Hmm, lemme see here.
Principles:
1. What gain, bandwidth and center frequency are you looking for?
2. What input and output impedance? Any reactances or reflections to worry about?
3. Since this is for IF, is AGC needed? How will you address it?
I think I see #1 is "yes", "narrow" and 10.7MHz.
#2 seems like 50 ohms, but I think you're missing some important related information.
4.
Never use more bandwidth than you need! To do otherwise at least invites EMC and noise troubles; at worst, it simply won't work at all.
The BFQ19S is a
5 gigahertz transistor. It's also a [small] power transistor. Knowing that you intend to work at 10.7MHz, this is so far below fT that it's nothing but DC to this transistor. Which means gain is limited by hFE, which means, you will literally get the same or more gain from an MMBTH10 (fT ~ 1GHz, hFE > 60) or the like, without the potentially unmanageable GHz+ parasitic oscillations!
Bandwidth, in fundamental terms for amplifying devices, is kind of a hard-to-define thing:
BJTs tend to roll off pretty hard around fT. The definition of fT is in terms of currents, so it doesn't preclude real, usable power gain above fT -- though you're unlikely to get much in most situations, like a few dB at best. Most BJTs are designed so that other cutoff frequencies (such as recombination) fall in the same range, cutting off power gain as well. So for most BJTs, fT is a pretty reasonable limit to work from. Generally, you can assume hFE is constant from DC to a break frequency, then falling at -20dB/decade until reaching 1 at fT. Consequently, hFE has a phase shift, which is pretty much 90\$^\circ\$ between 'break' and fT.
FETs don't have much of a physical limit (the transit time across the channel is supposed to be something like fractional ns), but terminal resistances, and the triad of capacitances (G-D, G-S and D-S), work to limit bandwidth in a similar way, though often not as suddenly (there's no straightforward "fT" for a FET).
For example: a 2N7002 biased at 4V, 100mA, might have 20dB gain at 100MHz, 10dB at 200MHz, and a few dB at still higher frequencies. If the reduction is due to gate spreading resistance, then one would expect more of a diffusion characteristic (-10dB/dec) than a dominant pole (-20dB/dec) dropoff. (A dominant pole suggests a simple gain-bandwidth trade-off -- like the BJT's hFE(freq) trade-off.)
Both BJTs and FETs tend to be good for pulsed and wideband circuits, because their capacitances are comparable to their internal (resistance and physics) limitations. I think I'd say BJTs are a little better (a consequence of their higher current density, giving more gain, less capacitance, and lower load resistances), but MOSFETs can be designed quite nicely as well (Si MOS is available into the low GHz; RF Si BJTs are all but forgotten).
And if you're curious about vacuum tubes, they are limited by the transit time from cathode to grid to plate. The movement of the electron beam itself, is the motion of matter: extremely little matter, but at high enough frequencies, it has a large impact. By changing the grid voltage, you are literally performing work on the electron beam. At low frequencies, you "get back" the energy, i.e., the beam exhibits capacitance. At high frequencies, the energy simply propagates away and gets dissipated in the plate as heat (part of the reason why tubes must be derated at high frequencies). This gives a physics-limted cutoff frequency in the fractional GHz for most UHF tubes, up to a dozen or so GHz for the most advanced planar triodes.
Because the current density is so poor, tubes are quite awful at pulsed/wideband applications (even Tektronix strained to achieve >85MHz bandwidth in their early 1960s scopes). While capacitances are comparable to other devices (some pF), the voltages are high and the load current and gain are low, so load impedances are high (rarely under a few kohms). Thus, tube circuits are almost always "narrowband", with such examples as Tektronix's distributed amplifiers being some of the exceptions.
So those are the basic devices. With BJTs or FETs, you may still need to use a narrowband design, because of other necessities like power efficiency (or simply because you need a bandpass characteristic, which is the case here). But you definitely don't want a screaming fast part, that's just inviting trouble. A pedestrian 2N3904 would even serve well here, maybe in cascode, or using two stages in cascade, to kick the gain up a little more (since, with fT ~ 300MHz and hFE > 200, it will be in the hFE-dropping-off region).
As for your design process, I see:
- IC is pretty arbitrary. Do you actually need 20mA worth of gain, or power? Do you need the noise figure? Is the noise figure even optimal with respect to IC, or the frequency you'll be operating at? (The datasheet says it's measured at 900MHz and 1.8GHz. Curves with respect to Vce, IC are not provided.) Are you matched into the optimal impedance? (Note they didn't happen to say what Zs(opt) was!)
- Your VE is pretty arbitrary; it's correct as a maximum (but then, your Vc(pk) is arbitrary too, perhaps just as an example..?), but it's undesirable as an operating condition. More typical is VE = 1-2 x Vbe, enough to stabilize DC bias (on the order of, say, 10-20%) against thermal and manufacturing error, without impacting the Vc(sat) figure too badly. Higher Vce is also desirable in reducing Ccb (though Ccb vs. Vce is not documented in the datasheet, so it's unclear how much difference this might make).
- Your base divider is correct, otherwise.
- gm is wrong, because RE1 is undefined at this point, actually! The value is about right, if RE1 = 0 and CE1 is sufficiently large. But you may wish to add emitter degeneration here, to prevent oscillation perhaps (at the expense of reduced gain, but with such a hot transistor, that would probably be a good idea).
- You also noted Rpi here, but didn't do anything with it.
Presumably that needs to be matched with RS, perhaps using a tapped coil, or a narrowband matching network (series C, parallel L?). Beware that Zin tends to be lower than Rpi (or even negative!) due to parasitic effects; though with fT so large, it should look pretty much like DC...
- You defined Rd for a power match, then took a side track with an arbitrary winding for L2...
