MIMOSystems

Waveform Diversity and Design Conference, Slide #1
Short Course on MIMO Systems

Diversity in Communications
Raviraj S. Adve
University of Toronto
Dept. of Elec. and Comp. Eng.
Toronto, Ontario, Canada M5S 3G4
rsadve@comm.utoronto.ca
Introduction and Overview
Basic Digital Communications
Basic Wireless Communications
Receive Diversity
Transmit Diversity
MIMO Information Theory
Course Summary
END
Multiple Input Multiple Output Systems
MIMO Systems: the use of an antenna array at the receiver
(Multiple Output) and/or the transmitter (Multiple Input) in
wireless communications
Outline of this Course:
s
Basic digital and wireless communications
s
Diversity on Receive
s
Diversity on Transmit
s
Multiplexing and data rate
Detailed notes available at
http://www.comm.utoronto.ca/ rsadve/teaching.html
Receive Diversity
Transmit Diversity
Course Summary
END
I like this research area because...
MIMO Systems bring
together....
s
Antenna array theory
s
Probability theory
s
Linear algebra
s
Optimization
s
Digital communications
....and it is useful!
(borrowed from The Economist,
April 28th - May 4th 2007)
qBER
qInfo Theory
Receive Diversity
Transmit Diversity
Course Summary
END
A Digital Communication System
Transmitter
Analog
Source
Bits
Sample
and D/A
Channel
Encoder
Symbols
Pulse
Modulator
Transmitter
Receiver
Bits
Source
Decoder
D/A
Symbols
Detector
Demodulator
and sample
x(t)
Channel
Analog
Output y(t)
x(t) = s(t) +n(t)
s
The noise term, n(t), is usually modelled as additive, white,
Gaussian noise (AWGN)
qBER
qInfo Theory
Receive Diversity
Transmit Diversity
Course Summary
END
Performance Measure : BER
s
Performance of the system is generally measured via the bit
error rate (BER)
x
BER is a function of signal-to-noise ratio (SNR)
SNR =
E{|s(t)|
2
}
E{|n(t)|
2
}
where E{} is the expectation operator.
s
Binary Phase Shift Keying (BPSK): bit = 0 symbol = 1, bit
= 1 symbol = -1
s
For an AWGN channel and BPSK
BER = Q
_
2SNR
_
Q(x) =
1
2
_

x
e
t
2
/2
dt
x
Q(x) < (1/2)e
x
2
/2
, i.e., for an AWGN channel BER falls
off exponentially with SNR.
qBER
qInfo Theory
Receive Diversity
Transmit Diversity
Course Summary
END
Basic Information Theory
s
Channels are characterized by channel capacity C
s
Shannon says: Given a channel with capacity C, one can
nd a coding scheme to transmit at a data rate R < C
without error. Furthermore, one cannot transmit without error
at a data rate R > C.
x
C acts as the effective speed limit on the channel
s
R is generally measured in bits per channel use.
s
For an AWGN channel (with complex inputs and outputs)
C = log
2
(1 +SNR) (bits)
s
Note that C is a non-linear function of SNR
x
At low SNR, C SNR
x
At high SNR C log
2
(SNR)
qFading Channels
qSmall Scale Fading
qRayleigh/Rician Fading
qFlat/Selective Fading
qSlow/Flat fading
qSummary
qSISO Data Model
qChannel Fluctuation
qError Rate
qOutage Probability
Receive Diversity
Transmit Diversity
Course Summary
END
A Wireless Communication System
Transmitter
Data
Source
Bits
Source
Encoder
Channel
Encoder
Symbols
Transmitter
Pulse
Modulator
RF
Modulator
Receiver
Format
Data
Bits
Source
Decoder
Channel
Decoder
Symbols
Receiver
Detector
Demodulator
and sample
Channel
Impulse
Response
A wireless communication system is fundamentally limited by
the random channel, i.e., fading.
qFading Channels
qSmall Scale Fading
qSlow/Flat fading
qSummary
qSISO Data Model
qError Rate
qOutage Probability
Receive Diversity
Transmit Diversity
Course Summary
END
Fading Channels
Due to the unknown location of the mobile station and the
unknown medium between the transmitter and receiver, the
wireless channel is best characterized as random.
s
Fading has three components:
Overall fading = (Distance Attenuation) (Large Scale
Fading) (Small Scale Fading)
s
Distance attenuation: fall off in power with distance
x
In line-of-sight conditions, received power 1/d
2
(where
d is the distance between transmitter and receiver)
x
In non line-of-sight conditions, received power 1/d
n
,
where n is the distance attenuation parameter
s
1.5 < n < 4.5.
s
In some scenarios n < 2 due to tunnelling
s
n large in urban areas, 4.5
qFading Channels
qSmall Scale Fading
qSlow/Flat fading
qSummary
qSISO Data Model
qError Rate
qOutage Probability
Receive Diversity
Transmit Diversity
Course Summary
END
Fading Channels (cont...)
s
Large Scale Fading: Occurs due to the attenuation each time
a signal passes through an object
Large Scale Fading = 10
(x
1
+x
2
+...)
= 10
x
where x
i
is the attenuation due to object # i.
s
By the Cental Limit Theorem, x = (x
1
+x
2
+. . .) is a
Gaussian random variable
large scale fading can be modelled as log-normal, i.e., the
log of the fading term is distributed normal:
Large Scale Fading = 10
x
;
x N(0,
2
h1
)
s
Large scale fading varies slowly, remaining approximately
constant over hundreds of wavelengths
x
There is not much one can do about large scale fading or
distance attenuation, just power control
qFading Channels
qSmall Scale Fading
qSlow/Flat fading
qSummary
qSISO Data Model
qError Rate
qOutage Probability
Receive Diversity
Transmit Diversity
Course Summary
END
Small Scale Fading
s
Occurs due to multipath propagation
x
Each signal arrives over many many paths
Transmitter Receiver
s
Each path has slightly different length slightly different
time of propagation
with different phase that is effectively random
qFading Channels
qSmall Scale Fading
qSlow/Flat fading
qSummary
qSISO Data Model
qError Rate
qOutage Probability
Receive Diversity
Transmit Diversity
Course Summary
END
Small Scale Fading (cont...)
s
Total received signal is the sum over all paths
Received Signal = Transmitted Signal Distance Attenuation
Large Scale Fading (h
1
+h
2
. . .)
h
i
=
i
e
j
i
;
i
is random
s
If all
i
are approximately the same (Rayleigh fading),
Small Scale Fading = h = (h
1
+h
2
+. . .)
h CN(0,
2
h
)
s
If one of the components is dominant (Rician fading)
h CN(,
2
h
)
s
CN(,
2
h
) represents the complex Gaussian distribution with
mean and variance
2
h
.
x
We will focus on Rayleigh fading, i.e., h CN(0,
2
h
)
x
Note: Rayleigh fading is just one of many fading models
qFading Channels
qSmall Scale Fading
qSlow/Flat fading
qSummary
qSISO Data Model
qError Rate
qOutage Probability
Receive Diversity
Transmit Diversity
Course Summary
END
s
Frequency at versus frequency selective fading:
x
Symbol period: T
s
; Channel time spread: T
c
T
2T
s
s
h(t)
T
c
T
2T
s
s
h(t)
T
c
Modelled As
s
Channel modelled as a train of impulses
x
If T
c
< T
s
, h(t) = (t) H(j) = constant (Frequency
at fading)
x
If T
c
> T
s
, h(t) =
(t T
s
) H(j) = constant
(Frequency selective fading)
s
Note: CN(0,
2
h
)
qFading Channels
qSmall Scale Fading
qSlow/Flat fading
qSummary
qSISO Data Model
qError Rate
qOutage Probability
Receive Diversity
Transmit Diversity
Course Summary
END
s
If the mobile is moving with radial velocity v, the channel
changes as a function of time
x
Rate of change = Doppler frequency (f
d
= v/)
s
Symbol rate = f
s
= 1/T
s
x
If f
d
f
s
, channel is effectively constant over several
symbols (slow fading)
x
If f
d
> f
s
, channel changes within a symbol period (fast
fading)
s
We shall focus on slow, at, fading
qFading Channels
qSmall Scale Fading
qSlow/Flat fading
qSummary
qSISO Data Model
qError Rate
qOutage Probability
Receive Diversity
Transmit Diversity
Course Summary
END
In summary...
s
Fading has three components
x
Distance attenuation 1/d
n
x
Large scale fading; modelled as log-normal; constant over
hundreds of
x
Small scale fading; modelled as Rayleigh, i.e., complex
normal; uctuates within fraction of
s
Assume power control for distance attenuation and large
scale fading.
x
This is all that can be done!
qFading Channels
qSmall Scale Fading
qSlow/Flat fading
qSummary
qSISO Data Model
qError Rate
qOutage Probability
Receive Diversity
Transmit Diversity
Course Summary
END
Data Model
s
With at, slow fading:
x = hs +n
s
x: received signal, h: channel, s: transmitted complex
symbol, n: noise
x
h CN(0,
2
h
)
x
Instantaneous channel power: |h|
2
(1/
2
h
)e
|h|
2
/
2
h
(exponential)
s
channel is often in bad shape
x
Note:
2
h
= E{|h|
2
} = average power in channel
s
Set
2
h
= 1 for convenience (channel does not
introduce power)
qFading Channels
qSmall Scale Fading
qSlow/Flat fading
qSummary
qSISO Data Model
qError Rate
qOutage Probability
Receive Diversity
Transmit Diversity
Course Summary
END
Impact of Fading
Average SNR = E{|h|
2
}
E{|s|
2
}
2
=
2
h
=
= Instantaneous SNR = |h|
2
E{|s|
2
}
2
= |h|
2
;
1
e
/
s
Note that the average SNR has not changed
s
The uctuation in power due to the fading seriously impacts
on the performance of a wireless system
0 10 20 30 40 50 60 70 80 90 100
25
20
15
10
5
0
5
10
Sample #
I
n
s
t
a
n
t
a
n
e
o
u
s

