FFT is fast, but limited

FFT is limited to equally spaced, equally weighted, cyclic data without any missing point. B-FFT does the same, for not equally spaced, not equally weighted, cyclic/not-cyclic data with missing items, at speed and noise sensitivity on par with commercial FFT routines. (FFT is 5*Nlog2(N). General solution is N*N*N/4; BFFT ~ (5-8)*FFT). The theory is presented in Articles, while computer routines (libraries for Matlab and C/C++) are available for download as Shareware: pay AFTER you try. Everyone may brows the Forum; only those who payed may write or open new thread.

Each data item input has its own tolerance; The B-FFT can handle any set of data. Reliability and error-bars in results change with the available data. Part of the solution is listing the expected error-bar at each position and for each Fourier component.

In the extreme case where data is too scarce, and computation rounding errors render the results unreliable, the routines - automatically - decompose the results to reliable regions and unreliable ones, and report it. The threshold is user-controlled