1

CAPACITY AND SCHEDULING IN CAPACITY AND SCHEDULING IN

MULTI MULTI- -USER MIMO SYSTEMS USER MIMO SYSTEMS
Carles Antn-Haro, PhD
Centre Tecnolgic de Telecomunicacions de Catalunya (CTTC)
ACE Course on MIMO Communication Systems and Antennas, KTH, Stockholm, Sept 5-9, 2005.
A cross-layer approach
2
The sources of this seminar are mainly:
T.M. Cover and J.A. Thomas, Elements of Information Theory, Wiley
Series in Telecommunications, John Wiley & Sons, New York, 1991.
D. Tse, P.Wiswanath, Fundamentals of Wireless Communications,
Cambridge University Press, 2005. Chaps 5,6,10.
A. Goldmsith et al, Capacity Limits of MIMO Channels, IEEE Trans. on
Selected Areas in Communications, Vol. 21, No. 5, June 2003.
H. Boche and M. Wiczanowski, Queueing Theoretic Optimal Scheduling
for Multiple Input Multiple Output Multiple Access Channel, ISSPIT 2003,
Darmstadt, Germany.
REFERENCES REFERENCES
2
3
MOTIVATION MOTIVATION
Best way for multiple users to transmit over a shared medium? Or Best way for multiple users to transmit over a shared medium? Orthogonal thogonal
access? Simultaneous? access? Simultaneous?
Differences between uplink (multiple access) and downlink (broad Differences between uplink (multiple access) and downlink (broadcast) cast)
channels? channels?
Impact of multiple transmit and/or multiple receive antennas? Impact of multiple transmit and/or multiple receive antennas?
In multi In multi- -user systems, can we take advantage of fading? user systems, can we take advantage of fading?
Can the scheduling process be enhanced with channel Can the scheduling process be enhanced with channel- -related information? related information?
Combined use of queue and channel information for scheduling? Combined use of queue and channel information for scheduling?
Information theory approach: keep it general !! Information theory approach: keep it general !!
4
OUTLINE OUTLINE
Motivation Motivation
A review of capacity issues in single A review of capacity issues in single- -user systems user systems
Definition, Capacity for MIMO systems.
Capacity issues in multi Capacity issues in multi- -user systems user systems: :
Broadcast (BC) and Multiple Access (MAC) channels.
Capacity regions for SISO BC & MAC. Sum capacity. Symmetric capacity.
Multi-user diversity. Channel-aware scheduling.
Fairness issues: Proportional Fair Scheduling
Slow-fading channels: Opportunistic Beamforming
Channel Channel- - and queue and queue- -aware scheduling aware scheduling
Motivation
Q&A Q&A
3
5
OUTLINE OUTLINE
Motivation Motivation
A review of capacity issues in single A review of capacity issues in single- -user systems user systems
Definition, Capacity for MIMO systems.
6
A REVIEW OF CAPACITY ISSUES IN A REVIEW OF CAPACITY ISSUES IN
SINGLE SINGLE- -USER SYSTEMS USER SYSTEMS
4
7
CAPACITY IN LINEAR TIME INVARIANT SYSTEMS CAPACITY IN LINEAR TIME INVARIANT SYSTEMS
( )
n
,K N 0 ~CN
H
H
Y
X` X
( ) P
x
K tr
LTI (ISI, MIMO)
GAUSSIAN
NOISE!!
GAUSSIAN
NOISE!!
Information capacity of an AWGN channel with power constraint P:
( ) Y X I C
x f
x
; max
) (
=
( ) P t s
x
K tr . .
Mutual information maximized for GAUSSIAN input:
( )
x
,K X 0 ~CN
( ) ( ) ( ) ; , I X Y h X h X Y =
( ) ( ) ( ) log
x x
h X E f x =
( ) ( ) ( )
,
, log ,
xv x v
h X Y E f x v =
with
Definition of mutual information
8
In these conditions, maximizing mutual information amounts to:
( ) ( )
n
H
x n
x f
x x x
Y X I Y X I C
K
H HK K
K K
+
= = = log max ; max ; max
) (
( ) P t s
x
K tr . .
Signal
Noise + interference
Remarks:
In general, K
x
depends on H and what information is available @ Tx side (partial, full, none).
Units: bits/s/Hzwhen log = log
2
Interpretation (Shannons Channel Capacity Theorem): For every data rate R
Below capacity (R<C): There exists a channel code allowing transmission with an
arbitrarily small error probability
Above capacity (R>C) : Such a code does NOT exist.
Information capacity (C) provides an upper bound of the
achievable data rates (R)
Information capacity (C) provides an upper bound of the
achievable data rates (R)
Assumptions: Gaussian input symbols & ideal channel coding (and decoding)
Useful equivalence:
H
x n
n
H
x n
x x
C H HK K I
K
H HK K
K K
1
log max log max

