RF transistor amplifier design, any suitable literature?

[G0HZU]

It's worth considering why the packaged MMICs are so successful in the first place.  For many years they were noisier than properly-designed amplifier stages built with discrete transistors.  That is still true to some extent but the margin is now relatively small, with more cheap sub-1 dB NF MMICs becoming available. 

IMHO almost no one working at 50 ohms should be doing discrete small-signal microwave amplifier designs anymore.  The industry apparently agrees. 

Yes, the way RF stuff is designed these days is changing in many ways. It's becoming more and more like LEGO

Yansi : + way higher power output levels available. (it is such a pain to find a MMIC gain block with more than 10dBm P1dB at few GHz frequencies!  For example almost impossible to find a MMIC gain block that will drive a +13dBm level MIXER at 6GHz).

There are quite a few MMICs that operate up at medium or high power levels at 6GHz. I had a look here at home in my fairly modest stash of MMICs and found an old eval board for the Hititte HMC788A and the Macom MAAM28000-A1. The MACOM part is old and quite expensive but there are MMICs from various manufacturers that cover a wide range of power levels up to 6GHz.  eg there are wideband MMICs that can go to >+40dBm at 6GHz.

[Yansi]

HMC788 at 20$+ a pop.  Maybe  good  LEGO for a Company, not good enough for an individual hobbyist (pricewise), let alone a student. Maybe I could just shut up, if it would be possible to sample one or two of these for a project, but as it is not possible, screw it! Transistors can do it for a tenth of a cost (and they did it for the last decade), if one knows how to use them, wisely - which I unfortunately don't. It is not possible to sample even anything cheaper, like a decent mixer, or a VCOPLL.  And then I read how someone sampled a whole lot of RF EVAL boards and made something out of them, using a spectacularly craptacular bodginess. But hey, thats the problem of the fucking country I live in. Nobody cares if you don't buy billions a year. (An then they moan there is not enough smart people interested in stuff... when almost nobody cares to support anything technical here)... But that would make a whole completely new different thread.

Now back to RF amps please. Rant over.

[CopperCone]

can you give an example of some wideband 10w chips?

[Yansi]

So, seems the matching networks are based on the S11 and S22 respectively.

To continue with my example  of 1627MHz amplifier with 2SC5773 (datasheet):
Thanks to the thorough colleagues from Hitachi, we have at least some  S parameter tables in the datasheet present.
So for example picking from the table on page 9 (Vce=3V, Ic=50 mA, Zo=50ohm), at 1600MHz we obtain S11 of MAG 0.482 ANG 141.

Now using the formula that z = (1+Gama) / (1-Gama), I can calculate the input impedance of the transistor to be zs = 0.387+j0.306

So, this should be at least a crude (but still better than nothing) estimate of an input impedance of the amp.  In absolute value, |zs| = 0.493.  The reference Z0 is 50ohm, so if I understand right, the amp has an input impedance of 0.493*50 = 24.7ohm.   Interesting. No idea if this is correct or how far from reality, but at least gives idea, what kind of matching network would be needed.

As a matching network for the amplifier input, it seems a simple LC lowpass network should be enough, with the series element connected against the input terminal and the shunt capacitor against the transistor. How to calculate these values still I have to figure out, so gimme a while...

EDIT: Well, I made it the other way round. The series element should be against the smaller impedance, while the shunt against the larger one
EDIT2: Oh... it may not work at all that way either. Doh!

[Yansi]

Last time I calculated the input impedance of the transistor to be 0.387+j0.306 in the normalized values or 19.35+j15.3 ohm denormalized.

Using the PDF attached, I tried to figure out the input matching, using chapter "Impedance Matching of Complex Terminations" on page 11.  Assuming the following model:



I choosed the method to absorb the inductance.  I calculated first the Q required to match the 50ohm into the 19.3ohm real impedance, leaving the jX=15.3 ohm to be absorbed in the low pass matching network.

