Need help with PA 2-4GHz 10W

[Waldmann]

Greetings!
I am a beginner RF designer and my current task is to design an RF power amplifier (PA). This will be my first PA.
Specification is following:
Architecture: Balanced PA
Frequency range: 2 to 4 GHz
Output power (p4dB): 10 W
Gain: 40±4 dB
Supply voltage: 50 V
I’ve  designed a block diagram of my PA and picked up some RF elements for it (see attached picture).
First two stages are 50 Ohm internally matched MMICs HMC609LC4 and HMC637A.
The third and final stage is two discrete transistors NPTB00004A assembled in a balanced scheme with two quadrature couplers. External matching networks between transistors and couplers are to be designed.
So my petition to community is to estimate this diagram and its entire design. Is it OK? Will it work this way? Will I get my 10W on the output with given elements?

[KJDS]

The NPTB models don't match the transistors, so designing matching networks becomes little more than guesswork.

[rfeecs]

Your problem is going to be matching the discrete output stage from 2 to 4 GHz.  You are using the datasheet numbers for 2.5GHz.  You will get much lower gain at 4GHz.  Those numbers are also the optimum for single frequency.  Broad band you will have degraded performance.

The package parasitics are going to make it difficult to get good broadband performance.  There is a reason people use chip and wire for broadband discrete power amplifiers.  You would be better off using a MMIC for the output stage.  (Except for the cost of course.)

[dmills]

The '637 is a wire bond bare die sort of affair, are you really set up to mount that (In a way that gets the heat out)?

Package parasitics will kill you for broadband matching those output devices over a full octave, and in any case consider the thermal issues, they would be right on the ragged edge @ 10W.

I would be reaching for something with at least partial internal matching, and then printing a few hairpins or stubs or something to get what I needed in terms of gate match, and careful selection of final supply rail will let you set the resistive part of the drain impedance to 50R @ 10W which may prove convenient.

W1GHZ has a paper "Simple Broadband Solid-State Power Amplifiers" which discusses the use of resistive feedback to get the input impedance of GaAs fet amplifiers in the low microwave, it may be useful.   

Regards, Dan.

[Waldmann]

Thank you very much for your replies.

The NPTB models don't match the transistors, so designing matching networks becomes little more than guesswork.

I don’t understand what is “NPTB models” and why should they match the transistor. Did you mean that this NPTB00004A is not internally matched?

Your problem is going to be matching the discrete output stage from 2 to 4 GHz.  You are using the datasheet numbers for 2.5GHz.  You will get much lower gain at 4GHz.  Those numbers are also the optimum for single frequency.  Broad band you will have degraded performance.

Well, yes, I understand this. But I simply failed to find any another non-obsolete output active device within this band and power. I searched on Mauser and Digikey. There is yet HMC7885 but it costs 10,000 dollars which is unacceptable.

The package parasitics are going to make it difficult to get good broadband performance.  There is a reason people use chip and wire for broadband discrete power amplifiers.  You would be better off using a MMIC for the output stage.  (Except for the cost of course.)

I don’t understand some of this. Package parasitics are parasitic capacitances and inductances which are too high in this particular NPTB00004A package? Did I get it right? And “wire” is a sort of RF conductor medium which has round cross-section? It is more fitting to a broad band transmission than a convenient planar PCB lines? Right?

The '637 is a wire bond bare die sort of affair, are you really set up to mount that (In a way that gets the heat out)?

Well I didn’t think about it yet. Am I getting it right that this die demands very special techniques and equipment to mount, heat sink and keep working in comparsion to standart packages, right? I have no idea how complicated it can be so I maybe find some more convenient package if this ones demands too much efforts.

Package parasitics will kill you for broadband matching those output devices over a full octave, and in any case consider the thermal issues, they would be right on the ragged edge @ 10W.

Could you please specificate what parasitics and what particular package do you mean? And what “ragged edge” do you mean? The edge of HMC637 die faced towards quadrature coupler or what?

I would be reaching for something with at least partial internal matching, ...