+ =
+
=
CAPACITY IN LINEAR TIME INVARIANT SYSTEMS CAPACITY IN LINEAR TIME INVARIANT SYSTEMS
5
9
MIMO CHANNEL MODEL MIMO CHANNEL MODEL
Simplest model:
Channel: Flat fading (frequency), static / independent Rayleigh fading (time)
Noise: Gaussian (spatially) white ( ) ( )
R
n o n
I ,N , 0 ~ 0 ~ CN CN W K N
n
T
Tx
antennas
n
T
Tx
antennas
n
R
Rx
antennas
n
R
Rx
antennas
w
1
.
.
.
.
.
.
+
+
+
x
1
x
2
x
n
T
v
1
v
2
v
n
R
w
n
R
w
2
h
11
h
21
h
n 1
R
h
12
h
n 2
R
h
22
h
1n
h
n n
R
h
2n
T
T
T

(
(
(
(
(

+
(
(
(
(
(

(
(
(
(
(

=
(
(
(
(
(

R T T R R R
T
T
R
n n n n n n
n
n
n
w
w
w
x
x
x
h h h
h h h
h h h
v
v
v

2
1
2
1
2 1
1 22 21
1 12 11
2
1
w Hx y + =
10
CAPACITY OF MIMO SYSTEMS (LTI) CAPACITY OF MIMO SYSTEMS (LTI)
H
x w
x
C H HK K I
K
1
log max

+ =
R
n o w
N I K =
LOG in power
LIN in antennas
LOG in power
LIN in antennas
Asympt growth
i.e. n bits/s/Hz
every 3 dB
MIMO, no CSI at Tx Isotropic transmission:
T
n
T
x
n
P
I K =
H
T o
n N
P
C HH I + = log

=
+
n
i
i
T
n N
P
n C
1
2
0
log log
( ) SNR 1 log 1 log
2
+ =
|
|
.
|

\
|
+ =
o
N
h P
C
( )
o w
N = K
P
x
= K
SISO, Shannon Capacity
MIMO, full CSI at Tx Waterfilling over channel eigenmodes (SVD):
( )
H
n x
P P V V K
1
diag =
( )
H
n
V U H
1
diag =
[ ]

+ =
n
i
i i
P N C
1
2 1
0
1 log
Power allocation (Lagrange): ( ) ( )
R T
n
i
i
i
i i
n n n P P n i
N
P , min 1
1
2
0
= = =
|
|
.
|

\
|
=

=
+

LOG in power
LOG in power
Asympt growth
i.e. 1 bits/s/Hz
every 3 dB
6
11
CAPACITY OF MIMO SYSTEMS (LTI) CAPACITY OF MIMO SYSTEMS (LTI)
H
T o
n N
P
C HH I + = log

