Single transistor mixer

use a DDS chip... you can simply code it from a cheap uC...

That should fit your needs..

Jamie

A single-transistor multiplier would probably be awful.

You don't need to multiply sine waves. Just multiply the signal alternately by +1 and -1 and then lowpass filter.

That can be done, for example, by alternately multiplexing between SIG+ and SIG- with a good fast analog mux. That will always have residual errors, but could be pretty good.

I guess the best lock-in would use a good ADC and do the multiply digitally, in an FPGA. Noise dither maybe.

Assuming that your criteria are "low cost and works" then I'm not sure that you'll be happy with a single-transistor mixer.

Be that as it may, stick your LO signal on the emitter and your RF signal on the base and see what comes out the collector.

Do you need something that gives an output down to DC, without bias? If so, you're going to be vastly unhappy with a single-transistor mixer.

Check out the SA602. It's so old it creaks, but it works and with the right signal conditioning on its outputs it gives a fairly decent output down to DC.

Or, look at the 1000 or so results in DigiKey's "RF mixers" section.

With today's parts you may be able to do this with a set of fast analog switches and an op-amp. Perhaps even a 744051 or whatever?

Something like FSA3157 or ADG604 or one of those T3USB gadgets might be better at 1 MHz.

The classic single-opamp +1/-1 circuit might work with a small phemt as the grounding switch.

But no single transistor!

For RF, a simple diff-pair makes an excellent mixer... think half of a Gilbert multiplier (you don't need the half that eliminates DC from the source... tank loads do that). ...Jim Thompson

Single diodes have been used as mixers for nearly a century. A transistor biased near cutoff with a strong signal applied to the base at one frequency and a weak signal at another will produce sum and difference frequencies with some gain over a simple diode.

The conceptual information and enough math to give you a handle on behavior are here:

A CMOS exclusive-or gate (74HC86) is useable as a mixer as well.

The classic one-transisor mixer is a dual-gate MOSFET. If your 'lock-in amp' has a squarewave reference, analog-switch polarity reversal can be very effective, too. The 'switch' elements can be diodes, or FETs, or any other kind of transistor; you could even consider photocouplers, at your relatively low frequency range.

I'm OK with generating the sine wave. I need to demodulate.

ChesterW

Thanks, I'll check that out!

ChesterW

The Faulkner and Harding phase sensitive detector works fine at audio frequ encies as well (though it was claimed that it worked up to 30MHz).

E.A.Faulkner and D.W. Harding J.Sci. Instrum. volume 43 page 97-99 (1966)

Faulkner improved it a bit

E.A.Faulkner and J.B.Grimbleby Electronic Engineering volume 39 page 565-67 (1967)

Both references are from my Ph.D. thesis.

The original Faulkner and Harding phase sensitive detector turned the - off

-set - signal into a modulated direct current at the collector of a transis tor constant current source, and fed that into a long-tailed pair of transi stors whose bases were driven by the demodulating signal - ideally a pair o f anti-phase square waves at the frequency of interest.

The integrated difference in the currents coming out of the two collectors was the demodulated signal - an operational amplifier subtractor could refe r that to 0V. I just stuck a moving coil ammeter across the collector resis tors.

You have to add enough voltage offset to the original AC signal to make sur e that the current going into the long-tailed pair never drops to zero, and the mark-to-space ratio (which should be exactly 50%) of the demodulating drive determines how effectively this offset is rejected.

It's three transistors rather than one, but it's a remarkably good circuit

- even better if you used a matched transistor pair for the long-trailed pa ir.

I was afraid I might hear that.

Yep, even inexpensive micros come with +1/-1 multiplier :). Unfortunately using this method means the DC mixing product will include noise from all of the odd harmonics. The beauty of a lock-in is the narrow noise bandwidth, and it is compromised with the +1/-1 method.

Or I could DC offset the signal and just do the +/- in the micro. The DC component will cancel out.

That's my benchmark. I can use something like a microchip dsPic that has a hardware multiplier and mix the signals in the chip. The customers product is high volume though, so pennies matter, and the dsPic costs around $1.60. If I could implement a reasonable analog mixer I could use a more modest micro for something around $0.50.

I've taken a look at the raw signal and I've got plenty of noise for bit-dithering. I only need a TC of a second or two, so even for an 8 bit ADC digitization noise will not be a factor.

ChesterW

I need to include a microcontroller in the circuit for other unrelated reasons, so I might as well use it for the signal processing. If I demodulate using a square wave, then the baseline noise in the signal at all of the odd harmonics of the square wave will shift to DC.

The customer originally tried to do this measurement at DC and never got it to work. I guess they never heard of 1/f noise.

I may have to live with the extra noise from demodulating with a square wave, but it would be nice if I could figure a not-too-expensive way to demodulate with a sine and thus keep my low noise bandwidth.

ChesterW

I certainly like the price of that FSA3157.

To be clear, 1 Mhz is not a requirement, I can pick the frequency. I was planning to look for a clean spot in the spectrum maybe around 40 kHz - something where I can get a reasonable over-sample on the A to D if I go with signal processing in an inexpensive micro.

ChesterW

Yes, I've been getting that idea from the responses here. You guys probably saved me a lot of time trying to figure out how to make it work.

Might be a nice benchmark anyhow, even if it's a dead end.

I'm assuming you mean does the input signal need to go down to DC. No. To be clear, I'm generating the carrier and running it through the DUT, then recovering the modified signal and demodulating it using the carrier. I can pick the carrier to be anything I want within reason. It might be nice to give the system the smarts to scan for a clean spot in the spectrum and change the carrier accordingly, but certainly no need to do anything but demodulate at whatever frequency is used to probe.

It's too expensive. Also I'd be afraid of it going obsolete. My benchmark for cost right now is a dsPic at around $1.60. It has a hardware multiplier and so I could do a pretty good demod inside the chip. So the idea is to try to get rid of the need for the multiplier by using an analog multiplier, thus enabling the use of a more modest micro.

ChesterW

I've never heard of a dual-gate mixer, so thanks for the tip!

In an ideal world I'll use a sine wave for my carrier and demodulate with a sine. That way my noise bandwidth will be lower I think than using a square wave.

My DUT doesn't give a linear response to a square wave, but does to a sine wave. The carrier used to demod the signal can be thought of as a sort of matched filter, so since my signal is a sine the best demod carrier is also a sine.

ChesterW

Thanks for the idea. I think using a diode will cause a lot of my desired mixing products to be well removed from DC and so not available to add to my signal. Also, noise in the spectrum of the input signal far removed from the carrier frequency will end up translating down to DC and further decrease my SNR.

ChesterW

I thought of the +1/-1 circuit, but I wasn't sure how it'd do with a

DigiKey lists 74VHC4052 parts from ON that -- if they're as good as they claim -- should be good to past 1MHz. But they're 22 cents each in quantity from Digi-Key.

A good quality four quadrant multiplier will demodulate the sinusoidal component if used to multiply the signal being demodulated by a good good quality sine wave at the desired frequency.

The Analog Devices AD734 is a very good quality multiplier, but it isn't cheap - $A45.18 in small quantities from the local supplier.

In practice the noise advantage of demodulating with a sinewave rather than a square wave is rarely dramatic. Getting rid of 1/f noise tends to be what matters.