Lecture 2

Download as pptx, pdf, or txt
Download as pptx, pdf, or txt
You are on page 1of 70

EE360: Multiuser Wireless Systems and Networks

Lecture 2 Outline
 Announcements
 HW 0 due Monday (high-impact papers – you decide)
 Bandwidth Sharing in Multiuser Channels
 FD, TD, CD, SD, Hybrid (OFDMA)
 Overview of Multiuser Channel Capacity
 Capacity of Broadcast Channels
 AWGN, Fading, and ISI
 Capacity of MAC Channels
 Duality between BC and MAC channels
 MIMO Channels
Review of Last Lecture
 Overview of multiuser wireless systems
and networks
 Multiuser systems
 Cellular, Ad-Hoc, Cognitive, and Sensor
Networks
 Random Access Techniques
 ALOHA
 CSMA-CD
 Reservation Protocols
 Random access works poorly in heavily
loaded systems
Deterministic Bandwidth
Sharing in Multiuser Channels
Deterministic Bandwidth
Sharing
Code Space
 Frequency Division
Time

Frequency Code Space


 Time Division
Time

Frequency
Code Space
 Code Division
 Multiuser Detection Time

Frequency
 Space Division (MIMO)
 Hybrid Schemes (OFDMA)
What is optimal? Look to Shannon.
OFDMA and SDMA
 OFDMA
 Combines OFDM and frequency division
multiplexing
 Different subcarriers assigned to different users

 SDMA (space-division multiple access)


 Different spatial dimensions assigned to different
users
 Implemented via multiuser beamforming (e.g. zero-
force beamforming)
 Benefits from multiuser diversity
Code Division via
Multiple Access SS

 Interference between users mitigated by code


cross
T
b
correlation
xˆ (t )   1s1 (t ) sc21 (t ) cos 2 (2f c t )   2 s2 (t   ) sc 2 (t   ) sc1 (t ) cos( 2f ct ) cos( 2f c (t   )) dt
0
Tb

 .51d1  .5 2 d 2  sc1 (t ) sc 2 (t )dt  .5d1  .5d 2 cos( 2f c ) 12 ( )


a
0

 In downlink, signal and interference have


a
same received power
 In uplink, “close” users drown out “far” users
(near-far problem)
Multiuser Detection
 In all CDMA systems and in TD/FD/CD cellular
systems, users interfere with each other.
 In most of these systems the interference is
treated as noise.
 Systems become interference-limited
 Often uses complex mechanisms to minimize impact
of interference (power control, smart antennas, etc.)

 Multiuser detection exploits the fact that the


structure of the interference is known
 Interference can be detected and subtracted out
 Better have a darn good estimate of the interference
Ideal Multiuser Detection
Signal 1 - =
A/D
A/D Signal 1
A/D Demod
A/D

Iterative
Multiuser
Detection
Signal 2
Signal 2
Demod

- =
Why Not Ubiquitous Today? Power and A/D Precision
Capacity Limits of
Broadcast Channels
Multiuser Shannon Capacity
Fundamental Limit on Data Rates
Capacity: The set of simultaneously achievable rates {R1,…,Rn}
with arbitrarily small probability of error R 3

R2
R2 R3
R1
R1

 Main drivers of channel capacity


 Bandwidth and received SINR
 Channel model (fading, ISI)
 Channel knowledge and how it is used
 Number of antennas at TX and RX

 Duality connects capacity regions of uplink and downlink


Broadcast Channel Capacity
Region in AWGN
 Model
 One transmitter, two receivers with spectral noise
density n1, n2: n1<n2.
 Transmitter has average power Pand total bandwidth
B.

 Single User Capacity:  P 


Ci  Brate
 Maximum achievable 1  asymptotically
log with  small Pe
 ni B 

 Set of achievable rates includes (C1,0) and (0,C2), obtained


by allocating all resources to one user.
Rate Region: Time Division
 Time Division (Constant Power)
 Fraction of time t allocated to each user is varied

U  R 1  C1 , R2  (1   )C 2  ;0    1

 Time Division (Variable Power)


 Fraction of time t and power si allocated to each user
is varied
   1   2  
U R1  B log 1  , R2  (1   ) B log 1   ;
   n1 B   n2 B   
 
  1  (1   ) 2  P, 0    1. 
Rate Region: Frequency Division

 Frequency Division
 Bandwidth Bi and power Si allocated to each user is varied.

