Signals of low frequency sampled at a highly enough rate, tend to yield adjacent samples of similar numerical value. The reason is that low frequency essentially means slow variation in time and so the numerical values of a low frequency signal tend to change slowly or smoothly from sample to sample. By the subtraction, we remove the part of the samples that did not change in relation to its adjacent samples (what adjacent means is specified by an exponential window parametrized by $\alpha$) and so what remains is the part of the signal that changes rapidly, i.e. its high-frequency components.
↧