I-1: Review of DT Signal & System Concepts

1/30

Notational Conventions (P-1.2)

Sets of Numbers: R = real numbers C = complex numbers Z = integers ( , -3, -2, -1, 0, 1, 2, 3, ) Proakis & Manolakis dont follow this

Signals: x(t), y(t) = continuous-time (CT) signals x[n], y[n] = discrete-time (DT) signals

DT Convolution:

{x * y}[n ] =

m =

x[ m ] y [ n m ]
N 1 m =0

Proakis & Manolakis dont follow this

0 n N 1

DT Circular Convolution:

{x y}[n ] =

x[m] y[(n m) mod N ]

Recall: We dont want Circular convolution it is what we get if we try to implement Convolution in the frequency domain by multiplying DFTs (but not DTFTs) But we can fix it so that it works. We use proper zero-padding!!!
2/30

Transforms (P-2)
Fourier Transform for CT Signals
X F ( ) = {Fx}( )

Proakis & Manolakis dont follow this Notation

Fourier Transform for DT Signals
X f ( ) = {Fx}( )

x (t )e

jt

dt

n =

x[n]e jn

Discrete Time Fourier Transform

1 x (t ) = 2

F jt X ( )e d

1 x[ n ] = 2

( )e jn d

Set z = ej
Z Transform for DT Signals
X z ( z ) = {Zx}( z ) =
n =

Discrete Fourier Transform for DT Signals

X [k ] = {Dx} k ] [
d

Discrete Fourier Transform

k = 0,1,2,..., N 1

x[n]z n

N 1 n =0

x[n]e j 2kn / N
N 1 k =0

Inverse ZT done using partial fractions & a ZT table

1 x[ n ] = N

X d [k ]e j 2kn / N

n = 0,1,2,..., N 1
3/30

Fourier Transforms for DT Signals (P-2.7,P-4, PM-4, PM-7)

DTFT = In Head/On Paper tool for analysis & design DFT = In Computer tool for Implementation/Analysis&Design (In both cases you generally strive to make the DFT behave like DTFT) DFT for Implementation: Compute DFT of signal to be processed and manipulate DFT values May or may not need zero-padding (depends on application) DFT for Analysis & Design: Compute DFT of given filters TF to see its frequency response Generally use zero-padding to get fine grid to get approximate DTFT DFT/DTFT Properties you Must Know: Periodicity. (DFT&DTFT both periodic in FD) . (IDFT gives periodic TD signal) Time Shift.. (Time Shift in TD Linear Phase Shift in FD) Frequency Shift. (Mult. by complex sine in TD Freq Shift in FD) Convolution Theorem (Conv. in Time Domain Mult. in Freq Domain) Multiplication Theorem.. (Mult. in Time Domain Conv. in Freq Domain) Parsevals Theorem. (Sum of Squares in TD = Sum of Squares in FD)
4/30

Summation Rules (P-1.3)

Make sure you are familiar with rules for dealing with summations, as shown in Section 1.3 of Porat. In addition, a particular summation formula that shows up often is the Geometric Summation:
N1 N 2 , if 1 1 N 2 1 n = N N , n = N1 if = 1 1 2

Special Case: N1 = 0
1 N 2 , if 1 N 2 1 1 n = N , n =0 if = 1 2

5/30

Eulers Formulas
e j + e j cos( ) = 2 e j e j sin( ) = 2j
Im

e j = cos( ) + j sin( ) e j = cos( ) j sin( )

Im

j sin( )

j sin( )

cos( )

Re

2jsin()

e j

e j

2 cos( )

Re

e j

e j
6/30

Discrete-Time Processing (P-2, PM-5)

x[n] Xf() Xz(z) h[n] Hf() Hz(z) y[n] = h[n]*x[n] Yf() = Hf() Xf() Yz(z) = Hz(z) Xz(z)

h[n] = systems Impulse Response It is response of system to input of [n] Hf() = systems Frequency Response It is DTFT of impulse response Hz(z) = systems Transfer Function It is ZT of impulse response

7/30

Discrete-Time System Relationships

Time Domain
Block Diagram Inspect ZT (Theory) Inspect (Practice)
q

Z / Freq Domain
Pole/Zero Diagram Roots
H ( z) =
z

Inspect Difference Equation

y[k ] = ai y[n i ] + bi x[n i ]
i =1 i =0 p

Transfer Function

b0 + b1 z 1 + 1 + a1 z 1 +

+ bq z q + a pz p

ZT DTFT

Unit Circle z = ej Frequency Response

Impulse Response h[k]

H f ( ) = H z ( z ) |z = e j
8/30

Poles and Zeros of Transfer Function

Consider Here Only p>q
H ( z) = = = =
z

b0 + b1 z 1 + 1 + a1 z 1 +

+ bq z q + a pz p + bq z q ) + bq ) (z q ) + a p z p )

