EEVblog Electronics Community Forum
Electronics => Projects, Designs, and Technical Stuff => Topic started by: kronos on May 02, 2021, 07:17:10 pm
-
Hello, newbie here.
I am building an RF amplifier. Intended for the low Megahertz range, should be broadband, moderate-low gain (for now).
It is based on a cascode amplifier, with 2N2369A.
I have built and simulated a schematic in LTSpice: (https://i.imgur.com/4YUPloa.png)
The frequency response should be good, up to about 44 MHz: (https://i.imgur.com/nIQnPEO.png)
I built it on a breadboard, and tested it. Source is a DDS (AD9834), 220 ohm output impedance. The frequency response of the amplifier is bad, of course, up to about 5 MHz. It is a breadboard. I simulated it on ltspice (adding lots of 7pf parasitic capacitors), and the simulation and reality were close to each other.
I then built it on copper, Manhattan style (is this the correct name?). (https://i.imgur.com/2ps5L7A.jpg)
I expected full performance, as there are now no parasitic capacitors. However, it is the same as in the breadboard!
Here you can see the scope output:
(https://i.imgur.com/HaVE93t.png)
The yellow trace is the output, the blue trace is the input. The source is the DDS, performing a linear frequency sweep from 1MHz to 15MHz, in steps of 100kHz each millisecond. As you can see, the frequency response is up to about 5.3 MHz (I know the blue trace is not perfectly constant over frequency).
I expected much better. And I do not know what the reason is for the poor performace. Any help is appreciated.
-
Probe tip capacitance? That can be 10~20pF for common 10x probes and it's effectively in parallel with R4.
-
Thank you, this is a relevant point. I simulated the circuit with 10pF at the output, and the frequency response drops to 8.5 MHz. My probes have indeed 10 pF at x10.
What does this mean? The amplifier is working, but I cannot look at it?
May a buffer (an emitter follower) as next stage be the solution?
-
Do the real and sim'd N1 and N2 points match?
-
Yes, a low impedance output, as a 50 ohm load or from an emitter follower, is a good idea. Obviously this will reduce gain and/or increase supply current, which is understandable. 2N2369 can go up to 100mA or so.
When the load is an RC equivalent network, some peaking can be done to extend it, using a series inductor or tapped coil. Values on the order of L = R_L / (2 pi Fc) are a good starting point.
Tim
-
What does this mean? The amplifier is working, but I cannot look at it?
May a buffer (an emitter follower) as next stage be the solution?
Yep. That means you either have to buffer the output, or use an active probe (with typically a much lower input capacitance.)
If your output impedance is so high than a 10pF load will ruin the bandwidth, then surely your amplifier will have very limited use anyway.
-
Do the real and sim'd N1 and N2 points match?
Yes. N1 is approx 6 dB attenuation, over the frequency range (except for the small decay due to the DDS). N2 is almost zero, as it is AC ground.
-
Yep. That means you either have to buffer the output, or use an active probe (with typically a much lower input capacitance.)
If your output impedance is so high than a 10pF load will ruin the bandwidth, then surely your amplifier will have very limited use anyway.
The output impedance is 1500 ohm, right? The collector resistance of the cascode BJT. This gives a pole at 10.6 MHz (with 10 pF), seems to match (although I get -3dB at 5.5 MHz in real world).
If I lower that resistance, the gain vanishes, due to the relation with R3. Is there any way to reduce the output impedance?
I connected another probe (x10) at the output, in parallel. This reduces the BW to 3.5 MHz, it confirms that the BW reduction is due to this.
-
Yes, a low impedance output, as a 50 ohm load or from an emitter follower, is a good idea. Obviously this will reduce gain and/or increase supply current, which is understandable. 2N2369 can go up to 100mA or so.
When the load is an RC equivalent network, some peaking can be done to extend it, using a series inductor or tapped coil. Values on the order of L = R_L / (2 pi Fc) are a good starting point.
I designed an emitter follower, (with Ie 3.8 mA), followed by a 50 ohm load. The predicted BW (LTspice) is 16 MHz. Much much lower than my original 44 MHz :(
I added an inductor between R4 and power supply. With the proper value (I tried several in the sim, too much peaks the gain curve), in that emitter follower configuration, it extends the BW to 25 MHz. Much better, still much lower than 44 MHz. Did I put it in the correct place?
I will try the emitter follower in practice. But both gain and BW drop a lot when simulating it.
-
Yes, that sounds right. If you aren't hard set on current, just halve the emitter and collector resistors, and that'll get you there.
Also try a peaking capacitor on the bottom emitter (to ground), say 47pF.
Tim
-
Yes, that sounds right. If you aren't hard set on current, just halve the emitter and collector resistors, and that'll get you there.
Also try a peaking capacitor on the bottom emitter (to ground), say 47pF.
A capacitor does not help in the simulation, reducing the emitter resistance (which is high) does not either.