What is “partial internal matching”? How is it possible? As far as I know device is whether internally matched or not. Can you please give me a part number or a link to such partially internally matched device?

...and then printing a few hairpins or stubs or something to get what I needed in terms of gate match, and careful selection of final supply rail will let you set the resistive part of the drain impedance to 50R @ 10W which may prove convenient.

It’s about input and output matching networks of output power amplifier? Gate match = input match; drain match = output match?
Does it refer to my NPTB00004A transistors or to some another device which I should pick instead of NPTB00004A?

W1GHZ has a paper "Simple Broadband Solid-State Power Amplifiers" which discusses the use of resistive feedback to get the input impedance of GaAs fet amplifiers in the low microwave, it may be useful. 

Nice document, I found it useful.

I’m sorry for asking silly questions, I’m only making my first steps in RF design.

[dmills]

By "right on the ragged edge" I mean I think the things will cook if you try for 10W, that is clearly more like a 3-4W part and you have to allow for matching and combiner losses (You will not manage a conjugate match over an octave).

It might be ok at low duty cycle in class C or something, but as you have not specified duty cycle or linearity requirements, I can only assume 100% and reasonably linear (Which would have me reaching for **2** 10-15W parts (By the gibberish they tend to print on the front page of any datasheet) if I was building a 10W amp. I would also be trying very hard to find parts in proper RF packages with proper thermal design for this use.

It is kind of difficult to pick parts given that we don't know the above, but hit someone like www.rf-microwave.it  or richardson and see what might be available (The usual mouserkey sort of full service suppliers are hopeless for the more interesting sort of RF power parts).

I would suggest that a broadband microwave power design is NOT a good place to start as a beginner, I would start by building something UHF and fairly narrowband (70cms band maybe?), many lessons about stability, parasitics and layout dependence that are far easier (and far less critically) learned there before you relearn them in a properly microwave project. 

Regards, Dan.

[KJDS]

Thank you very much for your replies.

The NPTB models don't match the transistors, so designing matching networks becomes little more than guesswork.

I don’t understand what is “NPTB models” and why should they match the transistor. Did you mean that this NPTB00004A is not internally matched?


I’m sorry for asking silly questions, I’m only making my first steps in RF design.

You want to use an NPTB transistor. The easy way of using any transistor in the GHz region is to ask the manufacturer for a mathematical model of the device. You can then use that model in your circuit simulator of choice to get schematic of a matching circuit. Then use that circuit to create a shape that you put into an EM simulator which you adjust until you get the required performance.

If you're only going to need a few of these, buy something from MiniCircuits. If you need a lot then hire a contractor, but this isn't a task for an RF newbie unless you've got several years to get something that's good.

[TheUnnamedNewbie]

Can you clarify what tools you have at your disposal? Are you just a hobby designer? From your wording I get the impression you work at a company. Do you have access to 'proper' tools for this (2.5/3D EM simulators, RF signal chain tools (ADS/SystemVue), etc..) or do you have to do it with a simple PCB tool like Altium/OrCad/whatever?

In the latter case, you are in for a lot more challenge.
However, in the first case, if you can get good models of the components and have time to delve into the functionality of the software, you can get almost first-time-good design from the modeling - heck, we do this with ICs all the time where I work (granted - we've built some oscillators now and then when we didn't intend to..)

Macom also advertise the following: ''For load-pull data and non-linear models, please send an e-mail to applications@macom.com''. If you can get a proper x-parameter or better model, you really can get a first-time-good design (using good RF pcb materials etc that is).

Keep in mind that these are GaN devices, not silicon - you can push them to 200 C or something.

[KJDS]

Macom's models don't match the transistors, so no matter what simulator is available, unless you can rerun the load pull then there's no chance of a first time working design. I tried this design, gave up and used something from Ampleon, but Cree and Qorvo also have models that give first time results.

[Waldmann]

You can then use that model in your circuit simulator of choice to get schematic of a matching circuit. Then use that circuit to create a shape that you put into an EM simulator which you adjust until you get the required performance.

