en>Citation bot 1 at 08:17, 11 March 2011

2011-03-11T08:17:18Z

New page

In [[digital signal processing]] (DSP), the continuous time variable, '''t''', with units of ''seconds'', is replaced by the discrete integer variable, '''n''', with units of ''samples''. More precisely, the time variable, in ''seconds'', has been normalized (divided) by the sampling interval, '''T''' (''seconds/sample''), which causes time to have convenient integer values at the moments of sampling. This practice is analogous to the concept of [[Natural units]], meaning that the natural unit of time in a DSP system is ''samples''. Accordingly, the natural unit of frequency is ''cycles/sample'', known as '''normalized frequency'''.

The normalized value of a frequency variable, <math>\scriptstyle f</math> (''cycles/sec''), is <math>\scriptstyle f/f_s,</math>  where <math>\scriptstyle f_s = 1/T</math>  is the [[sampling rate]] in ''samples/sec''.  The maximum frequency that can be unambiguously represented by ''digital'' data is <math>\scriptstyle f_s/2</math>  (known as [[Nyquist frequency]]) when the samples are real numbers, and <math>\scriptstyle f_s</math>  when the samples are complex numbers.<ref>See [[Aliasing]]</ref>  The normalized values of these limits are respectively 0.5 and 1.0 ''cycles/sample''. This has the advantage of simplicity, but (similar to [[natural units]]) there is a potential disadvantage in terms of loss of clarity and understanding, as these constants <math>\scriptstyle T</math> and <math>\scriptstyle f_s</math> are then omitted from mathematical expressions of physical laws.

The simplicity offered by normalized units is favored in textbooks, where space is limited and where real units are incidental to the point of a theorem or its proof. But there is another advantage in the DSP realm (compared to physics), because <math>\scriptstyle T</math> and <math>\scriptstyle f_s</math> are not "universal physical constants". The use of normalized frequency allows us to present concepts that are universal to all sample rates in a way that is independent of sample rate. An example of such a concept is a digital filter design whose bandwidth is specified not in [[hertz]], but as a percentage of the sample rate of the data passing through it. Formulas expressed in terms of <math>\scriptstyle f_s</math>  and/or <math>\scriptstyle T</math>  are readily converted to normalized frequency by setting those parameters to 1. The inverse operation is usually accomplished by replacing instances of the frequency parameter, <math>\scriptstyle f ,</math>  with <math>\scriptstyle f/f_s</math>  or  <math>\scriptstyle f\cdot T.</math><ref>{{cite book|last=Carlson|first=Gordon E.|title=Signal and Linear System Analysis|year=1992|publisher=©Houghton Mifflin Co|location=Boston,MA|isbn=8170232384|pages=469,490}}</ref>

== Alternative normalizations ==

Some programs (such as [[MATLAB]]) that design filters with real-valued coefficients use the Nyquist frequency (<math>\scriptstyle f_s/2</math>) as the [[normalization constant]]. The resultant normalized frequency has units of ''half-cycles/sample'' or equivalently ''cycles per 2 samples''.

Sometimes, the unnormalized frequency is represented in units of [[radians per second|''radians/second'']] ([[angular frequency]]), and denoted by <math>\scriptstyle \omega.</math>  When <math>\scriptstyle \omega</math> is normalized by the sample-rate (''samples/sec''), the resulting units are ''radians/sample''. The normalized Nyquist frequency is π ''radians/sample'', and the normalized sample-rate is 2π ''radians/sample''.

The following table shows examples of normalized frequencies for a 1 kHz signal, a sample rate <math>\scriptstyle f_\mathrm{s}</math> = [[44.1 kHz]], and 3 different choices of normalized units. Also shown is the frequency region containing one cycle of the [[discrete-time Fourier transform]], which is always a periodic function.

{| class="wikitable"
|-
| '''Units'''
| '''Domain'''
| '''Computation'''
| '''Value'''
|-
| cycles/sample
|  [-½, ½]  or  [0,1] 
| 1000 / 44100
| 0.02268
|-
| half-cycles/sample
|  [-1,1]  or  [0,2] 
| 1000 / 22050
| 0.04535
|-
| radians/sample
|  [-π,π]  or  [0,2π] 
| 2 ''π'' 1000 / 44100
| 0.1425
|}

== Notes and citations ==
{{reflist}}

==See also==
*[[Prototype filter]]

{{DEFAULTSORT:Normalized Frequency (Digital Signal Processing)}}
[[Category:Digital signal processing]]
[[Category:Units of frequency]]

Electron capture ionization - Revision history

en>Citation bot 1 at 08:17, 11 March 2011