Beruflich Dokumente
Kultur Dokumente
X
2
+ Y
2
is a Rayleigh r.v. if X and Y are independant Gaussian r.v.
with mean 0 and variance
2
Then the pdf of R is
f
R
(r ) =
r
2
exp
r
2
2
2
.
Luc Deneire 4
MISO/SIMO : (Spatial) Diversity
Diversity best illustrated on Rayleigh Fading Channel
MISO/SIMO : Spatial Diversity
Rayleigh Fading Channel
y = h.x +
where
SNR
_
.
||h||
y = ||h||x +
h
||h||
l =1
|h
l
|
2
is a Chi-square distributed r.v. with 2.L degrees of
freedom, whose density is given by f (x) =
1
(L1)!
x
L1
e
x
, x 0.
After computation, the Error Probability is computed as :
BEP =
_
2L 1
L
_
1
(4.SNR)
L
Luc Deneire 8
MISO/SIMO : (Spatial) Diversity
Diversity best illustrated on Rayleigh Fading Channel
MISO/SIMO : Spatial Diversity
Bit Error Probability on AWGN and Rayleigh Fading Channel
Luc Deneire 9
MISO/SIMO : (Spatial) Diversity
Diversity best illustrated on Rayleigh Fading Channel
MISO/SIMO : Spatial Diversity
Capacity of a SISO channel
Systme SISO
TX RX
h
h
x
n
y
S
E
C = log
2
(1 + ||h||
2
) en Sh/2D
avec =
E
s
N
0
.
Pour le canal de Rayleigh, la capacit est une variable alatoire
CNAM Cours ELE 203. p.14/77
C = log
2
(1 +||h||
2
SNR) bits/s/Hertz/dimension
Luc Deneire 10
MISO/SIMO : (Spatial) Diversity
Beam-Forming / Beam-Steering
MISO/SIMO : Spatial Diversity
Capacity of a MISO channel
Systme MISO
TX RX
TX 1
TX N
t
TX 2
RX 1
CNAM Cours ELE 203. p.15/77
Systme MISO
1
h
1
x
n
y
S
t
E
N
2
h
2
x
Nt
h
Nt
x
S
t
E
N
S
t
E
N
C = log
2
(1 +
N
t
N
t
i=1
||h
i
||
2
)
CNAM Cours ELE 203. p.16/77
C = log
2
(1 +
SNR
N
t
N
t
l
||h
l
||
2
) bits/s/Hertz/dimension
Luc Deneire 11
MISO/SIMO : (Spatial) Diversity
Beam-Forming / Beam-Steering
MISO/SIMO : Spatial Diversity
Beam Steering / Beamforming as a MISO channel
instant
=
E
s
(
N
r
i=1
||h
i
||
2
)
2
N
0
(
N
r
i=1
||h
i
||
2
)
=
E
s
(
N
r
i=1
||h
i
||
2
)
N
0
=
N
r
i=1
||h
i
||
2
C = log
2
(1 +
N
r
i=1
||h
i
||
2
)
CNAM Cours ELE 203. p.18/77
Considering a MRC (Maximum Ratio Combining)
C = log
2
(1 + SNR
N
t
l
||h
l
||
2
) bits/s/Hertz/dimension
Luc Deneire 13
MISO/SIMO : (Spatial) Diversity
SIMO : Multiple antenna receiver
MISO/SIMO : Spatial Diversity
Example of a Space-Time SIMO Receiver operation
24 CHAPITRE 3. CONTRIBUTIONDTAILLE
E(z)
Receiver
t=0
t=
FIG. 3.5 La multiplication du signal reu par une matrice polynomiale permet de combiner de faon optimale
les diffrentes rpliques du signal mis.