What circuit and EM simulators can you recommend?
A have Altium Designer now at my disposal.

Do you really need full coverage from 2-4 GHz @ 10 Watts?

Yes, I need it full

Can you clarify what tools you have at your disposal? Are you just a hobby designer? From your wording I get the impression you work at a company. Do you have access to 'proper' tools for this (2.5/3D EM simulators, RF signal chain tools (ADS/SystemVue), etc..) or do you have to do it with a simple PCB tool like Altium/OrCad/whatever?

Ive got Altium currently and looking for getting a student version of ADS.
It's not a hobby, but yet i don work in any company. My prospective employer gave me this test job to check if I fit the bill as at least a junior.

Thank you for your advices gentlemen. I am really grateful.
I keep working @ let you know about any success.

[TheUnnamedNewbie]

Can you clarify what tools you have at your disposal? Are you just a hobby designer? From your wording I get the impression you work at a company. Do you have access to 'proper' tools for this (2.5/3D EM simulators, RF signal chain tools (ADS/SystemVue), etc..) or do you have to do it with a simple PCB tool like Altium/OrCad/whatever?

Ive got Altium currently and looking for getting a student version of ADS.
It's not a hobby, but yet i don work in any company. My prospective employer gave me this test job to check if I fit the bill as at least a junior.

I wish you the best of luck getting that license. They didn't give it to me when I was actually a verified student at a university that works together with Keysight a lot.. but perhaps it is a location thing.

Macom's models don't match the transistors, so no matter what simulator is available, unless you can rerun the load pull then there's no chance of a first time working design. I tried this design, gave up and used something from Ampleon, but Cree and Qorvo also have models that give first time results.

Ah, that sucks. In that case, I would just ditch Macom and switch to manufacturers that have good models available for them. This really simplifies design effort a lot.

[OwO]

I'm using a pirated Keysight ADS in a VM. It exists if you look hard enough

[holzi]

Hi,
requesting a 1 year student license from ADS worked fine for me. As I remenber the license is bound to the MAC Adress of your network card. Maybe it´s location dependent if they grant a license to you. My studnent e-mail adress was from "Technical University of Munich" in Germany.

I really think the best way to get this task done is to use a rf simulation software like Keysight ADS oder AWR Microwave office. These tools have bulit in Load Pull techniques which simplifies the design a lot. There are also very good videos on youtube covering this thopics. The one from Cree helped me a lot with my first design. For a beginner this is a pretty tough task.



How are you supposed to build the prototype? You will need a pcb with rf-substrate which can be quite expensive and a heatsink. Maybe (considering the limited time for a job application)you are better to go using a MMIC like the CMPA2560025F which does not cover your complete frequency band of interest. Maybe there are similiar devices which are suitable for application from another manufacturer as Ampleon, NXP... But these devices are also quite expensive. Will the company pay for your design? This is a lot of work to do for a job application...

[coppercone2]

how do commercial amplifiers that work in this frequency range look like (of the solid state variety)?

I have a TWT for this range @ similar power levels, mainly because I suspected it would turn into a hairy mess to try to engineer a solution myself.

[Waldmann]

Thank you everybody for your help.
No more help is required, the device is cancelled by employer as unrequired.

[KJDS]

how do commercial amplifiers that work in this frequency range look like (of the solid state variety)?

I have a TWT for this range @ similar power levels, mainly because I suspected it would turn into a hairy mess to try to engineer a solution myself.

I'd squeeze some GaN devices between some Anaren couplers. I'd restrict myself to Ampleon, Quorvo or Cree for the transistors. The details would depend on size weight and power requirements, with the inevitable discussion with the customer that yes, I can design it with cheap parts but that will take much longer, so saving $50 on parts when you only want to build ten off is pointless as I'll charge an extra $50k dollars in design time. ... and no, I can't make it 100% efficient. ...and I can't magic away the 10W of heat that need to be dissipated somewhere. ... and I appreciate that you only allowed space for a 10W power supply, but that isn't enough to get 10W of RF out. ... etc