IQ modulator

I tweaked the function generator level and offset to make a pretty picture.

Adjacent lobes are of course of opposite RF phase.

But then it will be noisier than a hard-driven mixer, and parasitic emitter resistance will mean that it is much less linear with respect to the other input (not the LO). If that doesn't bother you then yes using it as an analog multiplier might be useful. Still, if you do that with a chip that is designed for hard switching (without the matched diodes on the LO input that linearise the LO transfer function) then it will be temperature-sensitive also, and not even all that linear with respect to the LO input. Maybe that doesn't bother you either, but it would bother me.

The LTC part starts to limit at about 10 mv RMS input, which is kinda low. As a typical RF part, it is poorly specified. I might use a 90 degree phase shifter and two analog multipliers.

Analog multipliers are going extinct. The best pick now seems to be AD835, which is over 20 years old.

The AD835 is the high frequency version. The AD734 is slower, a great more precise and horribly expensive.

I don't think that they are going extinct, but they are confined to niche applications that can afford high prices.

Not necessarily noisier; a tuned sinewave is narrowband and so is its noise. Parasitic emitter resistance is insignificant if you aren't saturating.

Thermostats aren't expensive.

, and not even all that linear with respect to

What does 'not even all that linear' mean?

Sometimes a sinewave has a lot of wideband noise, sometimes it does not. It depends on the signal source. Anyway I am not discussing the noise of the signal source that provices the LO, I was discussing the noise inherent to the mixer / multiplier.

I am talking about a mixer or multiplier in which current is steered using differential pairs, often called a Gilbert cell. If a differential pair is in a state where one base is at a substantially higher voltage than the other (e.g. higher by 1 volt), then the current entering at the tail will be entirely coming out of one of the collectors, with no current in the other collector. By applying Kirchoff's current law you can deduce that, aside from the noise in the base current (which is small if the transistors have high beta), the diff pair does not add noise to the output signal, where the output signal is the difference between the two collector currents.

If on the other hand the diff pair is biased with both bases at the same voltage (or nearly so) such that both transistors have significant collector current, then any slight change in the base voltages would cause a change in the output signal, and there will be more noise in the output signal. For example the Johnson noise of the base spreading resistance of the transistors in the diff pair will affect the differential output current.

Where it is not difficuit to make a circuit that can operate over a wide range of temperature, it is usually better to do so.

That would be unfortunate; it's how a square-wave LO operates, and exactly NOT how to bias for a sinewave mixer.

That's called gain; it's often a good thing

Yes, the base spreading resistance is a noise source. So, mixing against a square wave, its contribution is summed over the bandwidth of the post-filter, for each of the frequencies F, 3F, 5F, 7F... because the LO has those harmonics. For a sinewave, it's only the F contribution that gets through.

Again, not a small-signal input, so it isn't terribly relevant. The emitter parts of a transistor are the LOWEST resistivity in an IC process, you only put one ohm there when it's a power transistor which benefits from resistance to prevent current hogging/hot spots/second breakdown.

But the temperature sensitivity is just a few dB of gain change; anything other than a precision RF measurement wouldn't care about that. If you DO care, a thermostat is the practical fix. You might want one for your crystal reference source as well. In either case, there are also thermistor compensations possible. The square-wave approach has similar (saturation voltage and power supply) temperature sensitivities.

te:

u can ignore the odd harmonics of the LO. That's sometimes important.

noise.

That kind of mixer can be linear for one of the inputs - the current being steered. This is at the cost of rectifying all the in-phase odd harmonics o f the local oscillator, but you can filter a lot of them out of the input y ou are demodulating before you feed it into the demodulator.

I was in the process of explaining the properties of a mixer using square-wave LO, to contrast these with the case when a sine-wave LO is used. If you read a bit further you would see that.

Gain from a noise source is rarely a good thing in a mixer.

You missed my point entirely.

I agree that the Johnson noise of the base spreading resistance of the signal-input transistors, at all odd harmonics of the LO, will appear to some extent at the output. If you are designing a highly linear mixer then those transistors will be heavily degenerated using emitter resistors, so there won't be much gain from this noise source to the output. The noise contribution at higher harmonics will be quite small, due to the Fourier series of a square wave, and the fundamental component of the noise will be just as bad regardless of the LO waveform. That was not my point.

My point was that if the LO transistors are not hard-switched, then

You always (unintentionally) put resistance in the emitters, unless you have access to a superconducting IC process. If you make the LO transistors big enough that the parasitic emitter resistance does not affect the linearity when operating as a multiplier, then this will place a limit on the LO-frequency because the LO transistors would have to be large, with large capacitance, and low ft due to their low current density. This might be ok for low enough frequencies, but in many RF systems it would be a problem.

Having "a few dB" of unwanted gain change in some parts of a cellphone radio IC would be unacceptable, as one example.

Hehe, try convincing a cellphone manufacturer of that.

Of course it is better to avoid having to do that, wherever feasible.

No, the current-gain from the emitter to the collectors of a hard-switched pair of transistors is very close to 1, regardless of temperature.