Im{z}

z p q z q (b0 + b1 z 1 + z p (1 + a1 z 1 + z p + a1 z p 1 + ( z 1 )( z 2 ) z p q (b0 z q + b1 z q 1 + z p q ( z 1 )( z 2 )

Re{z}

+ ap (z p )

Numerator Roots define zeros Denominator Roots define poles

Poles must be inside unit circle for stability Poles & zeros must occur in Complex Conjugate pairs to give real-valued TF coefficients 9/30

In this case pq zeros at z=0

IIR and FIR Filters

H z ( z) = b0 + b1 z 1 + 1 + a1 z 1 + + bq z q + a pz p

FIR: Finite Impulse Response h[n] has only finite many nonzero values Filter is FIR: If ai = 0 for i = 1, 2, , p (It is IIR otherwise)

H z ( z ) = b0 + b1 z 1 +
bn , n = 0,1,..., q h[n ] = 0, otherwise

+ bq z q
TF coefficients are the impulse response values!!

10/30

Example System
Time Domain
Difference Equation ZT (Theory) Inspect (Practice)

Z / Freq Domain
Transfer Function

y (n) y (n 1) = x(n) y (n) = x(n) + y (n 1)

Recursion Form

Input-Output Form

Y (z) 1 z = X (z)
1

Y ( z) H ( z) = = 1 X ( z) 1 z
11/30

Example System (cont.)

Time Domain Z / Freq Domain
x(n)

y(n-1)

y(n)

z-1

Inspect

Block Diagram Inspect ZT (Theory) Inspect (Practice) Transfer Function

Difference Equation Recursion Form

y (n) = x(n) + y (n 1)

H ( z) = 1 z 1
12/30

Example System (cont.)

Z / Freq Domain

z = e j z 1 = e j
On Unit Circle

H ( z) = 1 z 1
z 1 = e j

Transfer Function Unit Circle Frequency Response

H ( ) = 1 e j ( , ]

13/30

Example System (cont.)

Plotting Transfer Function
= 1 e j 1 [ cos() j sin()] = [1 cos()] + j sin()
H ( ) = H () =
Eulers Equation

Group into Real & Imag

[1 cos()] + [ sin()]
2

sin() H () = tan 1 1 cos()

Standard Eqs for Mag. & Angle

14/30

Example System (cont.)

Unit Circle

Z/Freq Domain
Pole/Zero Diagram

Im{z}

Re{z} Roots Transfer Function

z H ( z) = = 1 1 z z

Zero Num. = 0 When z = 0 Pole Den. = 0 When z =

If || < 1: Inside UC If || > 1: Outside UC

15/30

Example System (cont.)

Time Domain
Inv. ZT Table (See Porat)

Z / Freq Domain

H ( z) = 1 z 1
Transfer Function ZT Unit Circle

h(n) =

Impulse Response

DTFT

Frequency Response

Inv. DTFT h( n) = d 1 e j

H ( ) = 1 e j ( , ]
16/30

Review of Standard Sampling Theory

17/30

Practical Sampling Set-Up ADC x(t)

Hold Sample at t = nT T = Sampling Interval Fs = 1/T = Sampling Rate
2 Signal Value 1 0 -1 -2 0 0.2 0.4 Tim e 2 Signal Value 1 0 -1 -2 0 0.2 0.4 Tim e 0.6 0. 8 0.6 0. 8

DAC
CT LPF
p x(nT ) ~(t nT )

x[n] = x(nT)

Pulse Gen
~ (t ) = xp

~ (t ) xr

n =

18/30

Sampling Analysis (#1) Goal = Determine Under What Conditions We Have: x Reconstructed CT Signal = Original CT Signal ~(t ) = x (t ) Simplify to Develop Theory: Use ~(t ) = (t ) p Why???? 1. Because delta functions are EASY to analyze!!! 2. Because it leads to the best possible case

~ (t ) xp

x p (t ) =

n =

x(nT ) (t nT )

xp(t) is called the Impulse Sampled signal Note: xp(t) shows up during Reconstruction, not Sampling!!
19/30

Sampling Analysis (#2) Now, the impulse sampled signal xp(t) is filtered to give the reconstructed signal xr(t)
x p (t ) =
n =

x(nT ) (t nT )
n =

x r (t ) = ???
CT LPF

= x (t )

(t nT )

H( f )

Xr ( f ) = H( f )X p( f )

Need to find Xp( f ). But First a Trick!!!

