MIMO Wireless Communications

Speaker: Sau-Hsuan Wu Date: 2008 / 07 / 1

Department of Communication Engineering NCTU

Outline
MIMO !ireless c"annels MIMO transcei#er MIMO preco$er

Outline
MIMO !ireless c"annels MIMO transcei#er MIMO preco$er

MIMO S'stem Mo$el

N () antennas an$ M *) antennas (flat-fading case)

&

&

: M x N MIMO c"annel matri+ : N x T (ransmit matri+ : M x T *ecei#e matri+

MIMO (ec"nolo,'
MIMO transmitter an$ recei#er
Space (ime Co$e -S(C. 1. High link /ualit' due to $ual -()-*). $i#ersit' 2. Create multiple signal copies thru MIMO channels Spatial Multiple+in, -SM. 1. High spectral e00icienc' due to spatial multiple+in, 2. Create spatial degrees of freedom for transmission of multiple data streams simultaneousl

(ransmission Sc"emes
2x2 (!"2 # $"2) Matrix & ('$C) %x% (!"% # $"%)

(uasi-Orth. ('$C)

(ransmission Sc"emes
2x1 (!"2 # $"1) Matrix ) ('$C 'M) %x% (!"% # $"%)

Matrix C ('M)

Space (ime Co$in, -S(C.

Space (ime Co$e

3re t"ere an' a$#anta,es 0or usin, space time co$e4

& practical performance criterion is the pro5a5ilit' o0 error *hich is particularl important *hen coding o+er a small num,er of ,loc-s *here the 'hannon capacit is .ero

Consi$er communications o#er *a'lei," 0a$in, c"annels

/or )0'1 o+er a 2a leigh channel *ith +ariance 34 25"1 0e 16 (%3,6!7) /or 80'1 /or coherent /'1 0e 16 (23,6!7) 0e 16 (23,6!7)

/or non-coherent /'1 0e 16 (3,6!7)

Space (ime Co$e

10 10

Space (ime Co$e

Ho! $o people usuall' $o to com5at 0a$in,4
Di#ersit' techni9ues
Frequency diversity maximal ratio combining (MRC) For BPSK with MRC: Temporal diversity Interleaving Spatial diversity u ing multi!le "# an$%or R# antenna Beam&orming 'ntenna election

11 11

Space (ime Co$e

12 12

Spatial plus temporal diversity u ing multi!le "# an$%or R# antenna an$ multi!le time lot S!ace(time co$e

Space (ime Co$e

Is t"ere a 0ormal $e0inition 0or $i#ersit'4
:e loo- at this from the sense of outa,e pro5a5ilit' $o achie+e capacit ; a coding is assumed to ta-e place across fading ,loc-s; *hich entails large dela s :e consider a coding scheme *hich is done <ust across one fading ,loc-

1% 1%

Def 1:

("e outa,e pro5a5ilit' 0or a transmission rate o0 R an$ a ,i#en transmission strate,' p-).

Space (ime Co$e

W"en one use a !"ite 6aussian co$e5ook

1& 1&

3t "i," S7*8 t"e outa,e pro5a5ilit' is t"e same as t"e 0rame error pro5a5ilit' -9:*. in terms o0 t"e S7* e+ponent -- ;< ="en, an$ D< 7< C< (se8 I:::-I(8 200%

Space (ime Co$e

Def 2:

1 1

3 co$in, sc"eme !"ic" "as an a#era,e error pro5a5ilit' >e -or an outa,e pro5a5ilit' >out. as a 0unction o0 S7* t"at 5e"a#es as

is sai$ to "a#e a $i#ersit' o0 or$er d

Space (ime Co$e

3 s'stematic !a' to e+ploit space-time $i#ersit' -- (arok" et al. I:::-I( 1999
Consider a code*ord se9uence + " =+$(7);>;+$(!)?; *here + (-)"=x1(-);> xMt(-)?$ @i+en the perfect C'I; the pair-!ise error pro5a5ilit' (>:>) ,et*een + and e can ,e ,ounded from a,o+e ,

