filtering-schoolofoceanandearthscienceand-FILTERING.pptVIP

FILTERING;A FILTER is a device or function that allows certain material to pass through it while not others. In electronics and signal analysis, a filter is a device or function that changes the shape of the phase and/or amplitude spectrum in the frequency domain, i.e. it attenuates energy at some frequencies relative to others, and it may also make considerable changes in phase, causing delays or advances in the time domain. A LINEAR TIME-INVARIENT (LTI) filter always behaves the same way regardless of the input or the time. For example a filter that attenuates 60 Hz energy by 48 dB, ALWAYS attenuates energy at 60 Hz by 48 dB. A LTI filter’s response is completely specified by its IMPULSE RESPONSE - the output when the input is an impulse.;A filter is CAUSAL if the output does not anticipate the input. A zero-phase filter (one that does not cause any delay) cannot be physically reliasable (except for a unit impulse). For example, consider a filter with the following impulse response: h=[.2 .2 .2 .2 .2] This filter is symmetric about its center (zero lag) and it will spread any signal at time t(i) over times t(i-2) to t(i+2). It thus sees AHEAD, and it can not be built with physical components. So it is noncausal. We can make this filter causal by adding zeros at the front, adding a 2-step delay: h=[0 0 .2 .2 .2 .2 .2] Now it spreads out the signal at t(i) from t(i) to t(i+4), and it does not see ahead. ;Smoothing filters. Often you will have a power spectrum that is very noisy with large changes from one spectral estimate to another. In the case below it just looks like spikes.;IF we want the underlying structure, we can convolve the power spectrum with a 25-point boxcar (each point =1/25) to obtain the heavy line below:;This type of boxcar filter is a moving average smoother. It is the same as taking a 25-point mean, moving one point, and taking another. Which domain you are in - time or frequency - is not important, smoothing filters work in both domains. Not