Mixer to Filter section-Check my understanding of common-gate FETs?

[Xnke]

Working on this at 3AM due to a bit of insomnia. I *think* I understand what's happening here. Wanna make sure I'm not being ignorant again.



So, RMS-2 mixer output at 9Mhz. Into a fairly broad diplexer centered on 9Mhz to provide a nice wide 50R impedance match for the mixer. Follow that up with a J310 in common-gate config, again, trying to maintain 50 ohms into the JFET. Not 100% sure that I've got the match into the source correct yet. Source/Drain resistors are intended to provide about 10mA Id, and the 1uH and 3.3uH inductors are just for RF chokes. I *think* the output impedance of the JFET should be around 150 ohms going into the pi-network, which pads it down to 50ohms and drops 6dB in the process. From there, the 5.7uH and 110pF capacitor network matches the roughly 560 ohms||25pF input impedance of the crystal filter. (datasheet says 1K ohms input, but VNA says 560||25pF...)

Now, the J310 common-gate amp, in conjunction with the 6dB pad, should give about 9dB power gain. That's enough, as I only need to make up the mixer's conversion loss and the 2-3dB lost through the filter chain. (1 or 2 of the KVG filters, at 1.25 to 2dB insertion loss each.) After the filter chain is the variable gain IF amplifier, which will then feed a constant level into another RMS-2 mixer to downconvert to audio, or upconvert to the TX chain.

The question is, do I really need that 6dB pad? I don't need the extra gain eliminating it provides, but since I could just up the output impedance of the J310 a bit, could it directly drive the filter instead? I chose to do it the way shown above, because if I use a different filter later on down the road I have a method of matching to whatever the new filter might be, instead.

Beyond that, are there any other glaring errors, and if so, what do I need to look at?

[Xnke]

Yeah, that pi-network pad is totally wrong. R12 should be 1.8K, R11 140R, and R13 68R, for a 150-ohm-to-50 ohm 10dB minimum-loss pad.

[T3sl4co1l]

9MHz is pretty low, so we can treat the transistor as resistances, and capacitances (of which only drain output capacitance is of much importance).

Q1's gm is >12mS, so the input resistance is <80Ω.  This isn't far from 50 so some padding may be needed to perfectly match things, depending on the exact JFET you get (gm might be up to 25, or more?).

C10 is -j800Ω and L9 is j56Ω, they don't seem to be filtering much of anything and surely only serve to unbalance the filter and add insertion loss?

L10 is fairly small (j186Ω in series with 330Ω) so doesn't do much of anything.  Suggest more like 10uH and little resistance (more dynamic range?), or just resistance because who cares.  (You might end up with the same for L9, i.e., using R15 for source bias directly, in which case you'd remove the bypass C9 of course.)

(Interestingly, as R15 rises, bias current and therefore gm falls, and eventually their parallel combination rises above 50Ω.  As R15 falls, gm rises to a point (no more than the zero-bias value, of course), and R15 comes to dominate; eventually their parallel combination falls below 50Ω.  Therefore, there will be some value where Rin = 50Ω.  And this has to be R15 > 50Ω, since the source conductance is nonzero.  If the JFET were well characterized, this could be calculated, but to get it perfect with any random JFET, you'll have to measure.  Attach a transmission line in place of the diplexing filter, and adjust for maximum return loss.)

Q1 can drive any load.  It's effectively a Norton AC source, i.e., a current source in parallel with some conductance and capacitance.  This is where drain capacitance (~2pF) comes into play.  Though at these frequencies, there's not much worry from a mere 2pF.

Indeed as drain conductance is very low, you could get massive gain out of this, still with pretty reasonable bandwidth given the low capacitance.  This is done by raising the impedance seen by the drain.  Like, 100uH RFC for bias, then either a matching network or transformer to bring the ~kΩ impedance down to whatever's needed (here, 500Ω, but it could be demonstrated into 50Ω for testing just as well).  The, er, gain in gain, isn't strong -- it goes as sqrt(impedance ratio) -- but it's nice when you can take it for free.