Q = Qs = Qp = sqrt (Z0/Zs -1) = sqrt (50/19.3 - 1) = 1.26 

Then proceeded to calculate the matching components:

L = Qs*Zs / 2pi*f = 1.26*19.3 / 2pi*1.627E9 = 2.38nH
C = Qp / 2pi*f*Zo = 1.26 / 2pi*1.627E9*50 = 2.47pF

So the reactive part of the transistor gets absorbed, leaving  jX / 2pi*f - L = 15.3/2pi*1.627E9 - 2.38 = 0.88nH for the external component (meaning one would not rather add anything, huh). Just an external shunt capacitance of 2.2puff will be needed to match the input of the amplifier to 50ohms at 1.627GHz.

Phew...  Having no practical experience in this field at all, I can make no better than just putting a prototype on PCB and having it measured on a VNA (which is what I am going to do, after I design the output network also)

[rfeecs]

can you give an example of some wideband 10w chips?

Qorvo (TriQuint), Wolfspeed (Cree) and Analog Devices (Hittite) all have several.
Some examples:
http://www.qorvo.com/products/p/TGA2578
http://www.qorvo.com/products/p/TGA2963
http://www.wolfspeed.com/cmpa0060025d
http://www.wolfspeed.com/cmpa2560025d
http://www.analog.com/en/products/rf-microwave/rf-amplifiers/power-amplifiers/hmc7885.html

You will never be able to make a wide band high frequency power amplifier with discrete packaged transistors.  The package parasitics are just too large.  The options are hybrid chip and wire or MMIC.  For mmwave, about the only viable option is MMIC.

[G0HZU]

Your numbers for the input match look OK to me but for your first attempts at this stuff you might find it best to try to use this transistor at a lower design frequency.

Looking at the s parameters for 3V 50mA the most (unconditionally stable) gain you can expect to get at 1600MHz is less than 9dB. In reality you might get 8dB. But if you add just a whiff of emitter parasitic inductance this number quickly drops by several dB. So you could easily end up with an amp with less than 5dB gain if you want it to achieve unconditional stability at 1600MHz.

Or maybe work with a better transistor? The BFP182 in your app circuit looks to be more suitable assuming you can find one. Maybe 12.5dB (unconditionally stable) gain at 1600MHz? However, I would want something even faster than the BFP182 for a design at 1600MHz...

[Yansi]

Thank you for quick checking. Also doesn't the "emitter whiff inductance" make the amp more stable, at the expense of gain of course?

Yes, the BFP182 looks really better, have at least bunch of them here do I could test those too. However I lack any detailed datasheets for this one, namely the S param tables.

I see only two ways obtaining those figures: Either estimate them from the hybrid pi model (which is I think way over what I am capable of mathematically), or just simulate them, if I can find a SPICE model for the BFP182 (but I saw something right in the datasheet I think). Using LTspice .net command should not be as complicated I think. But never done that before.

By the way what other readily available cheap transistors for 1.6G would you recommend?  (haven't done any research yet, just if there any favorite ones)

[CD4007UB]

Yansi, are you sure that you haven't left out some capacitors from your circuit diagram?

The UHF pre-amp circuits that I've been studying all use a capacitor to ground the top end of the load resistor (39ohm in your diag) that is connected to the drain (in your case, collector). That point is then an RF ground (but not a DC ground), and the series inductor (e.g., L2) further decouples it from the DC supply (so, the inductor is not part of the drain/collector load).

Maybe your circuit operates differently, but I'm pointing this out in case you've inadavertently missed out the decoupling capacitors.

[G0HZU]

Thank you for quick checking. Also doesn't the "emitter whiff inductance" make the amp more stable, at the expense of gain of course?

Yes, emitter inductance can be used to improve the input match and the stability but you will need a faster transistor to allow you to exploit this I think...

My advice on this would be to get hold of a linear simulator to help you. You can just about get away with something like RFSIM99. Its free to download even though it is very old. At work I usually use the Genesys and SonnetEM simulators combined as this allows the influence of the PCB layout to be simulated quite well. The two simulators work together and the results are processed and displayed in Genesys. But I think you could get by with RFSIM99 for initial fumbling about with this stuff.