0 4 8 12 16 20 24 28 32 36 40
!7
10
!6
10
!5
10
!4
10
!3
10
!2
10
!1
10
0
10
0 4 8 12 16 20 24 28 32 36 40
!7
10
!6
10
!5
10
!4
10
!3
10
!2
10
!1
10
0
10
0 4 8 12 16 20 24 28 32 36 40
!7
10
!6
10
!5
10
!4
10
!3
10
!2
10
!1
10
0
10
0 4 8 12 16 20 24 28 32 36 40
!7
10
!6
10
!5
10
!4
10
!3
10
!2
10
!1
10
0
10
SER
AWGN
Div : 4 Div : 2
Div : 1
Performance of QPSK in multipath Rayleigh fading channel
SNR per symbol
0 4 8 12 16 20 24 28 32 36 40
!7
10
!6
10
!5
10
!4
10
!3
10
!2
10
!1
10
0
10
0 4 8 12 16 20 24 28 32 36 40
!7
10
!6
10
!5
10
!4
10
!3
10
!2
10
!1
10
0
10
0 4 8 12 16 20 24 28 32 36 40
!7
10
!6
10
!5
10
!4
10
!3
10
!2
10
!1
10
0
10
0 4 8 12 16 20 24 28 32 36 40
!7
10
!6
10
!5
10
!4
10
!3
10
!2
10
!1
10
0
10
SER
AWGN
Div : 4 Div : 2
Div : 1
Performance of QPSK in multipath Rayleigh fading channel
SNR per symbol
FIG. 3.6 Le gain de diversit est la pente de la courbe du SER vs SNR (dans une chelle logarithmique)
Luc Deneire 14
MIMO : Spatial Diversity and/or Spatial Multiplexing
MIMO : Multiple antenna emitter and receiver
MIMO: Spatial Diversity and/or Spatial Multiplexing
Simple model of the MIMO channel
3.3. MULTIPLEINPUTMULTIPLEOUTPUT(MIMO) : DELADIVERSITAUMULTIPLEXAGESPATIAL25
3.3 Multiple Input Multiple Output (MIMO) : de la diversit au multiplexage
spatial
La diversit spatio-temporelle introduite ci-dessous se base implicitement sur une diversit au rcepteur. tant
donn le cot matriel induit par les antennes multiples (entre autre la duplication des front-ends analogiques),
il tait naturel dintroduire de la diversit lmetteur dans le cadre des communications cellulaires (les an-
tennes multiples tant alors introduites dans les stations de base). La diversit dmission [67, 75, 99] consiste
simplement mettre une version diffrente des donnes sur chaque antenne mettrice, charge pour le rcep-
teur de rcuprer les donnes utiles en sparant les signaux mis et en les combinant de la meilleure faon pos-
sible. Cette diversit dmission est principalement le domaine dapplication du codage spatio-temporel [95].
Enn, en combinant la diversit lmission et la rception, on arrive aux systmes MIMO, qui utilisent un
rseau dantennes lmission et la rception, multipliant ainsi le gain de diversit (gure 3.7).
T
X
H
X
R
FIG. 3.7 Un systme MIMO sans ls : exploitation de la diversit dmission et de rception.
Un autre manire dexploiter la structure des systmes MIMO est de transmettre des signaux de donnes
diffrents sur chaque antenne dmission et de rcuprer ces signaux sur le rseau du rcepteur, introduisant
donc un multiplexage spatial(gure 3.8).
T
X
H
X
R S
/
P
P
/
S
FIG. 3.8 Un systme MIMO sans ls : exploitation du multiplexage spatial
En effet, comme on le savait de par le SDMA (Spatial Division Multiplexing Access, e.g. [97]) et de la
littrature sur la sparation de sources(e.g. [69]), il est possible de sparer n sources dcorrles partir dun
y = Hx +n, with H =
_
_
_
h
11
. . . h
1N
t
.
.
.
.
.
.
.
.
.
h
N
r
1
. . . h
N
r
N
t
_
_
Luc Deneire 15
MIMO : Spatial Diversity and/or Spatial Multiplexing
MIMO : Multiple antenna emitter and receiver
MIMO: Spatial Diversity and/or Spatial Multiplexing
Capacity of the MIMO channel
Hypothesis :
Known channel
r min(N
t
, N
r
) : the rank of the H matrix
i =1
i
u
i
v
Dening x = V
x, y = U
y, n = U
w :
y = x + n
Luc Deneire 16
MIMO : Spatial Diversity and/or Spatial Multiplexing
MIMO : Multiple antenna emitter and receiver
MIMO: Spatial Diversity and/or Spatial Multiplexing
Capacity of the MIMO channel
y = x + n
translates to (gure from Tses book)
293 7.1 Multiplexing capability of deterministic MIMO channels
Figure 7.1 Converting the
MIMO channel into a parallel
channel through the SVD.
x
V
V
*
U
U
*
y
y
Pre-processing Post-processing
Channel
n
min
w
n
min
w
1
+
+
x
~
.
