Professional Documents
Culture Documents
reference sheet
by
Remco Litjens
(remco@aura.eecs.berkeley.edu)
Continuous-time Signals and Systems
General / Miscellaneous
!o = 2
1. To = 2fo angular or fundamental frequency
2. f (,t) = f (t) even function (e.g. cos t)
3. f (,t) = ,f (t) odd function (e.g. sin t)
= sinzz
4.
R
sinc z
1 sin ax
sinc-function
5. 1 bx dx = b area under a sinc-function
6. e,t (t) = e,0 (t) = (t)
p 2
7. ! = j! j
8. Hlpf (!) = 1 , Hhpf(!)
9. arctan( !0 ) = 2 sgn(!)
Trigonometry
1. sinx = cos(x , 12 )
2. sin2 x + cos2 x = 1
3. cos2x = 2cos2 x , 1 = 1 , 2sin2 x
4. sin2x = 2sinxcosx
5. cos( + ) = coscos , sinsin
6. sin( + ) = sincos + cossin
7. 2 cos cos = cos( + ) + cos( , )
8. ,2 sin sin = cos( + ) , cos( , )
d
9. d! sin! = cos!
d
10. d! cos! = ,sin!
d 1
11. d! arctan(!) = 1+!2
Complex Numbers
1.
p
j = ,1 imaginary constant
2. z = a + bj = Re[z] + jIm[z] complex number (rectangular or Cartesian form)
3. z = a , bj = Re[z] , jIm[z] complex conjugate of z
4. z = jzjej6 z
p complex number (polar form)
5. jzj = Re2 [z] + Im2 [z] magnitude of complex number z
6. 6 z = arctan( Im[z ] angle or phase of complex number z
Re[z] )
11.
R 1 (t)dt = 1
,1
12.
Rb
x(t)(t , a) = x(a)(t , a)
13.
Ra1x(xt)(s()t)(dtt ,=s)dsx(0)=; ifxa(t)< 0 < b; and 0; elsewhere
14.
15.
R,1
1 x(s)(n) (t , s)ds = (,1)n x(n) (t)
sifting or convolution property (integral)
,1
Convolution
R
1 h(t , )x( )d = 1 h( )x(t , )d
denition: h(t)
x(t) = ,1
R
,1
1. x1 (t)
x2 (t) = x2 (t)
x1 (t) commutativity property of convolution
2. x1 (t)
(x2 (t)
x3 (t)) = (x1 (t)
x2 (t))
x3 (t)) associativity property of convolution
3. x1 (t)
(x2 (t) + x3 (t)) = x1 (t)
x2 (t) + x1 (t)
x3 (t) distributivity property of convolution
General Systems
1. y(t) = H [x(t)] system H []
2. H [1 x1 (t) + 2 x2 (t)] = 1 H [x1(t)] + 2 H [x2 (t)] condition for linearity of H []
3. H [D [x(t)]] = D [H [x(t)]] condition for time-invariance of H []
4. y(to ) = H [x(t0)] depends only on x(t) for t = to condition for memoryless property of H []
5. y(to ) = H [x(t0)] depends only on x(t) for t to condition for causality of H []
Linear Systems
1.
2.
R
g(t; s) = H [(t , s)]
y(t) = ,1 1 g(t; )x( )d
3. check for causality: input (t , s) and check if for any s the output is zero 8 t > s:
4. H is time-invariant () g(t + ; s + ) = g(t;s)
if H [] is linear and memoryless, then H [] will be of the form H [] = (t)x(t):
BIBO-Stability
denition: system H [] is BIBO-stable i every bounded input results in a bounded output
Distortion
an LTI system is distortionless if the output y(t) = H [x(t)] = K x(t , ); a scaled and shifted replica of the input x(t) ()
() H (!) = K e,j! () jH (!)j is constant and 6 H (!) = ,! is linear in !:
an LTI system shows amplitude distortion when jH (!)j varies with !:
an LTI system shows phase distortion when 6 H (!) is non-linear in !; and then we dene g (!) = , d6 d! H (!) as the group delay.
