Part It :

Diversity Gain

*
Optimal detection in scalar AWGN

Y Xt W where x Eta a } CN to No )
-

,
w -
,
,

C MAP
Optimal detection :
Maximum a posterior )

-
a

fl x
-

-
-

at y ) 3 fix =
aly ) I MAP )
a

tax
HH
a
fly al
-

1x= a) apex -_

fly ) .pk a )
-

Ix
- -

a
-
-

3
-

⇐ - -
- -

Assume -
a

The -
a) enka ) fly IX
-

-
-

a ) 3 fly Ix
-

- a ) C ML )
a

Assume Cheat -
a 14 -

of
FIE
-

Wis Gaussian .

e
I .
e- It
a

A Minimum

( )
-

⇐ 14 -1 at 5 at
14
-

distance
a

- a

⇒ y 5 o

a
 I =
-
a
← I a call this region Da
Call
→
this
-

D-
region
-

,
a

,
y
-
a 0 a

threshold

PC error lxe -
a ) =P I y E Dal XI -
a )

f Da fly Ix = -
a ) dy
=
Pr I Y > o ,
y -
N I -
a , to I )

= Q ( h ) ( tail prob of Gaussian distribution )

Similarly pl Ix a) Q ( 1¥ )
=
error
-

-
,

Pe =P I error I x= -
a ) .

PCx= -

a ) t Pc error Ix -

-
a ) .

Plea )

exponentially
Q¥⇒ Qfzsiyp )
-
snip
= =
= e
with SNR !
decay

Ed!
'
x

fiveTmo ) l
-
-

ma
 * Coherent detection in
Kayleigh fading
means that we know the channel

y =
h x t W where x EL -

a. a ) her CNL oil )

, ,

w r CN to ,
No )

A I trivial ) sufficient statistics is

Re .

Y ) = lhlx
# t Z
,
where z -
Nco . )

Qfzlhisluk
( k¥1 ) )

=/
Hence , Tech =
Q =

exponential distribution
'
x=zlh¥K
"
pe-TEYhp.de ? Q(2lhtsNR )

¥¥
1h14 % snipe
h ]
.

e- dlhi µ
'

XL

e-
=L I ? fµFsµq
I
Maui
-

dx .

e
,
→

=
I ? Hsiu e
-

Waihi .
e- ¥d× x

'

Ira 194 ¥ue I E

-
-

-
-
e .

e ax

Hex Ets )
¥
¥15 Elkie ) ta
-

foo dx "
e
-

dx
= -

e =
-

's:i¥⇒ ,
=
I l I -
I I -

¥µztO⇐µ ) ) ) =¥µz linear in SNK ! !

Taylor expansion
 loss
Significant is introduced
by fading
!

reduced
.

why so ? SNK by a factor of Ihl

and Ihl be small with certain prob

can very .

very unlikely
In this situation it is that we
,

recover

normalized
can x .

-
To understand this phenomenon ,
hohsidlr the
following
channel model ( by Tmo ) :

y = h .
x t w ,
where
xtf-TE.rs ) ,
w - CN to , 1)

Noise level i I

Signal strength :
Ihf SNK
 below noise level it's hard
If signal strength is ,

to recover x from Y .

the ship
Deep fade event :

Prl deep fade ) =

Prllhicstupl Isnt
the factor of
This captures scaling Pe .

Question : How to improve ?

Answer provide many independent

To RX views
Diversity
:
.

of x .

Many ways
to
get diversity .
this course focuses

on antenna diversity .

Remark :
Although we only consider BPSK line , binary

hypothesis at receiver ) the and

testing , reasoning

results modulation whose detection

apply Av
general

problem can be
approximated by multiple binary

hypothesis testing .
* Receive diversity
Add antenna at RX ⇒
Single -

input
Muti -

Output ( SIMO )

T -
t
TX
EX

Ye Chi = helm ] X Cm ] t Wear ] ,

1=1,2 ,
-
-

It placed far C I ) then he

enough
antennas are >
,

are
independent of each other .
he - CN l 0,1 ) i . i. d .

wit

=!
-44T
Let 1=191 ,
Ya ,
-
-
.

Yi k -

-
[ hi ,
ha ,
- - .

he ] he =L w
, .ws ,
. . .