.
.
where w 0 N
0
I
n
r
has the same distribution as w (cf. (A.22) in
Appendix A), and x
2
= x
2
. Thus, the energy is preserved and we have
an equivalent representation as a parallel Gaussian channel:
y
i
=
i
x
i
+ w
i
i =1 2 n
min
(7.9)
The equivalence is summarized in Figure 7.1.
The SVD decomposition can be interpreted as two coordinate transforma-
tions: it says that if the input is expressed in terms of a coordinate system
defined by the columns of V and the output is expressed in terms of a coordi-
nate system defined by the columns of U, then the input/output relationship
is very simple. Equation (7.8) is a representation of the original channel (7.1)
with the input and output expressed in terms of these new coordinates.
We have already seen examples of Gaussian parallel channels in Chapter 5,
when we talked about capacities of time-invariant frequency-selective chan-
nels and about time-varying fading channels with full CSI. The time-invariant
MIMO channel is yet another example. Here, the spatial dimension plays the
same role as the time and frequency dimensions in those other problems. The
capacity is by now familiar:
C =
n
min
i=1
log
1+
P
i
2
i
N
0
bits/s/Hz (7.10)
where P
1
P
n
min
are the waterfilling power allocations:
P
i
=
N
0
2
i
+
(7.11)
with chosen to satisfy the total power constraint
i
P
i
= P. Each
i
corresponds to an eigenmode of the channel (also called an eigenchannel).
Each non-zero eigenchannel can support a data stream; thus, the MIMO
channel can support the spatial multiplexing of multiple streams. Figure 7.2
pictorially depicts the SVD-based architecture for reliable communication.
which in turn translates to the following capacity formula :
C = log
2
_
det
_
I
N
r
+
SNR
N
t
HH
__
bits/s/Hz
Luc Deneire 17
MIMO : Spatial Diversity and/or Spatial Multiplexing
MIMO : Multiple antenna emitter and receiver
MIMO
Capacity of MIMO systems
38 CHAPITRE 4. PROJET DE RECHERCHE
Dans le cas SIMO (plusieurs antennes au rcepteur), la capacit est donne par :
C = log
2
(1 +SNR
M
R
i=1
|h
i
|
2
) bits/s/Hz (4.2)
On notera que la capacit augmente logarithmiquement avec le nombre dantennes.
Dans le cas MISO (plusieurs antennes lmetteur), on a
C = log
2
(1 +
SNR
M
T
M
T
i=1
|h
i
|
2
) bits/s/Hz (4.3)
o la perte M
T
est due au fait quon transmet une puissance totale xe (en dautres termes, on diminue le
SNR au rcepteur dun facteur M
T
).
Dans le cas MIMO, Foschini donne la capacit par la formule [72] :
C = log
2
det
I
M
R
+
SNR
M
T
HH
bits/s/Hz (4.4)
pour des sources dcorrles et de puissance identique. Cette fois-ci, la capacit augmente linairement avec
m = min{M
T
, M
R
} (au lieu de logarithmiquement dans le cas SIMO). On appelle m le gain de multi-
plexage. La gure 4.1 illustre le gain en capacit de systmes MIMO 2x2 et 4x4. On notera que lon peut
difcilement obtenir un gain de multiplexage suprieur 4 en pratique, (au-del de 4, le produit HH
devient
mal conditionn).
0 2 4 6 8 10 12 14 16 18 20
1
3
5
7
9
11
13
15
17
19
0 2 4 6 8 10 12 14 16 18 20
1
3
5
7
9
11
13
15
17
19
0 2 4 6 8 10 12 14 16 18 20
1
3
5
7
9
11
13
15
17
19
SNR/bit
1 TX, 1 RX
2 TX, 2 RX
4 TX, 4 RX
bits/s/Hz
FIG. 4.1 Comparaison des capacits de systmes MIMO et SISO
Luc Deneire 18
MIMO : Spatial Diversity and/or Spatial Multiplexing
Diversity vs Spatial Multiplexing
Diversity or Spatial Multiplexing ?