But taking the simpler route, I would recommend either a ~500 ohm pullup, or RFC.  Maybe both, so that you get some damping against the filter as well as some inductive reactance to cancel the filter's capacitance?  Or just an RFC, sized to cancel the capacitance.

Note that the filter will only be ~500 ohms, heh well, if it has a lot of loss, or if its load is matched.  (Also, 5.7uF inductors?  Well, you did say it was 3AM... )

Tim

[Xnke]

Thanks, Tim. Now that I'm finished in the machine shop for the day, I'm back looking at this with a relatively fresh head.

Q1 being a J310, it was originally designed to present a 70 ohm input impedance in the common gate config (as per the original Siliconix datasheets) over a wide range of drain currents (5 to 30mA). That being the original Siliconix version, who knows what is going on with one today. Doesn't surprise me that it's somewhere near that point.

L9/C10, I don't think they're strictly needed, and I think either R15 or R10 can go, assuming I can get the input impedance of the amplifier dialed in to 50 ohms. C9 is right out at that point, as well. R10 was to terminate the pass-band of the diplexer, and so if the input impedance of the JFET source is matched down to 50 ohms, R10 is superfluous. L9 isn't serving a real purpose at that point anyway, so it can go. C9...I can't remember why I thought I needed the C9 bypassing cap.

L10 wasn't so much for loading, as it is for power supply decoupling. It's in the wrong spot, at a minimum needs to swap places with R14 and as you said, needs more inductance to be effective at decoupling. Originally, the old PCB I lifted the filters from used a 2n2222A emitter follower, as shown below.



The match into and out of the crystal filter is critical to the shape of the passband. If it's poorly matched, the filter goes pretty wonky. Putting the filter on the VNA needed the L8/C8 combo to be 110pf and 34 turns on a T50-2 toroid, which is 5.7uH. That input/output network flattened the passband nicely and minimized spurs out in the stop band. Contrast that with the values on the original board!

That said, some of these filters are KVG, and some are KVN. Same company, but some people measure the filters as roughly 560R||25pF, and some people get 1K||25pF. The spec sheet says 1K for the KVG filter, and so far no info on if the KVN filters are supposed to be different or not. I suspect that the boards may have different component values depending on which filter they got-I had three boards and didn't check but the one board for schematic values. (might still have one around unmolested, I bought four of them.)

At any rate, I don't want/need more gain in this JFET amp than to correct for the losses of the input bandpass filter (2dB loss) combined with the mixer conversion loss (5-6dB or so) and to get the impedance jump to drive the crystal filter. Adding more gain before getting through the roofing filter just adds to the noise.

AFTER this roofing filter, the first variable gain amplifier will go in, and its output will drive either no filter for AM modulation, the upper sideband filter and matching network, or lower sideband filter and matching network. This first IF amp will give about 3/4 of the overall gain in the IF strip. From there, a second variable gain amplifier will drive the second mixer. By breaking up the IF gain into two separate amps, I'm thinking I'll have better AGC range overall.

[T3sl4co1l]

Q1 being a J310, it was originally designed to present a 70 ohm input impedance in the common gate config (as per the original Siliconix datasheets) over a wide range of drain currents (5 to 30mA). That being the original Siliconix version, who knows what is going on with one today. Doesn't surprise me that it's somewhere near that point.

Makes sense, so the SWR at that point can't be too bad, and it can easily be tuned or padded to perfection if you're picky.

L9/C10, I don't think they're strictly needed, and I think either R15 or R10 can go, assuming I can get the input impedance of the amplifier dialed in to 50 ohms. C9 is right out at that point, as well. R10 was to terminate the pass-band of the diplexer, and so if the input impedance of the JFET source is matched down to 50 ohms, R10 is superfluous. L9 isn't serving a real purpose at that point anyway, so it can go. C9...I can't remember why I thought I needed the C9 bypassing cap.

Not quite -- R10 is on the trap, not GND.  At resonance, it's out of circuit; at other frequencies, it does its job.  I suppose any impedance out there will do -- you don't need a symmetrical diplexer when one port's impedance is fairly flat (the JFET) -- but the front components will have to be adjusted to compensate for the asymmetry.

