L3: Linear, Time-Invariant (LTI) Systems and Linear Distortion

(Carlson & Crilly, secs. 3.1-2)

Time Domain Properties of LTI Systems

Linearity: for any signals s1 (t), s 2 (t) and numbers k1, k 2 ,

T (k1 s1 (t)  k 2 s 2 (t))  k1T (s1 (t))  k 2T (s 2 (t)).

Time-Invariance: if T (s(t))  y(t), then for any time shift t0 ,

T s(t  t0 )   y(t  t0 ).

Not all channels are L or TI....

L3 1
 Impulse Response & Convolution

Impulse response - output when input is unit impulse at t  0,

h  T ( ).

Assertion: h is enough to calculate response of LTI system to any

input: 
 
s(t)   s( ) (t   )d  T s(t)   T  s( ) (t   )d ,
   
 

  s( )T  (t   )d   s( )h(t   )d : s(t)  h(t).

 
NB: s(t)  h(t)  h(t)  s(t).

L3 2
Convolving Aexp(-t)u(t) & tu(t)/T (Fig. 2.4-1,Carlson et al)

L3 3
Result of convolution (Carlson et al, Fig 2.4-2)

L3 4
 Causality

Observe that
 t 

y(t)   s( )h(t   )d   s( )h(t   )d   s( )h(t   )d .
  t

From 2nd integral, value of current output may depend on future

values of input!

For causal systems, y(t) can depend only on past & present input
values  h(t   )  0,   t.

I.e LTI, causal system  h(t)  0, t  0.

L3 5
 Frequency Response

Convolution in time domain  mult. in frequency domain.

 F y(t)  F h(t)  s(t)  H ( f )S( f ).

H ( f )= complex-valued transfer function.

Let input = unit phasor @ f 0 Hz. Then
Y ( f )  H ( f ) ( f  f 0 )  H ( f 0 ) ( f  f 0 ),
 y(t)  H ( f 0 )e j2f0t | H ( f 0 ) | e j2f0tH ( f0 ) .

I.e., output is phasor @ f 0 Hz.

Magnitude | H ( f 0 ) | & phase H ( f 0 ).
 H ( f ) characterises response to input phasor @ f Hz

L3 6
 Transients?

In circuit theory & Laplace transforms: output when phasor

applied to LTI system contains a transient  0 as t  .

Implicit assumption: input first applied infinitely long ago,


y(t)   s( )h(t   )d .
!

 Transient has died out by time t.

Response when s applied from t  0 onwards,

y(t)   s( )h(t   )d .
0

L3 7
 Step Response Method.

How to get h without using IFT?

Often difficult to calculate response of system to  directly.

1 when t  0,
Instead, apply u(t) : 
0 when t  0.

then find output g - step response.

u(t   )
As   (t   ),
t

L3 8
 Step Response, cont.

u(t   )
 

h(t)   h( ) (t   )d 


 h( )

t
d ,

d d
  h( )u(t   )d  g(t).
dt  dt
I.e., find step response & differentiate.

L3 9
 FT of Step Function
Q: what is FT of u?
 
-j2ft -j2ft
 e u(t)dt   e dt  undefined.
 0

Decompose as u(t)  0.5  0.5sgn(t),

where sgn(t) : sign of t (=1).

 U ( f )  0.5 ( f )  0.5F{sgn(t)}.
So, need to find FT of signum function.

L3 10
 Finding FT of Signum Function (advanced)

Trick: for any  0,

0  0 

F e - |t| sgn(t)  
 e t - j2ft
e dt  0 e t - j2ft
e dt   e (  - j2f )t
dt  0 e (   j2f )t
dt,
1 1 - j4f
   ,   0, f  R.
  j2f   j2f  2  (2f ) 2

- j4f
 F{sgn(t)}  lim F e - |t| sgn(t)  1 .
 0 (2f ) 2 jf

1
 U ( f )  0.5 ( f )  .
j2f

L3 11
 Ideal Channel

Ideal or distortionless channel:

y(t)  k  s(t  t0 ).

Output = input, except for constant (nonzero!) gain and delay.

