Lecture

 7:  MIMO  Capacity  and  

Multiplexing  Architectures

I-Hsiang Wang
ihwang@ntu.edu.tw

5/15, 2014
 Design  of  MIMO  Systems
• Regarding MIMO, what we have done so far:
- Established solid foundation on the statistical channel modeling
- Analyzed AWGN (no fading) MIMO capacity
• Indeed, MIMO is capable of the following:
- Multiplex multiple data streams simultaneously
- Provide spatial diversity
- Increase power gain
• What’s next:
- Derive MIMO capacity under fading
- Design transceiver architectures to extract multiplexing gain,
diversity gain, and power gain

2
 Plot
• Derive capacity of fading MIMO channel
- Fast fading: CSIR only and full CSI
- Slow fading: outage probability
• Discuss the nature of performance gains
• Introduce transceiver architectures for fast fading
- The V-BLAST family
• Introduce a transceiver architecture for slow fading
- D-BLAST

3
 Outline
• Capacity of fast fading MIMO
• V-BLAST
• Receiver architectures:
- Linear filters: decorrelator, matched filter, MMSE
- Combination with successive interference cancellation (SIC)

• Outage probability of slow fading MIMO

• D-BLAST

4
Fast  Fading  MIMO  Channel

5
 Fast  Fading  MIMO  with  Full  CSI
• Channel model: y[m] = H[m]x[m] + w[m]
- {H[m]}: random fading process which is stationary and ergodic

• With full CSI, Tx and Rx can perform pre- and post-

processing based on the SVD of H[m] at each time:

• H[m] = U[m]Λ[m]V[m]*

• Convert the fading MIMO into a fading parallel channel:

(nmin := min{nt,nr})
yei [m] = i [m]e
xi [m] ei [m], i = 1, 2, . . . , nmin
+w

• Water-filling to find the optimal power allocation

6
 Ergodic  Capacity  with  Full  CSI
1 [m] e1 [m]
w

x[m] y[m]

...
e[m]
x V[m] V*[m] U[m] U*[m] e [m]
y
nmin [m] enmin [m]
w

• Capacity via water-filling:

n
X min  ✓ 2 ⇤
◆
i P ( i)
CMIMO = E log 1 + 2
i=1

✓ ◆ n
"✓ ◆ #
2 + X min 2 +
P ⇤( ) = ⌫ 2
, ⌫ satisfies E ⌫ 2 =P
i=1 i

7
Transceiver  Architecture  with  Full  CSI
294 MIMO I: spatial multiplexing and channel modeling

AWGN {x~1[m]} {y~1 [m]}

Decoder
coder
.
.
n min {w[m]} .
.
information .
.
streams
~ [m]}
{x {y~nmin[m]}
AWGN nmin V H + U* Decoder
coder

{0}
.
.
.
{0}

Figure 7.2 The SVD architecture There is a clear analogy between this architecture and the OFDM system
for MIMO communication. pre-processing post-processing
introduced in Chapter 3. In both cases, a transformation is applied to convert a
matrix channel into a set of parallel independent sub-channels. In the OFDM
setting, the matrix channel is given by the circulant matrix C in (3.139),
defined by the ISI channel together with the cyclic prefix added onto the
input symbols. In fact, the decomposition C = Q−1 !Q in (3.143) is the SVD
decomposition of a circulant matrix C, with U = Q−1 and V∗ = Q. The
important difference between
8 the ISI channel and the MIMO channel is that,
 Receiver  CSI  Only:  V-­‐BLAST
• Tx cannot apply the pre-processing matrix V
MIMO II: capacity and multiplexing architectures
• V-BLAST architecture:
AST P1 AWGN coder w[m]
municating rate R1
nel. y[m] ··
·· x[m] Joint ··
·· Q H[m] + ··
decoder ··
Pnt AWGN coder
rate Rnt

- Tx prepares nt data streams, each encoded with a rate Ri coder

coordinate system given by a unitary matrix Q, not necessarily dependent on
- Generate x by multiplying them with a unitary matrix Q
the channel matrix H. This is the V-BLAST architecture. The data streams
- Rx carries jointly.
are decoded out joint
Thedecoding of the
kth data stream is streams
allocated a(eg.,
powerML)Pk (such that

• Wethewill
sum discuss
of the powers, + · · · + Pnt , is equal
RxP1architecture to P, the total transmit power
later
constraint) and is encoded using a capacity-achieving Gaussian code with rate
!nt
Rk . The total rate is R = k=1 Rk .
As special cases:
9
Capacity  of  Fast  Fading  MIMO  with  CSIR
• Using information theoretic arguments (or a sphere packing
argument), one can show that V-BLAST can achieve the
capacity, which is given by the following:
 ✓ ⇤
◆
HKx H
C= max E log det Inr + 2
Kx :Tr(Kx )P

• Kx := the covariance matrix of transmit signal vector x

• K
With V-BLAST, x = Q diag (P 1 , . . . , P nt ) Q ⇤

• The issue boils down to finding the optimizing Kx for a

given stationary distribution of H

10
 Multiplexing  in  the  Angular  Domain
• In V-BLAST, Q can be thought of as the coordinate
system onto which Tx modulates its data streams
- The question is, which coordinate system Q should be used?