Minimum Phase
A system H [] is said to be minimum phase, if all poles and zeros of H (s) lie in the open left half plane, so that H ,1 (s) is also minimum
phase; hence a causal system that is minimum phase, is also causally invertible.
Feedback
Given a feedback system block diagram, we can write the equations:
e(t) = x(t) y(t)
h2 (t) L! E (s) = X (s) Y (s)H2 (s) (error function);
y(t) = e(t)
h1 (t) L! Y (s) = E (s)H1 (s);
combine: y(t) = x(t)
h1 (t) y(t)
h2 (t)
h1 (t); L! Y (s) = X (s)H1 (s) Y (s)H1 (s)H2 (s) , XY ((ss)) = 1HH1 (1s()sH) 2 (s) :
in control problems, where we try to optimize feedback, we often use the nal value theorem for Laplace transforms;
Finding the Output of LTI Systems
1. y(t) = x(t)
h(t) straight convolution
2. , 1
y(t) = F [X (!)H (!)] using Fourier transforms
2. y(t) = L,1 [X (s)H (s)] using Laplace transforms
3. j! t
if x(t) = e ; then y(t) = H (!o)e
o j! o t using Fourier eigenfunctions
4. if x(t) = est; then y(t) = H (s)est using Laplace eigenfunctions
5. if x(t) = cos(!o t + ) and h(t) is real, then y(t) = jH (!o)jcos(!o t + + 6 H (!o))
steady-state sinusoidal response
6. if x(t) = sin(!o t + ) and h(t) is real, then y(t) = jH (!o)jsin(!o t + + 6 H (!o))
7. if x(t) =
P1n=,1 Xnejn!o t; then y(t) = P1n=,1 Xn H (n!steady-state
Psinusoidal
1
response
o)ejn!o t = n=,1 Xn jH (n!o)jej(n!o t + 6 H (n!o ))
using Fourier series
the transfer function and consequently the impulse response can be derived from any LDE by assuming zero initial conditions.
Remarks
in LTI systems it makes no dierence if we see h(t) as the input signal and x(t) as the impulse response
of the LTI system (follows directly from the commutativity property of convolution).
both h(t) and H (!) completely characterize an LTI system.
for an LTI system H [] : h(t) = 0 for t < 0 , H [] is causal.
a dierentiator (h(t) = _(t)) is causal.
an ideal dierentiator _(t) is a causal system.
H (!) real , h(,t) = h(t) ) H [] is non-causal (unless h(t) = 0 8t 6= 0) =) lters are non-causal
an LTI system converts a periodic input signal into a periodic output signal.
if H [] is LTI and memoryless, then h(t) = 0 for t 6= 0 and H [] will be of the form H [] = x(t):
Fourier Series
Denition of Fourier Series
We can decomposing any periodic power signal (with period To ) into discrete frequencies, each a multiple of !o ; yielding the Fourier series
representation of that signal:
x(t) = 1
P
1 R
n=,1 Xn e
Xn = To To x(t)e ,
jn!o t
jn!o t
synthesis equation
dt analysis equation
Some Examples
DC: x(t) = 1 FS! Xo = 1; Xn = 0 otherwise
cosine: cos !o t = 12 (ej!o t + e,j!o t )
sine: sin !o t = 21j (ej!o t , e,j!o t )
even square wave: period To ; x1 (t) = 1 for jtj < T1 and x1 (t) = 0 elsewhere in that period:
x1 (t) FS! Xn = n1 sin n!o T1
odd square wave: x2 (t) is shifted x1 (t)
even triangle wave: x3 (t) is running time integral of x1 (t)
x3 (t) FS! Xn = , n242 ; for n odd, and Xn = 0; for n even
pulse train: x4 (t) is same as x1 (t) but shifted by and with added DC (this changes X0 only):
x4 (t) = 1
P t,kTo , FS to ,jn!o sinc( nto )
even impulse train:
P k=,1 ( to ) ! Xn = To e
1 FS
x5 (t) = k=,1 (at , kT ) ! Xn = ja1jT
To
Fourier Transform
Denition and Interpretation
The Fourier transform X (!) is a decomposition of an (a)periodic signal x(t) into ej!t -type signals with ! real (x(t) and X (!) uniquely
dene each other); we write x(t) F! X (!) :