   P1   P2   
U R1  B1 log 1 
  , R2  B2 log 1  
 ;
   n1 B1   n2 B2   
 
 
 P1  P2  P, B1  B2  B 
 

Equivalent to TD for Bi=tiB and Pi=tisi.


Superposition Coding

Best user decodes fine points


Worse user decodes coarse points
Code Division
 Superposition Coding
 Coding strategy allows better user to cancel out interference
from worse user.

   P1   P2  

 U 
 R1  B log 1   , R2  B log 1  ;
 1 P  P2  P 

   n1 B   n2 B  S1   



U 
  RDS spread spectrum with spreading gain G and cross
1 
B
G
log

1 
P1 

n1 B / G 
, R2 
B
G
log

1 
P2 
;
 1
n2 B / G  S1 / G  
P  P2  P




    
correlation r12= r21 =G:


      
U R 
B
log 1 
P1
 , R 
B
log 1 
P2
 ; P1  P2  P 

  1
G n1 B / G  P2 / G 
2
G n2 B / G  P1 / G   
     


Capacity Limits of Fading
Broadcast Channels
Broadcast and MAC
Fading Channels
Broadcast: Wireless Wired
One Transmitter Gateway Network
to Many Receivers.
x g3(t)
Multiple Access: x g2(t)
Many Transmitters
x g1(t)
to One Receiver. R3
R2
R1

Goal: Maximize the rate region {R1,…,Rn}, subject to some minimum rate
constraints, by dynamic allocation of power, rate, and coding/decoding.
Assume transmit power constraint and perfect TX and RX CSI
Fading Capacity Definitions

 Ergodic (Shannon) capacity: maximum long-term rates


averaged over the fading process.
 Shannon capacity applied directly to fading channels.
 Delay depends on channel variations.
 Transmission rate varies with channel quality.

 Zero-outage (delay-limited*) capacity: maximum rate


that can be maintained in all fading states.
 Delay independent of channel variations.
 Constant transmission rate – much power needed for deep fading.

*Hanly/Tse, IT, 11/98


Two-User Fading Broadcast Channel
h1[i] n1[i]
x + Y1[i]
X[i]
x + Y2[i]
h2[i] n2[i]
At each time i:
n={n1[i],n2[i]}
n1[i]=n1[i]/h1[i]
+ Y1[i]
X[i]
+ Y2[i]
n2[i]=n2[i]/h2[i]
Ergodic Capacity Region*
 Capacity region:C ergodic( P )  P F C (P ) ,where
  
  
P ( n )
C (P )   R j  E n  B log1  ,
j
1  j  M}
  M 
 

n j B   Pi ( n )1[ n j  n ]
i 

 i 1

M
 The power constraint implies E n  Pj (n)  P
j 1

 Superposition coding and successive decoding


achieve capacity
 Best user in each state decoded last

*Li/Goldsmith, IT, 3/01


 Power and rate adapted using multiuser water-filling:
Zero-Outage Capacity Region*
 The set of rate vectors that can be maintained for all
channel states under power constraint P
C zero ( P )  P F nN C (P )
   
   
 P ( n ) 
C (P )   R j  B log 1  j , 1 j  M
 M 
  n j B   Pi (n)1[n j  ni ]  
  i 1  

 Capacity region defined implicitly relative to power:


 For a given rate vector R and fading state n we find
the minimum power Pmin(R,n) that supports R.
 RCzero(P) if En[Pmin(R,n)]  P
*Li and Goldsmith, IT, 3/01
Outage Capacity Region
 Two different assumptions about outage:
 All users turned off simultaneously (common outage Pr)
 Users turned off independently (outage probability vector Pr)

 Outage capacity region implicitly defined from the


minimum outage probability associated with a given rate
 Common outage: given (R,n), use threshold policy
 If Pmin(R,n)>s* declare an outage, otherwise assign this
power to state n.