A tradeoff
There are two points of view :
R
1
< log
_
1 +
P
1
N
o
_
R
2
< log
_
1 +
P
2
N
o
_
R
1
+ R
2
< log
_
1 +
P
1
+P
2
N
o
_
In the SDMA case :
R
1
< log
_
1 +
||h
1
||
2
P
1
N
o
_
R
2
< log
_
1 +
||h
2
||
2
P
2
N
o
_
R
1
+ R
2
< log
_
I
2
+
Hdiag(P
1
,P
2
)H
N
o
_
Luc Deneire 22
MU-MIMO : MultiUser Diversity
MU-MIMO : Single Antenna Mobiles : Spatial Division Multiplexing Access
MU-MIMO : Splitting the Receiver Antenna Array
MU-MIMO : Spatial Division Multiplexing Access : Capacity REGION
R2
R1
R1+R2
A
B
The exact values at point A and B depend on the structure of the receiver
uses.
When K users are present, the region is a polyhedron ...
Luc Deneire 23
MU-MIMO : MultiUser Diversity
MU-MIMO : Single Antenna Mobiles : Spatial Division Multiplexing Access
MU-MIMO : Splitting the Receiver Antenna Array
MU-MIMO : Spatial Division Multiplexing Access : MULTIUSER GAIN
Compare the capacity region with pure TDMA (similar results for general
orthogonal MA).
R2
R1
R1+R2
A
B TDMA
Luc Deneire 24
MU-MIMO : MultiUser Diversity
MU-MIMO : Single Antenna Mobiles : Spatial Division Multiplexing Access
MU-MIMO : Splitting the Receiver Antenna Array
MU-MIMO : Spatial Division Multiplexing Access : Some important results
let H = [h
1
, , h
K
] be the channel between an N
t
antennas Base
Station and K mobiles equipped with 1 antenna.
Suppose h
1
, , h
K
orthogonal (K N
t
)
we transmit x(m) =
K
k=1
x
k
(m)h
If the channels are not orthogonal, then transmit with a precoding vector u
k
in the subspace orthogonal to h
i
, i = k.
This is one example of decorrelator precoding .... lots of other things to say
on receivers, choosing u
k
to achieve capacity, ....
Luc Deneire 26
Massive MIMO
Uplink : Interference and noise vanishes
Massive MIMO
Simple Uplink Matched Filter Receiver
Suppose y = h
1
x
1
+h
2
x
2
+n (y : received signal at the BS with N antennas).
Suppose h
i
have i.i.d. entries with zero mean and unit variance, h
i
known at
the BS, n complex, zero mean and unit variance normal (hence
E(|h
i
|
2
) = SNR).
Apply a matched lter at the BS to detect User 1 :
1
N
h
H
1
y = x
1
1
N
N
i =1
|h
1i
|
2
. .
useful signal
+x
2
1
N
N
i =1
h
1i
h
2i
. .
interference
+
1
N
N
i =1
h
1i
n
i
. .
noise
Luc Deneire 27
Massive MIMO
Uplink : Interference and noise vanishes
Massive MIMO
Interference and noise vanishes
1
N
h
H
1
y = x
1
1
N
N
i =1
|h
1i
|
2
. .
useful signal
+x
2
1
N
N
i =1
h
1i
h
2i
. .
interference
+
1
N
N
i =1
h
1i
n
i
. .
noise
By the central limit theorem :
1
N
N
i =1
h
1i
h
2i
E{h
11
h
21
} = 0, for N
1
N
N
i =1
h
1i
n
i
E{h
11
n
1
} = 0, for N
Hence :
1
N
h
H
1
y x
1
E
_
|h
11
|
2
_
= x
1
Luc Deneire 28
Massive MIMO
Uplink : Interference and noise vanishes
Massive MIMO
SINR scales up with the number of BS antennas / SNR can be scaled down with N
SINR
MF
1
=
1
N
h
H
1
h
1
1
N
h
H
1
h
2
||h
1
||
2
+
1
N.SNR
h
H
1
h
2
||h
1
||
2
_
=
1
N
in TDD : use channel reciprocity (As many pilots as Users) BUT front
end calibration ([Kouassi, ICC 2013])
in TDD : use Uplink channel training and use the uplink channel
estimation on the downlink
Luc Deneire 32
Massive MIMO in the Downlink
Channel estimation in TDD
Classical training based MMSE estimator
Consider the two user case and use orthogonal pilot sequences (and
white Gaussian noise of unit variance) :
Y = [h
1
h
2
]
_
p
T
1
p
T
2
_
+N
i
||p
i
||
2
= h
i
+
n
i
where n CN(0, ||p
i
||
2
I
N
)
Luc Deneire 33
Massive MIMO in the Downlink
Channel estimation in TDD
Classical training based MMSE estimator
Suppose h
i
orthogonal and R
i
E
_
h
i
h
H
i
_
h
i
= R
i
_
R
i
+||p
i
||
2
I
N
_
1
Y
p
i
||p
i
||
2
The decomposition h
i
=
h
i
+
h
i
,
with E
_
h
i
h
i
H
_
= R
i
_
R
i
+||p
i
||
2
I
N
_
1
R
i
i
and E
_
h
i
h
i
H
_
= R
i
i
h
H
i
h
i
_
= 0.
