Bandlimited communication systems

Lecture 3 Vladimir Stojanovi

6.973 Communication System Design Spring 2006 Massachusetts Institute of Technology

Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

Passband channel example

Two-ray wireless channel (multi-path 1+0.9D)

Multi-path creates notching in frequency domain Just slide the frequency window to bb
Add single-sided noise
Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

Bandlimited channel example

Attenuation [dB]

pulse response

0 -10 -20 -30 -40 -50 -60 0 2 4 6 8 10 frequency [GHz]

1 0.8 0.6 0.4 0.2 0 0 1 2 3 ns

Tsymbol=160ps

Low-pass channel causes pulse attenuation and dispersion Notches cause ripples in time domain Makes it hard to transmit successive messages
Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

Inter-Symbol Interference (ISI)

1

Error!
0.8 Amplitude 0.6

0.4

0.2

0 0 2 4 6 8 10 12 Symbol time 14 16 18

Middle sample is corrupted by 0.2 trailing ISI (from the previous symbol), and 0.1 leading ISI (from the next symbol) resulting in 0.3 total ISI As a result middle symbol is detected in error
Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

Bandlimited communication systems

Block detector vs. symbol-by-symbol
1(KT - t) Block Detector

yp(t)

2(KT - t)

. . .

k(KT - t) t = KT yp(t) p(T - t)

SBS detector

Xk k = 0,...,K -1 yp(t)

t = kT, k = 0,...,K - 1

Block of K symbols MK messages

Figure by MIT OpenCourseWare.

MAP/ML detector complexity grows exponentially Simpler detector is Symbol-By-Symbol

Optimal for AWGN channel MK basis functions (branches in the matched filter) Sequence detection can bound that growth

Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

Symbol-by-symbol detection
x (t) xk
Input symbol at time k

n (t) h(t)
+

yk Matched filter Sampler R

Receiver

zk

mod

SBS detector

xk
Estimate of input symbol at time k

Bandlimited channel

Figure by MIT OpenCourseWare.

Suffers significantly from Intersymbol-interference (channel memory), so need to remove ISI to get almost AWGN channel Need to adapt basis functions to the particular channel, to avoid ISI Alternatively, use equalization to remove ISI
Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

Vector channel - revisited

np (t) xp (t) x(t) h(t) + np (t) xkn
n(t)

yp (t)

xn (t)

xp (t) h(t) + np (t) xp (t) yp (t)

xkn

pn(t)

yp (t)
Figure by MIT OpenCourseWare.

Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

ISI impact
xk pulse response p(t) xp,k xp(t)
||p|| jp(t)

np(t) S yp(t)
jp (-t)

y(t)

discrete time sample at receiver times kT

yk

xk

q(t) = jp(t)*jp(-t)
Figure by MIT OpenCourseWare.

Mean-distortion

Treat ISI as noise

Peak-distortion

Treat worst-case ISI as constellation offset

Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

Matched Filter Bound

You cant do better with successive transmissions than with one-shot Matched filter collects the pulse energy ||p||2 Then calculate performance as on AWGN

Example binary transmission

Will use MFB to compare different ISI compensation techniques

Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

Nyquist criterion 6.011 revisited

A channel specified by pulse response p(t) is ISI free if

Nyquist frequency:

w=pi/T or f=1/2T
10

Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

Raised-cosine pulses
Can have excess bandwidth as long as there is symmetry that fills the aliased spectrum flat

Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

11

Basic equalization concepts

Zero-forcing equalization
xk
Q(D)

yk

W(D)

xk

Flattens equalized channel transfer function

X Equalizer W(w)

Channel Q(w)

Equalized

H(D)=Q(D)*W(D)
=>

Wzfe(D)=1/(Q(D)||p||)

Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

12

Linear equalization
Zero-forcing not good on channels with nulls
Equalizer enhances noise

Remember, Pe depends on both noise and ISI Balance noise and ISI in the mean-square sense Minimizing MMSE wrt. Wk

Same as using the orthogonality principle

Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

13

ZFE vs. MMSE - LE

Q(e-jT ) WZFE (e-jT ) SNR ||p|| WMMSE-LE (e-jT ) 0
T

Figure by MIT OpenCourseWare.

Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

14

Example: ZFE vs. MMSE LE

1+0.9D channel
Equalizer response

zfe mmse

Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

15

Fractional equalizers
np(t) xk p xp,k p(t) xp(t) + yp(t)
anti-alias filter

y(t)
sample at times k(T/l)

yk

Fractionally-spaced equalizer

Wk

zk

gain T

Figure by MIT OpenCourseWare.

Oversampling in the receiver

Can merge matched filter and equalizer
Can reconstruct original signal from oversampled signal (as long as original is band-limited)

Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

16

ISI channel model

Oversampled channel representation (3x e.g)

Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

17

Finite length equalizer formulation

xk p yk w zk

Write convolution as multiply with Toeplitz matrix

zk = wPxk

Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

18

zk = xk = wPxk 1 = wzfe P wzfe = 1T PT ( PPT ) ,

1

Zero forcing equalizer (ZFE)

ZFE and MMSE solution

1 = [0 0...1...0 0]T

Minimum-mean square error (MMSE) equalizer

=>

Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

19

Decision feedback equalizer

xk p yk w zk
xk-
Amplitude

1 0.8 0.6 0.4 0.2 0 0 2 4 6 8 10 12 14 16 Symbol time 18

Feedback equalization

b Feed-forward equalizer Feed-back equalizer

Matched + whitening filter + remove pre-cursor ISI

Removes trailing ISI To get w, first puncture the channel matrix to emulate the effect of feedback on the equalized pulse response wP Then, get b from the causal taps of equalized pulse response wP

Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

20

MMSE DFE
Selects the feedback taps

Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

21

Basic multitone modulation

Bits/chan Gain Bits/chan

=
Frequency Frequency Frequency Bits/chan Gain RF Bits/chan

xtalk Frequency Frequency Frequency

Figure by MIT OpenCourseWare.

