Spectra of Digital Waveform

time-domain and frequency-domain

Prof. Tzong-Lin Wu Department of Electrical Engineering and Graduate Institute of Communication Engineering National Taiwan University

Basic Properties of Fourier Transform

1. periodic x(t ) F ( x(t ) ) = 1/ T

x(t ) =
n =-

x(t )e jn0t dt = Cn

C e
n

jn0t

Basic Properties of Fourier Transform

2. Cn = C- n* for real x (t )
3. F ( x1 (t ) + x2 (t ) ) = F ( x1 (t )) + F ( x2 (t ))

Basic Properties of Fourier Transform

4. F ( x(t ) ) = e
jn0

F ( x(t ) )

Basic Properties of Fourier Transform

5. F ( (t kT ) ) C n = (1/ T )e j0

k = 0, 1, 2, 3....

Basic Properties of Fourier Transform

6. F ( d k x(t ) / dt k ) Cn ( k ) = ( jn0 )k Cn

Basic Properties of Fourier Transform

Spectra of digital circuit waveforms

a. The periodic , trapezoidal pulse train representing clock and data signals of digital systems.
A A/2

A: Amplitude,

r : pulse rise time , f : pulse fall time

: pulse width ( 50 % )

c. To obtain the complex exponential Fourier series of this waveform

(1) x

A/ r

r
-A/ f

f
T t Next period

A/ r

A/ f
+ ( r f ) / 2

r
-A/ r

+ ( r f ) / 2
T t

-A/ f

 F ( x '') = A / T r - A e jn0 r / T r - A e + Ae = A / T * [ (e
jn 0 ( + ( r + f ) / 2) jn 0 r / 2

(2)

jn 0 ( + ( r f ) / 2)

/ T f

/ T f
jn 0 f / 2

/ r ) * ( e jn0 r / 2 - e jn0 r / 2 ) - e
jn 0 f / 2

- (e jn0 r / 2 e jn0 / f ) * ( e = j ( A / 2 n ) * ( n 0 ) 2 * e
(3)
jn 0 ( r + f ) / 2

)]

*[sin( n 0 r / 2) * e jn0 / 2 /( n 0 r / 2)

- sin( n 0 f / 2) * e jn0 / 2 /( n 0 f / 2) ]

F ( x ) = F ( x '') / ( jn0 ) 2 if

, n0

r = f sin(n0 / 2) sin(n0 r / 2) jn ( + e T n0 / 2 (n0 r / 2)

0 r

Cn = A

)/2

d. Expression as one-side spectrum

x(t ) = C0 + Cn+ cos(n0t +Cn )
n=1

when Cn+ = 2 Cn = 2 A sin(n / T ) /(n / T ) sin(n r / T ) /(n r / T ) , for n 0 with

r = f , C0 = A /T , and 0 = 2 / T

Note : (1) 50% duty cycle /T =1/2 no even harmonics (2) slight variations from an 50% even harmonics vary widely

Spectral bounds for trapezoidal waveforms

a. sin x / x 1 , for small x 1/ x , for small x

20 log 1/ x = 20dB/decade , for small x

sin x / x

20dB/decade

Effect of rise/fall time on Spectral Content

b. let f = n / T envolope = 2A* sin( f)/( f) sin( r f)/( r f) / T 20log(envolope) = 20log(2A / T ) + 20log sin( f)/( f) +20log sin( r f)/( r f)