=
+
n
i
i
T
n N
P
n C
1
2
0
log log
( ) SNR 1 log 1 log
2
+ =
|
|
.
|

\
|
+ =
o
N
h P
C
MIMO: LOG in power, LIN in antennas
MIMO: LOG in power, LIN in antennas
i.e. n bits/s/Hz every 3 dB
SISO: LOG in power
SISO: LOG in power
i.e. 1 bits/s/Hz every 3 dB
SNR = 3 dB
C = 1 bit/s/Hz (n=1)
C
2
/C
1
= 32/8
(n
2
/n
1
=4)
C = 4 bit/s/Hz (n=4)
D. Tse,P. Wiswanath, Fundamentals of Wireless Communications, Cambridge Univ. Press.2005
12
OUTLINE OUTLINE
Motivation Motivation
A review of capacity issues in single A review of capacity issues in single- -user systems user systems
Definition, Capacity for MIMO systems, time-varying systems.
Capacity issues in multi Capacity issues in multi- -user systems user systems: :
Broadcast (BC) and Multiple Access (MAC) channels.
Capacity regions for SISO BC & MAC. Sum capacity. Symmetric capacity.
Multi-user diversity. Channel-aware scheduling.
Fairness issues: Proportional Fair Scheduling
Slow-fading channels: Opportunistic Beamforming
Capacity regions for MIMO BC & MAC. Duality principle.
7
13
CAPACITY ISSUES IN MULTI CAPACITY ISSUES IN MULTI- -USER USER
SYSTEMS SYSTEMS
14
BROADCAST AND MULTIPLE BROADCAST AND MULTIPLE- -ACCESS CHANNELS ACCESS CHANNELS
Broadcast Channel (BC):
Downlink
One transmitter to many receivers simultaneously
Multiple Access Channel (MAC):
Uplink
Many transmitters to one receiver simultaneously
Remarks:
Users can be regarded as an antenna array
in a large area.
Cooperation among antennas within the
SAME location.
Multiple antennas in one location enable
Space Division Multiple Access or stream
Multiplexing.
x
x
x
x
h
1
h
21
h
22
h
3
x
h
1
x
x
h
3
h
22
x
8
15
MIMO BC MIMO BC and and MAC MAC CHANNEL MODEL CHANNEL MODEL
H
1
H
1

y
1
H
1
x n
1
n
1

y
K
H
K
x n
K
n
K
.
.
.
x
(n
R1
x n
T
)
(n
RK
x n
T
)
Broadcast Channel (BC)
BS Shared power constraint
H
1
H
1

y H
k
x
k
n
n
.
.
.
(n
T1
x n
R
)
x
1
x
K
k
Multiple Access Channel (MAC)
UE individual power constraints
(n
TK
x n
R
)
One base station (BS) equipped with n
T
(n
R
) antennas
K user equipments (UE) equipped with n
Rk
(n
T k
) antennas each
16
CAPACITY REGION FOR MAC CAPACITY REGION FOR MAC- -AWGN AWGN
Characterizes optimal trade-off achievable
by any MA scheme.
User 2 gets R
2
~0 while user 1 attains single-
user bound (A) !!
|
|
.
|

\
|
+
0
2
2
1 log
N
P
R
|
|
.
|

\
| +
+ +
0
2 1
2 1
1 log
N
P P
R R
|
|
.
|

\
|
+
0
1
1
1 log
N
P
R
P
1 P
2
SISO, MAC, AWGN channel, K=2 users:
[ ] [ ] [ ] [ ] m w m x m x m v + + =
2 1
CAPACITY REGION, C CC C !!
AWGN
Single user: Rate R achievable (with arbitarily low error rate) iff
RC C upper bound on performance
Multi-user: UEs communicate with BS in a shared bandwidth
trade-offs turning up!!
Set of achievable rates (R
1
,R
2
) with simultaneous communication??
|
|
.
|

