In digital signal processing (DSP), a normalized frequency is a ratio of a variable frequency (f) and a constant frequency associated with a system (such as a sampling rate, f_s). Some software applications require normalized inputs and produce normalized outputs, which can be re-scaled to physical units when necessary. Mathematical derivations are usually done in normalized units, relevant to a wide range of applications.

Examples of normalization

A typical choice of characteristic frequency is the sampling rate (f_s) that is used to create the digital signal from a continuous one. The normalized quantity, f' = \tfrac{f}{f_s}, has the unit cycle per sample regardless of whether the original signal is a function of time or distance. For example, when f is expressed in Hz (cycles per second), f_s is expressed in samples per second.[1]

Some programs (such as MATLAB toolboxes) that design filters with real-valued coefficients prefer the Nyquist frequency (f_s/2) as the frequency reference, which changes the numeric range that represents frequencies of interest from \left[0, \tfrac{1}{2}\right] cycle/sample to [0, 1] half-cycle/sample. Therefore, the normalized frequency unit is important when converting normalized results into physical units.

A common practice is to sample the frequency spectrum of the sampled data at frequency intervals of \tfrac{f_s}{N}, for some arbitrary integer N (see Discrete-time_Fourier_transform § Sampling_the_DTFT). The samples (sometimes called frequency bins) are numbered consecutively, corresponding to a frequency normalization by \tfrac{f_s}{N}.[2]: p.56 eq.(16)[3] The normalized Nyquist frequency is \tfrac{N}{2} with the unit 1⁄Nth cycle/sample.

Angular frequency, denoted by \omega and with the unit radians per second, can be similarly normalized. When \omega is normalized with reference to the sampling rate as \omega' = \tfrac{\omega}{f_s}, the normalized Nyquist angular frequency is π radians/sample.

The following table shows examples of normalized frequency for f = 1 kHz, f_s = 44100 samples/second (often denoted by 44.1 kHz), and 4 normalization conventions:

QuantityNumeric rangeCalculationReverse
f' = \tfrac{f}{f_s}  [0, 1⁄2] cycle/sample1000 / 44100 = 0.02268f = f' \cdot f_s
f' = \tfrac{f}{f_s / 2}  [0, 1] half-cycle/sample1000 / 22050 = 0.04535f = f' \cdot \tfrac{f_s}{2}
f' = \tfrac{f}{f_s / N}  [0, N⁄2] bins1000 × N / 44100 = 0.02268 Nf = f ' \cdot \tfrac{f_s}{N}
\omega' = \tfrac{\omega}{f_s}  [0, π] radians/sample1000 × 2π / 44100 = 0.14250\omega = \omega' \cdot f_s

See also

References

  1. ^ Carlson, Gordon E. (1992). Signal and Linear System Analysis. Boston, MA: ©Houghton Mifflin Co. pp. 469, 490. ISBN 8170232384.
  2. ^ Harris, Fredric J. (Jan 1978). "On the use of Windows for Harmonic Analysis with the Discrete Fourier Transform". Proceedings of the IEEE. 66 (1): 51–83. Bibcode:1978IEEEP..66...51H. CiteSeerX 10.1.1.649.9880. doi:10.1109/PROC.1978.10837. S2CID 426548
  3. ^ Taboga, Marco (2021). "Discrete Fourier Transform - Frequencies", Lectures on matrix algebra. https://www.statlect.com/matrix-algebra/discrete-Fourier-transform-frequencies.