11 11

*here 1m;n is the 2icean coefficient; Es"P 6 Mt

Space (ime Co$e

i are the eigen+alues of &(+;e)")(+;e)H)(+;e); and

17 17

9or *a'lei," 0a$in, c"annels -?m8n @0.8 !e "a#e

Space (ime Co$e

;et q $enote t"e rank o0 3-x8e.8 t"en !e "a#e

18 18

3ccor$in, to Def 2 0or t"e $i#ersit' or$er8 t"e $i#ersit' or$er usin, space-time co$in, is qMr
)(+;e) has to ,e 0ull ro! rank for an code*ords + and e to achie+e the ma+imum $i#ersit' Mt Mr

Space (ime Co$e

3 natural /uestion t"at arises is8 "o! man' co$e!or$s can !e "a#e t"at allo!s us to attain a certain $i#ersit' or$er4 Def 3:

12 12

3 co$in, sc"eme !"ic" "as a transmission rate o0 R( NR! as a 0unction o0 S7* is sai$ to "a#e a multiple+in, ,ain r i0

!"ere t"e constellation siAe is also allo!e$ to 5ecome lar,er !it" S7*

Space (ime Co$e

("eorem 1: /or ! A Mt B Mr -1; and 1 " min(Mt; Mr); the optimal tradeoff cur+e dC(r) is gi+en , the piece!ise linear 0unction connecting points in (-; dC(-)); - " 7;>;1; *here dC(-) " (Mr D -) (Mt D -) I0 r @ k is an inte,er8 t"is can 5e interprete$ as usin, -Mr B k. recei#e an$ -Mt B k. transmit antennas to pro#i$e $i#ersit'8 !"ile usin, k antennas to pro#i$e multiple+in, ,ain

20 20

S(C Matri+ 3
(ransmit Matri+
/or 2 x 2 MIMO s stem

21 21

/or % x % MIMO s stem

S(C Matri+ 3
*ecei#er Desi,n C1D

22 22

&t the recei+er:

Di#ersit' or$er is t!o E

S(C Matri+ 3 C1D

2% 2%

S(C Matri+ 3
'imilarl ; for 0our $E antennas; *e ha+e

2& 2&

Di#ersit' or$er is onl' t!oE

S(C Matri+ 3 -Simulation *esults.

2 x 2 '$C -Matrix &.
(2 x 2) II8 /ading Channel Model (/d$s " 7.771)

2 2

S(C Matri+ 3 -Simulation *esults.

2 x 2 '$C -Matrix &.
(2 x 2) II8 /lat /ading Channel Model (Code 2ate " 162)

21 21

S(C Matri+ 3 -Simulation *esults.

2 x 2 '$C -Matrix &.
(2 x 2) 'CM Channel Model (Fr,an; Marco; G7 1mh)

27 27

S(C Matri+ 3 -Simulation *esults.

% x % '$C -Matrix &.
II8 /lat /ading Channel Model (/d$s " 7.771)

28 28

S(C Matri+ 3 -Simulation *esults.

% x % '$C -Matrix &.
II8 /lat /ading Channel Model (Code 2ate " 162)

22 22

S(C Matri+ 3 -Simulation *esults.

% x % '$C -Matrix &.
'CM Channel Model (Fr,an; Marco; G7 1mh)

%0 %0

S(C Fuasi-Ort"o,onal Sc"eme

(ransmit Matri+
/or % x 1 MIMO s stem

%1 %1

S(C Fuasi-Ort"o,onal Sc"eme

*ecei#er Desi,n C2D
-1. Het *here and is an unitar matrix

%2 %2

-2. &t the recei+er:

S(C Fuasi-Ort"o,onal Sc"eme

%% %%

*here

and

-%. Het

S(C Fuasi-Ort"o,onal Sc"eme

-&. 8efine a filter

%& %&

!ote that the t*o su,s stems can ,e com,ined as

S(C Fuasi-Ort"< -Simulation *esults.

% x % '$C -(uasi-Orthogonal.
II8 /lat /ading Channel Model (/d$s " 7.771)

% %

S(C Fuasi-Ort"< -Simulation *esults.

% x % '$C -(uasi-Orthogonal.
II8 /lat /ading Channel Model (Code 2ate " 162)

%1 %1

S(C Fuasi-Ort"< -Simulation *esults.