I tried the emitter follower with an inductance at the cascode BJT (8.5 uH, measured). The simulated response is this:
(https://i.imgur.com/FOaLHBz.png?1)
It has a peak, undesirable, but useful: if I see it in the real circuit, it means the simulation matches reality.
Alas, what I see in real world (the emitter follower is on breadboard, though) is this:
(https://i.imgur.com/Pbi3bXd.png)
The good news: BW is up to 14.5 MHz (a bit more if we consider that the input level is decreasing with frequency).
The bad news: the simulated BW is way bigger, and the shape is plain wrong: no sign of peaking above 10 MHz.
There is something frustrating. It is not a very high frequency, and I cannot get it. The BJT should be OK (ft is 500 MHz).
Is there some source where I can learn methods and tricks for simple BJT designs up to 100 MHz?
-
Hm, probably source impedance is getting too high as well. Same story, you're driving a capacitance with a resistance, it can be peaked to extend a little further but if you need a lot more, you just need a lower source impedance, period.
Impedance is really the key here. You can have a transistor with almost nil capacitance, and still limited bandwidth, just because it's running too high an impedance. MMICs are all matched into 50 ohms or so, and get some GHz of bandwidth (several to 10s of, depending on type I suppose?). It's almost like the transmission line extends all the way up to the die.
Well, that's not even a lie, pre-matched transistors and MMICs for those bands really do have on-die matching networks. :D
Likewise, at some point those long spindly legs are going to be a problem. They give a characteristic impedance closer to 200 ohms, and relatively good coupling between each other. They're a liability both for matching to 50 ohms, and to oscillation. maybe not such an oscillation hazard with only 500MHz transistors, but it doesn't help, and it will be fatal when using faster types.
Also, I don't get the metal cans; if I might guess, I suppose old references are still just as available as ever? And 2N2222, 2N2369, 2N3055, 2N5179 and etc. are still available (at any price), and just, no one says there are better alternatives out there? Well FYI -- most of those you can get a plastic PNxxxx version, or an SMT MMBTxxxx version, for a fraction of the price. I'd recommend MMBTH10 (or the original MPSH10 if you insist on TO-92) here, you can make a wideband amp into the low 100MHz range without too much trouble, though it won't have much power (Ic ~ 20mA tops, whatever that is, peak, into 50 ohms).
The next step up, honestly I'm not sure offhand if there's an SMT equivalent of 2N3866. It might still be one of the best transistors, despite its metal can and boutique pricing. It's good for a couple watts RF. SMTs in the same range tend to be MMICs, and bleeding-edge (30GHz+ fT?) HBTs or PHEMTs or what have you, that are probably not something you want to play with just yet. (Middling products have tended to come and go very quickly on the market, to the detriment of all of us who just wanted a nice GaAsFET or whatever to play with, without having to worry about oscillation at frequencies that can burn ones' eyes out.)
For more power, might as well just get a proper RF MOSFET, with a few cheap (~$10 being cheap here) types coming in TO-247 and SMT flat pack styles, good for 10s or 100s of W, and at least as many MHz, to maybe a hair over a GHz. (This is a case where you really do need the right parts to succeed; general purpose power transistors are just so useless beyond 10MHz or so.) These parts either run at higher voltages (there are RF MOSFETs up to 1200V -- though they aren't as fast at that rating, as the ~100V ones are) for direct coupling to 50 ohms, or run at very low impedances -- hence the very wide terminals, soldered to equally wide microstrip transmission lines on the PCB -- and need a matching network to interface with 50 ohm sources and loads. (Which often ends up band-limited, say 2.45GHz +/- 10%, because it's easier to build an LC matching network, than trying to wind a wideband transformer for an impedance and bandwidth that extreme.)
Tim
-
The next step up, honestly I'm not sure offhand if there's an SMT equivalent of 2N3866. It might still be one of the best transistors, despite its metal can and boutique pricing. It's good for a couple watts RF. SMTs in the same range tend to be MMICs, and bleeding-edge (30GHz+ fT?) HBTs or PHEMTs or what have you
2SC5551 is cheap, fast (fT 3.5GHz) and 300mA Ic.
-
Ah, nice one! Cheap (well, cheaper than 3866), fast; but not so plentiful, looks to be EOL. Ah well, get them while you can I guess? :(
RF PNPs are also nearly or all obsolete. BFT92 used to be around (complementary to BFR92), twas the last way to make fast or low-power discrete circuits. MMBTH81 is probably not long for this world either...
Oh, as long as I have the catalog open, PCP1103-TD-H looks nice, it's I guess one of those low-sat types you'd use for small switching or gate driving, but boasts a refreshingly high fT. Curiously it doesn't seem to have a complement?
Tim
-
Book "Experimental Methods in RF Design" by Hayward etc al. Get a later edition if possible. One of Hayward's other books takes the "Experimental" out of the title and is more or less a college course in VHF and down, but in plain English/simple math. The Experimental book is far more Ham or Experimenter oriented, but you would have to pry it from my cold dead hands. ::)
Steve