Hence
h
i
CN(0,
i
) and
h
i
CN(0, R
i
i
).
Assumptions when N
Trace(R
i
) = c
i
N, for 0 < c
i
1 (linear energy growth)
liminf
N
1/Nrank(R
i
) > 0 (innite degrees of freedom)
limsup
N
||R
i
|| < (nite energy per degree of freedom)
Luc Deneire 34
Massive MIMO in the Downlink
Inuence of imperfect CSI ]
Inuence of imperfect CSI [Hassibi / Hochwald, Tr. Inf. Theory 2003]
BS projects y(t ) on
h
i
. The achievable rate is :
R
1
= (1 1/)E
_
_
_
log
_
_
1 +
||
h
1
||
4
h
1
(h
2
h
H
2
+ E
_
h
1
h
1
|
h
1
_
+ 1/SNRI
N
)h
1
_
_
_
_
_
For large N
R
1
(11/) log
_
_
_
_
_
_
_
_
_
1 +
(1/Ntr
1
)
2
1
N
2
tr(
1
R
2
)
. .
interference
+
1
N
2
tr(
1
(R
1
1
))
. .
ImperfectCSI
+
1
N
2
SNR
tr(
1
)
. .
noise
_
_
_
_
_
_
_
_
_
Luc Deneire 35
Massive MIMO in the Downlink
Inuence of imperfect CSI ]
Imperfect CSI can induce a loss of power efciency
if R
i
= I
N
then
R
1
(1 1/) log
_
_
_
_
_
_
_
_
_
1
1
N
..
interference
+
1
N||p
1
||
2
(2 +
1
SNR
)
. .
ImperfectCSI
+
1
NSNR
. .
noise
_
_
_
_
_
_
_
_
_
If ||p
1
||
2
N
a
and SNR N
b
with a + b < 1 : liminf
N
R
1
> 0.
If equal pilot and data power (a = b = 1/2) the power must be proportional at
least to
1
N
in contrast to
1
N
for perfect CSI.
Luc Deneire 36
Pilot contamination : due to linear processing
1
||p
1
||
2
= h
1
+h
2
+
n
1
h
1
= R
1
_
R
1
+R
2
+||p
1
||
2
I
N
_
1
Y
p
1
||p
1
||
2
E
_
h
1
h
1
H
_
= R
1
_
R
1
+R
2
+||p
1
||
2
I
N
_
1
R
1
Pilot Contamination : Estimate of h
1
correlated with h
2
Luc Deneire 37
Pilot contamination : due to linear processing
Pilot contamination can limit rate (Marzetta 2011)
As N , R
1
(1 /T) log(1 +
1
), with
1
1
When does pilot contamination become dominating?
0 20 40 60 80 100
200
190
180
170
160
150
140
Number of antennas N
P
o
w
e
r
(
d
B
m
)
Signal
Interference
Imperfect CSI
Noise
Pilot contamination
Powers of dierent parts of the received signal at the BS for R
i
= d
3.6
i
I
N
with
d
1
= 200 m, d
2
= 500 m, and kk
2
= SNR = 121 dB.
Jakob Hoydis (Bell Labs) Massive MIMO and HetNets IMSS13 34 / 96
Power of different parts of the received signal, power exponent of 3.6,
d
1
= 200m, d
2
= 500m
Luc Deneire 38
Pilot contamination : due to linear processing
Pilot contamination can be easily mitigated
Blind techniques
Phase Noise
IQ imbalance
Quantization noise
Non reciprocity
For Massive MIMO : need cheap, low-power, low-cost transceivers
Resistance to Hardware impairments is VITAL.