% x % '$C -(uasi-Orthogonal.
'CM Channel Model (Fr,an; Marco; G7 1mh)

%7 %7

S(C Matri+ G
(ransmit Matri+
/or % x 2 MIMO s stem

%8 %8

S(C Matri+ G
*ecei#er Desi,n
&t the recei+er:

%2 %2

-1. Consider the s m,ols X1 I X% at first

S(C Matri+ G

&0 &0

*here

S(C Matri+ G
and (Orthogonal matrix: P P " " P " P " I )

&1 &1

-2. Het the filter

and *e ha+e

S(C Matri+ G
-%. 'imilarl ; considering the s m,ols XJ I XK; *e ha+e

&2 &2

$herefore; XJ I XK can ,e detected , the same algorithm gi+en in step (1) and (2)

S(C Matri+ G -Simulation *esults.

% x % '$C -Matrix ).
II8 /lat /ading Channel Model (/d$s " 7.771)

&% &%

S(C Matri+ G -Simulation *esults.

% x % '$C -Matrix ).
II8 /lat /ading Channel Model (Code 2ate " 162)

&& &&

S(C Matri+ G -Simulation *esults.

% x % '$C -Matrix ).
'CM Channel Model (Fr,an; Marco; G7 1mh)

& &

Space (ime (rellis Co$e -S((C.

Co$e construction e+ample -8-state S((C !it" 2 transmit antennas -8->S?..

&1 &1

&ssuming that the input to the encoder at time # is the G input ,its ; the output of the encoder at time # is

*here $1k and $2k are the s m,ols transmitted at the first and second transmit antennas respecti+el

Space (ime (rellis Co$e -S((C.

3l,orit"m
M; $eco$er *here C : '$$C code ,oo- # % : sample si.e

&7 &7

MH decoder is implemented as a Hiter5i al,orit"m *here the trellis path *ith the smallest accumulated metric is chosen 3$#anta,e: L @ood tradeoff ,et*een $i#ersit' and co$in, ,ain L High code rate can ,e achie+ed S"orta,e: High decoding complexit

Space (ime (rellis Co$e -S((C.

:+ample C%D:

&8 &8

Space (ime (rellis Co$e -S((C.

Ot"er e+amples
%-0'1; Mt "2 at 2 ,6s6H.
& state 11 state %2 state

&2 &2

8 state

Space (ime (rellis Co$e -S((C.

9:* 0or t!o transmit antennas an$ one recei#e antenna Diversity order = 2

Space (ime (rellis Co$e -S((C.

9:* 0or t!o transmit antennas an$ t!o recei#e antennas Diversity order = 4

Space (ime (rellis Co$e -S((C.

K-0'1; Mt "2 at G ,6s6H.

32 state

16 state 8 state

Space (ime (rellis Co$e -S((C.

9:* 0or t!o transmit antennas an$ one recei#e antenna Diversity order = 2

Space (ime (rellis Co$e -S((C.

&

&

9:* 0or t!o transmit antennas an$ t!o recei#e antennas Diversity order = 4

Spatial Multiple+in, -SM.

SM Matri+ C
(ransmit matri+
/or 2 x 2 MIMO s stem /or % x % MIMO s stem

SM Matri+ C
*ecei#er Desi,n
/or % x % MIMO s stem; at the recei+er:

*here N is *hite @aussian distri,uted Ma+imum ;ikeli"oo$ (M;) recei+er (Optimal):

SM Matri+ C
Comments
Multi dimensional +ector se9uence search is re9uired 3xhausti+e search method is e+ponentiall' comple+ 2educed complexit solutions: )* Spherical decoding +* BLAST

SM Matri+ C -Sp"ere Deco$in,. C1D

3 tec"ni/ue 0or sol#in, ' @ 3+ Main i$ea o0 sp"ere $eco$in, is to re$uce computational comple+it' o0 M; $etector
'earches o+er onl lattice points in that lies in a certain h persphere of radius d around the recei+ed signal y 2e9uires onl pol nomial complexit

Sp"ere $etector e#aluates all & t"at satis0':

SM Matri+ C -Sp"ere Deco$in,.

("ere are t!o /uestions 0or sp"ere $eco$er:
Ho* to choose the radius d ? Ho* to tell *hich lattice points are inside the sphere M

10 10

C"oosin, d too lar,e lea$s to "i," comple+it'8 !"ile c"oosin, d too small results in an empt' sp"ere
& successi+e procedure ,ased on 5ack-su5stitution (e.g.; (2) is suggested

SM Matri+ C -Sp"ere Deco$in,.

