How it occurs
An ideal device is linear: the output is proportional to the input and no new frequencies are created. A real device has a small nonlinearity that can be described as a sum of terms of the signal raised to different powers. With two tones at frequencies f1 and f2, the cubic term generates, among others, the products 2f1−f2 and 2f2−f1.
Why it is a problem
If f1 and f2 are close together, these products lie at a distance from the signals equal to the spacing between them, that is, within or next to the band of interest. Unlike harmonics, which are far away and can be removed with a filter, these products cannot be separated by filtering. In a receiver, two strong interferers can create a product right on top of a weak signal that you want to receive.
How it grows with level
When the power of the input tones is raised by one decibel, the third-order products rise by three decibels, while the fundamental tones rise by one. On a plot in decibels, the lines have slopes of one and three. If they are extrapolated until they cross, the crossing point is called the third-order intercept point. It is a figure of merit: the higher it is, the better the linearity. It is an extrapolated value, because in a real device compression arrives before the lines cross.
How it is measured
- Two generators of the same level are combined, with isolation between them so that they do not intermodulate with each other.
- They are applied to the device and the output is observed on a spectrum analyzer.
- The level difference between the tones and the products is read.
- It is confirmed that the distortion is not produced by the analyzer itself, by adding attenuation at its input and checking that the difference does not change.
How to interpret the result
The difference between tones and products decreases as the level is raised. The same device changes its behavior with bias and with the spacing between tones, so the result must be reported together with the measurement conditions.