2A /T

20db / decade

40db / decade

1/

1/ r

Effect of rise/fall time on Spectral Content

1v

r = f =5 ns
f
6.37MHz 15.9MHz 63.66MHz

Effect of rise/fall time on Spectral Content

20ns

5ns

11th harmonic 110MHz 68dBuV 86.1dBuV

Effect of rise/fall time on Spectral Content

c. for example : 10MHz , r = f =20 ns and r = f =5 ns

1v

r = f =5 ns
f
6.37MHz 15.9MHz 63.66MHz

High frequency parts

Effect of rise/fall time on Spectral Content

Spectral bound prediction
20ns rise time

Level110MHz = 20 log10 (1V ) 20 log10 ( = 78.45dB V

15.9 MHz 110 MHz ) 40( ) 6.37 MHz 15.9MHz

5ns rise time

Level110MHz = 20 log10 (1V ) 20 log10 ( = 90.5dB V

63.7 MHz 110 MHz ) 40( ) 63.7 MHz 63.7 MHz

Effect of rise/fall time on Spectral Content

The exact value calculated by sinc function 20ns 73.8dBuV 5ns 90.4dBuV
Very close to the spectral bound prediction

The measured values are

20ns 68.8dBuV 5ns 86.1dBuV

Why are they lower than the value by the exact calculations?
What is the effect of the duty cycle on the spectral bound?

Effect of Repetition rate and Duty Cycle

What is the effect of the duty cycle on the spectral bound? Duty cycle:

D=

One-sided spectrum:
+ = 2 AD cn

sin( n D ) n r f 0 n D n r f 0

DC level 1st breakpoint

10

Effect of Ringing (Undershoot/Overshoot)

Assume = 0

cn = cn ( squarewave ) +

1 T

T /2

1 T /2 Ke t sin(r t + )e jn0t dt e jn0T / 2 Ke t sin(r t + )e jn0t dt T 0 1 T

T /2

= cn ( squarewave ) + (1 e jn0T / 2 )

Ke t sin(r t + )e jn0t dt

1 1 V0 sin( 4 n0T ) jn0T / 4 K sin( 4 n0T ) jn0T / 4 pr e e + p 2 + 2 p + 2 + r2 2 1 n T 2 1 n T 0 0 4 4 1 sin( n0T ) 2 2 2 V0 jn0T / 4 p + (2 + ( K / V0 )r ) p + ( + r ) 4 e = 2 1 n T p 2 + 2 p + ( 2 + r2 ) 0 4

Square wave components

where Bandpass

p = jn0

Effect of Ringing (Undershoot/Overshoot)

The spectrum of a square wave with undershoot/overshoot has a part of its spectrum enhanced or increased about the ringing frequency Wr. Quite often we see in the radiated emission profile a seemingly Resonant region of enhanced emissions in a narrow frequency band. One possible explanation for this is the undershoot/overshoot present on the digital waveforms.

11

Use of Spectral Bounds in Computing Bounds on the Output Spectrum of a Linear System

Spectrum Analyzer

12

Spectrum Analyzer
Bandwidth of SA

The displayed level at the center frequency of the bandwidth will be the sum of the spectral levels that fall within the bandwidth of the filter at that time. In order to obtain the lowest possible level on the SA display, we should choose as small a bandwidth as possible.

Spectrum Analyzer

EMC receiver regulation

13

Spectrum Analyzer

Spectrum Analyzer
Peak and Quasi-peak

Infrequently occurring signals will result in a measured Quasi-peak level that is considerably smaller than a peak detector would give. That infrequent events may be of sufficient magnitude distressingly large received levels on a SA that is set to peak detector function. The reason for the use of quasi-peak detector function relates to the intent of the regulatory limits, which is to prevent interference in radio and wire communication Receivers. Infrequent spikes and other events do not substantially prevent the listener from obtaining the desired information.

14

Representation of Random Signals

Digital data waveforms are obviously random signals, not deterministic signals

x(t ) =

1 X 0 [1 + m(t )] 2

Rx ( ) = x(t ) x(t + ) = lim

t

1 x(t ) x(t + )dt T T/ 2

T /2

stationary

1 2 X 0 [1 + m(t )][1 + m(t + )] 4 1 = X 02 [1 + m(t ) + m(t + ) + m(t )m(t + )] 4 1 = X 02 [1 + m(t )m(t + )] 4 = = 1 =0

for < T for > T

Representation of Random Signals

Power Spectral density

Gx ( f ) =

R ( )e
x

15