C
h
a
n
n
e
l

p
o
w
e
r

(
d
B
)
qFading Channels
qSmall Scale Fading
qSlow/Flat fading
qSummary
qSISO Data Model
qError Rate
qOutage Probability
Receive Diversity
Transmit Diversity
Course Summary
END
Bit Error Rate
s
Without fading, BER = Q(
2) exp()
x
Exponential drop off with SNR
s
With fading, instantaneous SNR = , i.e., instantaneous BER
= Q(
2)
Average BER = E
{Q(
_
2}
=
_

0
Q(
_
2)
1
e
/
d
=
1
2
_
1
_

1 +
_
x
At high SNR ( ),
BER
1

qFading Channels
qSmall Scale Fading
qSlow/Flat fading
qSummary
qSISO Data Model
qError Rate
qOutage Probability
Receive Diversity
Transmit Diversity
Course Summary
END
Bit Error Rate: Example
0 5 10 15
10
4
10
3
10
2
10
1
10
0
SNR (dB)
B
E
R
Comparing error rates with and without fading
Without fading
With Rayleigh fading
Note: In the case with fading, the BER v/s SNR plot (in log-log
format appears as a straight line)
qFading Channels
qSmall Scale Fading
qSlow/Flat fading
qSummary
qSISO Data Model
qError Rate
qOutage Probability
Receive Diversity
Transmit Diversity
Course Summary
END
Outage Probability
s
If channel is known, capacity: C = log
2
_
1 +|h|
2
_
s
If channel is unknown, true capacity = 0!!
x
One cannot guarantee any data rate
x
Therefore, dene outage probability for a rate R
P
out
= P(C < R)
P
_
log
2
_
1 +|h|
2
_
< R
_
= P
_
|h|
2
<
2
R
1
_
= 1 exp
_
2
R
1
_
x
IMPORTANT: as average SNR gets large ( ),
exp
_
2
R
1
_
1
2
R
1
P
out

1

qFading Channels
qSmall Scale Fading
qSlow/Flat fading
qSummary
qSISO Data Model
qError Rate
qOutage Probability
Receive Diversity
Transmit Diversity
Course Summary
END
Outage Probability: Alternate Denition
s
Note: Choosing a target rate R is equivalent to choosing a
SNR threshold,
s
s
Alternate denition of outage
P
out
= P [ <
s
]
=
_

s
0
1
e
/
d
= 1 e
s
/
s
Again, as ( )
P
out

1

qFading Channels
qSmall Scale Fading
qSlow/Flat fading
qSummary
qSISO Data Model
qError Rate
qOutage Probability
Receive Diversity
Transmit Diversity
Course Summary
END
So, what are we going to do about this?
Use multiple antennas!
which provide diversity
Receive Diversity
qIntroduction
qDiversity Types
qSelection Combining
qMaximal Ratio Combining
qDiversity Order
qEqual Gain Combining
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Introduction to Receive Diversity
SIMO : Single antenna at the transmitter, multiple at the
receiver
Transmitter
Receiver 1
Receiver 2
Receiver N
h
0
h
1
h
N
s
If for a single receiver, P
out
= 0.1, for two receivers
P
out
= 0.01
x
exponential gains in error rate with linear increase in
number of antennas
x
fundamental assumption: the error events are
independent, i.e, the channels are independent
s
Key: provide the receiver with multiple independent copies
of the message
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Diversity Basics
s
Assume for now that the channels are independent and
identically distributed (i.i.d.)
x
We will deal with the issue of correlation later
x
Channel to n
th
receive element = h
n
, i.e., h
n
is assumed
independent of h
m
for n = m
s
Signal to noise ratios are also i.i.d.:
n
is independent of
m
,
n = m. Also,
n