R
1 X (!)ej!t d!
x(t) = F ,1 [X (!)] = 21 ,1 synthesis equation
X (!) = F [x(t)] = ,1
R
1 x(t)e,j!t dt analysis equation
The Fourier transform X (!) exists when it's nite for all !: Allowing -functions, we can accept a broader class of signals: sucient conditions
for the existence of the generalized Fourier transform of x(t) are (1) x(t) is bounded; and (2) x(t) is periodic (E = 1) with nite power.
The interpretation of the Fourier transform is that X (!) contains the frequency content of the signal x(t) for each !: For example, consider
x(t) = 1 F! (!) (DC only), x(t) = ej!o t F! 2(! , !o ) (single frequency at !o ), and nally x(t) = cos !o t F! [(! + !o ) + (! , !o )]
(contains two frequencies: !o and ,!o ). Note: if the signal x(t) is time-limited, X (!) has innite bandwidth.
Note that the ULT only works for (real world) "causal" signals while the BLT permits us to consider any signal; furthermore, if x(t) = 0 for t < 0
(causal) then the ULT and the BLT are the same. Both the ULT and the BLT have regions of convergence (ROC) over which the transform
exists. A ROC is the range of values of s over which the integral converges. A ROC cannot contain a pole, in fact for causal signals/systems, the
ROC contains all s to the right of the rightmost pole. If x(t) is of nite duration, ROCx = CI :
Any signal x(t) that is of exponential order, i.e. jx(t)j does not increase faster than A ect with A and c real constants, can be transformed into
X (s): Examples of signals that are not of exponential order, consider x(t) = et2 u(t) and x(t) = ttu(t) = et ln t u(t):
Some Examples
h(t) L! H (s) transfer function
0 L! 0 _ (t) L! s all s
(t) L! 1 all s (t , ) L! e,s all s
u(t) L! 1s Re[s] > 0 ,u(,t) L! 1s Re[s] < 0
r(t) L! s12 Re[s] > 0 ,r(,t) L! s1 Re[s] < 0
(tnn,,1)!
1
u(t) L! s1n Re[s] > 0 , (tnn,,1)!
1
u(,t) L! s1n Re[s] < 0
e,t u(t) L! s+1 Re[s] > ,Re[] ,e,t u(,t) L! s+1 Re[s] < ,Re[]
(tnn,,1)! e u(t) L! (s+1)n
1 ,t
Re[s] > ,Re[] , (tnn,,1)! e u(,t) L! (s+1)n
1 ,t
Re[s] < ,Re[]
(sin !o t)u(t) L! s2!+o!o2 Re[s] > 0 (e,t sin !o t)u(t) L! (s+!)o2 +!o2 Re[s] > ,Re[]
(cos !o t)u(t) L! s2 +s!o2 Re[s] > 0 (e,t cos !o t)u(t) L! (s+s+)2+!o2 Re[s] > ,Re[]
Linear Dierential Equations
Solving LDE w/CC using Fourier transforms
One can solve LDE w/CC using Fourier transforms or Laplace transforms. Below we show how to do it with Laplace transform. For Fourier
transforms, the procedure is identical: nd h(t) = F ,1 [H (!)] with H (!) = XY ((!!)) and X (!) = F [(t)] = 1: Y (!) follows from using the
dierentiation property for Fourier transforms and some algebraic manipulations.