Luc Deneire 40
Massive MIMO seems to resist to Hardware impairments
An analysis of resistance to hardware impairments
An analysis of resistance to hardware impairments [Bj ornson et al,
Arxiv, 2013]
Uplink channel model
PRINCIPLE : model impairments with Gaussian r.v., with parameters
checked against real hardware
Uplink : y
BS
= h
_
x
UE
+
UE
t
_
+
BS
r
+n
BS
where E
_
|x
UE
|
2
_
: P
UE
, n
BS
CN(0, S); h CN(0, R)
Transmit distortion :
UE
t
CN(0,
UE
t
P
UE
)
Receive distortion :
BS
t
CN(0,
BS
r
P
UE
diag(|h
1
|
2
, , |h
N
|
2
))
Parameters in the range [0, 0.03].
Luc Deneire 41
Massive MIMO seems to resist to Hardware impairments
An analysis of resistance to hardware impairments
Channel estimation with hardware impairments
h = p
RZ
1
y
BS
where
Z = E
_
y
BS
(y
BS
)
H
_
= P
UE
(1 +
UE
t
)R + P
UE
BS
r
diag(R
11
R
NN
) +S
E
_
h
H
_
= R C
E
_
h
H
_
= C
C = R P
UE
RZ
1
R
Luc Deneire 42
Massive MIMO seems to resist to Hardware impairments
An analysis of resistance to hardware impairments
Channel estimation with hardware imperfections : simple case
If R = S = I
N
C =
_
1
1
1 +
UE
t
+
BS
r
+ 1/P
UE
_
I
N
,
Which tends to .... when P
UE
goes to innity
Hence, perfect channel estimation is not possible for innite pilot power.
Can be enhanced by increasing the number of pilots ...
Luc Deneire 43
Massive MIMO seems to resist to Hardware impairments
An analysis of resistance to hardware impairments
Numerical results
Channel estimation with hardware imperfections: Numerical results
0 5 10 15 20 25 30 35 40
10
4
10
3
10
2
10
1
10
0
Average SNR [dB]
R
e
l
a
t
i
v
e
E
s
t
i
m
a
t
i
o
n
E
r
r
o
r
p
e
r
A
n
t
e
n
n
a
LMMSE Estimator
Error Floors
UE
=
BS
=015
2
UE
=
BS
=01
2
UE
=
BS
=005
2
UE
=
BS
=0
Relative estimation error
tr C
tr R
for N = 50 over SNR = p
UE tr R
tr S
. The matrix R is generated
by the exponential model [22] with correlation coecient = 0.7 and S = I
N
.
Jakob Hoydis (Bell Labs) Massive MIMO and HetNets IMSS13 43 / 96
Luc Deneire 44
Massive MIMO seems to resist to Hardware impairments
An analysis of resistance to hardware impairments
Numerical results
Channel estimation with hardware imperfections: Numerical results
1 2 3 4 5 6 7 8 9 10
10
10
3
10
2
10
1
10
0
Pilot Length (B)
R
e
l
a
t
i
v
e
E
s
t
i
m
a
t
i
o
n
E
r
r
o
r
p
e
r
A
n
t
e
n
n
a
Fully Correlated Distortion Noise
Uncorrelated Distortion Noise
deal Hardware
5 dB
30 dB
Relative estimation error
tr C
tr R
over the pilot length B for for N = 50 and
UE
t
=
BS
r
= 0.05
2
.
Jakob Hoydis (Bell Labs) Massive MIMO and HetNets IMSS13 44 / 96
Luc Deneire 45
Massive MIMO seems to resist to Hardware impairments
An analysis of resistance to hardware impairments
Numerical results : impact of BS vanishes as N grows large
11
transceiver hardware that limits the DL and UL capacities as
N . Thus, the detrimental effect of hardware impairments
at the BS vanishes completely, or almost completely, when the
number of BS antennas grows large. This is, simply speaking,
since the BSs distortion noises are spread in arbitrary direc-
tions in the N-dimensional vector space while the increased
spatial resolution of the array enables very exact transmit
beamforming and receive combining for the useful signal. This
is a very promising result since large arrays are more prone
to impairments, due to implementation limitations and the will
to use antenna elements of lower quality (to avoid having
deployment costs that increase linearly with N). In contrast,
the UEs distortion noises are non-vanishing since they behave
as interferers with the same effective channels as the useful
signals.
