The conclusions are quite clear. Fast transient response and sharp spectral filters are not simultaneously achievable. The quickest settling is achieved by the plain old FFT (weighted or unweighted) at the expense of relatively poor spectral filtering. Another way of looking at this is to realise that the step response of the Sinx/x FFT filters is as near to an instant step as is achievable. Once 'brick wall' filters are introduced, the consequent transient response delays caused by the filters needing to fill up must be accepted. The sharper the filters, the longer the transients. If amplitude of the signal in a given bin is the main priority, then 'minimum phase' filters can help, but this will not improve (and may actually worsen) the adjacent channel rejection during the transient period. It will also introduce a non-linear phase response across each of the bins. Over-sampling may be useful in revealing more detail in the transient period, but it will not shorten it.
展开▼