\
|
+
+
0 2
1
1 log
N P
P
|
|
.
|

\
|
+
0
2
1 log
N
P
B
|
|
.
|

\
|
+
+
0 1
2
1 log
N P
P
|
|
.
|

\
|
+
0
1
1 log
N
P
A
HOW? Successive interference
Cancellation (SIC).
Reversing detection order leads to
different rate split (B) - fairness
R
1
R
2
9
17
Some performance measures (scalars) for a capacity region:
MEASURES OF INTEREST MEASURES OF INTEREST
Reached at AB segment (ANY point)
Points A,B achievable via SIC
Intermediate points in AB via time sharing or
rate splitting
Working point TBD according to fairness
constraints
R
1
R
2
|
|
.
|

\
|
+
+
0 2
'
1
1 log
N P
P
|
|
.
|

\
|
+
0
2
1 log
N
P
C
m

1
|
|
.
|

\
| +
+ +
0
2 1
2 1
1 log
N
P P
R R
|
|
.
|

\
|
+
0
1
1 log
N
P
|
|
.
|

\
|
+
+
0 1
2
1 log
N P
P
A
B
m

-
1
|
|
.
|

\
|
+
+
0 2
1
1 log
N P
P
OPTIMAL OPERATING POINTS
FOR SUM CAPACITY
2 1
) , (
sum
2 1
max : R R C
R R
+ =
C
Sum capacity
R C
R R C
=
) , (
sym
max :
Reached @ boundary (near/far) - C
Symmetric capacity
18
GENERAL CASE: MAC GENERAL CASE: MAC with with K K users users
R
1
R
2
R
3
|
|
.
|

\
|
+
0
1
1
1 log
N
P
R
|
|
.
|

\
|
+
0
2
2
1 log
N
P
R
|
|
.
|

\
|
+
0
3
3
1 log
N
P
R
|
|
.
|

\
| +
+ +
0
2 1
2 1
1 log
N
P P
R R
|
|
.
|

\
| +
+ +
0
3 2
3 2
1 log
N
P P
R R
|
|
.
|

\
| + +
+ + +
0
3 2 1
3 2 1
1 log
N
P P P
R R R
|
|
.
|

\
| +
+ +
0
3 1
3 1
1 log
N
P P
R R
|
|
.
|

\
|
+

0
1 log
N
R
R
k
k
k
k
S
S
2
K
-1 constraints
2
K
-1 non-empty subsets S of users
10
19
CAPACITY REGION FOR BC CAPACITY REGION FOR BC- -AWGN AWGN
P
2
h
1
h
1
P
1 2
P P P =
SISO, BC, AWGN channel, K=2 users:
[ ] [ ] [ ] m w m x h m v
1 1 1
+ = [ ] [ ] [ ] m w m x h m v
2 2 2
+ =
Assume: User 2 is the strongest and superposition coding
( )
2 1
h h [ ] [ ] [ ] m x m x m x
2 1
+ =
AWGN
BS communicates with UE in a shared bandwidth & shared
power (P) trade-offs turning up!!
How to MUX data for both users at the BS? x|m| ??
Set of achievable rates (R
1
,R
2
) with simultaneous comms.??
( )
|
|
.
|

\
|
+
+ =
0
2
1 1
2
1 1
1
1 log R
N h P P
h P ( )
|
|
.
|

\
|

+ =
0
2
2 1
2
1 log R
N
h P P
So apply SIC at the strongest (UE
2
) and
( ) ( )
2
UE
0
2
2 1
2
2 1
0
2
1 1
2
1 1
1
UE 1 1
SNIR
1 1
SNIR
x x
N h P
h P
N h P
h P
=
+

+
=
If x
1
decodable at UE
1
(weakest) in the presence of x
2
, so is at UE
2
(strongest) for all power
splits P
1,
P
2
(not possible if reversed order)
20
CAPACITY REGION FOR BC CAPACITY REGION FOR BC- -LTI ( LTI (contd contd) )
2
h
0
1
2
3
4
5
6
7
0 0,2 0,4 0,6 0,8 1
Rate of user 1
R
a
t
e

o
f

u
s
e
r

2
Superpo s itio n Co ding
Ortho go nal
( )
( )
( )
P
P
P P
P
N
h P
N
h P
1
2 1
1
opt
0
2
2 1
2
0
2
1 1
1
1
1
1 log 1 R 1 log R
=
+
= =
|
|
.
|