I would suggest you search for a few app notes for BJT designs for GPS (1575MHz) LNAs as a starting point. These will often give a realistic and sensible PCB layout, a parts list and some plots. Aim to reverse engineer what they did and see if you can replicate their results? Then maybe change the circuit to get a bit more power.

For example:
https://www.infineon.com/dgdl/AN155.pdf?fileId=db3a304319c6f18c0119ec027dee534f

There are quite a few others online. But note that these will be designed for low noise figure rather than signal handling etc. Successful RF amp design really does require you to draw up a set of requirements for what you need because some amps will be optimised for noise figure, some for signal handling and some for best match to 50R.

Also get some experience with RF (lumped) component engineering especially if you have access to a decent VNA to experiment with these components alongside the simulator.

The BFP420 is a decent device although you have to watch the low voltage limits. I've got a few here and they are cheap to buy at Farnell. But you will need to be wary of instability issues up to many GHz with this device. You might be better with something a bit tamer than the BFP420 and maybe you should try designing at a lower frequency before trying for 1600MHz. I've got discrete BJT design experience up at a few GHz but these days I either use PHEMT transistors or MMICs up at these frequencies.

I think a lot of the classic books covering this stuff are very dated. I think the best way to circulate/present this stuff (in 2017) is by video/DVD using a decent simulator and real test gear. Books are OK but they can't cram in enough words/pictures to compete with a real person talking in front of a simulator or a VNA. But the presenter has to have talent at 'presenting'!

Many years ago at my place of work we tried a video training course on RF design by Les Besser. Les Besser is a very experienced and clever RF designer. However, it was all presented in a very dull and dry manner (hour after hour) meaning it was hard to concentrate. It was on VHS videotape format so that shows you how old it was!

[Yansi]

Yansi, are you sure that you haven't left out some capacitors from your circuit diagram?

The UHF pre-amp circuits that I've been studying all use a capacitor to ground the top end of the load resistor (39ohm in your diag) that is connected to the drain (in your case, collector). That point is then an RF ground (but not a DC ground), and the series inductor (e.g., L2) further decouples it from the DC supply (so, the inductor is not part of the drain/collector load).

Maybe your circuit operates differently, but I'm pointing this out in case you've inadavertently missed out the decoupling capacitors.

Very sure about that. I have already posted a photo of the circuit I reverse engineered, I may post a very detailed macro photo of that board if you want to investigate yourself. As was said, it is a kind of negative feedback from the collector.  More Ic --> more drop across that feed resistor --> leads to less bias current, therefore a decrease of Ic. (But never seen this before too.)

[Yansi]

I see at least four ways of obtaining those figures.  #1 on the list involves looking a little more carefully, with #2-#4 being very distant next-resorts..
https://www.infineon.com/cms/en/product/rf-and-wireless-control/rf-transistor/low-noise-si-transistor-up-to-25-ghz/BFP182W/productType.html?productType=db3a3044243b532e0124c960d6f062e6#ispnTab7

Something is REALLY wrong if you don't end up with S-parameters from the manufacturer for a device that is specifically marketed as an RF transistor.  Most likely also SPICE, ADS, MWO etc. packages as well.

Yes, the BFP182 looks really better, have at least bunch of them here do I could test those too. However I lack any detailed datasheets for this one, namely the S param tables.

I see only two ways obtaining those figures: Either estimate them from the hybrid pi model (which is I think way over what I am capable of mathematically), or just simulate them, if I can find a SPICE model for the BFP182 (but I saw something right in the datasheet I think). Using LTspice .net command should not be as complicated I think. But never done that before.

Well that does make life a LOT easier, wouldn't you say? 

Thanks, didn't notice them. 

[CD4007UB]

OK, Yansi, but I wasn't querying the DC biasing/feedback, but the RF grounding. All I can see is a 1uF electrolytic capacitor on your supply, which won't act as an RF ground. Unfortunately, I don't seem to be able to post a circuit to make the point clear. But, as a general rule, with UHF circuits it's important to watch the RF grounding, as poor RF grounds produce unwanted coupling that can produce instability.

