Gilbert transconductance cell ( how does it work?)

Anybody know how this thing works? It seems to simulate ok but I'm unclear how 2 frequencies produce a difference frequency.with a gain of 30dB or more across the LC .The signal input is 640KHz at 1 millivolt and the oscillator frequency is 1095KHz at 500mV for a difference of 455KHz at 50mV.

Version 4 SHEET 1 912 680 WIRE -32 -224 -112 -224 WIRE 64 -224 -32 -224 WIRE 304 -224 64 -224 WIRE 432 -224 304 -224 WIRE -32 -192 -32 -224 WIRE 304 -192 304 -224 WIRE 432 -192 432 -224 WIRE 64 -176 64 -224 WIRE 64 -80 64 -96 WIRE 272 -80 64 -80 WIRE 608 -80 272 -80 WIRE 304 -48 304 -112 WIRE 304 -48 96 -48 WIRE 608 -48 608 -80 WIRE 608 -48 592 -48 WIRE 624 -48 608 -48 WIRE 304 -32 304 -48 WIRE 512 -32 304 -32 WIRE 592 -32 592 -48 WIRE 64 -16 64 -80 WIRE 96 -16 96 -48 WIRE 272 -16 272 -80 WIRE 304 -16 304 -32 WIRE -48 32 -240 32 WIRE 0 32 -48 32 WIRE 192 32 160 32 WIRE 208 32 192 32 WIRE 608 32 592 32 WIRE 624 32 608 32 WIRE 512 48 512 -32 WIRE 608 48 608 32 WIRE 608 48 512 48 WIRE 432 64 432 -112 WIRE 512 64 432 64 WIRE -112 80 -112 -224 WIRE 80 80 64 80 WIRE 96 80 80 80 WIRE 288 80 272 80 WIRE 304 80 288 80 WIRE 192 96 192 32 WIRE 432 96 432 64 WIRE 432 96 192 96 WIRE 432 112 432 96 WIRE -48 128 -48 32 WIRE 368 128 368 32 WIRE 368 128 -48 128 WIRE 512 128 512 64 WIRE 80 176 80 80 WIRE 288 176 288 80 WIRE -112 224 -112 160 WIRE 0 224 -112 224 WIRE 16 224 0 224 WIRE 400 224 352 224 WIRE -112 272 -112 224 WIRE 0 272 0 224 WIRE 80 288 80 272 WIRE 192 288 80 288 WIRE 288 288 288 272 WIRE 288 288 192 288 WIRE 400 288 400 224 WIRE 192 336 192 288 WIRE -112 448 -112 352 WIRE 0 448 0 336 WIRE 0 448 -112 448 WIRE 64 448 0 448 WIRE 192 448 192 416 WIRE 192 448 64 448 WIRE 64 464 64 448 FLAG -240 112 0 FLAG -32 -112 0 FLAG 400 368 0 FLAG 64 464 0 FLAG 432 192 0 FLAG 512 192 0 SYMBOL npn 0 -16 R0 WINDOW 0 118 -4 Left 0 WINDOW 3 -26 -14 Left 0 SYMATTR InstName Q1 SYMATTR Value 2N3904 SYMBOL npn 160 -16 M0 WINDOW 0 135 -42 Left 0 WINDOW 3 39 56 Left 0 SYMATTR InstName Q2 SYMATTR Value 2N3904 SYMBOL npn 208 -16 R0 WINDOW 0 40 32 Left 0 WINDOW 3 -29 -8 Left 0 SYMATTR InstName Q3 SYMATTR Value 2N3904 SYMBOL npn 368 -16 M0 WINDOW 0 38 32 Left 0 WINDOW 3 41 59 Left 0 SYMATTR InstName Q4 SYMATTR Value 2N3904 SYMBOL voltage -32 -208 R0 WINDOW 123 0 0 Left 0 WINDOW 39 0 0 Left 0 WINDOW 0 -73 58 Left 0 WINDOW 3 -64 86 Left 0 SYMATTR InstName V1 SYMATTR Value 6 SYMBOL voltage -240 16 R0 WINDOW 3 -161 -18 Left 0 WINDOW 123 0 0 Left 0 WINDOW 39 0 0 Left 0 WINDOW 0 -79 56 Left 0 SYMATTR Value SINE(3 .5 1095k 0 0 0 100) SYMATTR InstName V2 SYMBOL res 48 -192 R0 WINDOW 0 89 34 Left 0 WINDOW 3 53 35 Left 0 SYMATTR InstName R3 SYMATTR Value 3k SYMBOL res 288 -208 R0 SYMATTR InstName R4 SYMATTR Value 3k SYMBOL npn 352 176 M0 SYMATTR InstName Q5 SYMATTR Value 2N3904 SYMBOL npn 16 176 R0 SYMATTR InstName Q6 SYMATTR Value 2N3904 SYMBOL ind 608 -64 R0 SYMATTR InstName L1 SYMATTR Value .0003 SYMATTR SpiceLine Rser=5 SYMBOL cap 576 -32 R0 WINDOW 3 -48 5 Left 0 WINDOW 0 -35 40 Left 0 SYMATTR Value 400p SYMATTR InstName C1 SYMBOL res -128 256 R0 WINDOW 0 -32 34 Left 0 WINDOW 3 -31 61 Left 0 SYMATTR InstName R5 SYMATTR Value 10k SYMBOL voltage 400 272 M0 WINDOW 3 -195 166 Left 0 WINDOW 123 0 0 Left 0 WINDOW 39 0 0 Left 0 SYMATTR Value SINE(1.48 .001 640k 0 0 0 100) SYMATTR InstName V4 SYMBOL res 176 320 R0 SYMATTR InstName R8 SYMATTR Value 510 SYMBOL res 416 -208 R0 SYMATTR InstName R1 SYMATTR Value 10k SYMBOL cap -16 272 R0 WINDOW 0 41 34 Left 0 WINDOW 3 35 56 Left 0 SYMATTR InstName C2

