Designing an RF amplifier. Concept is, high power, complementary cascodes for the output stage (with heavy class A use, but being PP, class AB is an option). 50 ohm output, direct drive, say 10W level.
I happen to have a complementary pair that's not too slow (2SC2690A and
2SA1220A), and I'd like to maximize the bandwidth around that. The NPN side is fine, I have a 30V, 1A, 2GHz transistor that would pair very well with it. Don't have any such thing for the PNP side.
So, what if I fake a PNP, by wrapping, say, a BFT92 around the NPN?
formatting link
That'd be Q1 = BFT92, Q2 = 2SC4821, Q3 = 2SA1220A, and resistors for flavor, but probably roughly representative. (Ground wouldn't actually be ground-ground, but probably something like +40V, and "+12V" would be
+45V.)
The combination is still fast... ah, but Sziklai connections have a propensity for oscillation all their own, let alone in a cascode, plus whatever other machinations I might have for feedback around the thing. Game killer?
The other option would be folded cascode, which is understandably rather wasteful for a power stage!
Simple is usually better - it's easier to get unconditional stability when the PZ count is small, especially if you're contemplating feedback. Cascode helps with Miller capacitance but it introduces addition phase shift in the forward and reverse path - feedback becomes less attractive.
I'll register a contrarian view here. Sziklai pairs and other forms of local feedback can improve linearity out of all recognition.
Of course RF PAs are more vulnerable to stray inductance than the small signal stuff I generally do, but if you model that carefully you might make a real winner.
Does anyone use that admittance/Z parameter matrix stuff for RF amp design/stability analysis that I've read about in books like "Intro to RF Design" by Wes Hayward, or do they just mess with it in Spice and are then like "eh works well enough ship it"
It seems difficult to even find device data sheets that have those parameters evaluated at enough data points to be useful (so if you don't have a 20k network analyzer to check it yourself at the frequency of interest you're SOL), and I once searched for software that could deal with such things that didn't cost 5k and I couldn't really find much.
The classical approach has the great virtue that it tells you how well you _can_ do, so you don't wind up attempting recreational impossibilities on the one hand, or turning a silk purse back into a sow's ear on the other.
Not so much anymore - it's all wired into RF design packages - some bloody expensive, a couple free. SPICE input files can be used - if you have a well characterized SPICE model, the simulation engine will understand it. S-Parameter simulations are the norm. Generally, these packages will work with MATLAB/Octave, VHDL, Verilog... or have something similar built in - lots of post processing and modeling options.
SPICE sucks for modeling transmission lines. The RF/Microwave packages will work directly with dimensions and materials - just plug in the numbers or click on a library definition. Life is good.
Optimizers are generally available - select the free parameters, set up goals and constraints, go for coffee. Bandpass, gain, linearity, noise figure... can be optimized and often a layout is only a few clicks away.
Transient analysis can be done same as SPICE - current and voltage waveforms are available with some limitations. If you have a device defined with S-Parameters, the non-linear properties aren't available for the simulation - you'll need a circuit or behavioral model for that.
There's a "Harmonic Balance" solver in RF/Microwave packages (google Spectre RF - Ken Kundert
formatting link
) that is used for distortion and intermodulation analysis - function similar to transient analysis but done in the frequency domain with an FFT in the solution loop to achieve much higher accuracy and faster convergence.
The books are important however - it's important to understand what you're doing - RF design is not the same as boiling coffee water.
Look at QUCS - GPLed Design package if you want to get your feet wet. No layout capability but otherwise very functional. SPICE models can be added but the usual issues encountered with SPICE variants apply - PSPICE is not the same as the SPICE 3 derivatives, encrypted stuff isn't supported, you'll sometimes have to go through converting a netlist to a subcircuit schematic to get an opamp to work right etc.
formatting link
formatting link
If you decide to try it, install Octave (4.0.0 is current) first.
The windows port works but has some annoying issues that are probably related to specific device drivers - print formatting, display quirks, access rights etc. If you use Linux, there are tested binaries in the repositories for most distros which are generally very clean.
Or if you're a cheap SOB, too... :-) GNU Octave. Not quite as powerful as MATLAB, but the core functionality, and much of what you'd do with it, is all there.
At least 100MHz, prefer "as much as possible". Without going out of my way for, like, PHEMTs and shit I mean. ;-)
Gain at least 20dB, but that can be spread over several stages, and anyway, 1W input is pretty easily solved with a handful of other things (a
2N3866 would do that handily).
Some other junkbox items include video output transistors (something like
100V 200mA, fT's from 200MHz to ~1.2GHz), with the downside that I'll need to parallel several to achieve full output power. Parallel may be an advantage. I can use smaller 'input' side transistors for individual cascodes, and just wire all their outputs in parallel. Maybe even with some phase shift to get a distributed amplifier going.
Transformer coupling is okay, and I may just do so, to solve the PNP problem (and to keep R_L in the sweet spot for whatever transistors I've grabbed). In that case, LF limit is
Matlab has a lot of wizards and help facilities built in. With octave you end up typing lookfor a lot and then hunting for more information or worked examples to figure out how to make something happen. The latest release of Octave (4.0.0) is quite usable and does come with a GUI/IDE that makes life a lot easier.
That varies, sometimes in magical ways. There's always a Monte Carlo option. There have been simulated annealing methods in use but I haven't seen that mentioned for a while. Sometimes there are direct synthesis methods available for things like matching networks.
This one is GPLed and uses a parallel differential evolution method (works with QUCS and LTSpice):
A lot of moral relativists are much better than their principles, and I hope you're one. The principles themselves are so bad that it's scarcely possible for someone to be worse.
I try very hard to deal only with straight shooters. I've had to fire customers who weren't, and I certainly wouldn't knowingly work with someone whose moral ideas were controlled by his own convenience.
The real fork in the road is whether it needs to be wideband. A 100 MHz amp is a lot different from a DC-to-100 MHz amp. Narrowband, you can tune out parasitics, which is how people get tubes and mosfets to work at a GHz.
Wideband gets more interesting.
What's it for?
--
John Larkin Highland Technology, Inc
lunatic fringe electronics
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.