P  E n:P min ( R ,n ) s* P
min
( R, n ) 
 Pr  s* :
Power constraint dictates n:P min ( R ,n ) s*
p ( n)
Independent Outage
 With independent outage cannot use the threshold
approach:
 Any subset of users can be active in each fading state.

 Power allocation must determine how much power to


allocate to each state and which users are on in that state.

 Optimal power allocation maximizes the reward for


transmitting to a given subset of users for each fading state
 Reward based on user priorities and outage probabilities.
 An iterative technique is used to maximize this reward.
Broadcast Channels with ISI
 ISI introduces memory into the channel

 The optimal coding strategy decomposes the


channel into parallel broadcast channels
 Superposition coding is applied to each subchannel.

 Power must be optimized across subchannels


and between users in each subchannel.
Broadcast Channel Model
w1k m

H 1( w) y1k   h1i xk i w1k


i 1
xk w2k m

H2(w) y2 k   h2i xk i w2 k


i 1

 Both H1 and H2 are finite IR filters of length m.


 The w1k and w2k are correlated noise samples.
 For 1<k<n, we call this channel the n-block
discrete Gaussian broadcast channel (n-DGBC).
 The channel capacity region is C=(R1,R2).
Circular Channel Model
 Define the zero padded filters as:
~ n
{hi }i 1  (h1 ,..., hm ,0,...,0)

 The n-Block Circular Gaussian Broadcast Channel


(n-CGBC) is defined based on circular convolution:
n 1
~
y1k   h1i x(( k i )) w1k  xi  h1i  w1k
~
i 0
n 1 0<k<n
~
y2 k   h2i x(( k i )) w2 k  xi  h2i  w2 k
~
i 0

where ((.)) denotes addition modulo n.


Equivalent Channel Model
 Taking DFTs of both sides yields
~ ~
Y1 j  H1 j X j  W1 j
~ ~ 0<j<n
Y2 j  H2 j X j  W2 j

 Dividing by H~ and using additional properties


of the DFT yields
Y1j  X j  V1j
0<j<n
Y2j  X j  V2j

where {V1j} and {V2j} are independent zero-mean Gaussian


~ 2
random variables with lj 2
 n ( N l ( 2 j / n )/| H lj | , l  1,2.
Parallel Channel Model
V11
+ Y11
X1
+ Y21
V21 Ni(f)/Hi(f)

V1n f
+ Y1n
Xn
+ Y2n
V2n
Channel Decomposition
 The n-CGBC thus decomposes to a set of n parallel discrete
memoryless degraded broadcast channels with AWGN.
 Can show that as n goes to infinity, the circular and original channel
have the same capacity region

 The capacity region of parallel degraded broadcast


channels was obtained by El-Gamal (1980)
 Optimal power allocation obtained by Hughes-Hartogs(’75).

n 1

 E[ x 2
i ]  nP
 The power constraint i 0 onn 1the original channel is
 E[( X i) ]  n P
2 2

converted by Parseval’s theorem to i 0 on the


equivalent channel.
Capacity Region of Parallel Set
 Achievable Rates (no common information)
  j Pj    j Pj 
 R1  .5  log1     .5  log 1   ,
j: 1 j  2 j   1 j  j: 1 j  2 j
 (1   ) P   
 j j 1j 

 (1   j ) Pj   (1   j ) Pj 
R2  .5  log1    .5  log1  ,
    2 j 
j: 1 j  2 j   j Pj   2 j  j: 1 j  2 j 
0   j  1,  Pj  n 2 P 

 Capacity Region

For 0<b find {aj}, {Pj} to maximize R1+bR2+l SPj.
 Let (R1*,R2*)n,b denote the corresponding rate pair.
b
R2
 Cn={(R1*,R2*)n,b : 0<b  }, C=liminfn n1Cn .
R1
Optimal Power Allocation:
Two Level Water Filling
Capacity vs. Frequency
Capacity Region
Capacity Limits of
MAC Channels
Multiple Access Channel
 Multiple transmitters
 Transmitter i sends signal Xi with power Pi

 Common receiver with AWGN of power N0B


 ReceivedMsignal: X1
Y   Xi  N
i 1 X2 X3
MAC Capacity Region

 Closed convex hull of all (R1,…,RM) s.t.