1
n
/
x
Note: every channel has the same average SNR
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Basics (cont...)
Output Signal
w
1
*
w
2
*
w
N
*
x
1
x
2
x
N
s
This appears to be beamforming! (for a single user!).
s
Write the received signal as a vector
x = hs +n
h = [h
1
, h
2
, . . . h
N
]
T
x
The output signal is given by:
y =
N
n=1
w
n
x
n
= w
H
x = w
H
hs +w
H
n
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Receive Diversity Techniques
s
The weight vector is w = [w
1
, w
2
, . . . w
N
]
T
s
Key: how are these weights chosen?
Receive Diversity Techniques:
s
Selection Combining
s
Maximal Ratio Combining (MRC)
s
Equal Gain Combining (EGC)
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Selection Combining
s
Choose, for further processing, the receive element with the
highest SNR
w
k
=
_
1
k
= max
n
{
n
}
0 otherwise
s
The output SNR is therefore the maximum of the receive
elements
Output SNR =
out
= max
n
{
n
}
s
Note: channel phase information not required in the
selection process
s
This is the simplest diversity scheme
x
Seems to waste (N 1) receivers
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Analyzing Selection Diversity
s
Outage probability : the output SNR is below threshold
s
if
all receive elements have SNR below
s
:
P
out
= P [
out
<
s
]
= P [
1
,
2
, . . .
N
<
s
]
=
N
n=1
P[
n
<
s
]
P
out
=
_
1 e
s
/
_
N
x
P
out
=
_
1 e
s
/
for N = 1, i.e.,exponential gains in

outage probability
s
Note: at high SNR, as
P
out

_
1
_
N
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Performance: Outage Probability versus SNR
0 5 10 15 20
10
4
10
3
10
2
10
1
10
0
SNR (dB)
O
u
t
a
g
e

P
r
o
b
a
b
i
l
i
t
y

(

o
u
t

N=1
N=2
N=4
In this gure,
s
= 0dB
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Performance: Outage Probability versus
s
/
40 35 30 25 20 15 10 5 0 5 10
10
4
10
3
10
2
10
1
10
0
s
/ (dB)
F
s
c
s
)
N = 1
N = 2
N = 3
N =4
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Performance: Bit Error Rate
0 2 4 6 8 10 12 14 16 18 20
10
4
10
3
10
2
10
1
10
0
= Average SNR (dB)
B
E
R
SISO
Selection with 4 branches
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Analysis of Selection Combining (cont...)
Q: Are the gains in selection due to gains in SNR?
A: In fact, no!!
SNR analysis:
s
Note, P
out
= P (
out
<
s
) is also the cumulative density
function (CDF) of output SNR
x
the probability density function, f(
out
) = dP
out
/d
out
f(
out
) =
N
out
/
_
1 e
out
/
_
N1
Also, E {
out
} =
_

0
out
f(
out
)d
out
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
SNR Analysis (cont...)
E {
out
} =
N
n=1
1
n
,

_
C + lnN +
1
2N
_
,
s
The gain in SNR is only ln(N)!!!
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
SNR Analysis (cont...)
Q: So, where are the gains coming from?
A: Reduced variation in the channel
10 20 30 40 50 60 70 80 90 100
25
20
15
10
5
0
5
10
Sample #
C
h
a
n
n
e
l

M
a
g
n
i
t
u
d
e

(
d
B
)
No diversity
4 branch selection
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Maximal Ratio Combining
s
Selection is simple, but wastes (N 1) receive elements
s
Maximal Ratio Combining maximizes output SNR
out
x = hs +n
h = [h
1
, h
2
, . . . h
N
]
T
Output Signal = y =
N
n=1
w
n
x
n
= w
H
hs +w
H
n
Output SNR =
out
=
w
H
h
2
E{|s|
2
}
E
_
|w
H
n|
2
_
=
w
H
h
2
E{|s|
2
}
2
||w||
2
MRC: w
MRC
= max
w
[
out
] = max
w
_
w
H
h
2
||w||
2
_
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Maximal Ratio Combing (cont...)
s
Using Cauchy-Schwarz inequality
w h
s
Choose w = h,
Output Signal = y =
_
|h
1
|
2
+|h
2
|
2
+. . . |h
N
|
2
_
s +noise
=
_
N
n=1
|h
n
|
2
_
s +noise
Output SNR =
out
=
N
n=1
E{|s|
2
}|h
n
|
2
2
=
N
n=1
n
s
i.e., the output SNR is the sum of the SNR over all receivers
s
PDF of output SNR:
f(
out
) = f(
1
) f(
2
) f(
N
) =
1
(N 1)!
N1
out
N
e
out
/
,
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
MRC Results
s
Average SNR
out
=
N
n=1
n
E{
out
} = N
s
Outage probability
P
out
= P [
out
<
s
] = 1 e
s
/
N1
n=0
_
_
n
1
n!
s
At high SNR ( )
P
out

_
1
_
N
s
Similarly, bit error rate:
BER =
_

0
[BER/
out
] f(
out
)d
out

_
1
_
N

Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Performance: Outage Probability
40 35 30 25 20 15 10 5 0 5 10
10
4
10
3
10
2
10
1
10
0
s
/ (dB)
F
s
c
s
)
N = 1
N = 2
N = 3
N =4
Outage probability versus
s
/
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Performance: Bit Error Rate
0 2 4 6 8 10 12 14 16 18 20
10
4
10
3
10
2
10
1
10
0
= Average SNR (dB)
B
E
R
SISO
MRC with 4 branches
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Diversity Order
s
A fundamental parameter of diversity-based systems
s
Several times now we have seen that in the high-SNR regime
BER or P
out

_
1
_
N
log(BER)
log
= N (at high SNR)
s
i.e., at high SNR the slope of the curve in a log-log plot is N
x
this is the informal denition of diversity order
x
simulations show that SNR need not be very high for this
to hold
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Diversity Order (cont...)
s
Formal denition: The diversity order , D, is dened as
D = lim
SNR
log BER
log SNR
The diversity order measures the number of independent
paths over which the data is received
s
Can also use P
out
in the denition
s
Diversity order is (formally) a high-SNR concept
s
Provides information of how useful incremental SNR is
s
Sometimes diversity order is abused:
x
the high-SNR denition masks system inefciencies
x
Note: both selection and maximal ratio combining have
the same diversity order
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
The Gains are not due to SNR
10 5 0 5 10 15 20 25 30
10
6
10
5
10
4
10
3
10
2
10
1
10
0
N
B
E
R
N = 1
N = 2
N = 4
No Fading
BER versus output SNR. Note that even though output SNR is
the same, the BER is signicantly different.
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Equal Gain Combining
s
MRC requires matching of both phase and magnitude
x
Magnitude can uctuate by 10s of dB
x
Biggest gains are by the coherent addition
s
Equal Gain Combining: only cancel the phase of the channel
w
n
= e
jh
n
s
And so...
Output Signal = y = w
H
x =
N
n=1
w
n
x
n
= s
_
N
n=1
|h
n
|
_
+noise
s
...resulting in a small loss in SNR...
Average Output SNR =
_
1 + (N 1)
4
_

Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Comparing Diversity Schemes
Bit error rate:
0 2 4 6 8 10 12 14 16 18 20
10
4
10
3
10
2
10
1
= Average SNR (dB)
B
E
R
SISO
MRC
EGC
Note: MRC and EGC have similar performance
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Comparing Diversity Schemes (cont...)
Gains in SNR:
1 2 3 4 5 6 7 8 9 10
0
1
2
3
4
5
6
7
8
9
10
Number of elements (N)
G
a
i
n

i
n

S
N
R

(
d
B
)
Selection
Maximal Ratio
Equal Gain
Note: MRC and EGC have similar performance
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Comparing Diversity Schemes (cont...)
s
All these diversity schemes have same diversity order
s
Work by reducing uctuations in overall channel
s
Selection Combining
x
Simple to implement; only requires power measurement
x
Gain in SNR = ln(N)
s
Maximal Ratio Combining
x
Optimal in SNR sense
x
Gain in SNR = N
x
Requires knowledge of channel and matching over
several 10s of dB
x
Easiest to analyze
s
Equal Gain Combining
x
Small loss w.r.t. MRC
x
Very difcult to analyze, but may be a good trade off for
implementation
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
In Summary...
Diversity is based on providing the receiver with multiple
independent copies of the same signal
s
The key is the independence between the copies of the
same signal
x
The independence makes the gains in error rates
exponential with linear gains in number of elements
So, the question is ....under what circumstances can we
assume independence?
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
The Issue of Correlation
s
Correlation between the received signals reduces the
independence and hence the effective diversity order
s
The extreme case: if all elements were perfectly correlated
(e.g., line of sight conditions), diversity order = 1 (only SNR
gains)
For two receive antennas, with correlation of :
f(
out
) =
1
2||
_
e
out
/(1+||)
e
out
/(1||)
_
P
out
(
s
) = 1
1
2||
_
(1 +||)e
s
/(1+||)
(1 ||)e
s
/(1||)
_
s
Correlation arises because
x
Electromagnetic Mutual Coupling
x
Finite distance between elements
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Impact of correlation : Outage Probability
20 15 10 5 0 5 10
10
5
10
4
10
3
10
2
10
1
10
0
s
/ (dB)
F
s
)
= 0
= 0.25
= 0.50
= 0.75
= 1
Correlation below = 0.5 is considered negligible
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Mutual Coupling
Z
L
Z
L
Z
L
V
oc
= [Z +Z
L
] Z
1
L
V
s
Z: A mutual impedance matrix
s
Z
L
: Diagonal load matrix
s
V
oc
: Open circuits voltages that would arise without mutual
coupling
s
V: True received voltages
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Correlation due to Distance
s
If spacing between elements is d and signal arrives from
direction (, ) only, correlation is given by
(, ) = e
jkd cos sin
,
where k = 2/
s
Total correlation is therefore averaged over angular power
distribution
= E{(, )} =
_

0
_
2
0
e
jkdcossin
f
,
(, )dd,
where f
,
(, ) is the power distribution of the received
signals over all angles
s
So, the angular distribution is crucial
x
Depends on where the receiver is
s
at the mobile or the base station
s
the base station looks down on the mobile
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Correlation at the Mobile
s
The mobile is (usually) surrounded by many scatterers
s
In a dense multipath environment
f
,
(, ) =
1
2
(
0
)
=
_
,
1
2
(
0
)e
jkdcossin
dd
= J
0
(kd sin
0
)
s

0
= /2, < 0.5 if d > 0.24
x
Required distance increases as
0
decreases
x
Rule of thumb: d /2
x
At 1GHz, = 30cm, i.e., received signals independent if
d > 15cm
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Correlation at a Base Station
R
D
Mobile
BS
s
Signal arrives from a small angular region surrounding the
mobile
=
_

max
max
e
jkd cos(
+) sin
0
f
B
()d,
Receive Diversity
qIntroduction
qDiversity Types
qDiversity Order
qComparing Schemes
qSummary
qCorrelation
Transmit Diversity
Course Summary
END
Correlation at a Base Station (cont...)
s
Array along x-axis
s
For a uniform disk of scatterers,

= /2
=
2J
1
(kd sin
max
sin
0
)
kd sin
max
sin
0
.
s
R = 1.2km, D = 50m,
0
= 80
o
< 0.5 for d > 9
x
At 1GHz, received signals independent if d > 2.7m
(approx. 9ft.)
The required distance is therefore determined by the array
setting
Receive Diversity
Transmit Diversity
qTransmit Diversity
qSpace-Time Coding
qAlamoutis Scheme
qDesign Criteria
qExamples of Codes
qMultiplexing
qSummary
Course Summary
END
Transmit Diversity
s
So far, we had a SIMO situation: a single transmitter and
multiple receivers
x
Achieving diversity was relatively easy: each receiver
receives a copy of the transmitted signal
s
multiple receivers multiple copies
s
What about the MISO situation?
x
Very useful in the expected asymmetrical communication
scenarios with more trafc from base station to mobile
s
Base station is expensive, has more space, has multiple
antennas
s
Mobile is cheap, has little space, has one antenna
s
Users are downloading information, e.g., a webpage
Receive Diversity
Transmit Diversity
qTransmit Diversity
qSpace-Time Coding
qAlamoutis Scheme
qDesign Criteria
qExamples of Codes
qMultiplexing
qSummary
Course Summary
END
Transmit Diversity (cont...)
s
MISO: N transmit antennas, one receive antenna
s
Transmit diversity requires the time dimension. To see this,
consider if we did not use the time dimension. Each transmit
antenna transmits symbol s.
Received Signal = x =
N
n=1
h
n
s +noise
= hs +n
where h =
N
n=1
h
n
s
Received signal is a scalar, there is no diversity here!!
x
And hence the concept of space-time coding
Receive Diversity
Transmit Diversity
qTransmit Diversity
qSpace-Time Coding
qAlamoutis Scheme
qDesign Criteria
qExamples of Codes
qMultiplexing
qSummary
Course Summary
END
Space-Time Coding
s
Simple example: Transmit the same symbol over two time
slots (symbol periods)
x
On time slot 1, antenna n = 1 transmits symbol s
x
1
= h
1
s +n
1
x
On time slot 2, antenna n = 2 transmits the same symbol s
x
2
= h
2
s +n
2
s
At the receiver form a receive vector over the two time slots
x =
_
x
1
x
2
_
=
_
h
1
h
2
_
s +
_
n
1
n
2
_
= hs +n
s
Maximum Ratio Combining:
y = h
H
x =
_
|h
1
|
2
+|h
2
|
2
_
s +noise
and we would get order-2 diversity
Receive Diversity
Transmit Diversity
qTransmit Diversity
qSpace-Time Coding
qAlamoutis Scheme
qDesign Criteria
qExamples of Codes
qMultiplexing
qSummary
Course Summary
END
The (Famous) Alamoutis Code
s
The previous scheme wastes half the time
s
A more efcient approach: consider two symbols s
1
and s
2
s
In the rst time slot,
antenna n = 1 transmits s
1
and antenna n = 2 transmits s
2
x
1
= h
1
s
1
+h
2
s
2
+n
1
s
In the second time slot,
antenna n = 1 transmits s
2
while antenna n = 2 transmits s
1
x
2
= h
1
s
2
+h
2
s
1
+n
2
s
One subtle, but important point: each element transmits with
half the available power
s
Form the received data vector (note the conjugate on x
2
)
x =
_
x
1
x
2
_
=
_
h
1
h
2
h
2
h
1
__
s
1
s
2
_
+
_
n
1
n
2
_
Receive Diversity
Transmit Diversity
qTransmit Diversity
qSpace-Time Coding
qAlamoutis Scheme
qDesign Criteria
qExamples of Codes
qMultiplexing
qSummary
Course Summary
END
Alamouti (cont...)
Now work with x
x = H
_
s
1
s
2
_
+n
y = H
H
x = H
H
H
_
s
1
s
2
_
+noise
=
_
|h
1
|
2
+|h
2
|
2
0
0 |h
1
|
2
+|h
2
|
2
__
s
1
s
2
_
+ noise
i.e.,
y
1
=
_
|h
1
|
2
+|h
2
|
2
_
s
1
+noise
y
2
=
_
|h
1
|
2
+|h
2
|
2
_
s
2
+noise
we get order-2 diversity on both symbols!!
The key is that the effective channel matrix is orthogonal
Receive Diversity
Transmit Diversity
qTransmit Diversity
qSpace-Time Coding
qAlamoutis Scheme
qDesign Criteria
qExamples of Codes
qMultiplexing
qSummary
Course Summary
END
Alamouti Code : Performance
0 2 4 6 8 10 12 14 16 18 20
10
4
10
3
10
2
10
1
= Average SNR (dB)
B
E
R
SISO
Receive Diversity
Alamouti Scheme
Note: the diversity order of the Alamouti scheme and 2-branch
receive diversity is the same. Alamoutis scheme suffers a 3dB
loss because of the power splitting
Receive Diversity
Transmit Diversity
qTransmit Diversity
qSpace-Time Coding
qAlamoutis Scheme
qDesign Criteria
qExamples of Codes
qMultiplexing
qSummary
Course Summary
END
Code Design Criteria: What Makes a Code Good?
s
Alamoutis scheme is a space-time code
x
A careful organization of data (or a function of the data) in
space and time
s
Alamoutis scheme is specic to two transmit antennas...
x
...and, unfortunately, cannot be generalized to N > 2
s
Consider a code with K symbols over N antennas and L
time slots (code rate R = K/L)
C =
_
_
c
11
c
12
c
1L
c
21
c
22
c
2L
.
.
.
.
.
.
.
.
.
.
.
.
c
N1
c
N2
c
NL
_
_
s
here, c
nl
is a function of the K symbols transmitted in time
slot l using antenna n
x
e.g., in the Alamouti scheme, c
12
= s
2
Receive Diversity
Transmit Diversity
qTransmit Diversity
qSpace-Time Coding
qAlamoutis Scheme
qDesign Criteria
qExamples of Codes
qMultiplexing
qSummary
Course Summary
END
Design Criteria (cont...)
s
We want to minimize error rate, the probability that codeword
C was transmitted and