Corollaries 4 and 5 assumed that the inter-user interfer-
ence satisfy E{I
UE
} O(N
n
) and E{kQk
2
} O(N
n
),
respectively, for some n < 1. These conditions imply that the
interference terms only vanish asymptotically if the scaling
with N is slower than linear. This is satised by regular
interference (which has constant variance, n = 0), but there is
a special type of pilot contaminated interference in multi-cell
systems that scales linearly with N. This adds an additional
non-vanishing term to the denominators of (42) and (44). We
detail this scenario in Section VI.
Finally, we stress that the DL and UL capacity bounds
in Corollaries 2 and 3, respectively, have a very similar
structure. The main difference is that the UL is only affected
by UL hardware impairments (i.e.,
UE
t
,
BS
r
), while the DL
is affected by both DL and UL hardware impairments (i.e., all
-parameters) due to the reverse-link channel estimation.
C. Numerical Illustrations
Next, we illustrate the lower and upper bounds on the
capacity that were derived earlier in this section. We consider
a scenario without interference, Q = S = 0 and I
UE
= 0, and
dene the average SNRs as p
UE
tr(R)
N
2
BS
and p
BS
tr(R)
N
2
UE
in the UL
and DL, respectively. The SNRs are xed at 20 dB, while we
vary the number of antennas N and the levels of impairments.
We assume that the transmitter and receiver hardware of each
device are of the same quality:
BS
,
BS
t
=
BS
r
at the BS
and
UE
,
UE
t
=
UE
r
at the UE.
11
Furthermore, we assume
T
DL
data
T
coher
=
T
UL
data
T
coher
= 0.45. These assumptions make the bounds
for the DL and UL capacities become identical, thus we can
simulate the DL and UL simultaneously.
Fig. 6 considers a spatially uncorrelated scenario with
R = I for different levels of impairments:
UE
t
=
BS
r
{0, 0.05
2
, 0.15
2
}. The meaning of these parameter values was
discussed in Remark 1. The capacity with ideal hardware
grows without bound as N , while the lower and upper
bounds converge to nite limits under transceiver hardware
impairments. Recall that these bounds hold under any CSI
11
The transmitter and receiver hardware both involve converters, mixers,
lters, and oscillators; see [30, Fig. 1] for a typical transceiver model. The
main difference is the type of ampliers, thus the assumption of identical
levels of impairments makes sense when the non-linearities of the ampliers
at the transmitter are not the dominating source of distortion noise.
0 100 200 300 400 500
0
1
2
3
4
5
6
7
8
Number of Base Station Antennas (N)
S
p
e
c
t
r
a
l
E
f
f
i
c
i
e
n
c
y
[
b
i
t
s
/
c
h
a
n
n
e
l
u
s
e
]
Capacity: Upper Bounds
Capacity: Lower Bounds
Asymptotic Limits (Upper & Lower)
BS
=
UE
= 0.15
2
BS
=
UE
= 0.05
2
BS
=
UE
= 0
Fig. 6. Lower and upper bounds on the capacity. Hardware impairments
have a fundamental impact on the asymptotic behavior as N grows large.
conditions at the BS and UE; the lower bounds represent
no CSI in the decoding step and the upper bounds represent
perfect CSI. Although the gap between these extremes is large
for ideal hardware, the difference is remarkably small under
non-ideal hardware due to the nite capacity limit (caused by
distortion noise) and the channel hardening that makes random
inner product such as h
H
v become increasingly deterministic
as N grows large. Since a main difference between the lower
and upper bounds is the quality of the CSI, the small difference
shows that the estimation errors have only a minor impact on
the capacity; hence, the estimation error oors described in
Section III has no dominating impact in the large-N regime.
The asymptotic capacity limits in Fig. 6 are characterized
by the level of impairments, thus the hardware quality has a
fundamental impact on the achievable spectral efciency. The
majority of the multi-antenna gain is achieved at relatively low
N; in particular, only minor improvements can be achieved
by having more than N = 100 antennas. Larger numbers
are, however, useful for inter-user interference suppression and
multiplexing; see Section VI.