Example: if y(t) , y_ (t) , 30y(t) = x(t); then for x(t) = (t) we nd (j!)2 H (!) , (j!)H (!) , 30H (!) = 1 () H (!) = (j!)2 ,1(j!),30 ; so
h(t) = F ,1 [ (j!+5)(1 j!,6) ] = F ,1 [ ,j!1=+511 ] + F ,1 [ j!
1=11 ] = , 1 e,5t u(t) + 1 e6t u(t): Note that this system is BIBO unstable.
,6 11 11
In discrete-time, high frequencies lie around odd multiples of ; while low frequencies lie around even multiples of : Consequently,
an ideal low-pass lter with cut-o frequency
1 < has impulse response hlpf [n] =
1 sinc n
1 and a pulse train as frequency response.
The frequency response of a high-pass, bandpass or notch lter is simply a shifted and possibly duplicated version of the frequency
response of a low-pass lter.
Singularity Signals
delta function
P P
1. [n] = u[n] , u[n , 1] = 1; n = 0; and 0 elsewhere
2. u[n] = nk=,1 [k] = 1 unit step function
P
1
3. x[n] = k=,1 x[k] [n]
k=0 [n , k] = 1 n 0; and 0 elsewhere
sifting property (sum)
Convolution
denition: h[n]
x[n] = P1k=,1 x[k] h[n , k] = P1k=,1 x[n , k] h[k] often very easy to do!
example: u[n]
u[n] = P1k=,1 u[k] u[n , k] = Pnk=0 1 = (n + 1)u[n]
BIBO-Stability
denition: system H [] is BIBO-stable i every bounded input results in a bounded output
x[n] = F ,1 [X (
)] =
R
1
o +2
P 2
o X (
)ejn
d
synthesis equation
X (
) = F [x(t)] = 1 ,jn
analysis equation
n=,1 x[n]e
oo +2 X1 ()
X2 (
, )d periodic convolution
n-multiplication: n x[n] F! j dd
X (
)
upsampling: x(k) [n] F! X (k
) x(k) [n] = x[ nk ]; if n is a multiple of k
PARSEVAL's identity:
Pn=,1 jx[n]j2 = 21 R
o+2 jX (
)j2d
E = 1
and 0 otherwise
o
Some Examples
h[n] F! H (
) frequency response
sin
o n F! j
P1 ,1 [(
,
o , 2k) , (
+
o , 2k)]
cos
o n F!
P1kk==,1 [(
,
o , 2k) + (
+
o , 2k)]
0 F! 0 1 F! 2 1
Pk=,1 (
, 2k)
[n] F! 1 F
[n , no ] ! e,jno
u[n] F! 1,1ej
+
P1 ej
o n F! 2 1
P
1 sinc(
1 n ) F!
P1k=,1((
,2
,2k2k) ) k=,infty (
,
o , 2k)
k=,1 1
Z-Transform or DT Laplace Transform
Denition of Z-Transform:
The Z-transform X (z) is a decomposition of an (a)periodic signal x(t) into signals of the type z,n = r,n e,jn
with z = rej
complex.