• The choice of Q (and the power allocation {P1, … ,Pn}) depends on

the statistical property of H, so let’s focus on the angular
domain representation: Ha = Ur*HUt

• Under rich scattering, entries of Ha are statistically

independent and zero-mean ⟹ it is reasonable to
multiplex data on the coordinate system (indeed, optimal. HW.)
• Choose Q = Ut and hence Kx = Ut ΛpUt*.

• Still need to determine the power allocation Λp

11
 Uniform  Power  Allocation
• If further symmetry is present the random Ha, uniform
! nP! t Int ) turns out to be optimal
power allocation (!⇤! p =
• Sufficient condition: the nt column vectors of Ha are i.i.d.
- For example, i.i.d. Rayleigh faded Ha.

• Hence the covariance matrix Kx = Ut ΛpUt* = P

nt Int

• This gives us the capacity formula: (SNR := P/σ2)

 ✓ ◆
SNR ⇤
C = E log det Inr + HH
nt

• In this case, Q can be any unitary matrix; in particular, it

suffices to choose Q = Int

12
 V-­‐BLAST  under  i.i.d.  Rayleigh
MIMO II: capacity and multiplexing architectures

AST P1 AWGN coder w[m]

municating rate R1
nel. y[m] ··
·· x[m] Joint ··
·· I
Q H[m] +
decoder ··
··
Pnt AWGN coder
rate Rnt
P
P1 = P2 = · · · = Pn t = nt

• Effectively, each
coordinate system
the channelan
matrix
ofbythe
given Tx antennas,
a unitary
H. This is the V-BLAST
say, antenna
matrix Q, not necessarily
architecture.
dependenti,on
The data
transmits independent data stream of rate Ri streams
are decoded jointly. The kth data stream is allocated a power Pk (such that
• the sum
How of the powers, PR
to determine 1+
i?· · · + Pnt , is equal to P, the total transmit power
constraint) and is encoded using a capacity-achieving Gaussian code with rate
- For joint
Rk . The ML,
total rateit isdoes
!nt
R = not matter as long as ΣRi = the total capacity
k=1 Rk .

- ForAsother
specialRx
cases:
architectures (later), the individual rate depends on
the
•
effective channel it faces with, after the MIMO detector
If Q = V and the powers are given by the waterfilling allocations, then we
have the capacity-achieving architecture in Figure 7.2.
• If Q = Inr , then independent data streams are sent on the different transmit
13
 Receiver  CSI  vs.  Full  CSI
• Capacity formula can be rewritten using singular values
of the random matrix H: 1 2 ··· nmin 0
nP
min h ⇣ ⌘i
CCSIR = E log 1 + SNR
nt
2
i
i=1

- No CSIT ⟹ water-filling is not possible

• Recall the capacity with full CSI:

nP
min ⇥ ⇤
⇤
CFull CSI = E log 1 + SNR ( i ) 2
i
i=1
nP
⇣ ⌘+
min
⇤ +
SNR ( ) = ⌫ 1
2 , ⌫ satisfies E ⌫ 1
2 = SNR
i
i=1

14
 DoF  Gain  at  High  SNR
–10 10 20 30
8.2 Fast fading MIMO channel SNR (dB)

i.i.d. C (bits /s / Hz) 35 C (bits /s / Hz) 70

wer: 30 nt = nr = 1 60 nt = nr = 1
nt = 1 nr = 4 nt = 1 nr = 8
25 nt = nr = 4 50 nt = nr = 8

20 40

15 30

10 20

5 10

–10 10 20 30 –10 10 20 30
SNR (dB) SNR (dB)