Taking FT & using the time-shift prop.,
Y ( f )  ke-j2 ft S( f )  H ( f )  ke-j2 ft .
0 0

I.e. ideal channel has constant (flat) magnitude response and

linear phase response (modulo 180 deg)
Impulse response: h(t)  k   (t  t0 ).

In reality, channels are never ideal...

L3 12
 Amplitude Distortion

Occurs when | H ( f ) | is not constant.

From Fourier theory, any input  sum (integral) of phasors at

different frequencies.
Phasor @ f Hz is multiplied by magnitude response | H ( f ) | of
system.

| H ( f ) | not constant  each Fourier component experiences

different gain  Signal distorted. Attenuation @ f Hz is

L3 13
 Phase Distortion

Recall that when input is e j2f0t , output is

| H ( f 0 ) | e j2f 0t H ( f 0 ) | H ( f 0 ) | e j2f 0 [t H ( f 0 ) /(2f 0 )] .
H ( f 0 )
I.e., output is proportional to input delayed by  sec
2f 0
=Phase delay.

If phase response is not linear, this depends on f 0 .

I.e. system delays Fourier components of input by different
amounts, depending on frequency

 Signal again distorted!

L3 14
 Examples of Amplitude Distortion

Ke -j2 f  if | f | B   K rect  f  -j2 f 

Let H(f )     2B  e .
 0 otherwise   

- ideal low-pass filter (LPF), bandwidth B Hz, delay  , gain K

1)Let s(t )  cos( 2 f 0 t ) .

 S( f )  0.5 ( f  f 0 )  0.5 ( f  f 0 ).

Output = 0 if | f 0 | W .
Q: What if | f 0 | W ?

L3 15
 Amplitude Distortion Eg, cont.

2) s(t)   (t)  output = impulse resp.,

h(t)  2BK sinc(2B(t   )).

So, output is distorted.

Notice h(t)  0 when t  0.

 Ideal LPF is non-causal, thus not physically realisable!

Uncertainty principle: if spectral content of H is limited in extent,
then h must extend over entire time axis, including negative
times.
Also notice that as bandwidth W  , h   , as expected.

L3 16
Ideal Lowpass Channel & Impulse Response
(Fig 3.4-2, Carlson et al – here td   )

L3 17
 Amplitude Distortion, cont.

3) Consider now an input pulse s(t)  rect(t/T -0.5) to the ILPF

 S( f )  e-j fT sinc( fT )

If B  1/T , output is a slightly rounded pulse (why?) delayed by  .

If B  1/T , output  2B sinc 2B(t    0.5T )  - nothing like input!

Reason: Y  HS  SH , so when
1/T=BW(S)  BW(H)  B, we can treat h(t) as the “input” to a
nearly ideal channel S( f )  e-j fT .
Causes inter-symbol interference (ISI) (more on this later)

L3 18
Another example: Input x(t)  cos  0 t 1/ 3cos3 0 t  1/ 5cos5 0 t
(Fig 3.2-3, Carlson et al)

L3 19
 Effects of Amplitude Distortion on x(t)
a) Low freq attenuated by ½ ( more wiggles)
b) High freq. attenuated by ½ ( smoother)
(Carlson et al, Fig 3.2-4)

L3 20
Phase-distorted x(t): constant phase shift of -90 deg
Nothing like x(t)! Don’t underestimate phase distortion... (Carlson et al,
Fig 3.2-5)

L3 21
 A Circuit Example in Freq. Domain

Low-pass RC filter:
1
H( f )  .
1  j2RCf
1 1
 | H ( f ) |  ,
| 1  j2RCf | 1  (2RCf ) 2

H ( f )  (1  j2RCf )   arctan(2RCf ).

3dB bandwidth = 1/(2RC) Hz

Significant phase distortion at high frequencies.

Q: What is H as RC  0?

L3 22
 Circuit Example in Time-Domain

From circuit theory, the step-response is just

What happens as RC  0?

L3 23
Step Responses of 1st order RC and Ideal LP Channels
(Fig 3.4-9, Carlson et al)

L3 24
 Response of Ideal LP Channel to Pulse of Duration 
(Fig 3.4-10, Carlson et al). Resemblance between output
and input improves as B increases

L3 25