Generating model noise


Generating non-white random noise means producing arbitrary long time series with a given spectral density. Such time series are needed for example for the following purposes:

One way of doing this is to apply digital filters (FIR or IIR) to white input noise.
This approach is often used because of its simplicity but is has some disadvantages:

For complicated spectra a matching filter is not trivial to find. The filter has to be split into several simple sections. Via nonlinear optimization techniques optimized filters can be found. Those techniques however are non deterministic such that their convergence is not guaranteed. So although the resulting filter transfer function is well known, it may not perfectly match the given spectrum.
Moreover the filtering method requires a 'warm-up period'

A different approach is implemented in LTPDA as Franklin noise-generator.
It produces spectral densities according to a given pole zero model (see Pole/Zero Modeling) and does not require any warm-up period.




©LTP Team