EE359 – Lecture 13 Outline

 Annoucements
 Midterm announcements
 No HW due this week (will be usual HW next week)
 Adaptive MQAM: optimal power and rate
 Finite Constellation Sets
 Update rate
 Estimation error
 Estimation delay
 Midterm Announcements
 Midterm Wed Nov. 4, 8:45-10:45a in this rm.
 Open book/notes (bring textbook/calculators)
 Covers Chapters 1-7 (through today’s lecture)
 Review Session notes posted
 Extra OHs
 Me: Monday, Tuesday 2-3pm,
 Yao: Tues, 4-5pm in 380-380w

 Midterms from past 3 MTs posted

 10 bonus points for “taking” a practice exam (due at MT)
 Solns for all exams given when you turn in practice exam
 Review of Last Lecture
 Combining Techniques
 Selection Combining and its Performance
 Maximal Ratio Combining
Pb   Pb (  ) p(  )d     ... Pb (  ) p( 1 ) * p( 2 ) * ... * p( M )d 1d 2 ...d M

 MGF Approach for Performance of MRC

.5 M
1  g 
Pb    M i  2 ;  i d
 0 i 1  sin  

 EGC and Transmit Diversity

 Adaptive Modulation
 Change modulation relative to fading
 Parameters to adapt:
Constellation size
Transmit power
Instantaneous BER
Symbol time
Coding rate/scheme
Only 1-2 degrees of freedom needed for good performance

 Optimization criterion:
 Maximize throughput
 Minimize average power
 Minimize average BER
 Variable-Rate Variable-Power MQAM
One of the
log2 M() Bits
M() Points To Channel

Point M()-QAM
Uncoded
Data Bits Delay Selector Modulator
Power: P()

(t) (t)

BSPK 4-QAM 16-QAM

Goal: Optimize P() and M() to maximize R=Elog[M()]

 Optimization Formulation
 Adaptive MQAM: Rate for fixed BER

1.5 P( ) P( )

M ( )  1   1  K
 ln(5BER) P P

 Rate and Power Optimization

 P( ) 
max E log 2 [ M ( )]  max E log 2 1  K
P ( ) P ( )
 P 
Same maximization as for capacity, except for K=-1.5/ln(5BER).
 Optimal Adaptive Scheme
1
 Power Adaptation  0


P() 10 1K  K0 K 1
 K
P  0 else k 

 Spectral Efficiency
R

  
  log   p( ) d .
 
2
B  K K

Equals capacity with effective power loss K=-1.5/ln(5BER).

 Spectral Efficiency

K2
K1

K=-1.5/ln(5BER)

Can reduce gap by superimposing a trellis code

 Constellation Restriction
 Restrict MD() to {M0=0,…,MN}.
 Let M()=/*, where * is later optimized.
 Set MD() to maxj Mj: Mj  M().
 Region boundaries are j=Mj*, j=0,…,N
 Power control maintains target BER
M3
M()=/*
MD()
M3
M2
M2
M1 M1
Outage
0
0 1=M1K* 2 3 
 Power Adaptation and Average Rate
 Power adaptation:
 Fixed BER within each region
 Es/N0=(Mj-1)/K
 Channel inversion within a region

 Requires power increase when increasing M()

Pj ( ) ( M j  1) /(K )  j     j 1 , j  0

P  0   1
 Average Rate
R N
  log 2 M j p( j     j 1 )
B j 1
Efficiency in Rayleigh Fading
Spectral Efficiency (bps/Hz)

Average SNR (dB)

 Main Points
 Adaptive modulation leverages fast fading to improve
performance (throughput, BER, etc.)
 Adaptive MQAM uses capacity-achieving power and
rate adaptation, with power penalty K.
 Comes within 5-6 dB of capacity

 Discretizing the constellation size results in

negligible performance loss.
 Constellations cannot be updated faster than 10s to
100s of symbol times: OK for most dopplers.