Looking for design for multi-decade wideband (lab workbench style) RF amplifier

[wb0gaz]

I'd like build a workbench-style multi-decade wideband lab-style RF amplifier, ideally LF - 1.5 GHz (or some subset of that.) This would not be a single-chip design (example, Analog Devices ADPA9002 would be a candidate as it's DC - 10 GHz) as I'd like max CW output to be  at least 10 watts, at most 50 watts (gain is unspecified.) Device is unspecified at this point, but considering (not commited to) MRF9045 LDMOS as a candidate (not contemporary, but available inexpensively as I'd like to have spares on hand.) Use of ALC is expected (so there would be some kind of multi-decade wideband coupler and detector allowing controlling feedback to some prior gain stage.)

Looking for any project write-up (preferably including schematics or other build info) that I could use to get started.

I'm sure I've left off important criteria/requirements, so please feel free to challenge and/or question this!

Thanks,

[RoGeorge]

Is it possible for an RF power amplifier to work between DC and 10GHz?
(I always thought RF amplifiers can serve a single band, maybe a couple more bands nearby, but not from kHz to GHz.)

If of any help, seen some time ago a design exercise for an RF power amplifier (I think it was with LDMOS, but not sure):

How to Design, Build, and Test an RF Linear Amplifier
RF Man Channel
https://youtube.com/watch?v=GkOz4iJQ86A   (Part 1, Overview)
https://youtube.com/watch?v=rpDcuMvLs_M   (Part 2, LTSpice)
https://youtube.com/watch?v=0mg14eeu2uc   (Part 3, Input Board)
https://youtube.com/watch?v=aJHpAJ-IETQ   (Part 4, Input Board Con't)
https://youtube.com/watch?v=H40031nGLV0   (Part 5, Output Board)
https://youtube.com/watch?v=X2FWbiUtDEs   (Part 6, Output Board and Demonstration)

[wb0gaz]

Thank you RoGeorge, much appreciate the reply.

The example IC I cited (Analog Devices ADPA9002) is called "traveling wave amplifier", using a "distributed amplification" approach; it has a string of FETs that appear to be in some respects in parallel, but appears there is more to that. The ADPA9002 datasheet does show it has relatively flat gain across the whole passband. That got me wondering if there was a way to implement something like that using discrete (perhaps LDMOS) RF FETs.

Search continues...

[Bud]

It is input/output wideband impedance matching which is a problem, matching circuits are frequency dependent.

[G0HZU]

A lot has changed in the last 25 years or so in terms of wideband amplifier design. I've designed quite a few wideband amps for ECM use over the years and 25 - 35 years ago BJT and LDMOS/VDMOS designs were typically done over frequency ranges like 30 - 88 MHz, then 20 - 175 MHz and 20 - 512 MHz. Then Silicon Carbide and GaN arrived and suddenly it was possible to cover 10 MHz to several GHz at a few watts using GaN. The old rule was $7 per watt for GaN devices but I'm out of date now.

I've still got quite a few wideband dev amplifiers here but they are all obsolete now. I don't do wideband PA design anymore so I don't know much about the latest devices.

However, Qorvo do some nice wideband  (internally matched to 50 ohm) 15 to 25W GaN MMIC amplifiers with 20 dB gain that can cover 10 MHz to > 1 GHz. I've got a couple of eval boards here. To cover (say) 400 MHz to > 2500 MHz at > 25W I'd recommend using (IPP) hybrids to make up a wideband balanced PA using Cree/Wolfspeed (now Macom) GaN parts. This type of amplifier is great for bench use as it is rugged and provides a good input and output match. There are wideband GaN parts that can cover up past 6 GHz at 5-10W nowadays but a lot of these parts may still have ITAR restrictions. You may be able to buy them OK in the US.

Here at home, I cover 50 kHz to 80 MHz at 10W using an old Marconi TF-2167 class A PA. These old bench amplifiers are notoriously unreliable but can be improved with some internal rework. Mine now works well and has been very reliable.

I've got various evaluation board based GaN PAs (internally matched 50 ohm MMIC devices) that cover 10 MHz to about 1 GHz, plus I have the homebrew balanced GaN PA that can deliver about 50W across 400 MHz to over 2500 MHz. There are various legacy designs online for VHF through 6 GHz using Cree GaN parts at up to about 5 W but I've never tried to make one. I've got the parts here but never tried. I've also got various 100W wideband PAs here covering 10 MHz to 512 MHz based on application notes from PolyFET (LDMOS) and Semelab (VDMOS).

GaN is great in terms of ruggedness and bandwidth but not so good for linearity. If you want a fairly decent (as in fairly low distortion) wideband 10W GaN or LDMOS PA I'd recommend designing with 50-100W parts and accept that the efficiency will be really poor and the parts will be expensive assuming you can buy them OK in the US.

[geggi1]

you are able to make very wide band amplifiers using negative feedback.
W1GHZ have a pdf on this. https://www.w1ghz.org/small_proj/Simple_Broadband_Power_Amplifiers.pdf
You will however have a lot of heat generated in the negative feedback circuit.

[G0HZU]

In summary, for bench use I'd recommend breaking the requirement up into a few frequency bands rather than try to make an all in one wideband amplifier.

LF though 50 MHz
10 MHz to 1300 MHz
400 MHz to 2500 MHz
1800 MHz to 6000 MHz

Always try and use parts rated at higher power levels and try where possible to always run the amp output through an external wideband high power 3dB attenuator to protect the PA and to improve the source match. I use a decent 50W rated 3dB attenuator made by Bird for this stuff.

[G0HZU]

you are able to make very wide band amplifiers using negative feedback.
W1GHZ have a pdf on this. https://www.w1ghz.org/small_proj/Simple_Broadband_Power_Amplifiers.pdf
You will however have a lot of heat generated in the negative feedback circuit.

15-20 years ago the Cree CGH40010F 10W GaN device was the new kid on the block and there is (was) a Cree application note available to try and get it to work at something like 2-3W across VHF to 6 GHz using lots of feedback. I have experience using this part in various wideband PA designs at work as well as using the CGH40025 25W variant. A couple of my bench amplifiers here at home use this part in push-pull from 10 MHz to about 1 GHz. However, I didn't ever try to get VHF through 6 GHz operation.

[wb0gaz]

(so many follow-ups this morning! This is what I get for not bringing my laptop with me for breakfast!)

Thanks so much for the outstanding discussion! This is very helpful indeed...

The 10 watt CREE device appears to be current in US distribution (DC-6 GHz covers my original posting.) I have no hands-on GaN experience (so far). A quick initial attempt at surveying using mouser or digikey websites (critica is Fmin, Fmax, and P1dB or something like that) was fruitless as neither of their search pages appear to allow frequency specifications. Not a show-stopper, just a surprise...

Amplitude linearity/distortion is not important - I plan use this as last-stage output for my (very ancient) HP 8660C signal generator as a CW generator (not quite DC, but down to 1 MHz.) Some intermediate gain stages would be needed (expected) to reach the 10-50 watt output range (the 8660C is good to ~10 dBm to 1.3 GHz then engages a doubler in the RF section.)

The W1GHZ article citation will be helpful, as this is practical writeup aimed at someone that might build their own, so it's a good on-boarding resource.

Very poor DC conversion efficiency is expected and perfectly OK - it would be benchtop use so no battery concerns and discharging heat would be expected.

Covering a range in >1 band - probably a necessary trade-off to accomodate hobby-level resources...

Again very much appreciate the ideas and recommendations - helps the idea move forward and quickly!

Dave

[G0HZU]

If you download the s parameters for the CGH40010F (from the Macom website) you will quickly see that it's easy to get flat gain up into UHF using resistive negative feedback. My push-pull PA runs each CGH40010F at 25 ohms (defined by the negative feedback) and uses coaxial baluns to achieve 50 input and output ports. This isn't optimal for efficiency but it matches the devices really well and this suits a bench amplifier because the source impedance is quite good in this configuration for low to medium sized signals. To get wider bandwidth I did some tricks with printed shapes on the PA PCB but this was about 20 years ago so the design notes are long gone at work. The amp produced about 20 W flat out but the harmonic distortion was pretty bad as the push-pull PA resembles a square wave generator when driven really hard like this. It really does have to be backed off to just a few watts to get reasonably good linearity.

I included a driver and pre-driver stage on the second version so it could be driven by a VNA or a regular sig gen at about 0 dBm drive. I've got two of them here, but one doesn't have the pre-driver stage.

[G0HZU]

At work I was granted access to the official (best) non-linear models from Cree for these devices for use with Microwave Office 15-20 years ago. I had many face to face meetings with Cree back then and then later with Wolfspeed as I was involved in PA design quite a bit back then. I can't share these models but it might be possible to request a more basic non-linear model from Macom especially if you are in the US. Small signal s-parameter models are useful but nowhere near as useful as the non-linear models.

[G0HZU]

I had a read of the W1GHZ paper and you might want to consider newer (GaN) devices from Qorvo if you can still buy them. My old Eval MMIC boards I have here are for the (now obsolete) Qorvo QPD1000 and QPD1004 and these are internally matched to 50R across a huge bandwidth and offer about 20 dB gain across 30 - 1000 MHz and 15 -25W output power. In practice, they work OK over a slightly wider bandwidth than this. The same comments apply about linearity though. I would only run these parts at about 2W output power if the aim was to achieve reasonably low distortion levels. The PCB requirements are quite strict in terms of thermal management (using lots of tiny ground vias under the device) but I found ways around this when making a basic eval board for these devices. The thermal management is critical for reliability.

You don't have to run them at a 50V supply voltage if you don't need the full output power. I've run the QPD1000 device as low as 18V and I've run the QPD1004 at 35V and also a bit lower than this.
You do have to be really careful about biasing in terms of how much internal heat gets generated. It is possible to kill these devices through poor biasing and they also need to be biased in a sequence as they are depletion mode devices. Turn on Vdd without a negative bias at the input and the device will be destroyed in a fraction of a second. So always make sure the Vdd supply is current limited and the negative input bias needs to be turned on first and turned off last.

https://www.qorvo.com/products/p/QPD1004

[G0HZU]

Unfortunately, a lot of modern FET devices have the same biasing requirements and there are bias management chips available that try and make the power supply management of these devices as safe and reliable as possible. Otherwise, it's possible to design your own bias management circuit to prevent the PA devices from being destroyed during brown outs etc. One advantage of the old LDMOS parts is that they are enhancement mode devices rather than depletion mode. But they can't compete with GaN for bandwidth.

[wb0gaz]

The information/discussion is greatly appreciated!

It is fortunate when you (and others similarly situated) can bring professional experience forward to help those of us that lacked that opportunity (took normal BSEE MSEE degrees and spent that in cable tv operations as a procurement person, or so it seemed).

I have a learning curve ahead, but this discussion is enabling that!

As a fellow ham/engineer here (currently employed in the RF energy field) told me once a while back (I've forgotten the exact expression, but we were talking about an RF generator design he is party to that is evidently a (just) a VNA with "a source assembly of unusual size")...

"always 180 degrees away from total destruction"

Thanks again,

Dave

[nctnico]

I find this an interesting topic from this background: I have an 8W amplifier module from Aliexpress which goes to 1GHz. I have sort off levelled the output by characterising it through measurements and contol the drive level from an RF generator through software. My application is doing tests for radiated immunity which nowadays need to go to 6GHz.

Back to the topic: is there some kind of reference design available for creating an amplifier like this? I've seen there is an eval board for the CGH40010F but I guess this is not going to give you a really rugged amplifier ready for bench use.

[G0HZU]

I'm not aware of anything but my old push pull CGH40010F version ran at 24V and each device saw 25 ohms via the 1:1 balun. So the output power for each device would be in the order of (24^2)/(2*25R) = 11W. So with two devices in push pull it achieved about 20W flat out. But these CGH40010F devices are really old now and using 1:1 baluns limits the bandwidth quite a bit. The efficiency using this approach isn't great either.

If you want lots of bandwidth and good efficiency and maybe 12-20W power then go for modern high voltage GaN parts. I'm out of touch these days. It's much easier for todays designers to buy ready made COTs designs from companies like Empower or Aethercomm. But they are quite expensive and have restricted availability...

[G0HZU]

If you want to see some performance results from my old push pull PA I can post them up. This won't be for a while though. Maybe in a day or two. I've forgotten how poor the efficiency is and how poor the harmonic (and two tone) distortion is at (say) 4W output.

I can also test the Qorvo QPD1000 and QPD1004 MMIC eval boards but one of them might already be toasted, I can't remember if i repaired the toasted one.

Polyfet produced some reference designs for LDMOS parts about 20 years ago. eg the TB167 reference design is just one of dozens listed on their website. I worked closely with Polyfet for a few years on several PA designs before moving across to SiC and GaN.

https://www.polyfet.com/products/tb167d/

https://cdn.sanity.io/files/6jm0vstp/polyfet/8a32eb457bc918052d600361c3b3e7c228436e27.pdf

I've got a homebrew clone of the LDMOS TB-167 here that covers 20-512 MHz at 100W output. I rarely use it though. Older LDMOS parts like these can be fragile, especially if they manage to go unstable into a poor load.

[wb0gaz]

Wandered over to polyfet (I didn't realize they even existed at this point).

TB244 (after surveying listing of reference designs on their website) indicates 1-1000 MHz 16 dB 40W using a single GX3441 Single-ended 48V GaN. Seeing how this would have been implemented would be informative (

The link to reference design TB244 (cited at https://polyfet.com/products/tb244/) takes you to a photograph but not the PDF itself (tb244.pdf).

http://www.polyfet.com/tbplt/tb244.pdf doesn't resolve but a search on internet archive/wayback shows one capture from 2015 or so, however, there was no data actually captured.