\
|

+ =
|
|
.
|

\
|
+ =

Orthogonal multiple access

m -1
max
2 1
= + R R
Orthogonal multiple access strictly
suboptimal for all power splitsl!!
SC: low power for strong user (UE
2
) is
efficlently exploited (x
1
removed) and low
interference to weaker (UE
1
)
Remarks:
Strong assumption: DEGRADED BC
MIMO is non degraded.
Degradation not needed in UL
(centralized Rx & CSI).
Sum-rate: allocate ALL power to strongest
user (UE
2
)
at the expense of delays!!
SC boundary given by all P
1
/P
2
splits
Best policy in BC-SISO: ONE user at a time
(vs. MAC-SISO: ALL users simultaneously)
Best policy in BC-SISO: ONE user at a time
(vs. MAC-SISO: ALL users simultaneously)
11
21
Take the case with CSIT (i.e power allocation possible):
AWGN: Sum capacity maximized by transmitting to the BEST user
Fading: Schedule the BEST user at EACH time (greedy approach). Equivalent point-to-point channel
Assumptions:
Fading processes ( h
k
|m|} ): Independent and identically
distributed (symmetric case).
Power constraint (pooled power) :
BC CHANNEL WITH FADING BC CHANNEL WITH FADING
[ ] P m P E
K
k
k H
=
(

=1
2
.. 1
2
eq
max
k
K k
h h
=
=
( )
+
=
|
|
.
|

\
|
=
2
.. 1
0 *
max
1
k K k
h
N
P

h
P
k
h
1
h
1
P
k
P
fading
SISO, BC, fading channel, K users:
[ ] [ ] [ ] [ ] m w m x m h m v
k k k
+ =
( )
(
(

|
|
.
|

\
|
+ =
=
o
k K k
h
N
h P
C
2
.. 1
*
sum
max ) (
1 log E
h
How to allocate power? Temporal waterfilling for the equivalent P2P channel
22
Higher gain means higher rate !!
MUDiv
AWGN
0
SNR
N
P
=
D. Tse,P. Wiswanath, Fundamentals of Wireless Communications, Cambridge Univ. Press.2005
AWGN
Rayleigh
Ricean
dB 0 SNR =
D. Tse,P. Wiswanath, Fundamentals of Wireless Communications, Cambridge Univ. Press.2005
MULTI MULTI- -USER DIVERSITY ( USER DIVERSITY (MUDiv MUDiv) GAIN ) GAIN
2
.. 1
2
eq
2
1
max
k
K k
h h h
=
=
With K users FADING INDEPENDENTLY and OPPORTUNISTIC (DYNAMIC) SCHEDULING,
channel gain improves
Higher gain means higher (sum) rate!!
Gain wrt AWGN for K>1 (mid-high SNR)
=2.1 b/s/Hz
=1.5 b/s/Hz
The amount of MUDiv increases with pdfs tails: Rayleigh > Rice (=5, LOS, less random)
MUDiv gain increases with nr. of users (K): the stronger is the strongest channel
12
23
MULTI MULTI- -USER vs. CLASSICAL DIVERSITY USER vs. CLASSICAL DIVERSITY
Purpose:
Classical (time/frequency/space): Increase link reliability (slow fading)
MUDiv: Increase average cell throughput (fast fading)
but no rate guarantees in specific fading states
Means:
Classical: Counteract adverse fading effects.
MUDiv: Exploit independent fading (capture strongest user)
Scope:
Classical: Works at the link level
MUDiv: System-wide (active users)
24
REMARKS ON REMARKS ON MUDiv MUDiv
Signalling:
UEs: Track their link quality (common pilot)
BS: Access to quality measurements (delay-free feedback channel)
Delay in the feedback channel (ass.: delay&error free)
Mismatch actual channel-measured channel
FIX: scheduling slots signalling overhead selective MUDiv (f/b iff above threshold)
Fairness & delay:
Non-homogeneous user set in real-world networks (assumed so far)
Different statistics (Rayleigh, Rice,) average SNRs (near-far).. RESOURCE ALLOCATION ??
FIX: Proportional Fair Scheduler (PFS)
13
25
Proportional Fair Scheduler: Schedule user with peak rate with respect to its average rate
PROPORTIONAL FAIR SCHEDULING (PFS) PROPORTIONAL FAIR SCHEDULING (PFS)
Opportunistic-greedy is FAIR Opportunistic-greedy is UNFAIR Opportunistic-PFS is FAIR
T
1
T
2
Asymmetic case (SNR
avgk
differ)
T
1
T
2
Asymmetic case (SNR
avgk
differ)
[ ]
[ ]
[ ] m T
m R
m k
k
k
k
max
*
= [ ]
( ) [ ] ( ) [ ]
( ) [ ]