• High SNR regime: nmin-fold DoF gain

C (bits /s / Hz) 70
⇣ ⌘ !n2i2P
where is a !-square distributed random variable with
60 min ⇥ ⇤
nt = nr = 1
C
nt = 1 nCSIR
r=8
⇡ nmin log SNRfreedom.
+ E log 2
i
nt Note that the number of degrees
50 nt = nr = 8 i=1 of freedom is limited by
40 ⇣ of ⌘the number
nPmin
of transmit and the number of receive antennas
⇥ need multiple⇤ transmit and multiple rec
SNRa large capacity, we
30 CFull CSI ⇡ +
nmin log nminTo emphasize E fact,
this
2
log we also
i plot the capacity of a 1 by
20
i=1
Figure 8.2. This capacity is given by

• nmin :=
10
min{nt,nr} determines the high-SNR
! " slope
C = ! log 1 + SNR
nr
#
!hi !2
$%
bits/s/Hz"
–10 10 20 30
SNR (dB) 15
i=1
 –30 –20 –10 10

Power  Gain  at  Low  SNR

8.2 Fast fading MIMO channel
SNR (dB)

C (bits / s / Hz)
C1,1
capacities. C (bits / s / Hz) 8
a 4 by 4 C1,1
4
by 8 an 8 7
ty is a
channel 3.5 6

nt = 1 nr = 4 nt = 1 nr = 8
nt = nr = 4 nt = nr = 8 5
3
4
2.5
3

–30 –20 –10 10 –30 –20 –10 10

SNR (dB) SNR (dB)

• nr determinesC theLargepower gain

antenna array
C
regime under CSIR
(bits / s / Hz)
We saw that in the high SNR regime, the capacity increases line
nminP
1,1
SNR
8 ⇥ ⇤ SNRof the number of transmit
n7t E E [Tr (HH )] log e
2 minimum ⇤ and the number of receive an
CCSIR ⇡ i log2 e = nt 2 SNR regime, the capac
is a degree-of-freedom gain. In the low
i=1
linearly with the number of receive antennas. This is a power g
6 = n SNR log e
combined
r effect of the2 two types of gain yield a linear growth i
nt = 1 nr = 8 5
any SNR, as we scale up both nt and nr ? Indeed, this turns out to

• nt = nr = 8
If CSIT is available,
4
power gain is larger due to both us focus on the square channel nt = nr = n to demonstrate this.
With i.i.d. Rayleigh fading, the capacity of this channel is (cf
beamforming and3 dynamic power allocation !
"n #
# 2
$%
Cnn !SNR" = ! log 1 + SNR i $
i=1 n
–30 –20 –10 10 16
Nature  of  Peformance  Gain  (CSIR  only)
• High SNR (DoF-limited):
- min{nt,nr}-fold DoF gain
- capacity scaling linearly with nmin := min{nt,nr}
- MIMO is crucial
• Low SNR (Power-limited):
- nr-fold power gain
- capacity scaling linearly with nr
- Only need multiple Rx antennas
• At moderate SNR
- min{nt,nr}-fold gain
- Due to a combination of both effects

17
Receiver  Architectures

18
 Decoding  at  the  Receiver
• In the previous V-BLAST architecture, Rx uses ML:
- ML is optimal
- But the complexity grows exponentially with the # of data streams
• A natural approach:
- First separate the signal of each data stream from others with
certain linear operations
- Then decode each data stream using single-user decoder
• In the following we focus on Rx architectures that use
linear operations in the first step
- Assuming V-BLAST with Q = Int , that is, each Tx antenna sends
an independent data stream
- If not, we can just lump Q into the channel matrix H

19
 Decorrelator:  Interference  Nulling
• Rewrite the received signal vector y as follows:
X
y = h i xi + h j xj + w
j6=i

- xi denotes the signal sent from the i-th Tx antenna

- hi denotes the i-th column of channel matrix H, representing the
signal direction of xi .

• To decode xi , a simple idea is to use a decorrelator:

- First null out interference by projecting y onto the null space of
the directions of all interfering vectors {hj | j≠i}
- Then apply matched filter to the projected signal
• The Rx architecture consists of a bank of decorrelators.
• Also called interference nulling, zero forcing, etc.

20
 column of the pseudoinverse H† of the matrix H, defined by

Bank  of  Decorrelators

H† $= !H∗ H"−1 H∗ % (8.47)

Decorrelator ⇤
for stream 1
ye1 := (Q1 h1 ) (Q1 y) Rows of Qk form a
orthonormal basis
of the null space of
Decorrelator ⇤
for stream 2
ye2 := (Q2 h2 ) (Q2 y) {hj | j≠k}.

y[m] matched filter nulling

after projection
⇥ Pk
⇤
Rk = E log 1 + 2 ||Qk hk ||2

Decorrelator
Note: for successful decorrelation,
g for stream nt nt ≤ nr and hence nmin = nt

