Frequency estimation using Fourier coefficients from multiple series

On Dec 29, 8:55&#4294967295;am, Rune Allnor <all...@tele.ntnu.no> wrote:
> On 29 Des, 17:29, KP <ken.pra...@gmail.com> wrote:
>
> > On Dec 27, 10:49&#4294967295;am, Rune Allnor &#4294967295;wrote:
> > Because we need to estimate multiple frequencies per sequence, we were
> > using a DFT-based method which worked fairly well. &#4294967295;However, due to
> > noise, we were having to use sequence lengths longer than can be
> > optimally implemented in our processor (due to cache size, etc.).
>
> There is a significant difference between 'we can not get this to
> work' and 'we can not get this to work optimally'.
>
> If you can get the methods you already tested to work at all,
> even suboptimally, use them. If the operational constraints
> pose problems, these can be solved by means of faster, larger
> computers. Of course, that might take $$$ which might seem like
> a big hurdle up front, but on the other hand it *only* takes
> $$$ to get a larger faster computer. You don't risk aything
> in terms of operational feasability or robustness.
>
> If you can get the method to work, use it. If it works too
> slow for your application, document this fact and apply for
> the funding necessary to get more CPU power.

There are other reasons to use a shorter sequence length.  One is
platform motion: the longer the time taken to collect the sequence,
the greater the possibility that platform motion will degrade the
accuracy of the measurement.

Ken P.