 

iS
Ri  B log 1   Pi / N 0 B ,
 iS 
S  {1,..., M }
 For all subsets of users, rate sum equals that of 1
superuser with sum of powers from all users

 Power Allocation and Decoding Order


 Each user has its own power (no power alloc.)
 Decoding order depends on desired rate point
Two-User Region
Superposition coding
w/ interference canc.
Time division
C2 SC w/ IC and time
sharing or rate splitting

Ĉ2 Frequency division
SC w/out IC

 Pi  Ĉ1 C1
Ci  B log 1  , i  1,2
 N0 B 

ˆ  P1  ˆ  P2 
C1  B log 1  , C2  B log 1  ,
 N 0 B  P2   N 0 B  P1 
Fading and ISI
 MAC capacity under fading and ISI determined
using similar techniques as for the BC

 In fading, can define ergodic, outage, and


minimum rate capacity similar as in BC case
 Ergodic capacity obtained based on AWGN MAC
given fixed fading, averaged over fading statistics
 Outage can be declared as common, or per user

 MAC capacity with ISI obtained by converting to


Duality between
Broadcast and MAC Channels
Comparison of MAC and BC
P
 Differences:
 Shared vs. individual power constraints
 Near-far effect in MAC P1

 Similarities: P2
 Optimal BC “superposition” coding is also optimal for
MAC (sum of Gaussian codewords)
 Both decoders exploit successive decoding and
interference cancellation
MAC-BC Capacity Regions
 MAC capacity region known for many cases
 Convex optimization problem

 BC capacity region typically only known for


(parallel) degraded channels
 Formulas often not convex

 Can we find a connection between the BC and


MAC capacity regions?
Duality
Dual Broadcast and MAC Channels
Gaussian BC and MAC with same channel gains
and same noise power at each receiver
h1 (n)
z1 (n)
h1 (n)

x + y1 (n) x1 (n) x z (n)


( P1 )
+ y (n)
x(n) z M (n)
h M (n) h M (n)
(P )
x + yM (n) xM (n) x
( PM )

Broadcast Channel (BC) Multiple-Access Channel (MAC)


The BC from the MAC
C MAC ( P1 , P2 ; h1 , h2 )  C BC ( P1  P2 ; h1 , h2 )

h1  h2
P1=0.5, P2=1.5

Blue = BC P1=1, P2=1


Red = MAC

P1=1.5, P2=0.5

MAC with sum-power constraint

C BC ( P ; h1 , h2 )   C MAC ( P1 , P  P1 ; h1 , h2 )
0 P1  P
Sum-Power
MAC
C BC ( P; h1 , h2 )  C (
 MAC 1P , P  P ;
1 1h , h2 )  C Sum
MAC ( P; h1 , h2 )
0 P1  P

 MAC with sum power constraint


 Power pooled between MAC transmitters
 No transmitter coordination
P
Same capacity region!
MAC BC

P
BC to MAC: Channel
Scaling
 Scale channel gain by a, power by 1/a
 MAC capacity region unaffected by scaling
 Scaled MAC capacity region is a subset of the scaled BC
capacity region for any a
 MAC region inside scaled BC region for any scaling

P1 MAC
  h1  h1 +

P1
+  P2

h2 +
h2
P2
BC
The BC from the
MAC
  

h2
Blue = Scaled BC 
h1
Red = MAC
  0

P1
C MAC ( P1 , P2 ; h1 , h2 )   C BC (  P2 ;  h1 , h2 )
 0 
Duality: Constant AWGN
Channels
 BC in terms of MAC
C BC ( P ; h1 , h2 )   C MAC ( P1 , P  P1 ; h1 , h2 )
0 P1  P