21
 Performance  in  i.i.d.  Rayleigh
• Uniform power allocation ⟹ the overall achievable rate:
nP
min h ⇣ ⌘i
SNR
Rdecorr = E log 1 + nt ||Q k h k || 2
k=1

• At high SNR:
⇣ ⌘
SNR
nP
min ⇥ ⇤
Rdecorr ⇡ nmin log nt + E log ||Qk hk ||2
k=1
⇣ ⌘ nP
min ⇥ ⇤
SNR
CCSIR ⇡ nmin log nt + E log 2
i
i=1

• At low SNR: lose the power gain (HW)

• The decorrelator fully extracts the spatial multiplexing
gain of V-BLAST, but not the power gain

22
 straightforward to show that, just like the capacity, the total rate achievable

Performance  Gap  from  Capacity

by the bank of decorrelators scales linearly with the number of antennas (see
Exercise 8.21).

70
Constant gap due to
60 loss of power gain;
Capacity quite substantial
50 Decorrelator
bits / s / Hz

40

30

t of rate 20
the
nk for the 10
. Rayleigh
The capacity of 0
lso plotted for –10 –5 0 5 10 15 20 25 30
SNR (dB)

23
 Removing  the  Gap:  SIC
• Recall in Lecture 5, to achieve the uplink capacity, we
use successive interference cancellation (SIC) at Rx
• Similar idea can be applied here to remove this gap
• The only difference from the previous derivation
- ek
The nulling projection Qk is replaced by Q
- The rows of! Q!e k now form an orthonormal basis of the null space
of {hj | j > k}.

• Indeed this removes the gap at high SNR

24
 block length to be very large, so that streams are successfully cancelled

Successive  Interference  Cancellation

with very high probability. With this assumption the kth data stream sees
only down-stream interference, i.e., from the streams k + 1# $ $ $ # nt . Thus,

Decode Stream 1
Decorrelator 1
stream 1

Subtract Decode Stream 2

Decorrelator 2
stream1 stream 2

Subtract Decode Stream 3

y[m] Decorrelator 3
stream1, 2 stream 3

11 Decorrelator–SIC:
decorrelators with Subtract Decode Stream nt
stream Decorrelator nt
cancellation of stream nt
1, 2, ..., nt –1

25
 Zero  Forcing  vs.  Matched  Filtering
X
y = h i xi + h j xj + w
j6=i

• Decorrelator (zero forcing): remove all interference at the

expense of reducing the received SNR

• Matched filter: projecting onto hi to maximize the SNR

but SINR may be bad

26
 The decorrelator was motivated by the fact that it completely nulls out inter-
Zero  Forcing  vs.  Matched  Filtering
stream interference; in fact it maximizes the output SNR among all linear

R 1
C 0.9 Matched fillter
Decorrelator
0.8
0.7
0.6
0.5
.13 Performance (ratio
ate to the capacity) of
0.4
ched filter bank as 0.3
d to that of the 0.2
ator bank. At low SNR,
0.1
ched filter is superior.
osite is true for the 0
ator. The channel is –30 –20 –10 0 10 20 30
leigh with nt = nr = 8. SNR (dB)

• Low SNR: power-limited, interference not significant

• High SNR: interference-limited, power not significant

27
 Optimal  Linear  Filter:  MMSE
X
y = h i xi + h j xj + w
j6=i

• The optimal linear filter should maximize SINR at all SNR

• It offers the optimal compromise between ZF and MF
• Procedure:
P
- First whiten the aggregate interference + noise zi :=
j6=i
h j xj + w

- The apply MF since matched filter is optimal for white noise

⇣ ⌘⇤ ⇣ ⌘
•
1 1 ⇤
1
MMSE filter: ye1 := Kz1 h1
2
Kz 1 y = Kz 1 h 1
2
y

matched filter whitening

after whitening

28
 y = hx + z! (8.58)

Geometric  Picture
where z is complex circular symmetric colored noise with an invertible covari-
ance matrix Kz , h is a deterministic vector and x is the unknown scalar symbol

Decorrelator
Optimal filter

Signal direction
(matched filter)

Figure 8.14 The optimal filter

goes from being the Interference subspace
decorrelator at high SNR to
being the matched filter at low
SNR.

• Low SNR: MMSE → MF

• High SNR: MMSE → ZF

29
 Linear  MMSE:  Performance
• For the k-th stream, its data rate is
h ⇣ ⌘i
1 ⇥ ⇤
Rk = E log 1 + Pk ||K zk
2
hk || 2
= E log 1 + Pk h⇤k Kzk1 hk
P P
• