FET vs BJT - OIP3/IIP3?

[tec5c]

I was asked a question recently which I could not give a detailed answer to and I have been doing some research but have yet to find a decent response:

Why are FETs favoured for high-frequency radio applications where there are many signals closely spaced with interfering frequencies?

I am still learning about intermodulation and all of its properties so all I could think of was that because FETs are square law devices, but this doesn't really explain anything.

I'd appreciate any help or if you could point me in the direction of material to read/watch that could help me understand more about this.


Thanks.

[T3sl4co1l]

Just because they're cheaper?

Idunno.  Custom silicon and ASICs are almost exclusively CMOS, because it's cheapest to produce and has sufficient performance.  The main competitor I think is SiGe which also offers IC level integration, but is a bipolar process (at least, primarily so).  You can find MOSFETs, MESFETs, JFETs and BJTs with cutoff frequencies up to 60GHz, using the various processes and semiconductor materials (from Si to SiGe, GaAs, InP and GaN).

There isn't terrifically much difference in the fundamental electrical characteristics of those devices.  For sure, if nothing else, you can simply throw more transistors at it, to squeeze out more gain and bandwidth for less noise and distortion.  A single transistor, on a test jig, used for a particular purpose (like a tuned or wideband RF amplifier) might have that concern, but most generally, it's not a problem.

I think the compound semiconductors (GaAs, et al.) are preferred for CMOS (or NMOS alone), because they are direct bandgap (i.e., forward bias emits light -- e.g., green LEDs are ordinary GaP diodes).  So although there are all kinds of heterojunctions available in those materials, they aren't usually used for electrical effect.  (The most important use of that is in laser diodes, where the population inversion is created with heterojunctions, and the optical resonator is formed using alternating layers which exhibit slightly different indices of refraction, and can therefore be made into partially or fully reflective surfaces at the characteristic wavelength.)

Tim

[Howardlong]

I am not sure about the choice of whether to specifically state that a BJT or FET (of whatever flavour) is always better than the other in all cases, but I can recommend a superb text on the subject. "RF Power Amplifiers for Wireless Communications" by Steve Cripps, ISBN 0-89006-989-1.

Basically intermodulation distortion produces spectral regrowth in adjacent channels. Beyond that is above my pay grade these days, maybe a few years ago when I was heavily into linearisation of nonlinear PAs using digital techniques I would have been able to offer more, that's a huge subject in itself with lots of IP around it.

[TimFox]

A fundamental reason why FETs may be superior for RF distortion is the different non-linear response.
To first order, the output current of a BJT is exponential in the input voltage (for signal levels too strong for a linear response, but too low to saturate the device).  The exponential function, expressed as a polynomial, contains terms of all orders (exponents).
To first order, the output current of a FET is a quadratic function of the input voltage, containing only the linear and V^2 term. 
For RF purposes, the 3rd order distortion (measured with IP3) is normally the most important, since the distortion of two closely-spaced frequencies (two-tone test) will be close in frequency to the desired frequencies, and cannot be filtered without losing the desired signals.
With 2nd order distortion, the distortion frequencies will be near zero and the second harmonic of the desired signals, and can be filtered in a communication system.

[G0HZU]

Normally simple FET transmitter PAs are worse than simple BJT PAs for (multitone) linearity tests including IP3 tests for various reasons.

But if you operate certain types of FET PAs in class AB you can find a bias point where the various contributors to the 3rd order IMD terms can cancel and you enter a special zone where the IMD tones go DOWN as you increase drive level. Obviously, this can only happen under a limited range of drive levels and at the optimal bias point but a class AB FET can have lower IMD than a class A amplifier under these conditions.

i.e. with the FET the two test tones get increased slowly and the 3rd order IMD terms (kind of) follow the expected rate of increase but then the IMD terms stop rising and start going DOWN as the optimal point is reached for bias point and drive level. How useful this effect is depends on the application but it does get exploited in commercial applications where this form of 'linearising' is much cheaper than other methods. eg feedforward or adaptive predistortion methods.

[G0HZU]

When it comes to mixer design then single FETs make very good (low cost) mixers for radio receivers because of their square law response. They also make very good passive switching mixers and can give very high IP3 (and so low IMD) if the LO signal can switch them very quickly between states.

But simple single ended FET amplifiers generally tend to suffer poor large signal linearity due to the square law response. So BJTs can often outclass then here. (except in special cases like in my previous post)

So a lot depends on the application. There are some radio applications where BJTs are better and some where FETs are better.