[G0HZU]

It's hard to tell from the picture but the 1uF cap looks like an SMD tant cap. We can see that they haven't tried to connect this cap directly to the signal trace. So even though some 1uF tants can still present a fairly low impedance at 1GHz I think there may be another ceramic cap on the other side of the PCB or maybe the 5V rail is arranged as an inner 'power' layer in the multilayer PCB that will have its own self capacitance to the ground layer(s).

[Yansi]

There really is NONE other decoupling caps. I have already reverse engineered the whole PLL VCO part of the board.
Yes, the board is an 8 layer custom and exotic stackup, whole board being 3mm thick. Uses blind vias (which makes a pain in the as job to trace some signals!). On the other side of the board, theres the PLL, VCO and LDOs. Nothing is related to the other side. Yes the 1uF cap is tantalum. On the other side, there is only another 10uF tantalum after a LP2951 regulator. Nothing else powered from it, everything having its own local regulator, the whole board having I think 18 LDOs.

I have cut out the PLL VCO part of the board and already try to power it to have some fun out of it, and it works. The cut exposed the internal stackup. Unfortunately all layers on the side of the shielding rings do present only GND, so no idea if under the the UHF circuit may be a whole layer of V+ for the RF amps. It is possible.

[G0HZU]

Getting back to +13dBm 6GHz MMICs I had another rummage in my MMIC eval graveyard here at home and found an old milled test board for the Stanford/Sirenza/RFMD/Qorvo SBB-5089Z. I've used this part a lot over the years, but never at +13dBm at 6GHz.

My test board originally was designed for use from 20MHz to 3GHz but I changed the caps and bias choke to suit 6GHz operation. I retested it this morning and could get +13dBm at 6GHz without having to drive it very hard. It was beyond P1dB but still gave +13dBm at 6GHz.

Can you get these in your country? They are only $4 USD

https://www.digikey.com/product-detail/en/rfmd/SBB-5089Z/599-1007-1-ND/936981

These parts are Qorvo parts but they are quite old and date right back to the days of Sirenza at least.

http://www.qorvo.com/products/p/SBB5089Z

Obviously, to get good gain performance up at 6GHz you would need to make a PCB with a very good layout with lots of ground vias and use decent ceramic capacitors and a well behaved bias choke at 6GHz. Otherwise the performance will drop away well before 6GHz. You also need very good test gear and test accessories to prove the RF performance with reasonable certainty. I can show you the performance plots/data I got if it helps?

[G0HZU]

I had a tinker on the Genesys/Sonnet simulator to see if I could compete with the 5089Z at 6GHz by trying to make something with a cheap but fast BJT from my stash of transistors here at home. I found that if I used a fast SiGe HBT/BJT (BFP650F £0.50ea) the s parameter data suggested maybe just over 11dB gain was possible up at 6GHz. But the simulator predicted this would drop to less than 9dB with a real PCB layout with the various losses and parasitics. By changing to a SiGe part with a lower Pdiss spec I could get maybe 10.5dB gain according to the simulator. But I don't have that transistor here.

However, I can only achieve this with a fairly narrow bandwidth. Probably only about 1GHz (-1dB) BW at 6GHz? I think the P1dB would be >+15dBm and the input and output match would be OK but this amp would only be suitable for something like ISM or WiFi or maybe the 5.6GHz ham band. Certainly not a general replacement for a wideband MMIC but would something like this still be of interest?

The ATF-54143 PHEMT would work here too and would maybe give 10-11dB gain but these things seem to be very expensive today at about £4 each. I have quite a few of them here but I don't know if you can buy the ATF-54143 in your country. Can you buy SMD BJT transistors in the BFPxxx range?

[Yansi]

Hello! Sorry for not being able to respond quicker, have not much free time during the work days.

The reason why I mentioned the 6GHz MMIC would be for a whole separate thread. I head a few interesting ideas and design challenges for me, that I could try. The MMIC should serve as an LO buffer to drive a 13 level MMIC mixer, but in a rather large 1 octave range (approx 3 to 6GHz). But I need first to make some more babysteps in RF design, before I proceeding to such designs.