[G0HZU]

Polyfet were almost exclusively LDMOS manufacturers maybe 20 years ago but they also design GaN parts now as well. When I dealt with them 20 years ago I mainly spoke with Jerome.C and Charles.G at Polyfet and they were always very helpful. So you might be able to email Jerome to ask about that eval design if he still works there. I think Charles left/retired a few years ago.

Just google Jerome and Polyfet to get his full name.

[wb0gaz]

Thanks! Found his email address, will get an email out later today.

Thank you from (very abnormally hot!) Denver, Colorado...

Dave

[G0HZU]

Looking at the TB244 pdf image, it may well be fed between a couple of wideband 4:1 coaxial TLTs and there appears to be two resistors in series providing the feedback. Maybe 33R + 100R. It all looks to be quite scruffy and probably a one off prototype.

The output transformer core must get quite hot up at UHF! I'm not sure how that works over 1-1000 MHz though. 10 MHz to 520 MHz would be more like it but maybe there's something else going on with the matching.

[wb0gaz]

The heat (from low overall electrical efficiency) would have to go somewhere - that means passives probably need attention from a thermal perspective!

I'll let you know if I hear anything back about TB244.

Thanks again!

[G0HZU]

In terms of what is possible at the moment using packaged parts have a look at what Qorvo currently lists for wideband GaN PA parts. I have experience of a couple of the packaged parts on this list.

https://www.qorvo.com/products/amplifiers/power-amplifiers/parametric-table

Some are packaged and some are bare die. To reliably use the packaged parts would normally require an exotic PCB layer stack using modern via hole techniques. One workaround is to simply order an eval board from Qorvo but this won't be cheap.

The older QPD1000 (15W) and QPD1004 (25W) 50 ohm MMIC gain block parts I have here are fitted to crude homebrew eval boards using techniques aimed at prioritising thermal management. So they are very fiddly to assemble but work really well. But these parts only cover 10-1000 MHz or maybe a bit higher. A crude rule of thumb would be to bias them at maybe 0.7x the normal Vdd voltage and run them at a fairly high drain current and then run them at least 7dB backed off from their P1dB point. This can give reasonably good linearity. I can demo this at some point.

Otherwise, just driving a typical GaN amp to within 3 to 4 dB of P1dB will generate lots of (two tone IMD and harmonic) distortion compared to what you might be used to seeing from a classic BJT based wideband PA from decades ago that runs at a relatively small idle current.

[wb0gaz]

While researching - I have an intuitive concept type question -

For a very wideband and "flat" passband power amplifier, input matching seems to sometimes be with series and feedback resistance  vs. an RF transformer --- presuming the input of (any?) FET would be capacitive with input impedance going from ~infinity (DC) to very low (microwave frequency, ignoring miller effects) derived from 1/Xc.

Is there any intuitive way to understand how wideband impedance matching into the gate is done (is it like an op amp where gain is based on Rfeedback/Rinput)?

Thanks again!

[G0HZU]

I think it's going to depend on what the goals are and also on the device itself and what the intended frequency range is. For a typical 10W GaN part, the input is going to look like a few ohms in series with maybe 10-20 pF and this would hold true up into V/UHF. The parallel equivalent will be a very large resistance in parallel with 10-20 pF.

By adding resistive (negative) feedback from the drain back to the gate this large parallel resistance will reduce down to a few tens of ohms over a wide bandwidth up into UHF. For larger GaN parts it might drop below 10 ohms. The capacitance will also reduce due to the feedback so it will look resistive at the input and also at the output. The feedback will flatten the gain over a huge bandwidth.

So for amplifiers where flat gain from a few MHz upwards is important, it can be really useful to add the resistive feedback. The downside is reduced gain and it can also degrade the efficiency and output power.

I suspect that the PolyFET TB244 circuit uses fairly aggressive negative (resistive) feedback and this will reduce the input and output resistances to be quite low. So this is probably why there are wideband TLT transformers at the input and the output. So this amp uses both feedback and transformers to define the frequency response.

Up at UHF, resistive feedback isn't so useful as there isn't much gain to trade vs the feedback. So wideband amps starting at high VHF or low UHF often don't use resistive feedback like this. The matching and frequency response will be defined with printed shapes on the PCB that can mimic a transformer. GaN devices are really attractive here as it's easy to get a reasonable match. Generally, the aim would not be to match perfectly to 50R but to match such that the efficiency is good whilst achieving a fairly flat frequency response. Up at very large signal levels the resistive form of feedback will misbehave anyway as the circuit becomes less linear. So it isn't so easy to understand what is happening unless you have access to the factory non-linear models for the GaN device.