C was decoded
P(C

C) exp
_
d
2
_
C,

C
_
E
s
4
2
_
s
E
s
is the available energy, = E
s
/
2
s
d
_
C,

C
_
is the effective distance between C and

C
d
2
_
C,

C
_
= h
H
EE
H
h
E =
_
_
c
11
c
11
c
12
c
12
c
1L
c
1L
c
21
c
21
c
22
c
22
c
2L
c
2L
.
.
.
.
.
.
.
.
.
.
.
.
c
N1
c
N1
c
N2
c
N2
c
NL
c
NL
_
_
s
h = [h
1
, h
2
, . . . h
N
]
T
is the channel vector
Receive Diversity
Transmit Diversity
qTransmit Diversity
qSpace-Time Coding
qAlamoutis Scheme
qDesign Criteria
qExamples of Codes
qMultiplexing
qSummary
Course Summary
END
Design Criteria (cont...)
E =
_
_
c
11
c
11
c
12
c
12
c
1L
c
1L
c
21
c
21
c
22
c
22
c
2L
c
2L
.
.
.
.
.
.
.
.
.
.
.
.
c
N1
c
N1
c
N2
c
N2
c
NL
c
NL
_
_
s
Tarokh et al. developed two design criteria based on this
error matrix:
x
The Rank Criterion: The maximum diversity order is
achieved if the rank of the error matrix is maximized (N)
s
If M receiving antennas, total diversity order available is
NM
x
The Determinant Criterion: The error rate is minimized if
the determinant of EE
H
is maximized over all code pairs
C,

C
Receive Diversity
Transmit Diversity
qTransmit Diversity
qSpace-Time Coding
qAlamoutis Scheme
qDesign Criteria
qExamples of Codes
qMultiplexing
qSummary
Course Summary
END
Examples of Space-Time Codes
s
Space-Time Trellis Codes: Design a trellis code over space
and time
s
Orthogonal Block Codes: Codes in which each symbol can
be independently decoded
x
Independent decoding is very convenient
x
Alamoutis code is orthogonal for N = 2
x
Rate-1 orthogonal codes cannot exist for N > 2
x
Rate-1/2 orthogonal codes are always available
s
rate-3/4 codes are available for N = 3, 4, e.g.,
G
3
=
_
_
_
_
_
s
1
s
2
s
3
s
2
s
1
s
4
s
3
s
4
s
1
s
4
s
3
s
2
_
_
_
_
_
,
Receive Diversity
Transmit Diversity
qTransmit Diversity
qSpace-Time Coding
qAlamoutis Scheme
qDesign Criteria
qExamples of Codes
qMultiplexing
qSummary
Course Summary
END
Examples of Codes (cont...)
s
Linear Dispersion Codes: Minimize error based on mutual
information directly, not the design criteria
s
Codes based on algebra: Algebraic codes, e.g., TAST etc.
s
In all cases, space-time codes provide diversity by giving the
receiver independent copies of the same message
So far we have focused on diversity order (reliability) only.
What about data rate?
Receive Diversity
Transmit Diversity
qTransmit Diversity
qSpace-Time Coding
qAlamoutis Scheme
qDesign Criteria
qExamples of Codes
qMultiplexing
qSummary
Course Summary
END
Space-Time Coding and Multiplexing
s
Transmit more than one data stream (multiplexing)
x
Requires multiple receive antennas as well
s
Instead of transmitting only a single data stream, transmit Q
data streams in parallel.
s
M receive, N transmit antennas. Divide the transmit
antennas into Q groups, N = N
1
+N
2
+. . . N
Q
x
Data stream q uses N
q
antennas
x = Hc +n,
=
_
_
h
11
h
12
h
1N
1
h
1(N
1
+1)
h
1N
h
21
h
22
h
2N
1
h
2(N
1
+1)
h
2N
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
h
M1
h
M2
h
MN
1
h
M(N
1
+1)
h
MN
_
_
_
_
c
1
c
2
.
.
.
c
Q
_
_
+n
Receive Diversity
Transmit Diversity
qTransmit Diversity
qSpace-Time Coding
qAlamoutis Scheme
qDesign Criteria
qExamples of Codes
qMultiplexing
qSummary
Course Summary
END
STC and Multiplexing (cont...)
s
The q
th
data stream uses a space-time code over N
q
antennas
s
How do you isolate individual streams?
x = H
1
c
1
+