Compared to the discrete-time Fourier transform, the extra (real) exponential r,n allows us to treat a broader class of signals. The r,n part
is used to model growth or decay, while the e,jn! part models the oscillatory behavior of x(t): Again, x[n] and X (z) uniquely dene each
other and here we write x[n] Z! X (z) :
P1
unilateral ZT: X (z) = Z [x[n]] =
( bilateral ZT: X (z) = Z [x[n]] =
P1kk==0,1x[nx][zn,]zn,n )
Note that the UZT only works for (real world) "causal" signals while the BZT permits us to consider any signal; furthermore, if x[n] = 0 for
n < 0 (causal) then the UZT and the BZT are the same. Both the UZT and the BZT have regions of convergence (ROC) over which the
transform exists. A ROC is the range of values of z over which the integral converges. A ROC cannot contain a pole, in fact for causal signals/
systems, the ROC contains all z on the outside of the circle that crosses the outermost pole. If x[n] is of nite duration, ROCx = CI ; except
possibly for z = 0 and z = 1:
Any signal x[n] that is of exponential order, i.e. jx[n]j does not increase faster than A cn with A and c real constants, can be transformed
into X (z): As an example of a signal that is not of exponential order, consider for instance x[n] = 2n2 u[n]
Some Examples
h[n] Z! H (z) transfer function
0 Z! 0
[n] Z! 1 all z [n , M ] Z! z,M all s except 0 (1) if m > 0 (m < 0)
u[n] Z! 1,z1,1 jzj > 1 u[,n , 1] Z! 1,z1,1 jzj < 1
n u[n] Z! 1,z1 ,1 jzj > jj ,n u[,n , 1] Z! 1,z1 ,1 jzj < jj
nn u[n] Z! (1,zz,,11 )2 jzj > jj ,nn u[,n , 1] Z! (1,zz,,11 )2 jzj < jj
[cos
o n]u[n] Z! 1,[21,cos[cos
o]z,1 jzj > 1 [sin
o]z,,11 ,2
[sin
o n]u[n] Z! 1,[2cos jz j > 1
o ]z,1 +z,2,1
o]z +z ,1
[rn cos
o n]u[n] Z! 1,[21rcos
,[rcos
o]z
o]z,1 +r2 z,2 jzj > jrj [rn sin
o n]u[n] Z! 1,[2rcos
[rsin
o,]z1 2 ,2
o]z +r z jzj > jrj
Linear Dierence Equations
Solving LDE w/CC using Z-transforms
We show how the unilateral Z-transform can be used to solve linear dierential equations with constant coecients. In particular, the delay
property of the unilateral Z-transform is utilized. The procedures are not shown for the general case, but rather for an example.
Example: Assume we are given the following LDE w/CC that we are trying to solve:
y[n] , 2y[n , 1] = x[n] + x[n , 1]
with initial conditions y[,1] = 1 and x[,1] = 0:
First, take the unilateral Z-transform of both sides, using the delay property:
Y (z) , [2z,1 Y (z) + 2y[,1]] = X (z) + [z,1 X (z) + x[,1]] ()
,1
() Y (z) = (1 , z,11+)(1z , 2z,1 ) + x[,11],+2z2,y[1,1] ;
wherein inverse Z-transform of the rst part equals the zero-state-response:
,1 z,1
yzsr [n] = Z [ (1 , z,11+)(1z , 2z,1 ) ] = Z [ (1 , z,1 )(1
1 1
, 2z,1 ) ] + Z [ (1 , z,1 )(1 , 2z,1 ) ] = ([n] + [n , 1])
Z [ (1 , z,1 )(1 , 2z,1 ) ] =
= ([n] + [n , 1])
Z [ 1 ,,z1,1 + 1 , 22z,1 ] = ([n] + [n , 1])
(,u[n] + 2 2n u[n]) = ,u[n] , u[n , 1] + 2n+1 u[n] + 2n u[n , 1]:
Secondly, we have the zero-input-response:
yzir [n] = Z [ (1 , 22z,1 ) ] = 2 2n u[n] = 2n+1 u[n]:
The actual output y[n] is the sum of the two:
y[n] = yzir [n] + yzir [n] = ,u[n] , u[n , 1] + 2n+1 u[n] + 2nu[n , 1] + 2n+1 u[n] = ,u[n] , u[n , 1] + 2n+2 u[n] + 2n u[n , 1]:
X (!) = F [x(t)] = F [
1 X
Xn ejn!o t ] =
1 X
Xn F [ejn!o t ] = 2
1 X
Xn (! , n!o ):
n=,1 n=,1 n=,1
How to get from FT to FS (if the signal is periodic)? If a periodic signal x(t) is written as a Fourier Series,
Xn = T1 X (!) j !=n!o :
o
THIS SECTION IS FINISHED NOR CLEAR TO MYSELF; APOLOGIES.