11 11

Gack-su5stitution proce$ure: Gasic I$ea: 8etermine all lattice points in a sphere of dimension N *ith radius d , successi#el' $eterminin, all lattice points in sp"eres o0 lo!er $imensions 1828I8NJ1 !it" ra$ius d :+ample -Step 1. 8ecompose H " F*; *e ha+e

SM Matri+ C -Sp"ere Deco$in,.

*here ' is an unitar' matri+; and R is an upper trian,ular matri+ -Step 2. 'tarts *ith the last la'er searc" (1-$im) and successi#el' $etecte$ from the lo*er to upper la ers

12 12

-Step %. 'tops until the first la er (7 $imension) is detected

SM Matri+ C -Sp"ere Deco$in,.

1% 1%

G:* per0ormance e#aluation o0 sp"ere $etector !it" 0our transmit an$ recei#e antennas C1D

SM Matri+ C -G;3S(.
G;3S( (ranscei#er al,orit"m C&D C D

1& 1&

-1. Ordering -2. Interference nullin, (e.g.; MM'3 or N/) -%. Interference cancellation (e.g.; 'IC or 0IC)

SM Matri+ C -G;3S(.
1st Iteration
Inter0erence 7ullin, (decoupling N transmitted signals)

1 1

a. 8ecorrelating (N/) detector: 5. MM'3 : S'm5ol Decision: &ssume that E1 is the detected s m,ol

SM Matri+ C -G;3S(.
Inter0erence Cancellation

11 11

2n$ Iteration
Inter0erence 7ullin, a. Nero /orcing : 5. MM'3 :

SM Matri+ C -G;3S(.
S'm5ol Decision: &ssume that X2 is the detected s m,ol

17 17

(:ithout the interference X1 More >recise O) Inter0erence Cancellation

%r$8 &t" iterations are similar an$ &%8 && are $etecte$ E

SM Matri+ C -Simulation *esults.

2 x 2 'M !it" P)H&'$ recei#er -Matrix C.
II8 /lat /ading Channel Model (/d$s " 7.771)

18 18

SM Matri+ C -Simulation *esults.

2 x 2 'M !it" P)H&'$ recei#er -Matrix C.
II8 /lat /ading Channel Model (Code 2ate " 162)

12 12

SM Matri+ C -Simulation *esults.

2 x 2 'M !it" P)H&'$ recei#er -Matrix C.
'CM Channel Model (Fr,an; Marco; G7 1mh)

70 70

SM Matri+ C -Simulation *esults.

2 x 2 'M !it" MH recei#er -Matrix C.
II8 /lat /ading Channel Model (/d$s " 7.771)

71 71

SM Matri+ C -Simulation *esults.

% x % 'M !it" P)H&'$ recei#er -Matrix C.
II8 /lat /ading Channel Model (/d$s " 7.771)

72 72

SM Matri+ C -Simulation *esults.

% x % 'M !it" P)H&'$ recei#er -Matrix C.
II8 /lat /ading Channel Model (Code 2ate " 162)

7% 7%

SM Matri+ C -Simulation *esults.

% x % 'M !it" P)H&'$ recei#er -Matrix C.
'CM Channel Model (Fr,an; Marco; G7 1mh)

7& 7&

SM Matri+ C -Simulation *esults.

% x % 'M !it" MH recei#er -Matrix C.
II8 /lat /ading Channel Model (/d$s " 7.771)

7 7

71 71

Git Interlea#e$ Co$e$ Mo$ulation

S'stem Mo$el
Glock $ia,ram

77 77

MIMO OFDM

Con#olutional Co$e
:nco$er -Ginar' rate K con#olutional co$e. C7D 6enerator pol'nomial

78 78

Glock $ia,ram

Con#olutional Co$e
Deco$er -Hiter5i $eco$er. (rellis Mo$ule

72 72

Con#olutional Co$e
(rellis $ia,ram

80 80

Con#olutional Co$e
Deco$erLs (ask 2eco+er length-( code*ord code*ord )

81 81

from the gar,led

Con#olutional Co$e
Hiter5i al,orit"m 'uppose there are M code*ords e9ual pro,a,ilit @i+en the recei+ed se9uence ); decoded output if