H
1
_
_
c
2
c
3
.
.
.
c
Q
_
_
+n
s
Now, let H
1
be the null space of

H
1
(H
H
1

H
1
= 0)
x
this is possible if M > N N
1
. H
1
is size
M (M N +N
1
)
y
1
= H
H
1
x =
_
H
H
1
H
1
c
1
+noise
which is a space-time coded system with N
1
transmitters
and (M N +N
1
) receivers
x
Diversity order = N
1
(M N +N
1
)
Receive Diversity
Transmit Diversity
qTransmit Diversity
qSpace-Time Coding
qAlamoutis Scheme
qDesign Criteria
qExamples of Codes
qMultiplexing
qSummary
Course Summary
END
Multiplexing (cont...)
s
Clearly we can apply the same idea for q = 2, . . . , Q
x
Data stream q would achieve diversity order of
N
q
(M N +N
q
).
s
Can we do better? Yes!! Use interference cancellation...
x
...since c
1
has been decoded, subtract it!
x
Data stream 2 sees less interference (N
1
interfering
transmissions are eliminated). Stream 2 can get diversity
order of N
2
(M N +N
2
+N
1
)
x
Similarly, the q
th
data stream can achieve diversity order
of N
q
_
M N +
q
p=1
N
p
_
s
BLAST: Bell Labs Layered Space-Time
x
N
q
= 1
x
Requires M N (at least as many receivers as
transmitters)
x
Achieved (in lab) spectral efciency of 10s of b/s/Hz!
Receive Diversity
Transmit Diversity
qTransmit Diversity
qSpace-Time Coding
qAlamoutis Scheme
qDesign Criteria
qExamples of Codes
qMultiplexing
qSummary
Course Summary
END
In Summary...
s
Transmit diversity requires the time dimension
x
Space-time coding is the careful arrangement of data (or
function of data) in space and time to achieve
s
Greatest diversity order (rank criterion)
s
Minimum error rate (determinant criterion)
s
Several space-time code families are available
x
We focused on the simplest family of orthogonal
space-time block codes
s
Can also use the spatial degrees of freedom to multiplex
x
BLAST is one (famous) example
Receive Diversity
Transmit Diversity
qBasics
qUnknown Channel
qKnown Channel
qWaterlling
qMIMO Systems
qExamples
qSummary
qDMT
Course Summary
END
s
We wish to investigate the fundamental limits of data transfer
rate in MIMO wireless systems
s
Remember:
x
A channel is fundamentally characterized by (and the data
rate limited by) its capacity C
x
In the SISO case,
C = log
2
(1 +SNR)
s
Our MIMO system: N transmit and M receive antennas
y = Hx +n
s
y: the length-M received signal vector
s
H: the M N channel
s
x: The length-N transmit data vector
x
Let S
x
= E{xx
H
} be the covariance matrix of x
Receive Diversity
Transmit Diversity
qBasics
qUnknown Channel
qKnown Channel
qWaterlling
qMIMO Systems
qExamples
qSummary
qDMT
Course Summary
END
Channel Unknown at Transmitter
s
It is not too hard to show
C = log
2
det
_
I +
1
2
HS
x
H
H
_
s
If channel H is known at the transmitter, S
x
can be chosen to
best match the channel
s
However, lets start with the case that the channel is not
known. The best choice is
S
x
=
E
s
N
I
NN
C = log
2
det
_
I +
1
2
E
s
N
HH
H
_
s
Since H is not known, again, one cannot guarantee a data
rate and the true capacity is zero!
x
Again, talk of an outage probability and/or expected
capacity
Receive Diversity
Transmit Diversity
qBasics
qUnknown Channel
qKnown Channel
qWaterlling
qMIMO Systems
qExamples
qSummary
qDMT
Course Summary
END
Unknown Channel (cont...)
C = log
2
det
_
I +
1
2
E
s
N
HH
H
_
s
Since we are assuming Rayleigh fading, the entries of H are
complex Gaussian
x
Experts in STAP will recognize HH
H
as following the
Wishart distribution
s
Let the eigenvalues of HH
H
be
2
m
, m = 1, 2, . . . , M
Eigenvalues of
_
I +E
s
/(N
2
)HH
H
=
_
1 +
E
s
(N
2
)
2
m
_
Receive Diversity
Transmit Diversity
qBasics
qUnknown Channel
qKnown Channel
qWaterlling
qMIMO Systems
qExamples
qSummary
qDMT
Course Summary
END
Unknown Channel (cont...)
C = log
2
M
m=1
_
1 +
E
s
N
2
2
m
_
=
M
m=1
log
2
_
1 +
E
s
N
2
2
m
_
s
The distribution of
m
is known (HH
H
is Wishart)
x
Without ordering, these eigenvalues are independent and
identically distributed
x
There are r = min(M, N) eigenvalues
s
Therefore, on average
E{C} = E
{
m
}
_
M
m=1
log
2
_
1 +
E
s
N
2
2
m
_
_
= min(N, M)E
_
log
2
_
1 +
E
s
N
2
2
__
s
We get linear gains in capacity, not just power gains
x
As if we have min(N, M) parallel channels!
Receive Diversity
Transmit Diversity
qBasics
qUnknown Channel
qKnown Channel
qWaterlling
qMIMO Systems
qExamples
qSummary
qDMT
Course Summary
END
Unknown Channel: Ergodic Capacity
Note: Ergodic capacity for xed SNR (from Telatar (1999)).
Here r is the number of elements in the transmitter and
receiver (r = M = N)
Receive Diversity
Transmit Diversity
qBasics
qUnknown Channel
qKnown Channel
qWaterlling
qMIMO Systems
qExamples
qSummary
qDMT
Course Summary
END
Unknown Channel: 1-(Outage Probability)
Note: Success rates (probability that capacity is above target)
from Foschini and Gans (1998). Comparing success rates for
SISO and a N = M = 2 system.
Receive Diversity
Transmit Diversity
qBasics
qUnknown Channel
qKnown Channel
qWaterlling
qMIMO Systems
qExamples
qSummary
qDMT
Course Summary
END
Channel Known at the Transmitter
s
What if the channel is known at the transmitter?
x
Usually obtained via feedback (frequency division duplex -
FDD) or via reciprocity (TDD)
x
The transmitter can tune the covariance matrix S
x
to
match the transmitter
x
This may be via power allocation
S
opt
x
= max
S
x
C
= max
S
x
log
2
det
_
I +
1
2
HS
x
H
H
_
Constraints: S
x
must be positive-denite
S
x
must satisfy a power constraint
s
So, how do you optimize over a matrix?
x
Lets start with a simpler (and very instructive) system:
parallel channels
Receive Diversity
Transmit Diversity
qBasics
qUnknown Channel
qKnown Channel
qWaterlling
qMIMO Systems
qExamples
qSummary
qDMT
Course Summary
END
Parallel Channels
1
2
3
N
s
Each channel is independent of the other
s
On the n
th
channel
y
n
= h
n
x
n
+noise
s
The transmitter has one important constraint - a total
available energy constraint of E
s
x
Since the transmitter knows the channel values,
h
n
, n = 1, 2, . . . , N it can allocate power to maximize the
overall capacity
Receive Diversity
Transmit Diversity
qBasics
qUnknown Channel
qKnown Channel
qWaterlling
qMIMO Systems
qExamples
qSummary
qDMT
Course Summary
END
Parallel Channels (cont...)
s
The transmitter allocates power E
n
to channel n
C =
N
n=1
log
2
_
1 +|h
n
|
2
E
n
2
_
s
Intuitively, the transmitter should allocate all its power to the
strongest channel, right?
x
Strangely enough, wrong!
s
This is because...
C = log (1 +SNR)
s
At high SNR, C log (SNR)
s
At low SNR, C SNR
x
There are diminishing marginal returns in allocating power
Receive Diversity
Transmit Diversity
qBasics
qUnknown Channel
qKnown Channel
qWaterlling
qMIMO Systems
qExamples
qSummary
qDMT
Course Summary
END
Parallel Channels (cont...)
The problem formulation:
_
E
opt
n
_
= max
{E
n
}
N
n=1
log
2
_
1 +
E
n
|h
n
|
2
2
n
_
N
n=1
E
n
E
s
E
n
0
The solution:
_

