Let's suppose I have a discrete signal $x[n]$ with a certain frequency $f_x$.

Now, due to various unimportant reason, I want to obtain a new signal $y[m]$ with a new frequency $f_y = \frac{f_x}{k}$, with, for example $k = 10$.

Now, there are various ways to do it, I am for the moment interested in two:


$y_d[m] = x[m* k + (k - 1)]$


$y_a[m] = \frac{1}{k}\sum_{i=m*k}^{m*k+k-1}{m[i]}$

What are the differences between the two, from a theoretical standpoint? What effect on the spectrum of $y_d$ and $y_a$?

I have limited knowledge of signal processing, so sorry for if something confuses you. A brief explanation with some references for further reading would be perfect.


What you are calling decimation is in fact downsampling: keeping one of every $k$ samples from the original signal.

Proper decimation involves a digital low-pass filter in order to eliminate all components above $\pi/k$ (or $f_s/(2k)$ in continuous time) since these components cannot be represented at the new lower sampling frequency (and will result in aliasing).

After low-pass filtering, you downsample your signal keeping one out of every $k$ samples.

The second method you propose is actually a low-pass filter followed by downsampling. But in this case, the low-pass filter you use is a moving-average filter, which is very simple, but it is not a very good low-pass filter. Gain in the pass band is very variable, and attenuation in the stop band is poor.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.