 MAC in terms of BC
P1
C MAC ( P1 , P2 ; h1 , h2 )   C BC (  P2 ; h1 , h2 )
 0 

What is the relationship between


the optimal transmission strategies?
Transmission Strategy
Transformations
 Equate rates, solve for powers
h12 P1M h12 P1B
R1M  log(1  M 2
)  log(1  2
)  R1
B
h2 P2   

h22 P2M h22 P2B


R2M  log(1  )  log(1  )  R2
B
2 h22 P1B   2
 Opposite decoding order
 Stronger user (User 1) decoded last in BC
 Weaker user (User 2) decoded last in MAC
Duality Applies to Different
Fading Channel Capacities

 Ergodic (Shannon) capacity: maximum rate averaged


over all fading states.

 Zero-outage capacity: maximum rate that can be


maintained in all fading states.

 Outage capacity: maximum rate that can be maintained


in all nonoutage fading states.

 Explicit
Minimum transformations
rate capacity: between
Minimum transmission strategies
rate maintained in
all states, maximize average rate in excess of minimum
Duality: Minimum Rate
Capacity

MAC in terms of BC
Blue = Scaled BC
Red = MAC

 BC region known
 MAC region can only be obtained by duality
What other capacity regions can be obtained by
Broadcast duality?
MIMO Channels
Capacity Limits of
Broadcast MIMO Channels
Broadcast MIMO Channel
t1 TX antennas
n1
(r1  t ) r11, r21 RX
H1 y1  H1x  nantennas
1

x n2 Perfect CSI at TX and RX


(r2  t )

H2 y2  H2x  n 2

n1 ~ N(0, I r1 ) n 2 ~ N(0, I r2 )

Non-degraded broadcast channel


Dirty Paper Coding (Costa’83)
 Basic premise
 If the interference is known, channel capacity same as if
there is no interference
 Accomplished by cleverly distributing the writing
(codewords) and coloring their ink
 Decoder must know how to read these codewords

Dirty Dirty
Paper Paper
Coding Coding

Clean Channel Dirty Channel


Modulo Encoding/Decoding
 Received signal Y=X+S, -1X1
 S known to transmitter, not receiver

 Modulo operation removes the interference effects


 Set X so that Y[-1,1]=desired message (e.g. 0.5)
 Receiver demodulates modulo [-1,1]

-1 0 +1

… …
-7 -5 -3 -1 0 +1 +3 +5 +7

S X
-1 0 +1
Capacity Results
 Non-degraded broadcast channel
 Receivers not necessarily “better” or “worse” due to
multiple transmit/receive antennas
 Capacity region for general case unknown

 Pioneering work by Caire/Shamai (Allerton’00):


 Two TX antennas/two RXs (1 antenna each)
 Dirty paper coding/lattice precoding (achievable rate)
 Computationally very complex
 MIMO version of the Sato upper bound
 Upper bound is achievable: capacity known!
Dirty-Paper Coding (DPC)
for MIMO BC
 Coding scheme:
 Choose a codeword for user 1
 Treat this codeword as interference to user 2
 Pick signal for User 2 using “pre-coding”
 Receiver 2 experiences no interference:
R 2  log(det(I  H 2 2 H 2T ))
 Signal for Receiver 2 interferes with Receiver 1:
 det(I  H1 (1   2 ) H1T ) 
R 1  log  T


 det(I  H  H
1 2 1 ) 
 Encoding order can be switched
 DPC optimization highly complex
Does DPC achieve capacity?
 DPC yields MIMO BC achievable region.
 We call this the dirty-paper region

 Is this region the capacity region?


