balun or coupling network is required to do this.

Lumped element networks might be able to synthesize a combining/matching structure, but what you specifically need depends on the mode of operation.

If both amplifiers conduct (class A) then a balun type structure is absolutely needed to perform phase rotation, combining and matching functions. If the amplifiers conduct over a half cycle (class B) then you may need a network that appropriately couples each output to the load, but doesn't (somehow) mess up the match when one device is off.

I am much more familiar with the former situation than the later. I have combined two class AB 180 degree outputs using lumped element networks to rotate the phase, combine the signals and provide 50 ohm match to the load.