2
|h
n
|
2
+E
n
_
= , n = 1, . . . , N
E
n
=
_

2
|h
n
|
2
_
+
,
where (x)
+
= 0 if x < 0 and (x)
+
= x if x 0
Receive Diversity
Transmit Diversity
qBasics
qUnknown Channel
qKnown Channel
qWaterlling
qMIMO Systems
qExamples
qSummary
qDMT
Course Summary
END
Waterlling
s
Note that is a constant
s
Channel sees an effective noise variance of
2
n
=
2
/|h
n
|
2
Power Level: Difference
Channels 1 to N
s
Note that the better channels do get more power
s
Some channels are so bad that they do not get any power
Receive Diversity
Transmit Diversity
qBasics
qUnknown Channel
qKnown Channel
qWaterlling
qMIMO Systems
qExamples
qSummary
qDMT
Course Summary
END
MIMO Systems with Known Channel
s
So far, we have focused on parallel channels.
So far, y =
_
_
h
1
0 0
0 h
2
0
.
.
.
.
.
.
.
.
.
.
.
.
0 0 h
N
_
_
x +n
s
What does this tell us about a regular MIMO system?
s
We have
y = Hx +n
s
N transmitters, M receivers, H is the M N channel
H =
_
_
h
11
h
12
h
1N
h
21
h
22
h
2N
.
.
.
.
.
.
.
.
.
.
.
.
h
M1
h
M2
h
MN
_
_
Receive Diversity
Transmit Diversity
qBasics
qUnknown Channel
qKnown Channel
qWaterlling
qMIMO Systems
qExamples
qSummary
qDMT
Course Summary
END
MIMO Systems (cont...)
s
One can use the singular value decomposition of H
H = UV
H
s
U is the matrix of eigenvectors of HH
H
s
V is the matrix of eigenvectors of H
H
H
s
is a diagonal M N matrix of singular values
=
_
1
0 0 0 0
0
2
0 0 0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0 0
R
0 0
0 0 0 0 0
0 0 0 0 0
_
_
,
s
R is the rank of H
s
The zeros pad the matrix to match the M N dimensions
Receive Diversity
Transmit Diversity
qBasics
qUnknown Channel
qKnown Channel
qWaterlling
qMIMO Systems
qExamples
qSummary
qDMT
Course Summary
END
s
How does this help? We have
UU
H
= U
H
U = I
MM
VV
H
= V
H
V = I
NN
y = Hx +n
= UV
H
x +n
U
H
y = V
H
x +U
H
n
s
y = U
H
y, x = V
H
x, n = U
H
n
y = x + n
s
We have a set of R parallel channels!
Receive Diversity
Transmit Diversity
qBasics
qUnknown Channel
qKnown Channel
qWaterlling
qMIMO Systems
qExamples
qSummary
qDMT
Course Summary
END
Therefore,
Channel
H
S/P
Routputs
V
N-R zeros
N outputs
U
M inputs
H
Decoder
x
~
y
~
x
y
data
data
Power
Allocation
Routputs
s
The transmitter precodes the transmitted signal using matrix
V. This matches the transmission to the eigen-modes of
the channel
s
The transmitter also waterlls over the singular values
n
as
the equivalent channel values as parallel channels (called h
n
earlier)
s
The receiver decodes using the matrix U
s
IMPORTANT: This diagonalization process (transmission on
eigen-modes is a fundamental concept in wireless
communications
Receive Diversity
Transmit Diversity
qBasics
qUnknown Channel
qKnown Channel
qWaterlling
qMIMO Systems
qExamples
qSummary
qDMT
Course Summary
END
Improvement in Capacity
0 2 4 6 8 10 12 14 16 18 20
3
4
5
6
7
8
9
10
11
12
= Average SNR (dB)
C
a
p
a
c
i
t
y