 We use duality, dirty paper coding, and Sato’s upper
bound to address this question

 First we need MIMO MAC Capacity


MIMO MAC Capacity
 MIMO MAC follows from MAC capacity
formula   
C MAC ( P1 ,..., Pk )   ( R1 ,..., Rk ) :  Rk  log 2 det  I   H k Qk H ,
H
k
 kS  kS 
S  {1,..., K } 
 Basic idea same as single user case
 Pick some subset of users
 The sum of those user rates equals the capacity as if
the users pooled their power

 Power Allocation and Decoding Order


MIMO MAC with sum
power
 MAC with sum power:
 Transmitters code independently
 Share power
P
C Sum
MAC ( P)   CMAC ( P1 , P  P1 )
0 P1  P
 Theorem: Dirty-paper BC region equals the dual
sum-power MAC region

C DPC
BC ( P)  C Sum
MAC ( P)
Transformations: MAC to
BC
 Show any rate achievable in sum-power MAC also
achievable with DPC for BC:
DPC BC Sum MAC
DPC
C BC ( P )  C MAC
Sum
( P)
 A sum-power MAC strategy for point (R1,…RN) has a given input
covariance matrix and encoding order
 We find the corresponding PSD covariance matrix and encoding
order to achieve (R1,…,RN) with DPC on BC
 The rank-preserving transform “flips the effective channel” and
reverses the order
 Side result: beamforming is optimal for BC with 1 Rx antenna at
each mobile
Transformations: BC to
MAC
 Show any rate achievable with DPC in BC also
achievable in sum-power MAC:
DPC BC Sum MAC
C DPC
BC ( P)  C Sum
MAC ( P)

 We find transformation between optimal DPC strategy


and optimal sum-power MAC strategy
 “Flip the effective channel” and reverse order
Computing the Capacity Region

C DPC
BC ( P)  C Sum
MAC ( P)
 Hard to compute DPC region (Caire/Shamai’00)
 “Easy” to compute the MIMO MAC capacity
region
 Obtain DPC region by solving for sum-power MAC and
applying the theorem
 Fast iterative algorithms have been developed
 Greatly simplifies calculation of the DPC region and the
associated transmit strategy
Sato Upper Bound on the
BC Capacity Region
 Based on receiver cooperation
n1

+ y1
H1 Joint receiver
x
H2 n2

+ y2

 BC sum rate capacity  Cooperative capacity

sumrate max 1
CBC (P, H)  log | I  HΣ x H T |
x 2
The Sato Bound for MIMO
BC
 Introduce noise correlation between receivers
 BC capacity region unaffected
 Only depends on noise marginals

 Tight Bound (Caire/Shamai’00)


 Cooperative capacity with worst-case noise correlation
inf max 1
sumrate
CBC (P, H)  log | I  Σ z1/2HΣ x H T Σ z 1/2 |
z x 2

 Explicit formula for worst-case noise covariance


 By Lagrangian duality, cooperative BC region equals the
sum-rate capacity region of MIMO MAC
MIMO BC Capacity Bounds
Single User Capacity Bounds
Dirty Paper Achievable Region

BC Sum Rate Point


Sato Upper Bound

Does the DPC region equal the capacity region?


Full Capacity Region
 DPC gives us an achievable region

 Sato bound only touches at sum-rate point

 Bergman’s entropy power inequality is not a tight


upper bound for nondegraded broadcast channel

 A tighter bound was needed to prove DPC optimal


 It had been shown that if Gaussian codes optimal, DPC
was optimal, but proving Gaussian optimality was open.

 Breakthrough by Weingarten, Steinberg and


Enhanced Channel Idea
 The aligned and degraded BC (AMBC)
 Unity matrix channel, noise innovations process
 Limit of AMBC capacity equals that of MIMO BC
 Eigenvalues of some noise covariances go to infinity
 Total power mapped to covariance matrix constraint

 Capacity region of AMBC achieved by Gaussian


superposition coding and successive decoding
 Uses entropy power inequality on enhanced channel
 Enhanced channel has less noise variance than original
 Can show that a power allocation exists whereby the
enhanced channel rate is inside original capacity region
Illustration

Original

Enhanced
Main Points
 Shannon capacity gives fundamental data rate limits for
multiuser wireless channels

 Fading multiuser channels optimize at each channel instance


for maximum average rate

 Outage capacity has higher (fixed) rates than with no outage.

 OFDM is near optimal for broadcast channels with ISI

 Duality connects BC and MAC channels


 Used to obtain capacity of one from the other

You might also like