82 82

*ith

is chosen as the

Con#olutional Co$e
&ssume the channel is memor)le**# *e ha+e

8% 8%

Con#olutional Co$e
Distance (Metric) ,et*een t*o ad<acent nodes

8& 8&

& path from a to + a code*ord code 8istance from a to + for :

of con+olutional

Con#olutional Co$e

8 8

is chosen , the trellis path from a to 5 *ith the smallest accumulate$ metric

Con#olutional Co$e
So0t-input Hiter5i al,orut"m
'oft input -to t"e Hiter5i $eco$er.:

81 81

*here *e defined the lo,-likeli"oo$ ratio as the metric:

Con#olutional Co$e
Hog-li-elihood ratio can 5e $eri#e$ 5' t"e $emo$ulator:

87 87

Con#olutional Co$e
Sli$in,-!in$o! $eco$in, !it" truncation len,t" ;

88 88

9i#e constraint len,t"s rule:

If the truncation length is larger enough ( 5m) then the sur+i+ors *ill ,e of common tail (m: 1.(otal enco$er memor' " 2.M o0 s"i0t re,ister)

Con#olutional Co$e
Sur#i#or (runcation /inite memor si.e /inite decoding dela ("ree strate,ies 0or sur#i#or truncation: 1< 9i+e$ state truncation 2< MaNorit'-#ote truncation %< Gest state truncation

82 82

Con#olutional Co$e
:+ample: ;@& Deco$e t"e 0irst 5ranc":

20 20

Con#olutional Co$e
3pproac" 1: /ixed 'tate $runcation 2egard a 0i+e$ state as the start of the trace ,ac 8ecide the output ,it , a trace *ith minimum accumulated distance

21 21

Con#olutional Co$e
3pproac" 2: Ma<orit Pote $runcation 2egard all states as the start of the trace ,ac 8ecide the output ,it , ma<orit +ote (7 or 1)

22 22

Con#olutional Co$e

2% 2%

3pproac" %: )est 'tate $runcation 2egard a state *ith the minimum accumulate$ metric (Gest state) as the start of the trace ,ac 8ecide the output ,it , a trace *ith minimum accumulated distance

Con#olutional Co$e
:+ample: ;@& Deco$e t"e 0irst 5ranc":

2& 2&

Con#olutional Co$e
Deco$e t"e secon$ 5ranc":

2 2

Con#olutional Co$e
Deco$e t"e k-t" 5ranc":

21 21

Con#olutional Co$e
Deco$e t"e -(otal len,t" B ;+1. to 0inal 5ranc"

27 27

Simulation *esults

28 28

Comparison 5et!een So0t $ecision O Har$ $ecision

&:@! Channel Model

Simulation *esults
Comparison 5et!een $i00erent truncation len,t"s
&:@! Channel Model

22 22

Simulation *esults

100 100

Comparison 5et!een $i00erent truncation strate,ies

&:@! Channel Model

Glock Interlea#er

101 101

& ,loc- interlea+er is defined , a t*o step permutations C7D:

(1) $he first ensures that ad<acent coded ,its are mapped onto nonad<acent su,carriers

(2) $he second permutation ensures that ad<acent coded ,its are mapped alternatel onto less or more significant ,its of the constellation; thus a#oi$in, lon, runs of lo*l relia,le ,its

Glock Interlea#er

102 102

$he de-interlea+er; *hich performs the in+erse operation; is also defined , t*o permutations:
(1) /irst permutation:

(2) 'econd permutation:

7ote: $he first permutation in the de-interlea+er is the in+erse of the second permutation in the interlea+er; and con+ersel

Glock Interlea#er
>ermutation >arameters
Ncpbs Ncpc s k, j mk jk mj kj !um,er of coded ,its per the allocated su,channels per O/8M s m,ol

10% 10%

!um,er of coded ,its per su,carrier; i.e.; 1; 2; % or Q for )0'1; (0'1; 1Q-(&M; or Q%-(&M; respecti+el " ceil( Ncpc 6 2 ) Index of the coded ,it ,efore the first permutation Index of that coded ,it after the first and ,efore the second permutation Index after the second permutation; <ust prior to modulation mapping Index of that ,it after the first and ,efore the second permutation Index of that ,it after the second permutation