L10 wasn't so much for loading, as it is for power supply decoupling. It's in the wrong spot, at a minimum needs to swap places with R14 and as you said, needs more inductance to be effective at decoupling. Originally, the old PCB I lifted the filters from used a 2n2222A emitter follower, as shown below.

Ah, that'll do.  Maybe not with the greatest noise factor, but after the preamp, it's not the worst sin.

Nice thing about both (that case and yours), you can basically achieve the same result in exactly complementary means -- that is, they did Thevenin equivalent, you're doing Norton equivalent.  The electrons don't care.

The match into and out of the crystal filter is critical to the shape of the passband. If it's poorly matched, the filter goes pretty wonky. Putting the filter on the VNA needed the L8/C8 combo to be 110pf and 34 turns on a T50-2 toroid, which is 5.7uH. That input/output network flattened the passband nicely and minimized spurs out in the stop band. Contrast that with the values on the original board!

Ahhhh.  Yup, crystals, ceramic filters whatever, can be a bit sloppy; best results after measurement and adjustment.

That said, some of these filters are KVG, and some are KVN. Same company, but some people measure the filters as roughly 560R||25pF, and some people get 1K||25pF. The spec sheet says 1K for the KVG filter, and so far no info on if the KVN filters are supposed to be different or not. I suspect that the boards may have different component values depending on which filter they got-I had three boards and didn't check but the one board for schematic values. (might still have one around unmolested, I bought four of them.)

And the impedance that gives the best compromise between band shape and insertion loss, need not be a perfect impedance match, true.

AFTER this roofing filter, the first variable gain amplifier will go in, and its output will drive either no filter for AM modulation, the upper sideband filter and matching network, or lower sideband filter and matching network. This first IF amp will give about 3/4 of the overall gain in the IF strip. From there, a second variable gain amplifier will drive the second mixer. By breaking up the IF gain into two separate amps, I'm thinking I'll have better AGC range overall.

Nice.  I wonder if more amps might not be desirable for dynamic range or linearity purposes, but mind I'm less experienced on the nuances of IF design, and indeed the exponential of the B-E junction covers many decades.  (Last IF strip I made used three pentodes, with AGC, and an FM detector/limiter at the end, for multimode operation.  Who cares about dynamic range when your plate voltage is 100V, right? )

Tim

[Xnke]

Still got that Class-D/A 38HE7 amp?

Part of why it's been so hard for me to build confidence in solid state circuits-when you goof up one of these, you don't get a screen grid glowing to warn you you're screwing up before the magic smoke comes out. But, gotta learn. (And there's a pair of 2E26 plate modulating a pair of 2E26 in push-pull out on the workbench in the radio room right now!)

Here's where I got this evening. I'm still not sold on R15 and R14's values-currently the gate-source voltage should be around -0.68v if I have managed to get it to draw 10mA, but the datasheet says it should be around -1.6 volts to get that kind of drain current. The actual current isn't so important, so long as it's not running hot enough to melt the solder holding it to the board. (It's kinda like a triode, so it should kinda have ratings, rather than the sometimes vague suggestions of a triode, right?)



I've given up the thought of the output of this amp being 50 ohms and matching it up to 1K, that doesn't need to happen since the output impedance of this amp can easily be whatever I need it to be. R11/C22 can be selected to give best match into the filter.

L1 and L2 will get wound on a small T37 size cores, probably out of 43 material. C1 and C2 are actually going to be split into two caps, probably a 330pF and 22pf combo.

Am I getting farther away, or starting to home in on the right questions?

[T3sl4co1l]

Hah yep, it's on the shelf, haven't lit it up in ages...

R15 sounds a bit small to get to 50 ohms, but again that's just a matter of adjustment.  Hm, which also shifts bias, so it might be better to use a large RFC (L10 say 47uH) and hold drain voltage ~constant.

SOT-23 is good for a few 100 mW, more than 300 is a concern.  Check the datasheet.  Pin 3 is usually substrate and can be heatsunk to help out.  10mA at 12V is only 120mW so you'll be hard pressed to get into trouble.