Then ,
I = he .

x t
he ,
he - CN to ,
I )

Optimal detector :

.
a

fl x
-

-
-

at y ) 3 fix = a ly ) I MAP )
- -

toy a)
a
fly at
-

1x= a ) .pk fly )

-

apex a
-

Ix
- -
-
-
-

I
-

a
-

Assume

TIX : -
at epix -

- a ) x
-

-
-

al
-1111
Ia
fly Ix
-

- a ) C ML )
 -
a '
akil
÷
" E
3
-

I tahir
In
It e-
⇒ e-
a

- a

(
' '

⇒ lakh
111 -

5 111
Aku -
minimum distance )
a

kaiak
Alternative view :

Only things happening as = -

ah
.

on the subspace cutting t

→ x
Hates
midpoint
through KA and his 2

matters

Ht As
the
-

direction hy
Projecting I along I = a
GBH
HA -

11h11

does not incur

any loss .
( i.e.
,
it is a sufficient statistics )

⇒ 5 =

→
,
.

,
text
Helix
he
-

-
+ I where E -
cnn.IS

(, so tropic )
( This is called maximum ratio
combining )
the scalar AWGN and
The problem reduces to

FLYQ¥vI
Hall
? SNR

Pelle =Q( )
"
= ) = e-
 that
'
Note
'
Hall n
X ( 2L )

[
" "
④ nkhtelk] =
( ) -

µ÷Ey decrease polynomial ly .

We can
similarly define the deep fade event .

Deep fade event :

the" Cst ) 'T

'

Pr =

scaling
L
! correct
 I
'

Half .

SNR =L .

SNR .
11h11

① ②

① Power gain or
array gain

=L Ethel
'

hell
'

②
Diversity gain
:

Ill
.

be
.

In order to make the small all of them have

average ,

to be small .

Prob . he is in
deep fade Kstp
t
Prob .
all hi ,
ha ,
-
- .
he are in
deep fade =¥z )

* Transmit diversity
°
In
practice it is a lot easier to add antennas
,

at base station than at devices

.
Lost can be amortized over devices .

Add antenna at TX ⇒ Multi -

input single
-

output ( MISO )

+ →
- t
y
.

Y Cm ] = h , Cm ]XCm]thim ] Xan ] t Wcm ]

 Coding send
'

over
space
:
All antennas same X

Y cm ]
-

-
them
] them ] ) xcm ] t won ]

- CN 10,4 ,
NO
diversity gain
-

Coding over time :

Multi -

Shot scheme

Each time use one antenna

only and send same u

the
him ] yin ] Emei ] yantis achieves
diversity of 2

Maximum MEL Time

ratio
combining L ) space I 2

This to
repetition coding ⇒
I U 0

2
O u
Code Alamo Code

corresponds
-

Space -
time :
uti

symbols slots
Sending two Ui us over two time
,

Time
I 2
space

I Xia ]
-

-
U , X.cz ]= -
UE

Z Xzci ]
-

-
Us Xd ]
-

-
u ,*

Assume channels constant over symbol times

stay
two

Yi ] h tha Ust WED

{
=
,
U ,

ya , = -
h , UE-ihau.tw CD
 4¥ .
:* Hi rise :
.

y = H .

K t
he
-

't 't 't

E H
I H H K t H w
-
-
= .

::X :* :*a
.
a:*: vis

I
" ""
I in HI It I
-

. " .

Chi
'

where ht ,
WT - CN lo
,
't Iha I )
I .
No
)

÷.m⇒ a
-

f I:! ) where wi .
win cnn.ms

order 2
diversity
 We view the transmitted signal across L antennas

and N times as a codeword matrix X .

In multi -
shot scheme : X =

( ! ]

( !I ) orthogonal
In Alamo code matrix !
uti :
X =

Normalize X that symbol is I

average energy per
so

i. e.
,
SNK =
¥
For a pair of codeword XA and XB a sufficient
,

statistics is HICXA -

XB ) and

Prt XA →
XB thy =
Q
"¥%H
2ft )
XB )*k
( ))
-

PRIX ,
→ XB )
-
-

I Q SNKTEHAJXBJCXA

"

Note that ( XA -

XB ) ( XA XB )
. -

is Hermitian ,
its

eigenvalue decomposition
exists .

"

UE Uf

%
ie,
( XA -

XB ) .

l XA -

XB )
-

-
Lee I -

-
Lith
Unitary
,
diagonal
" "

all
( )
> o

.
.

,
 prlxa-sx.f-EFQIINE.EE#eidi) ) '

, iii.
+ sue

4 2

Ie
<
it all de > o
shift
-

a ,
.