I have already looked a bit at the RFSIM99. It seems it will be a very powerful tool, thanks for the tip. Haven't played with it yet, but definitely will. I have also here a copy of Ansys Designer student version, which I obtained mainly to design and simulate microstrip filters. (already made a few of these physically, and measured them partially) Maybe the Ansoft Designer can also simulate using a "black box" two port component with loaded S parameters into it. That would become veeery handy too!

I see some ATF-54143 on Aliexpress for reasonable prices. (the BFP420 is there too)  Probably some NOS, I think one could probably trust those, maybe.  Some BFP devices are in stock at Mouser (Mouser is the very available large distributor here, Digikey unfortunately is not, they want a utter nonsense money for shipping. Farnell is here available easily too, but they want nonsense money for the components instead. Their pricing is mostly very noncompetitive).

For completeness and for those interested, am attaching a photo of my last microstrip bandpass etch test and a measurement of its S11.  Didn't measure anything else (apart from having it "noised" on a spectrum analyzer to see a sort of S21), as the VNA I have access to has only single port available for measurement. (Do not have the other connector converter for the crazy HP RF connectors :-/) The filter should have been 100MHz wide, however measures about 200 meg. Dunno why that happened, but the main variable is the mostly unknown material I made it on. I basically made an well educated guess of the permitivity (and haven't been very far off! The filter center is 1.46GHz, but should have been 1.5). Loss tangent of course unknown.

[G0HZU]

Your BPF looks tidy. I don't see any scalpel marks or soldering iron burns on it yet though One thing of note is that the (1.5dB?) return loss in the stopband makes it look like the material is very lossy or maybe you measured it with some cable inline after you calibrated the VNA?

RFSIM99 is going to struggle a bit up at 1600MHz because it's really just aimed at lumped element design. So it isn't easy to model microstrip unless you import it as a 2 port model or if you use the basic transmission line model. But at a stretch you can use something like Sonnet Lite to model and export short lengths of microstrip as a two port model for use in RFSIM99 as a two port data file.

RFSIM99 would have been fine for designing and simulating an amplifier at 145MHz. There are several classic ways to (reliably) make a broadband 50R amplifier with a BJT at 145MHz but I guess you are more interested in higher frequencies? But if you have a decent BJT, a 145MHz amplifier with quite good 50R matching and good stability and flat gain is quite easy to make. You can almost do it all with a pocket calculator with no need for s parameter data but having a simulator and a 2 port model of the transistor makes it easy to simulate and predict the performance.

At work I have access to the latest SW from Microwave Office and Agilent Genesys and Sonnet but I still like to use the old version of Eagleware from 2004 for quick and dirty design work here at home.

When making one off designs for amps or oscillators I like to use as few critical components as possible so for the simulation of that BFP650F amplifier the only critical parts were the BFP650F and two (Kemet HiQ CBR 0603) ceramic caps. But these were still low cost parts. To keep the design simple I used radial stubs for the bias feeds and simulated it in Genesys and Sonnet combined together as in the image below. This does mean it is quite big. The whole PCB is about 25mm x 25mm in size. I did use a decent PCB material but this wouldn't be that critical for this design. But the PCB does need to be quite thin to minimise the inductance of the through vias.

As you can see it only has about a 1GHz -1dB bandwidth at best so no good as a replacement for a decent MMIC. But it was fun trying to get a simple PCB layout that preserved the gain as much as possible up at 6GHz.

[KE5FX]

There are quite a few MMICs that operate up at medium or high power levels at 6GHz. I had a look here at home in my fairly modest stash of MMICs and found an old eval board for the Hititte HMC788A and the Macom MAAM28000-A1. The MACOM part is old and quite expensive but there are MMICs from various manufacturers that cover a wide range of power levels up to 6GHz.  eg there are wideband MMICs that can go to >+40dBm at 6GHz.