SYMBOL res 416 96 R0 SYMATTR InstName R6 SYMATTR Value 10k SYMBOL cap 496 128 R0 WINDOW 3 49 35 Left 0

SYMATTR InstName C3 SYMBOL res -128 64 R0 SYMATTR InstName R2 SYMATTR Value 30k TEXT -328 432 Left 0 !.tran .0001s

--- news://freenews.netfront.net/ - complaints: snipped-for-privacy@netfront.net ---

Reply to
Bill Bowden
Loading thread data ...

I didn't bother to look at the schematic. A Gilbert cell is an analog multiplier. You multiply signals and you get sidebands. It is simply math.

Gray and Meyer covers the Gilbert cell. Without doing the math, a simple hand waving explanation of a Gilbert cell is you modulate the current in a long tail pair. This modulates the transconductance of the BJTs. The gain of the long tail pair changes, hence multiplication.

Reply to
miso

But how about we all start calling it the Jones cell, because, he invented it (1963).

Gilbert added a linearization diodes, which is great for VCAs, but essentially, no one uses the Gilbert add on in general mixer applications.

I consider it a bit of a travesty that Howard Jones is not given the universal credit for this brilliant topology.

Kevin Aylward

formatting link
formatting link
- SuperSpice

Reply to
Kevin Aylward

Jones is given credit in the wiki page but the circuit is still called the Gilbert cell.. Interesting it was invented in 1963. I was thinking it might have been done earlier with vacuum tubes.

formatting link

--- news://freenews.netfront.net/ - complaints: snipped-for-privacy@netfront.net ---

Reply to
Bill Bowden

It works with e.g. 3 x 7360s. I've seen the circuit before, though I don't think it was in an ARRL handbook where you might expect such a circuit to end up at.

Tim

--
Seven Transistor Labs 
Electrical Engineering Consultation 
 Click to see the full signature
Reply to
Tim Williams

Err.. it was actually me that, effectively, wrote the text for that page :-)

The schematics and an outline were posted on the talk page, but no one seemed up to putting it in the main page. I reformatted the description, and added the calculation bit.

Kevin Aylward

formatting link
formatting link
- SuperSpice

Reply to
Kevin Aylward

The exponential (or log) behavior of a transistor's input diode is essential to clean analog multiplication in a Gilbert cell. It is a different story with the grid non-linearity of a tube.

The 7360 beam-deflection tube was a good analog to a well overdriven Gilbert cell, but the linearity of the deflection mechansim was not too good for linear four-quadrant multiplication.

(A Gilbert cell is not sensitive to magnetic fields, but 7360 really is).

--

Tauno Voipio
Reply to
Tauno Voipio

I saw a tube version in an ARRL publication maybe 45 years ago. Must have been an older handbook though. Maybe 60 years old by now.