Glock Interlea#er
Operatin, parameters 0or 5lock interlea#er

10& 10&

Glock Interlea#er
Glock siAes o0 t"e Git Interlea#er

10 10

*ate 3$aptation

101 101

*ate a$aptation is a -e functionalit to cope *ith time+ar ing channel conditions and to maximi.e the throughput of point-to-point *ireless lin-s 8esign a channel code *ith code rate # 6 n (e.9. (n;-) con+olutional code) is complex and impractical $o generate code rates of 26G; G6%; J6Q; and R6K; the rate 162 channel encoder outputs shall ,e puncture$

*ate 3$aptation
>uncturin, / Depuncturin,

107 107

:+ample: -So0t-input Hiter5i al,orit"m.

So0t metric -lo,-likeli"oo$ ratio. for each ,inar input:

*ate 3$aptation

108 108

>uncture patterns an$ serialiAation 0or con#olution co$e C7D

*ate 3$aptation

102 102

Supporte$ mo$ulations an$ inner -(CM. co$e rates

*ate 3$aptation

110 110

) ad<usting the constellation siAe of modulation; co$e rate of channel coding and puncture patterns; a tuna,le multi-rate communication can ,e reali.ed )ased on a re9uired packet error rate (>:*); there exists a tradeoff ,et*een
Constellation 'i.e (Modulation scheme) Code 2ate 2ecei+ed '!2

S(C Matri+ 3 -Simulation *esults.

2 x 2 '$C -Matrix &.

111 111

(2 x 2) II8 /lat /ading Channel Model (Code 2ate " 162)

S(C Matri+ 3 -Simulation *esults.

2 x 2 '$C -Matrix &.

112 112

$radeoff ,et*een co$e rate and constellation siAe under a fixed )32 (/32)

S(C Matri+ 3 -Simulation *esults.

% x % '$C -Matrix &.
II8 /lat /ading Channel Model

11% 11%

S(C Fuasi-Ort"< -Simulation *esults.

% x % '$C -(uasi-Orthogonal.
II8 /lat /ading Channel Model

11& 11&

S(C Matri+ G -Simulation *esults.

% x % '$C -Matrix ).
II8 /lat /ading Channel Model

11 11

SM Matri+ C -Simulation *esults.

2 x 2 'M !it" P)H&'$ recei#er -Matrix C.

111 111

II8 /lat /ading Channel Model (Wit" C"annel Co$in,)

SM Matri+ C -Simulation *esults.

% x % 'M !it" P)H&'$ recei#er -Matrix C.
II8 /lat /ading Channel Model

117 117

*e0erence
=1? '. M. &lamouti; S& simple transmit di+ersit techni9ue for *ireless communications;S I333 T'&C; +ol. 1Q; no K; pp. 1%J1-1%JK; Oct. 1UUK

118 118

=2? H. Tafar-hani; S& (uasi-Orthogonal 'pace-$ime )loc- Code;S I333 $rans. on Communications; +ol. %U; no. 1; pp. 1-%; 2771 =G? P. $aro-h; !. 'eshdri and &. 2. Calder,an-; V'pace-time codes for high data rate *ireless communication: 0erformance anal sis and code construction;W I333 $rans. Inform. $heor ; +ol. %%; no. 2; pp. R%%-RQJ; Mar. 1UUK

*e0erence

112 112

=%? @. T. /oschini; VHa ered space-time architecture for *ireless communication in a fading en+ironment *hen using multiple antennas;W )ell Ha,s ' st. $ech. T.; +ol. 1; pp. %1-JU; &utumn 1UUQ =J? 0. :. :olnains- ; @. T. /oschini; @. 8. @olden and 2. &. Palen.uela; V8etection algorithm and initial la,orator results using P-)H&'$ space-time communication architecture;W3lectronic Hetters; +ol. GJ; no. 1; Tan. 1UUU =Q? 3. Nimmermann; :. 2a+e and @. /ett*eis; VOn the complexit of sphere decoding;W in 0roc. :0MC 277%

*e0erence

120 120

=R? K72.1Q; I333 'tandard for Hocal and metropolitan area net*or-s; 0art 1Q: &ir Interface for /ixed )road,and :ireless &ccess ' stems =K? 8igga+i et al; V@reat expectations X $he +alue of spatial di+ersit in *ireless net*or-s;W 0roceeding of I333 277%