Remarks :

(( XA XB ) (XA XB )* )
i
) Note that Dee -
. -
=

①
To achieve full
diversity order ,
all the > o

This is called the determinant criterion

full order
ii )
For two codes achieving diversity ,
the

one with
larger min Dee XA -

XB @
) l XA -

X , )* ) is better
XA , XB

iii ) Alamo hei code is better than multi -

shot ( repetition )

To have a fair comparison .

Consider 1375k for Alamo uti

and 4- PAM for

repetition .

Also we normalize the power to be I per antenna .

A lamont :
'

.
Ui E I - I
,
I
}

I
'

Repetition I 3b b b 3
b) b =
: U E -

,
-

, ,
,
 ! ) ! )
-

( Ks
(
-

Alamo XA
-

nei :
-

dee ( Xa
B) -
X
HAXB )* -
=
dee ( to I) -

-
16

!! )
→

!! )
-

Repetition XA ( XB
(
-

: = -

ok
! )
"
det ( B)
Xa -
X
HAXB -

)* =
dee
( ¢
= 16104--16 .

=
6¥ c 16

ID Receiver of Alamo
very simple
structure uti code is
②
't
H This called the linear
I I is receiver
property
= .

V
) bodes having both ① and ② can be constructed

Hurwitz
based orthogonal design of Radon
-

on .

( Taro Kh -
Jafar khani -
Calder bank 99 )
if EQ and
L only exists in 2 in
For it n hi
-

n
-
-

n -

- 2
,
4 ,
8 if Ui ERI .
It exists for
every
h 72L .
 MIMO : 2×2 example

TX I Y t EX '
E

24
Distance 3

Ll
his TX j to Ex
2
:
i
TX I 4 EX 2
has i. i. d distributed

WED
w.cz
,UEt
hill
]=
Alamonei code
Apply

Time
I 2

.mn#l:::::.:::::.ii:i:::i::.
space
Y '
" I =
,
t Ust
I Xia ]=U , X.cz ]= -
UE
Yiu -

hi U # t ]

. . . . .

:* ::÷,
:i*:i÷i÷÷÷÷ : "

I = H K t he

't

tf I
-

- H Hattie diversity order ¢

I &
'

lhijl

=/ ) the
, , '

° E .
thin
 Alamo nti Lode :

slots
.

Sending 2 symbols over 2 time

Achieving diversity order of 4 .

what
Is it optimal ? In sense ?

*
Spatial Multiplexing
.

Another
way
to benefit from multiple antennas .

send independent data

Always streams
.

Vertical Bell Space Time

Laboratories Layered
-
-
or

V -

BLAST architecture

When Mr receive we show

having antennas can
,

Nr

AHA -7431 E
( II ,
it i. s

For V -

BLAST over 2×2 MIMO each codeword is a

,

two -
dimensional vector I

(
"

, -7¥

Prix
Isnake E

+ ,
.

¥1144
) '
5M¥" ¥ -

* "
"
* V -
BLAST :

I slot
.

Sending
2
symbols over time

order of
Achieving diversity 2
.

what
Is it optimal ? In sense ?

exploiting
Alamo code channel diversity
uti is
good at
"

exploiting
while V -

BLAST is good at spatial degrees of

freedom .

.
A unified view is called for .

admits how that

Alamo code a
homplexity receiver
-
.
uti

decouples streams
effortlessly .

V BLAST ML detection
requires joint
. -

.
*
V -
BLAST with Zero -

Forcing

The
complexity of V -

BLAST detection
grows

exponentially as the number of antennas .

A low Zero I ZF )
complex receiver :
Forcing
-

hit his
I = H .
K t k
,
H =
( ha , hw ) -
- ( ht .
he )

- -
I -
'
I ⇒ decoupled streams
⇒
I
=
H I =
I + H he = x x
with correlated noise .

then the correlation and perform ML

We
ignore separately .

=µ⇐T II! i!
-

)
'
Note that
-

'

hi
1h44 that
YT = X , t where ,
~ Nco No )
I haha huh
,

ha
,
-
,

histha
1h11
has
#
Eta
-

5
,
⇒
) z
=
x +

f¥
.
,

that

YI.h.i-j://.IE)
vector amines y to

E-
⇒
the direction that is

orthogonal to he

similarly ,
§ ,
will
project I to direction orthogonal to hi

slot order of I
Sending
2 symbols per time , diversity .