x p (t ) = x (t )

n =

(t nT )
e j 2kt / T
s

periodic use FS

1 X p( f ) = T 1 = T

1 = x (t ) T 1 = T

k = k =

F {x(t )e

j 2kFs t

k =

X ( f kFs )
20/30

k =

x(t )e j 2kF t

Sampling Analysis (#3) x(t) ADC

Hold Sample at t = nT X( f ) B B Xp( f ) 2Fs Fs H( f ) Fs 2Fs f

DAC x[n] = x(nT)

Impulse Gen
x p (t )

CT LPF

x r (t )

Xr( f )

Xr( f ) = X( f ) If Fs 2B

f
21/30

Sampling Analysis (#4) What this says: Samples of a bandlimited signal completely define it as long as they are taken at Fs 2B Impact: To extract the info from a bandlimited signal we only need to operate on its (properly taken) samples Use computer to process signals

x(t)
Hold Sample at t = nT

x[n] = x(nT)

Computer

Extracted Information

22/30

Sampling Analysis (#5) FT of Impulse Sampled Signal gives view of Original FT & FT of Impulse Sampled Signal = DTFT of Samples DTFT gives view of FT of original signal
Lets see this

F {x p (t )} = F x ( nT ) (t nT ) = x ( nT ) F { (t nT )} n = n = e jnT

n =

x ( nT )e jnT =

n =

x[n ]e jn

= T

F X p ( / T ) = X f ( )

23/30

Sampling Analysis (#6)

X( f ) B B DTFT 2Fs 4 4 Fs Fs/2 2 2 Fs/2 Fs 2 2Fs 4 f f

= T
2 = f Fs

DT Freq (rad/sample) / Normalized DT Freq

Read notes on web on Concept of Digital Frequency

24/30

Introduction to EE521 Case Study

25/30

EE521 Case Study: Emitter Location

Processing Tasks
Intercept RF Signal @ Rxs

Sample Signal Suitably for Processing Detect Presence of Emitters Signal Estimate Characteristics of Signal Use Estd Chars to Classify Emitter Share Data Between Rxs Cross-Correlate Signals to Locate Tx

Radio Freq Transmitter (Tx) Communications Radar

Data Link Data Link

Receiver #3 (Rx3) Receiver #2 (Rx2)

Receiver #1 (Rx1)

26/30

Rx1
Antenna

Overall Processing Set-Up

Detect Signal Digital Front-End Estimate Signal Parameters Classify Emitter Compress & Data Link

Analog Digital RF Front-End Sampling Sub-System

Rx2
Antenna Analog Digital RF Front-End Sampling Sub-System Digital Front-End Estimate Signal Parameters

RF Data Link

Data Link & Decompress Cross Correlation

Detect Signal

Classify Emitter
27/30

Processing Tasks Overview (#1)

RF Front-End (Analog Circuitry): RF = Radio Frequency Selects reception band; may need to scan bands Amplifies signal to level suitable for ADC, etc. Frequency-shifts signal spectrum to range suitable for ADC Technology trend is toward requiring less shifting Unfortunately: noise introduced by analog electronics (Random Signals)
Topic Well Cover

FT of Signal Before RF-FE

1 GHz 3 GHz 4 GHz

Antenna

RF Front-End

FT of Signal After RF-FE

200 MHz 28/30

Processing Tasks Overview (#2)

Sampling Sub-System (Mixed Analog/Digital Circuitry): Converts analog CT signal into digital DT signal (Bandpass Sampling) Converts real-valued RF signal into complex LP signal (Equiv. LP Signals) Digital Front-End (Digital Processing): Topics Convert to complex LP signal (Equiv. LP Signals) Well Equalizes response of analog front-end (DFT-Based Filtering) Cover Magnitude and Phase Bank of digital filters splits signal into sub-bands for further processing (Filter Banks) Reduce sampling rate after filtering to sub-bands (Multi-Rate Processing)
Signals are Oversampled

DTFT of Signal Before Digital FE

Digital Front-End
rad/sample

29/30

Processing Tasks Overview (#3)

Detect Presence of Signal (Digital Processing): Has signal been intercepted or only noise in subband (DFT-Based Proc.) Estimate Parameters of Signal (Digital Processing) (see also EE522): Estimate frequency of signal (DFT-Based Processing) Classify/Model Signal (Digital Processing): What type of signal is it? (Spectral Analysis of Random Signals)

PSD Model

Px ( ) =

2 b0

+ b1 z 1 + b2 z 2 +

1 + a1 z 1 + a 2 z 2 +

+ bp z p q + a q z j z =e

Use PSD Model parameters {bi} and {ai} to classify signal type Compress/De-Compress Signal (Digital Processing) (see also EE523): Exploit signal structure to allow efficient transfer (DFT-Based Proc./ Filter Banks / Spectral Analysis) Cross-Correlate Signals (Digital Processing): Compute relative delay and Doppler (DFT-Based Proc.& Multi-Rate Proc.)
30/30