MAAM-011206 looks pretty good, P1dB about +18 dBm at 6 GHz, usable to 15 GHz, about $11/1 at DigiKey.  Quite a bit cheaper than the HMC788A.

[Yansi]

The VNA was uncalibrated in the first place. The dudes operating are well... how to say it. Corporate employees.  And the nice dualport 3GHz VNA is used for singleport measurement of 13.56MHz loop antennas for NFC. So if there is something, that does not have to be done, they do not do it, they do not have the tools for it. They said something like the machine has a metrology certificate, but only up to 30MHz.  So currently am searching for anyone else in the "neighborhood" who has a DC-3G dualport VNA, that can be calibrated at least using the short-open-50-through.
There was also a 20cm SMA-SMA cable between the VNA and the filter, so a bit of loss might be introduced by that and the PCB should be a variant of FR4, but FR4 should not have very high loss at 1.5GHz either.

I have the Ansoft Designer here to simulate higher frequencies. I didn't experiment with that part of the software yet, so I do not know its limitations, but from what I know about it so far, it should work. I have also the Sonnet Lite here, but I must admit I hate that. Its craptacular GUI drives me nuts and the memory limit prevents me to simulate anything more complicated than a line with few stubs. 

Yes, you guess right, my interest is at higher frequencies, but nothing especially high. The current goal I set for myself was to make at least somewhat usable receiver for the 3.4GHz amateur radio band.  I already made a few experiments, designs based on very old books that were designed in a way they might work only by a random chance. So quickly reverted back to basics, to learn a lot first and make the things more properly with more modern components.

But currently I still need to get a good understanding of the impedance matching, and make a few experiments with that to confirm the theory works. At least finish the experiment with the 2SC5773 at 1627MHz and maybe make some more for different frequencies with different transistors. Currently need to get some free time to try design the collector matching circuit.

[G0HZU]

I couldn't resist actually building the 6GHz BJT amplifier so I milled the PCB and there are only a handful of parts on the board as I bias it manually with a second supply. So It was quick and easy to make.

See below for a quick measurement with my VNA and an image of the PCB. The response is shifted down slightly to about 5.6GHz and I think this is partly because I rounded up the capacitor values from the ideal. I haven't bothered to change the caps as it's probably more use to me on the 5.7GHz band anyway Plus I didn't want to inflict any rework scars on the PCB and spoil its photoshoot. However, I didn't want to waste any of my ultra fine PCB rivets on this board so the via holes are all just wired/soldered. So it looks a bit clunky.

But the plots below show that even up at 6GHz the s parameter models for these BFP transistors seem to be quite good and it shows how powerful Genesys and Sonnet are when combined. Like you I don't like the user interface of Sonnet but when combined with Genesys all the user inputs are done in Genesys and Genesys launches and controls Sonnet remotely and exports the PCB to Sonnet and then re imports the simulation data. It then uses this data to produce the graphs for S21 etc.

It's the first time I've used the cheap Kemet caps. I would normally use ATC 600S caps here but they are expensive. I don't know how accurate the Kemet caps are up at 6GHz and I think they are the reason it is slightly off frequency.

But it still looks good. I got about 9.4dB gain at the peak. Slightly more than the simulation but that's partly due to the extra solder around the device and this reduces the emitter inductance a bit I think.
I also tested it for P1dB and got about +15dBm P1dB on a decent lab power meter and the OIP3 was about +28dBm. But I could have biased it harder to improve this I think. I can't measure the noise figure very reliably here at home. Maybe one day I'll measure the NF at work but that won't be for some time.

Just happen to have a picture of SBB5089

Thanks. Looks like there's a lot going on inside that device!

MAAM-011206 looks pretty good, P1dB about +18 dBm at 6 GHz, usable to 15 GHz, about $11/1 at DigiKey.  Quite a bit cheaper than the HMC788A.

Every year the MMICs just get better and better... Nearly 30 years ago when I first started work on RF converter design the WJ A25-1 amplifier was considered to be very special. But it was always a bit weedy in terms of OIP3 and P1dB and it only worked to about 1500MHz and the noise figure was just under 4dB. But it was also expensive! There are amplifiers available now that would have been beyond fantasy back in those days

[Yansi]

Fakin hell! The firefox crashed when I was writing a lengthy response. 