WB-FM is much more noise immune than AM, since noise mainly eects the amplitude
of a signal and not so much the (instantaneous) frequency; furthermore, the amount of
used bandwidth is much larger for WB-FM, leading to better noise immunity.
modulation: H [m(t)] = xu (t) = m(t) cos !c t , m^ (t) sin !c t F! Xu (!) = M (! + !c ) u(,! , !c ) + M (! , !c ) u(! , !c );
demodulation: y(t) = xu (t) cos !c t = = 21 m(t) + 12 m(t) cos 2!ct , 12 m^ (t) sin 2!ct ,! LPF ,! z(t) = 12 m(t);
modulation: H [m(t)] = xl (t) = m(t) cos !c t + m^ (t) sin !c t F! Xl (!) = M (! + !c ) u(! + !c ) + M (! , !c ) u(,! + !c );
demodulation: y(t) = xl (t) cos !c t = = 12 m(t) + 21 m(t) cos 2!c t + 12 m^ (t) sin 2!c t ,! LPF ,! z(t) = 21 m(t);
Remarks on SSB-AM
Hilbert transformer is a system with hHT (t) = t1 and HHT (!) = ,j sgn(!) = e,j 2 sgn(!) (phase shift); x^(t) = x(t)
hHT ;
power ecient: no carrier transmitted;
bandwidth ecient: now the modulated signal x(t) takes up the same amount of bandwidth as the message signal: BX = BM :
you can demodulate SSB-AM using imperfect carrier cos(!c t + ) : received signal is 12 cos() m(t) : ne as long as cos() 6= 0 (see notes 15);
you can demodulate SSB-AM using imperfect carrier cos([!c + !]t) (see notes 15);
SSB-AM: power-ecient, no waste of bandwidth, requires synchronized reception.
B: FM:
(t) = m(t) is the instantaneous phase;
R
x(t) = A cos[!c t + (t)]; with (t) = ! ,1t m(s)ds; with ! the frequency deviation constant;
!(t) = dtd [!c t + (t)] = !c + ! m(t is the instantaneous frequency;
we speak of NARROWBAND angle modulation when j(t)j 1; hence ej(t) 1 + j(t); so x(t) cos !c t , (t) sin !c t;
note that in narrowband angle modulation a LARGE CARRIER cos !c t is transmitted (inecient!)
modulation: = m(t); so H [m(t)] = x(t) = cos !c t , m(t) sin !c t F! X (!) = [(! , !c ) + (! + !c )]+
+ 2j [M (! , !c ) , M (! + !c )] (similar to DSB-AM-LC);
NARROWBAND FREQUENCY R t
MODULATION (NB-FM)
modulation: = ! ,1 m(s)ds; with j(t)j 1; hence m(t) cannot have a DC component, hence M (!)j!=0 = 0;
H [m(t)] = x(t) = cos !c t , (t) sin !c t F! X (!) = [(! , !c ) + (! + !c)] + 2j [(! , !c ) , (! + !c)];
with (!) = ! Mj!(!) ; so X (!) = [(! , !c ) + (! + !c )] + !2 [ M!(!,,!!c c ) , M!(!++!!c c ) ];
Examples of Systems and their Properties
time BIBO
system linear invariant causal memoryless stable
, @@
,
, @ and y(t) = H (!)ej!t ;
solve for H (!)
h(t) H (!)
Fourier transform
H (!) = jH (!)jej6 H (!)
jH (!)j
6 H (!)
,@
LDE
substitute x(tL) L! X (s)
, @
,, @@ and y(t) ! Y (s);Y (s)
solve for H (s) = X (s)
h(t)
Laplace transform
H (s)
write H (s) = N (s)
D(s) and
solve N (s) = 0 ,! Zeroes
solve D(s) = 0 ,! Poles
Poles
Zeroes