= +
=
*
*
/ 1 1
, / 1 / 1 1
k k m T t
k k m R t m T t
m T
k c
k c k c
k
PFS vs. greedy opportunistic schedulers:
Both channel-dependent (vs.round-robin, vs. queue-based). PFS implemented in IS-856.
Greedy: No short-term fairness, captures MUDiv, maximizes average sum-rate.
PFS: No short-term fairness, long-term fairness (same # access), captures some MUDiv, loss in
average sum-rate.
Latency time scale (t
c
), a design parameter: if larger, larger averaging period, higher latency
(schedule when hitting a really high peak)
D. Tse,P. Wiswanath, Fundamentals of Wireless Communications, Cambridge Univ. Press.2005
Symmetic case (SNR
avgk
=SNR
avg
)
T
1
T
2
26
REMARKS ON REMARKS ON MUDiv MUDiv
Signalling:
UEs: Track their link quality (common pilot)
BS: Access to quality measurements (delay-free feedback channel)
Delay in the feedback channel (ass.: delay&error free)
Mismatch actual channel-measured channel
FIX: scheduling slots signalling overhead selective MUDiv (f/b iff above threshold)
Fairness & delay:
Non-homogeneous user set in real-world networks (assumed so far)
Different statistics (Rayleigh, Rice,) average SNRs (near-far).. RESOURCE ALLOCATION ??
FIX: Proportional Fair Scheduler (PFS)
Limited and slow fluctuations (ass: high & fast)
Limited: poor scattering/LOS Slow : low mobility environment
Result: low cell throughput (peaks) - Delay requirements not met.
FIX: Opportunistic beamforming.
14
27
OPPORTUNISTIC BEAMFORMING OPPORTUNISTIC BEAMFORMING
Slow fading hurts: If all users fade slow like K=1 user no MUDIv
Limited fluctuation hurts: lower peak rates
TRICK (MISO): Induce fast and high fluctuations by transmit beamforming with a time-
varying common set of random weights (e.g circularly symmetric Gaussian):
h
1k
(t)
.
.
.
x
x
x
h
2k
(t)
h
n k
(t)
x (t)
q
1
(t)
q
2
(t)
q
n
(t)
user k
T
T
[ ] [ ] [ ] [ ]
T
k n k k
m h m h m
T

