How to Choose Cut-off Frequency

I'd have thought you could base cut off frequency on the response time of the system? Say 10 times the max frequency the system can respond at.

Pure cat gut feeling

I always cannt help to smile when see this avatar. I was impressed by it. I guess you are a student studying in some university, but graduate soon.

It seems always only me answer your question I wonder why no others interest in and give you a hand?

now back to the topic.

I wish you put out more condition in order that to answer.

it might be simple, but on other side, its complex.

it depends on the system you dealt with.

if you would use digitalize data, the cutoff fq should be double higher than your simple frq. this is famous Nequist law. usually we use 3-5 times higher cut off frq.

if you wish system is stable, the feedback frq of cuff shoiuld be seelct at that of which amplitude across zero less than 180⁰

^{if you} have goal bandwide, the cut off will be end at that wide. and what data are you dealing with?

if take account of human movement by shot move pictures, the seep can be up to 500 - 1000 frames per second to see clear the details of movement.

if you deal with data analysis, there are many kindsof filter parameter for you to choise, except cutoff frq. for example, rectangualr coeffcience, roll off speed, leakage amount, or even window function etc.

Wish others can offer more details.

I would not go as far as CNPOWER since the basic frequency of human movements is in the range of Hz in unloaded conditions. The fastest movement is the forearm movement.

His comment is correct the Nyquist value is good but it is better to have at least 10 points for the maximal possible frequency which could occur. If we consider it to be 2...3 Hz your cut off should be at least > 3Hz and your sampling frequency > 30 Hz.

The 2 have not to be mixed. 500 to 1000 frames/s is too much, you should count of about 50 to 100.

For the upper limit of neural signal spikes I see a figure of 8 khz, in the poster "VLSI for Multimodal Neuromonitoring" Murari et al, that's somewhere in the 2006 niw workshops report here

http://www.ninds.nih.gov/research/npp/ws/agenda_index.htm

Case to consider, a method employed in audio editing software called "De-noising":

A sample on the unwanted noise is taken, and subtracted from the overall signal, to produce the desired result with "cut-off curve" removed, instead of a "cut-off frequency" curve roll-off, as usually done with an equaliser:

.