This is where you should've entered bandwidth. (Review the specs I led in with..
)
If you go with Rd = 250 ohms (not too arbitrary, a fine starting point if nothing else), then you need L and C such that the bandwidth is satisfied. The Q factor is Fc/BW, or (picking an arbitrary 200kHz BW as you didn't provide a spec), 53.5. For a series resonant network, that's Zo = 13375 ohms; parallel, 4.67 ohms. The corresponding L and C will be impractical in either case...
So you should probably choose a tapped resonator (as pictured!). The LC impedance Zo might be 10-2000 ohms (a not-unreasonable range for easy-to-find L's and C's). Picking 1000 ohms, parallel resonant, as an example, you'd get 53.5kohms at the top (for the required Q). You need to pick a tap some sqrt(250/53500) of the way down, or a turns ratio of 14.6:1. The total L needed is 14.87uH (tapped at 70nH, but mind the coupling coefficient as well), and C = 14.8pF (which is a reasonable value for a variable capacitor).
You might decide on a lower Zo, resulting in a more favorable tapping ratio, and more capacitance (up to 500pF would be reasonable; you're pushing the limits on available trimmer caps with any more).
Note that the transistor's capacitance has basically no effect on the resonant circuit, because it's on such a low tap. Or equivalently: ~2pF is comparable to 250 ohms at ~320MHz, which is 1.5 orders of magnitude beyond your center frequency (= affects tuning), and almost 3 orders of magnitude beyond your bandwidth (= affects operation)! (With a tube circuit, you'd at least have to take it into account, because the plate load resistance might be 10kohms, and the plate capacitance ~4pF, suggesting BW < 4MHz. They had it hard, back then!)
You could also add a tap to this winding for the output (you've got the correct ratio for 50 ohms output), but that's an even tighter ratio than we've got already, so it might be very tricky indeed.
You might instead opt for a "coupled resonator" topology.
In this case, L3 resonates with a capacitor (series or parallel). It has a Q (due to RL) of the same value (53.5), and a coupling factor (to L1 and L2) of about 1/Q (~0.02). The turns ratio doesn't matter, as long as everything comes out right. The filter becomes second order, which means it can be tuned for a sharp but sloppy response, or an optimally flat response (for different meanings of "flat": the usual Bessel / Butterworth / x dB Chebyshev prototypes apply!), or overcoupled (two peaks with a valley inbetween).
You'd probably want L3 to be series resonant with RL, in which case L3 = 39.8uH and C = 5.56pF (on the small side, but reasonable). Mind that L3 needs to be made with small parasitic capacitance, otherwise you'll blow well past the C figure. (A tapped resonator here would probably be good, again, too.)
This approach gives you better selectivity (sharper attenuation, far from Fc) without compromising bandwidth.
If the inductors are wound as regular solenoids, with height about equal to diameter, then when positioned axially, the end-to-end distance will be a bit more than the same distance, for this coupling factor. (You might get lucky with building such a coupling with SMT inductors of various types, but really, that's the biggest downside to a coupled resonator design, it's practically not manufacturable, it takes a lot of tweaking and hand labor.)
Speaking of bandwidth:
Note that the IF strip probably needs a shitload of gain (~120dB including limiting, if you're doing broadcast FM as I'd guess). No single transistor can provide that (nor would you want it to; it would oscillate under all conditions!). You'll be cascading multiple stages to achieve that. Each stage you cascade, the bandwidth gets narrower. Consider if you have one stage of 200kHz BW: it has -3dB points at 10.6 and 10.8MHz. The -3dB points of two cascaded stages come at the
-1.5dB points of each stage. And so on. This gets more complicated if you include higher-order filters (you can over-couple the filter, to place its peaks over the shoulders of surrounding stages, making things wider and flatter), or staggered tuning (by tuning stages around Fc, you lose some gain, but can keep things flat without the complexity of higher order filters).
And then you have to concern yourself with coupling. If you want to normalize your amp stages to 50 ohms input and output, I would suggest adding some attenuation or damping. This allows the input and (proceeding stage) output to operate into a more nominal load, without their reactances interacting and screwing up your carefully tuned frequency responses. It also helps to address feedback and reactance of the amplifier itself (perhaps avoiding neutralization). Obviously, this comes at the expense of gain, as usual.
It also helps to keep the center frequency, and frequency response, consistent. If you apply AGC to the IF strip, you'll have to deal with all the transistor junction capacitances changing. If nothing else, Rpi rises when bias is reduced, raising the Q factor.
If you aren't using AGC, then sooner or later, one or more stages will go into saturation, which means Vce(sat) is being pushed... which means varying capacitances again (and also a severe input loading, because Rpi drops to ~RE1 in saturation!).
And a final word about matching and reactance: did you really think an SBL1 is a perfect 50 ohms? (It would be a pretty shitty mixer if it were, actually.)
https://www.minicircuits.com/pdfs/SBL-1+.pdfNote the VSWR of the ports: RF isn't too bad (~1.2 over most of the range), but the LO is about 3, meaning more than half the incident power is reflected back -- it's more like 100 ohms (or maybe it's 25 ohms, or reactive -- who knows, the phase isn't defined). Note also that these figures assume 50 ohm systems on all ports -- if your R/L port(s) aren't quite resistive at all frequencies (includes at LO and RF and sum and difference frequencies, and probably harmonics too!), expect things to be even worse. It's often good practice to put attenuator pads around mixers, to help out with this. The attenuators burn dynamic range, but at this point, you've got all that you can handle (if your IF strip is an FM limiter, then you don't care!), and you typically want to avoid mixers at the front end (which seems to be the case here), so it's not affecting your noise floor, either.
Tim
Home
Help
Search
Login
Register