1
= h
[ ] [ ] [ ] [ ]
T
n
m q m q m
T

1
= q
[ ] 1
2
= m q
with
[ ] [ ] [ ] ( ) [ ] [ ] m w m x m m m v
k
T
k k
+ = q h [ ]
[ ] [ ]
0
2
SNR
N
m m
m
T
k
k
q h
=
measure at UE
k
feedback to BS
Random weights
???
When are SNR peaks reached?: When beam points at user k
[ ] [ ] m m
k
*
// h q OPPORTUNISTIC BEAMFORMING
28
OPPORTUNISTIC BEAMFORMING ( OPPORTUNISTIC BEAMFORMING (contd contd) )
Before opportunistic
beamforming
After opportunistic
beamforming
How fast should q|n| change?: Design parameter:
Fast enough to induce fast fading
Slow enough for reliable channel estimation, timely feedback, stable loop.
D. Tse,P. Wiswanath, Fundamentals of Wireless
Communications, Cambridge Univ. Press.2005
15
29
OPPORTUNISTIC BEAMFORMING ( OPPORTUNISTIC BEAMFORMING (contd contd) )
Slow fading [ ] [ ] n Iluctuatio high & Fast : Constant :
2
*
m h h m h
k k k
q =
YES
NO
YES
Additional power for FAST fluctutations
No additional fluctutations
D. Tse,P. Wiswanath, Fundamentals of Wireless Communications, Cambridge Univ. Press.2005 D. Tse,P. Wiswanath, Fundamentals of Wireless Communications, Cambridge Univ. Press.2005
DOES OPPORTUNISTIC BEAMORMING ALWAYS HELP?
[ ] [ ] Gaussian i.i.d. : Gaussian i.i.d. :
2
*
m h m h
k k
q
i.e. identical distribution Ior ANY distribution oI q
Fast Rayleigh fading:
[ ] [ ] [ ] [ ] [ ] [ ] [ ]
2
* *
2
*
m m m m m m h h m h
w k k kw k k
q h q h q h + = + =
Fast Ricean Fading:
30
Opportunistic vs. coherent beamforming:
REMARKS ON OPPORTUNISTIC BEAMFORMING REMARKS ON OPPORTUNISTIC BEAMFORMING
Performance: Comparable for high K
(always a user to point at)
CSIT needs:
Opp.: SNR only (Opp.)!!!
Coherent: full CSI
Multiple transmit antennas just for inducing fluctuations? Can we do better?
MULTIPLE ORTHOGONAL
RANDOM BEAMS
YES
+ Still inducing fast fading
+ Additional spatial multiplexing gain (SDMA)
- Extra overhead for SNR measurements &
feedback
D. Tse,P. Wiswanath, Fundamentals of Wireless Communications, Cambridge Univ. Press.2005
16
31
OUTLINE OUTLINE
Motivation Motivation
A review of capacity issues in single A review of capacity issues in single- -user systems user systems
Definition, Capacity for MIMO systems.
Capacity issues in multi Capacity issues in multi- -user systems user systems: :
Broadcast (BC) and Multiple Access (MAC) channels.
Capacity regions for SISO BC & MAC. Sum capacity. Symmetric capacity.
Multi-user diversity. Channel-aware scheduling.
Fairness issues: Proportional Fair Scheduling
Slow-fading channels: Opportunistic Beamforming
Channel Channel- - and queue and queue- -aware scheduling aware scheduling
Motivation.
Q&A Q&A
32
CHANNEL CHANNEL- - AND QUEUE AND QUEUE- -AWARE AWARE
SCHEDULING SCHEDULING
17
33
ASSUMPTIONS REVISITED ASSUMPTIONS REVISITED
Implicit assumptions so far
Ass. 1: Infinite transmit buffer size:
Users can be delayed without bound (to maximize sum-rate).
Did not care much about packet arrival rates.
Ass.2 : Scheduled user(s) always have data to transmit
BUT in realistic scenarios
Finite buffer size:
When close to buffer overflow, user should be scheduled regardless of channel conditions.
If too many packets arrive, buffer bound to explode.
Traffic is bursty: no point in scheduling a user with empty buffer!
CONCLUSION: Channel and queue (buffer) information must be jointly
considered in the scheduling process (i.e. cross-layer)
34
QUESTIONS ? QUESTIONS ?