Look what I have found! Less than $3 on Mouser: TRF37A75: http://www.ti.com/lit/ds/symlink/trf37a75.pdf
+12dBm OP1dB at 6GHz.  Not as bad.

So I need to make some progress with the discrete transistor amplifier experiment. I need to design the collector circuit, however am unsure how to proceed with that. I guess using an RF choke is a preferred way. But how to pick a value? What impedance should it present at the design frequency? For example to achieve 1kohm, I'd need about 100nH at 1.627GHz. That seems rather lot.  Still the impedance will have to be matched to 50ohm, so maybe a lower impedance of the choke make be sufficient. I guess I have to combine the S22 of the transistor with the impedance of the collector circuit being in parallel and then the resulting impence will be used in the impedance matching design process. Is this the correct approach?

Also, some times I see also a small resistor (tens of ohms) in series with the collector choke. What is the purpose of this resistor? Could not find a definitive answer for that. (Even the amplifier I have reverse engineered has that).

I still use common whatever SMD 0603 capacitors I have available, the small values being generic NP0/C0G.  I think these are still good enough until couple GHz, aren't they?

Thanks,
Yansi

[G0HZU]

The simplest starting point when choosing the collector components is to decide what output power you want and what DC operating point you want for the transistor. The stuff I've written below is just a crude/ballpark analysis and it may contains nuts/typos as I'm in a bit of a rush. I've got a few things to do tonight but hopefully all the stuff below is OK...

If something like the BFP420F is used it will typically be run at 3V Vce so there isn't much in the way of voltage swing available at the collector. So if you want to produce +13dBm (20mW) with fairly low distortion, you would probably want to operate the transistor at 3V and about 30mA bias current and present it with a collector load of 100 ohms or maybe a bit less. This is the load the transistor sees look outwards into the L match. Not the same as looking back inwards obviously.

Here is why 100R is a reasonable target for +13dBm power...

Power = (Vpk*Vpk)/(2*Rload)

With 3Vce at the BFP420F you will get maybe 2Vpk (4Vpkpk) of voltage swing before severe distortion and limiting kicks in so in your case with a 100R load seen by the transistor...

Power = (2*2)/(100*2) = 0.02W = +13dBm. Bingo.

If you present it with too high a load then you need lots of voltage to get +13dBm and the lack of (available) voltage swing from the BFP420F will limit your power (before distortion). So you won't achieve +13dBm power.

If you present it with too low a load resistance then the 30mA bias current will limit the (distortion free) power to a level below +13dBm.

A reasonable target would be Vce 3V, Ic 30mA and a load of about 100R if you want +13dBm. So a typical L match from 100R to 50R at 1620MHz would have 10nH as the collector choke and a series 2pF cap to your 50 ohm output. But these values are only approximate and you need to play with your simulator and model for your transistor to optimise it all.

Putting a 10R resistor in series with the collector choke will help with stability and also supply isolation/filtering and can also help with matching and it won't cost much in terms of loss. It should be OK to use cheapo ceramic 0603 COG/NPO caps here but there will be some variability between manufacturers I guess. Worth a try and they should be fine at 1620MHz as long as the overall inductance within the structure is low.

If I was designing the amp I'd do things a bit differently than this simple/classic approach but it should work OK.

[Yansi]

Well thanks a lot for that explanation, I completely missed the voltage x current point 

So for a 100 ohm collector load impedance, I need 10nH.  I will then combine that with the S22 of the transistor and calculate the match. Let me try that, I am looking forward to what I can come up with. Only paper, pencil and my trusty old calculator. Just need to exercise my math skills a bit.  Then I'll try to work out the Ansoft, if I can simulate that maybe including layout.

I will try to finish the design experiment with the 2SC5773, as I have these available right now. Even though the C5773 is not very good at 1.6GHz.  (The BFP420 and others will be on the way.)