Answer 3—A median-value filter is a class of nonlinear filter that’s implemented as follows. A series of data samples is fed into a FIFO buffer, usually of odd length. As each sample is added (and the oldest discarded), the entire buffer is sorted numerically, and the output of the filter is the sample that ends up in the middle of the sorted sequence. This is known as the median value because there are as many samples below its value as above.

This type of filter is good at filtering glitches of limited duration (but arbitrary amplitude) from a stream of sampled data. As long as a glitch is less than half the buffer length, it will be removed from the output stream. The underlying waveform of the data will be well preserved. The major drawback of this type of filter is the delay, which equals half of the buffer length as in any symmetric FIR filter.

Contributor: David Tweed