Techniques for calibrating non-linearities of ADCs are described, which can be applied whether or not the non-linearities change with frequency. When the non-linearities do not change (are static), the frequency of a calibrating signal is first estimated coarsely in a calibration mode, then a fine estimate is determined using the coarse estimate. These estimates are then used to predict the sinusoidal signal using a linear predictor. A Look Up Table (LUT) containing corrections to the ADC is derived from this result. The LUT is then used in a normal operating mode to correct the output of the ADC. In a case where the characteristics of the non-linearities of the input signal are dynamic and thus change with frequency, a frequency spectrum of interest is broken into several regions. In each of these regions, a frequency is identified and used as a calibrating signal to generate the corresponding LUT. During normal operation of the ADC, in a first method, the bin corresponding to dominant frequency of the signal is identified using a short-length FFT. This bin is used to select the appropriate LUT for operating on the output of the ADC to provide the calibrated output. In a second method used when dynamic input is expected, a single LUT is developed using the averages values from the LUTs determined from the various regions.