?-)

Reply to
josephkk

Looks better to me. Of course you're not going to get large output current ratios in either case (though the Gilbert cell probably is better at that anyway), but the meat of the deflection range is still fairly linear:

formatting link
page 9.

Although, you can clearly see the slope changes with cathode current, and not quite proportionally so.

BJTs having the advantage that, the derivative of an exponential is an exponential, so they work over essentially the *entire* operating range, period. No worries about reduced input range (how the curves 'knee' at lower voltages for lower currents) or stuff like that.

The graph on the proceeding page, I think, gives the pertinent figure: deflection-plate transconductance versus grid bias. It would be nice if it were in terms of cathode current. Curiously, it compresses above -1.1V (the 'crossover' graph only goes up to -1V as well), which should be about the cathode space charge voltage.

Tubes of course exhibit exponential tails as well, but the transition from exponential to power law occurs very early (within the normal operating range, as with RF MOSFETs). And you don't get much exponential tail before leakage (grid and plate emission, for example) swamp your signal. So there's not much value there in regards to multiplication.

Wouldn't be surprised if an optimized structure would be capable of overcoming these limitations, maybe not to such an extent as semiconductors, but improving upon the 7360 at least -- but obviously enough, the 7360 is expensive enough as it is, so who cares?

Tim

--
Seven Transistor Labs 
Electrical Engineering Consultation 
 Click to see the full signature
Reply to
Tim Williams

Speaking of oddball tubes, I've always wondered, occasionally you run across something like "two control grid, three plate" tubes for "waveform generation and special purposes", or the like. Anyone know anything about those sorts of tubes and applications?

Examples: RC26 page 101, "complex wave generators": multi-plate triodes and tetrodes (and combos) 12FQ8, 6FA7, 6KM8, 6FH8.

Tim

--
Seven Transistor Labs 
Electrical Engineering Consultation 
 Click to see the full signature
Reply to
Tim Williams

Maybe not so much as you suggest. The Barrie Gilbert addition of input linearization does depend on the transistor curve, but the basic differential-current-source and two differential current-steering circuits: (1) creates no DC from either input, A or B (2) cancels out direct A and B terms in the output (at Fa and Fb frequencies) (3) cancels out even (square-law) terms from either input, leaving 3F and higher harmonics only (4) leaves an A*B term in.

So, if you treat the system as a small-signal model, the lowest order term (in small A and B amplitudes, the dominant term) is the A*B term. And the frequency output spectrum contains Fa+/- Fb frequencies only.

That means that any matched-pair three terminal devices will produce most of the behavior of the 'Gilbert cell' and almost all the features of Jones' version.

Reply to
whit3rd

In sufficiently low performance applications, yes. The MC1496 is still very useful in low SNR situations.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
 Click to see the full signature
Reply to
Phil Hobbs

I used a LM1496 years ago in a DC application so that we could get the sum of the inputs and nothing else.

I was going to use trannies to make the cell but I found using the

1496 was simpler, since it was already a close match. A Dc mixer block.

With uC today and ADC's it becomes easy to do that in software.

Jamie

Reply to
Maynard A. Philbrook Jr.

called the

Not true. Spice it up if you don't believe it. The technique must act the same at DC as AC.

frequencies)

and higher

term

of Jones' version.

Reply to
josephkk

[about a Gilbert cell]

I'm only talking about the small-signal model; That is to say, d(output)/d(A) with B= 0 (which is the small-signal case) is zero.

Reply to
whit3rd

act

d(output)/d(A) with

Same deal. The model should reflect the actual circuit behavior and that includes DC multiplication. They are also used for amplitude control in various cases with a DC like control input.

?-)

Reply to
josephkk

Yes, of course it does reflect the actual circuit behavior. The result of multiplication by zero is ... zero. Which is what the small-signal response has as a proportional-to-A term. When you allow B to be nonzero, and get a nonzero result, it's the MIXED partial derivative d^2/(dA dB) that is involved, and that is what gives rise to the nonvanishing multiplication term.

Reply to
whit3rd

What's with all the double-talk?

Ignoring real-world imperfections a linearized Gilbert cell simply outputs...

V(OUT) = V(A)*V(B)/K

Where K is usually 10 to keep the output voltage inside the rails ;-)

Good for ALL frequencies including DC. ...Jim Thompson

--
| James E.Thompson                                 |    mens     | 
| Analog Innovations                               |     et      | 
 Click to see the full signature
Reply to
Jim Thompson

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.