Decibel scale with signal processing applications 1.8 2004/12/02 03:18:32 US/Central 2005/01/24 05:10:08.166 US/Central Anders Gjendemsjo gjendems@NO-SPAM.iet.ntnu.no Anders Gjendemsjo gjendems@NO-SPAM.iet.ntnu.no Decibal Processing Signal Introduces the decibel scale and shows typical calculations for signal processing applications.

Introduction The concept of decibel originates from telephone engineers who were working with power loss in a telephone line consisting of cascaded circuits. The power loss in each circuit is the ratio of the power in to the power out, or equivivalently, the power gain is the ratio of the power out to the power in. Let Pin be the power input to a telephone line and Pout the power out. The power gain is then given by Gain Pout Pin Taking the logarithm of the gain formula we obtain a comparative measure called Bel. Gain Bel 10 Pout Pin This measure is in honour of Alexander G. Bell, see .

Alexander G. Bell

Decibel Bel is often a to large quantity, so we define a more useful measure, decibel: Gain dB 10 10 Pout Pin Please note from the definition that the gain in dB is relative to the input power. In general we define: Number of decibels 10 10 P Pref If no reference level is given it is customary to use Pref 1 W , in which case we have: Number of decibels 10 10 P Given the power spectrum density (psd) function of a signal xn, Sxx f . Express the magnitude of the psd in decibels. We find Sxx dB 10 10 Sxx f .

More about decibels Above we’ve calculated the decibel equivalent of power. Power is a quadratic variable, whereas voltage and current are linear variables. This can be seen, for example, from the formulas P V 2 R and P I 2 R . So if we want to find the decibel value of a current or voltage, or more general an amplitude we use: Amplitude dB 20 10 Amplitude Amplituderef This is illustrated in the following example. Express the magnitude of the filter H f in dB scale. The magnitude is given by H f , which gives: H dB 20 10 H f . Plots of the magnitude of an example filter H f and its decibel equivalent are shown in .

Magnitude responses.

Some basic arithmetic The ratios 1,10,100, 1000 give dB values 0 dB, 10 dB, 20 dB and 30 dB respectively. This implies that an increase of 10 dB corresponds to a ratio increase by a factor 10. This can easily be shown: Given a ratio R we have R[dB] = 10 log R. Increasing the ratio by a factor of 10 we have: 10 log (10*R) = 10 log 10 + 10 log R = 10 dB + R dB. Another important dB-value is 3dB. This comes from the fact that: An increase by a factor 2 gives: an increase of 10 log 2 ≈ 3 dB. A “increase” by a factor 1/2 gives: an “increase” of 10 log 1/2 ≈ -3 dB. In filter terminology the cut-off frequency is a term that often appears. The cutoff frequency (for lowpass and highpass filters), fc, is the frequency at which the squared magnitude response in dB is ½. In decibel scale this corresponds to about -3 dB.

Decibels in linear systems In signal processing we have the following relations for linear systems: Y f H f X f where X and H denotes the input signal and the filter respectively. Taking absolute values on both sides of and converting to decibels we get: The output amplitude at a given frequency is simply given by the sum of the filter gain and the input amplitude, both in dB.

Other references: Above we have used Pref 1 W as a reference and obtained the standard dB measure. In some applications it is more useful to use Pref 1 mW and we then have the dBm measure. Another example is when calculating the gain of different antennas. Then it is customary to use an isotropic (equal radiation in all directions) antenna as a reference. So for a given antenna we can use the dBi measure. (i -> isotropic)

Matlab files filter_example.m