Best performance if basis functions are tailored to the channel

Use each tone as a basis function Each tone transmits narrow QAM signal and satisfies Nyquist criterion i.e. no ISI per tone Put less energy where channel is bad or where there is more noise
Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

22

A bit of history
1948 Shannon constructs capacity bounds
AWGN channel with linear ISI effectively uses multi-tone modulation

Analog multi-tone
1958 Collins Kineplex modem (first voiceband modem) analog multitone 1964 Holsingers MIT thesis modem that approximates Shannons water-filling 1967 Saltzberg, 1973 Bell Labs, 1980 IBM

Digital multi-tone ~ 1990s

DMT for DSL - Major push by prof. Cioffis group at Stanford Use DSP power to improve the robustness and algorithms for discrete multi-tone modulation We will mostly focus on this type of modulation
Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

23

Basic multitone transmission

X0 0(t) e j2f1t X1 E N bits C O D X N-1 E R XN 1(t) . . . N-1(t) X + h(t) n(t) +
p h a s e s p l i t

0(-t) e-j2f1t X 1(-t) Yn = Hn . Xn+ Nn e-j2fN-1t X e-j2fN t p,n(t) = n(t) * h(t) bn = X N (-t)

Y0

Y1

e j2fN-1t X e j2fN t

N = 2N

D E C bits T=kNT' =kT O D YN-1 E N-1(-t) R

N (t) n(t) =

YN

1 . sinc t T T

((

1 log 1+ SNRn 2 2

(
Figure by MIT OpenCourseWare.

Each tone sees AWGN channel (no ISI)

N QAM-like symbols (complex) 1 PAM symbol
Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

24

The effect of the channel

X( f ) input X0 X1 f1 X2 f2 N = 2N XN-1 fN-1 XN fN

H( f )

Yn Hn . Xn

Y( f ) output Y0 Y1 f1 Y2 f2 YN-1 fN-1 YN fN

Figure by MIT OpenCourseWare.

Each channel can be treated as AWGN

With only one basis function hence simple symbol-by-symbol detector is optimal
Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

25

Gap review
3 2.5 bits/dimension 2 1.5 1 0.5 0 dB 3 dB 6 dB 9 dB

b SNR for Pe = 10-6 (dB) 22b 1 (dB) (dB)

.5 8.8 0 8.8

1 13.5 4.7 8.8

0 2 4 8 10 12 SNR (dB) Achievable bit rate for various gaps 6 14 16

20.5 26.8 32.9 38.9 11.7 18.0 24.1 30.1 8.8 8.8 8.8 8.8

Illustration of bit rates versus SNR for various gaps.

Figure by MIT OpenCourseWare.

Table of SNR Gaps for Pe = 10-6 .

Figure by MIT OpenCourseWare.

Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

26

Example simplified multitone

1+0.9D channel
With gap 4.4 dB

Put unit energy per dimension (simply guessed) Data rate 1bit/dimension SNRmfb=10dB
Same as baseband DFE Re-calculate the necessary SNR margin SNRmultitone=8.8dB (with more tones to better approx no-ISI case) SNRdfe=7.1dB Can do even better with multitone, if allocated energy properly
Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

27

Water-filling derivation
Find optimum energy allocation that maximizes b for given total energy constraint
b is a convex function in energy/dimension

Use Lagrange multipliers to solve for n

=>

d d n

Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

28

Water-filling spectrum
Flip the channel and pour in energy like water
Energy
n

+ g = constant n

2 n = constant Hn 2

constant

E0

E1

E2

E3

g0
0

g1
1

g2
2

g3
3

g4
4

g5
5

Channel

Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

L
Subchannel index
Figure by MIT OpenCourseWare.

29

Water-fill loading algorithms

Rate maximization

Margin maximization

Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

30

Rate-adaptive loading
Solve through matrix inversion

Solve iteratively

Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

31

Water-filling example (rate-adaptive)

1+0.9D again (Gap=1, so calculating capacity)
1.29 1.29 1.29 1.29 1.29 1.29 1.29

Energy

E0
1.24

E1
1.23

E1
1.23

E2
1.19

E2
1.19

E3
.96

E3
.96

1 19.94

1 17.03 1

1 17.03 2

1 10 3

1 10 4

2.968 5

2.968 6

1 .0552 7

Try 8 dim first Try 7 dim next

Figure by MIT OpenCourseWare.

=> Capacity=1.55bits/dim
6.973 Communication System Design 32

Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

Summary
Bandlimited communication
Block vs. symbol-by-symbol detector Use complex block detectors to orthogonalize basis functions (MAP) Simplify with equalization+sbs detector Generate basis functions that dont loose orthogonality when passing through frequency selective channle (multitone modulation) ZFE removes ISI but enhances noise Trade-off by designing MMSE equalizer DFE removes trailing ISI without noise enhancement Optimal transmission with proper allocation of energy/dimension (waterfilling)

Try to make bandlimited channel look AWGN

Equalization

Multitone

Next practical loading algorithms and DMT/OFDM, Vector coding

Cite as: Vladimir Stojanovic, course materials for 6.973 Communication System Design, Spring 2006. MIT OpenCourseWare (http://ocw.mit.edu/), Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

6.973 Communication System Design

33