Fig. 7 considers the same scenario as in Fig. 6 but with
a xed level of impairments
UE
= 0.05
2
at the UE and
different values at the BS. As expected from the analysis, the
lower and upper capacity bounds increase with
BS
, but the
difference is only visible at small N since the curves converge
to virtually the same value as N . This validates that the
impact of impairments at the BS vanishes as N grows large.
Finally, we consider the capacity behavior for different
channel covariance models, namely the four propagation
scenarios described in Section III-B. The lower and upper
capacity bounds are shown in Fig. 8 for
BS
=
UE
= 0.05.
The upper bound is identical for all the models, since it
only utilizes the diagonal elements of R. However, there are
clear differences between the lower bounds. The spatially
uncorrelated covariance model provides the highest perfor-
mance, while the strongly spatially correlated one-ring model
from [42] with 10 degrees angular spread gives the lowest
performance. This stands in contrast to Section III-B, where
the highly correlated channels gave the lowest estimation er-
rors. However, the differences between the channel covariance
models vanish asymptotically as N .
Luc Deneire 46
Massive MIMO seems to resist to Hardware impairments
An analysis of resistance to hardware impairments
Hardware impairments : conclusions
h h)
1
SNR
Channel estimation is the issue !
At Mobiles : R
R
= I
R R
T
; r = rankR (rank as number of non negligible eigenvalues)
e
j k
T
(+)(u
m
u
p
)d
q
D
D
r
Scattering ring around
the mobile
Luc Deneire 50
Massive MIMO and FDD : classical beamforming
System Model
y = H
H
Vd +n = H
H
x +n (V is the N s precoding matrix,
= E
_
Vdd
H
V
H
_
is the s-rank input covariance.
Luc Deneire 51
Massive MIMO and FDD : classical beamforming
System Model
Principle of JSDM : divide ut impera
g
U
H
g
.
= K/G mobiles, N
= N/G BS antennas, s
+ k user k in group G.
1/2
g
w
gk
pre-beamforming : B C
Nb
, only depending on second order
statistics {U
g
,
g
} and b such that b
= b/G s
Hence y
g
= H
H
g
B
g
P
g
d
g
+
=g
H
H
g
B
g
P
g
d
g
+n
g
= g.
R
g
varies slowly (much more slowly than H
g
)
U
g
can be tracked by subspace tracking algos.
Luc Deneire 54
Massive MIMO and FDD : classical beamforming
JSDM with Eigen-Beamforming
Capacity is optimal with reduced CSI
1/2
g
P
g
d
g
+n
g
where W
g
is a r K
Global pre-beamforming : B = [B
1
, B
G
] (N b)
Block Diagonalisation : U
H
g
B
g
= 0
g
= [U
1
, , U
g1
, U
g+1
, U
G
] of rank r (G 1)
its SVD being
g
= [E
1
g
, E
0
g
]S
g
F
H
G
,
where E
0
g
is the null space of
g
(orthogonal to all other groups).
1/2
g
w
g
k
) on this orthogonal space :
h
g
k
= (E
0
g
)
H
U
g
1/2
g
w
g
k
h
g
k
=
U
g
1/2
g
w
g
k
with
U
b
g
containing the b
dominant eigenvectors
g
Luc Deneire 56
Massive MIMO and FDD : classical beamforming
JSDM with Eigen-Beamforming
Simulation Results (Caire)
M = 100, G = 6 user groups, Rank(R
g
) = 21, effective rank r
g
= 11.
We serve S
= 10, r
= 6 and r
= 12.
For r
g
= 12: 150 bit/s/Hz at SNR = 18 dB: 5 bit/s/Hz per user, for 30 users
served simultaneously on the same time-frequency slot.
0 5 10 15 20 25 30
0
50
100
150
200
250
300
350
SNR (in dBs)
S
u
m
R
a
t
e
Capacity
ZFBF, JGP
RZFBF, JGP
ZFBF, PGP
RZFBF, PGP
0 5 10 15 20 25 30
0
50
100
150
200
250
300
350
SNR (in dBs)
S
u
m
R
a
t
e
Capacity
ZFBF, JGP
RZFBF, JGP
ZFBF, PGP
RZFBF, PGP
Luc Deneire 57
Massive MIMO and FDD : classical beamforming
JSDM with Eigen-Beamforming
Order of magnitude gain in training
With Full CSI : 100 30 channel matrix : 3000 complex per block 100
100 Unitary pilot matrix for downlink channel estimation