INTEGRATION OF THE POWER SPECTRAL DENSITY FUNCTION Revision B

By Tom Irvine
Email: [email protected]
March 18, 2000
_____________________________________________________________________

Introduction
Random vibration is represented in the frequency domain by a power spectral density
function. The overall root-mean-square (RMS) value is equal to the square root of the
area under the curve. The purpose of this tutorial is to explain the integration procedure.

A power spectral density specification is typically represented as follows:

An example is shown in Figure 1.

POWER SPECTRAL DENSITY

0.1
ACCELERATION (G /Hz)
2

0.01

0.001
10 100 1000 2000

FREQUENCY (Hz)

Figure 1.

1
 2
Note that the power spectral density amplitude is represented in units of (G /Hz). This is
2
an abbreviated notation. The actual unit is (GRMS /Hz).

Derivation
The equation for each segment is

y 
y( f ) =  1  f n (1)
 f1n 

The starting coordinate is ( f 1, y 1 ) .

The exponent n is a real number which represents the slope. The slope between two
coordinates ( f 1, y 1 ) and ( f 2 , y 2 ) is

y 
log 2 
 y1 
n= (2)
f 
log 2 
 f1 

The area a 1 under segment 1 is

f 2  y1  n
a 1= ∫
f1
 n  f df (3)
 f 1 

There are two cases depending on the exponent n.

The first case is

 y1   1  n + 1 f 2
a1 =   f , for n ≠ − 1 (4)
n   +
 f 1  n 1 f1

y 
a1 =  1 
n
 f 1 
[
 1  n+ 1
 n + 1 f 2 ]
− f 1n + 1 , for n ≠ − 1 (5)

2
The second case is

f 2  y1  − 1
a1 = ∫ f1
 − 1  f df , for n = − 1 (6)
 f 1 

∫f 1 [y1f1 ] f
f2 df
a1 = , for n = − 1 (7)

[ ]()
f2
a 1 = y1f 1 ln f , for n = − 1 (8)
f1

[ ]
a 1 = [y1 f1 ] ln( f 2 ) − ln( f 1 ) , for n = − 1 (9)

  f 
a 1 = [y1 f1 ] ln 2   , for n = − 1 (10)
  f 1 

In summary, the area under segment i is


[ ]
 yi   1 
  n   f i + 1n + 1 − f i n + 1 , for n ≠ − 1
  f i   n + 1

a i=  (11)
  f 


[ i i ]   f   , for n = − 1
y f ln  i + 1

   i  

The overall level L is

m
L = ∑ ai (12)
i= 1

where m is the total number of segments.

3
Example
Consider the power spectral density function in Figure 1. The breakpoints are given in
Table 1.

Table 1.
Power Spectral Density
Freq Level
(Hz) (G 2/Hz)
10 0.002
100 0.04
1000 0.04
2000 0.02

Consider the first pair of coordinates:

2
f 1 = 10 Hz y 1 = 0.002 G /Hz
2
f 2 = 100 Hz y 2 = 0.04 G /Hz

Calculate the slope.

 0.04 
log 
 0.002 
n= (13)
 100 
log 
 10 

n = 1.3 (14)

Substitute into equation (11).

 0.002 
a1 = 
101.3 

 1
1.3 +

1
[
1001.3+ 1 − 101.3+ 1 ] (15)

 0.002   1 
a1 = 
10   
1.3 [
2.3
  2.3  100 − 10
2.3
] (16)

a 1 = 1.726 G 2 (17)

4
Consider the second pair:

2
f 2 = 100 Hz y 2 = 0.04 G /Hz
2
f 3 = 1000 Hz y 3 = 0.04 G /Hz

Calculate the slope.

 0.04 
log 
 0.04 
n= (18)
 1000 
log 
 100 

n = 0. (19)

Substitute into equation (11).

 0.04   1 
a2 =   
100   0 + 1
0 [
10000 + 1 − 1000 + 1 ] (20)

 0.04  1 
a2 =     10001 − 1001  (21)
 1  1 
 

a 2 = 36.000 G 2 (22)

Consider the third pair:

2
f 3 = 1000 Hz y 3 = 0.04 G /Hz
2
f 4 = 2000 Hz y 4 = 0.02 G /Hz

Calculate the slope.

 0.02 
log 
 0.04 
n= (23)
 2000 
log 
 1000 

n = -1. (24)

5
Substitute into equation (11).

  2000  
a 3 = [( 0.04)(1000)] ln  (25)
 1000  


a 3 = 27.726 (26)

Now substitute the individual area values into equation (12).

L= (1.726 + 36.000 + 27.726)G 2 (27)

The overall level is

L = 8.09 G RMS (28)

Additional information on slopes is given in Appendix A.

6
 APPENDIX A

Introduction to dB/octave Slopes

0.1
Overall Level = 6.0 grms
2
0.04 g / Hz

+3 dB / octave -3 dB / octave
PSD ( g / Hz )
2

0.01

0.001
20 80 350 2000

FREQUENCY (Hz)

Figure A-1.

The task is to determine the coordinates of the endpoints.

Derivation
2
Assume that a1 and a 2 each has an amplitude in G /Hz. The difference in dB between
a1 and a 2 is

a 
∆ dB = 10 log  2  (A-1)
 a1 

7
Furthermore,

a 2 = a1  10 ∆ dB / 10  (A-2)
 

Additional equations are needed.

The slope N between two coordinates ( f1, a1 ) and ( f 2 , a 2 ) in a log-log plot is

a 
log  2 
N=  a1  (A-3)
f 
log  2 
 f1 

Solve for a 2 .

f  a 
N log  2  = log  2  (A-4)
 f1   a1 

 N
 f 2   a 
log     = log  2  (A-5)
  f1    a1 

Take the anti-log.

N
a 2  f 2 
 =   (A-7)
 a1   f1 

Thus,
N
f 
a 2 = a1  2  (A-8)
 f1 

8
Now consider a one-octave frequency separation.

f 2 = 2 f1 (A-9)

Substitute equation (A-9) into (A-3).

a 
log  2 
N=  a1  (A-10)
log[]2

Substitute equation (A-1) into (A-10).

∆ dB / 10
N= (A-11)
log[]
2

Note that ∆ dB represents the dB/octave slope in equation (A-11). Again, equations
(A-10) and (A-11) assume a one-octave frequency separation.
Now substitute equation (A-11) into (A-8).

 ∆ dB / 10 
 f 2   log[]2 

a 2 = a1   (A-12)
 f1 

Example
Calculate the amplitude at 2000 Hz for the power spectral density in Figure A-1. The
slope is -3 dB/octave.
Note

f1 = 350 Hz
f 2 = 2000 Hz
a1 = 0.04 G 2 / Hz

Substitute into equation (A-12).

9
  − 3 dB / 10 
2  2000 Hz  log[]
 2


a 2 = 0.04 G / Hz   (A-13)
 350 Hz 

a 2 = 0.007 G 2 / Hz at 2000 Hz (A-14)

Now calculate the amplitude at 20 Hz for the power spectral density in Figure A-1. The
slope is +3dB/octave.
Note

f1 = 80 Hz
f2 = 20 Hz
a1 = 0.04 G 2 / Hz

Substitute into equation (A-12). Note that this equation allows f 2 < f1 .

 + 3 dB / 10 
 20 Hz   log[] 2 

a 2 = 0.04 G 2 / Hz   (A-15)
 80 Hz 

a 2 = 0.01 G 2 / Hz at 20 Hz (A-16)

10