Frequency estimation using chromatic derivatives

Changes in version 2: (1) now no need to specify the SNR in the detection part of the algorithm it is determined automatically (by a rough but reasonably accurate hack which will be improved later) from the noisy signal alone, assuming that there are at least a few legitimate fragments; (2) counters in the previous version were off and showed more errors than actually present; (3) much faster implementation with much less memory used. To do: (1) to clean up and include a new part of the code which increases noise robustness of the algorithm in a way explained further in the tutorial; (2) to implement colored noise environment; (3) to implement all of this and morre in Matlab, hopefully in August after I come back from a trip home.