Ah, 350mW limit and 60mA max Idss.  So you'll want some R15, and it'll probably be in a reasonable range either to match input, or set bias or gm as desired.

Anyway, with an RFC pullup (L10, no C10, small or no R14), just couple it to whatever.  No need to waste R11 in series, use it in parallel instead -- can be in parallel with L10, or C8, doesn't matter.

The one optimization you can do, is: if bias is low enough, and the shunt resistance is low enough, you can use it for supply instead of L10.  Just resistance couple the drain to the filter.  If it needs to be 1k and bias is 10mA, that's not great, you drop 10V; but if it's 5mA say, that's perfect.  Just don't want to run out of drain voltage headroom.

Tim

[Xnke]

More work on this.



The section of the schematic from C11 through C34, fed from and terminated by a 50R source and load, gives very good matching (-28db or better return loss on the input from 1Mhz to 50Mhz) when built on pad-per-hole board. The pi-network is pretty critical, probably better to do a transformer coupled output to get back to 50R. As drawn here, the amplifier from C11 through C34 provides 12dB of gain, and a noise factor of 1.2dB. I don't know how to measure noise factor so I took it to the RF engineer I know and had him look at it-he did the measurements and gave the good news on the matching and noise factor. Their lab gear is much, MUCH more spendy than what I've got...and the engineer is very bored with the 315Mhz wireless thingamawhatzits that they make.

And yes, now I need to make input and output 50R. I screwed up and forgot the CMOS bus switches that are being used to switch the filters can't take the 1K impedance level. Matching to 1K before the switches would mean that on a strong signal the bus switches could see voltage that exceeds their smoke limit. They do have excellent isolation in the off state, and extremely low insertion loss in the on state up to about 100Mhz.

Don't need the diplexer for the wideband match now, so it'll go away.

Now...the CMOS bus switches...Just going by other RF projects on the web who've used them, these things got 100nF DC blocker caps to keep from robbing the input bias provided by the voltage dividers. That's definitely gonna screw up the output pi-network on the amplifier, so I guess that's the next thing to work on.

[T3sl4co1l]

Ah, then R14 probably isn't needed, as the pi network + termination has an impedance of its own (it then needs to be designed from a open circuit / singly terminated filter prototype).  Gain and noise sounds fine so it won't make much difference.

The switches, looks like they switch to ground?  Makes an annoying pop when switched, dunno if that'll matter (AM detector may find it annoying?).  Why not use one coupling cap and one resistor divider, and why not a larger value that loads it less (say 10 or 100k's)?

Good old fashioned CD4051 or CD4066 is good for higher voltage and impedance, or TC4W53FU,LF is, probably pretty much a higher voltage substitute?, or the good old DG419 or the like.  YMMV on cost and high frequency isolation.

I don't think I'd worry about shorting the unused filters (with SPDT per tap), just use a pair of 1-to-N muxes.  If crosstalk ends up being a problem (noticeable notch in pass/transition band?), could just ground the unused ones with 2N7000s, say?

Tim

[Xnke]

Not sure about if I can use a single divider to bias the inputs of the switches. Definitely would be using a higher value pair of resistors, just haven't gotten directly to optimizing that stage yet.

The switches are configured so that the filter is on the common side of the switch, and the switch is either "open"/high impedance, or is connected to the filter. Filters that are not selected are instead grounded on both input and output sides of the filter, to eliminate the chance of excitation through capacitive or inductive means.

I had looked at using CD4066, but I've already got these SC-70 package switches from another project-and they're extremely good for this when working under 100Mhz. Over that, isolation and distortion start creeping up. They're still usable at 500Mhz, though. At less than a quarter each, they're pretty cheap too.

I've also looked at some of the Skyworks RF switches-an SP3T on each end would allow a single switch on each end of the filters and they are two-wire controlled. Saves 1 I/O line, but losses are higher. Would definitely solve the "IF Strip switches can't handle more than 0dBm signal" though as they're rated for 5W through power. No DC blocking caps or bias resistors needed for them, either.