Overview of Noise Calculations

Parseval's Theorem1:
The mean square value of a random variable v(t) or i(t) is related to its (one-sided) Power
Spectral Density Sv(f) or Si(f) by
∞ ∞
2 2
v (t ) = ∫ Sv ( f ) df and i (t ) = ∫ S ( f ) df
i
0 0

White Noise

For "white €
noise" sources including the thermal noise of resistors or shot noise associated
with pn junctions, the power spectra are independent of frequency:

Sv(f) Si(f)

4kTR 2qI

f f
Resistor Thermal Noise Shot Noise
Power Spectrum Power Spectrum

The noise power in a frequency band between f1 and f2 Δf = f2 - f1 is given by

f2 f2
2 2
v (t ) = ∫ 4kTR df = 4kTRΔf and i (t ) = ∫ 2qI df = 2qIΔf
f1 f1

The bandwidth Δf is also often called B in many publications, B = f2 - f1:

f2 f2
2 2
€ v (t ) = ∫ 4kTR df = 4kTRB and i (t ) = ∫ 2qI df = 2qIB
f1 f1

Addition of Noise Sources

For vs (t ) = v1(t ) + v2 (t ) ,
€
2
[ ]
vs2 (t ) = v1 (t ) + v2 (t ) = v12 (t ) + v22 (t ) + 2v1 (t )v2 (t )

v1 (t )v2 (t )
vs2 (t ) = v12 (t ) + v22 (t ) + 2C12 v12 (t )v22 (t ) with C12 = = correlation coefficient
v12 (t )v22 (t )

€ 1
R. D. Thornton, D. Dewitt, E. R. Chenette and P. E. Gray, Characteristics and
Limitations of Transistors, SEEC Vol. 4, Wiley: 1966.

1 RCJ - 8/26/07
 Overview of Noise Calculations

C12 represents a measure of the correlation between v1 and v2, and -1 ≤ C12 ≤ +1. For
uncorrelated noise sources, C12 = 0, and

2
For uncorrelated noise sources : [ ]
v s2 (t ) = v1 (t ) + v2 (t ) = v12 (t ) + v22 (t )

Example of Noise Calculations for Resistor Combinations

€

R
vn2 2
R1 R R2
1

vo vo

vn1 vn1 vn2

Series connection:
vo = v n1 + vn2 where vn12 = 4kTR1B and 2
v n2 = 4kTR2 B
Since the resistors are physically independent,
we expect the individual noise sources to be uncorrelated, and
2
vo2 = (v n1 + vn2 ) = vo12 + v o2
2
= 4kTR1B + 4kTR2 B = 4kT ( R1 + R2 ) B
Parallel connection:
R2 R1
€ vo = v n1 + vn2
R1 + R2 R1 + R2
Again, assuming the individual noise sources to be uncorrelated,
2
2
 R2 R1  2
 R2 2 2  R1 2
v = v n1
o + vn2  = vo1  + v o2  
 R1 + R2 R1 + R2   R1 + R2   R1 + R2 
R1 R22 + R12 R2 R1 R2
vo2 = 4kTB 2
= 4kTB
R1 + R2
(
= 4kT R1 R2 B )
(R + R )
1 2

2 RCJ - 8/26/07
 Overview of Noise Calculations

Noise Calculations in Frequency Dependent Networks

For a linear frequency dependent network, the output power spectral density Svo(f) is
equal to the input power spectrum Svi(f) multiplied by the square of the magnitude or the
network transfer function H(f):

Svi(f) H(f) Svo(f) C vo

vn

2
Svo ( f ) = Svi ( f ) H ( f )

Consider the low-pass filter in the figure above

€ Vo ( s) 1 2 1
Svi = 4kTR H (s ) = = H( f ) = 2
Vi ( s) 1+ sRC 1+ (2πRCf )
4kTR ∞
4kTR kT
Svo ( f ) = 2
vo2 = ∫ 2
df =
1+ (2πRCf ) 0 1+ (2πRCf ) C

Effective Noise Bandwidth B

€ v o2
For a given noise power, the effective noise bandwidth is defined as ENB =
4kTR
1
For the single - pole low - pass filter above with cutoff frequency ω o = ,
RC

kT C 1 π
ENB = = = fo
4kTR 4RC 2

Available Noise Power

€ The maximum power available from a resistor is v n2 = kTB
In a 1- Hz bandwidth at room temperature, the available power corresponds to
( )
vn2 = 1.38x10−23 J / K (300K )(1Hz) = 4.14x10−21W
 4.14x10−21W 
This value is often expressed in dBm as : 10 log  = −174dBm
 0.001W 

€ 3 RCJ - 8/26/07