(
b
i
t
s
/
c
h
a
n
n
e
l

u
s
e
)
Without Waterfilling
With Waterfilling
Comparing capacities for N = M = 4
Receive Diversity
Transmit Diversity
qBasics
qUnknown Channel
qKnown Channel
qWaterlling
qMIMO Systems
qExamples
qSummary
qDMT
Course Summary
END
Some Illustrative Examples
s
Example 1: SIMO System, 1 transmitter, M receivers
H = h =
_
_
h
1
h
2
.
.
.
h
M
_
_
U =
_
h
||h||
h
_
V = [1]
= [||h||, 0, . . . , 0]
T
s
Note that we have only a single parallel channel
C = log
2
_
1 +
E
s
2
||h||
2
_
s
Effectively, all channel powers added together
Receive Diversity
Transmit Diversity
qBasics
qUnknown Channel
qKnown Channel
qWaterlling
qMIMO Systems
qExamples
qSummary
qDMT
Course Summary
END
Some Illustrative Examples (cont...)
s
Example 2: MISO System, N transmitters, 1 receiver
H = h = [h
1
, h
2
, . . . , h
N
]
U = [1]
V =
_
h
||h||
h
_
U = [1]
= [||h||, 0, . . . , 0]
s
Again we have only a single parallel channel
C = log
2
_
1 +
E
s
2
||h||
2
_
s
This is the same as the SIMO case!
s
Note: without channel knowledge,
C = log
2
_
1 +
E
s
N
2
||h||
2
_
Receive Diversity
Transmit Diversity
qBasics
qUnknown Channel
qKnown Channel
qWaterlling
qMIMO Systems
qExamples
qSummary
qDMT
Course Summary
END
s
Example 3: MIMO System, line of sight conditions
s(
r
) =
_
1, z
r
, z
2
r
, . . . , z
M1
r
T
, z
r
= e
jkd
r
cos
r
s(
t
) =
_
1, z
t
, z
2
t
, . . . , z
N1
t
T
, z
t
= e
jkd
t
cos
t
H = s(
r
) s
T
(
t
)
=
_
_
1 z
t
z
2
t
z
N1
t
z
r
z
r
z
t
z
r
z
2
t
z
r
z
N1
t
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
z
M1
r
z
M1
r
z
t
z
M1
r
z
2
t
z
M1
r
z
N1
t
_
_
s
This is a rank-1 matrix!
x
This one singular value =
1
=
NM
C = log
2
_
1 +
E
s
N
2
NM
_
= log
2
_
1 +
E
s
2
M
_
Receive Diversity
Transmit Diversity
qBasics
qUnknown Channel
qKnown Channel
qWaterlling
qMIMO Systems
qExamples
qSummary
qDMT
Course Summary
END
s
Example 4: MIMO System, M = N, rich scattering
conditions, full rank channel
x
We have N parallel channels
s
For convenience, assume all singular values are equal
x
Let this singular value =
1
x
Since all parallel channels are equally powerful, power
allocation is uniform
C =
N
n=1
log
2
_
1 +
E
s
N
2
2
1
_
= N log
2
_
1 +
E
s
N
2
2
1
_
s
Note the huge difference from line of sight scenario
x
The number of transmit or receive elements is outside the
log term; we get linear gains in capacity
If N = M C = min(N, M) log
2
_
1 +
E
s
N
2
2
1
_
Receive Diversity
Transmit Diversity
qBasics
qUnknown Channel
qKnown Channel
qWaterlling
qMIMO Systems
qExamples
qSummary
qDMT
Course Summary
END
Summary of Information Theoretic Analysis
s
MIMO systems allow for huge increases in capacity
x
If the fading is independent then one can achieve linear
gains in capacity over the SISO case
x
Notice that this inherently requires the concept of a
diversity of paths
s
The concept of diagonalization or transmission on
eigen-channels and the associated concept of waterlling
are fundamental
s
Waterlling allocates more power to better channels
x
Note that this is, initially, counter-intuitive. Generally, if we
have a poor channel, we add power, not reduce power
Receive Diversity
Transmit Diversity
qBasics
qUnknown Channel
qKnown Channel
qWaterlling
qMIMO Systems
qExamples
qSummary
qDMT
Course Summary
END
Diversity-Multiplexing Tradeoff
Consider a system with N transmit and M receive antennas in
a rich scattering, Rayleigh fading environment
s
Diversity: We have seen that through space-time coding and
receive diversity we can achieve a diversity order of NM.
s
Multiplexing: In the information theoretic analysis we saw
that we could get a pre-log factor of min(M, N). Also, at high
SNR
C min(M, N) log
2
(SNR)
Q: Can we get both diversity and multiplexing (rate) gains?
A: Yes! But, there is a trade-off between the two!
Receive Diversity
Transmit Diversity
qBasics
qUnknown Channel
qKnown Channel
qWaterlling
qMIMO Systems
qExamples
qSummary
qDMT
Course Summary
END
DMT (cont...)
s
As before, dene the diversity order as:
D = lim
SNR
_
log P
out
log SNR
_
x
D tells us how fast the error rate falls with increases in
log(SNR)
s
Dene a multiplexing gain r as
r = lim
SNR
_
R
log SNR
_
x
R is the rate of transmission
x
r is the rate at which the transmission rate increases with
log(SNR)
Receive Diversity
Transmit Diversity
qBasics
qUnknown Channel
qKnown Channel
qWaterlling
qMIMO Systems
qExamples
qSummary
qDMT
Course Summary
END
DMT (cont...)
s
Diversity Multiplexing Tradeoff: The optimal tradeoff curve,
d
(r) is given by the piecewise-linear function connecting the

points (r, d
(r)), r = 0, 1, . . . , min(M, N), where

d
(r) = (M r)(N r)
s
Note: d
max
= MN and r
max
= min(M, N).
s
At integer points, r degrees of freedom are used for
multiplexing, the rest are available for diversity
Note the linear interpolation

between integer points
Receive Diversity
Transmit Diversity
Course Summary
qCourse Summary
qWhats Left?
END
Course Summary
s
We have explored uses of the spatial dimension in wireless
communications
s
Performance of wireless systems is fundamentally limited by
fading
x
Fading makes the received signal to be a random copy of
the transmitted signal
x
MIMO systems are based on independent fading to/from
multiple elements
s
We covered three major concepts:
x
Receive Diversity
x
Transmit Diversity
x
s
Receive Diversity:
x
Selection, maximal ratio and equal gain combining
s
Trade off between complexity and performance
Receive Diversity
Transmit Diversity
Course Summary
qCourse Summary
qWhats Left?
END
Course Summary (cont...)
s
Receive Diversity (cont...)
x
Notion of diversity order
s
Measures the number of independent paths the signal is
received
x
Impact of correlation
s
Array elements must be some minimum distance apart
s
The key is to create independence
s
Transmit Diversity:
x
Requires the time dimension: space-time coding
x
An important example of block codes: Alamoutis Code
s
Unfortunately, only rate 1/2 codes are guaranteed for
complex data constellations
x
Good codes designed using the rank and determinant
criteria
Receive Diversity
Transmit Diversity
Course Summary
qCourse Summary
qWhats Left?
END
Course Summary (cont...)
s
Multiplexing:
x
Transmitting multiple data streams simultaneously
x
Can combine space-time coding with multiplexing
s
A famous example: the BLAST scheme
s
We should emphasize that this is only one possible
multiplexing scheme
s
MIMO Information Theory:
x
Capacity via determinant of channel
x
Ergodic capacity and outage probability if channel is
unknown
x
Transmission on eigen-channels and using waterlling if
channel is known to transmitter
s
Creates r = rank (H) parallel channels
x
There is a fundamental trade-off between diversity and
multiplexing
Receive Diversity
Transmit Diversity
Course Summary
qCourse Summary
qWhats Left?
END
We could have done so much more...
s
Channel Estimation: Throughout we assumed the receiver
knows the channel
x
The receiver has to estimate and track a time varying
channel
s
Frequency selective channels: We focused on at channels,
frequency selectivity is becoming important
x
Everyone assumes 4G will be based on OFDM -
MIMO-OFDM is a hot topic
s
Error control coding to achieve the capacity of MIMO
systems
s
Feedback to inform the transmitter of the channel
x
Low data rate schemes, error bounds, impact of error
Receive Diversity
Transmit Diversity
Course Summary
qCourse Summary
qWhats Left?
END
What else? (cont...)
s
Multiuser Communications: Transmitting to/receiving from
multiple users simultaneously
x
Is another form of multiplexing; requires interference
cancellation, provides exibility
x
Is getting more important, both theoretically and in
implementation
s
Cooperative Communications:
x
Probably the hottest research area now
x
Nodes with a single antenna share resources to act like a
MIMO system
s
via relaying, forwarding, cooperative diversity
x
Especially applicable to the new modern kinds of
networks. Also, distributed signal processing.
s
Sensor Networks: Large scale networks of small, cheap
nodes
s
Mesh Networks: Networks of access points
Receive Diversity
Transmit Diversity
Course Summary
END
END
That all folks!
Again, detailed notes available at
http://www.comm.utoronto.ca/ rsadve/teaching.html

MIMOSystems

Загружено:

Сведения о документе

Исходное описание:

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

MIMOSystems

Загружено:

Авторское право:

Доступные форматы

Waveform Diversity and Design Conference, Slide #1

Short Course on MIMO Systems

Introduction and Overview

Introduction and Overview

Introduction and Overview

for N = 1, i.e.,exponential gains in

Introduction and Overview

(r) is given by the piecewise-linear function connecting the

(r)), r = 0, 1, . . . , min(M, N), where

Note the linear interpolation

Вам также может понравиться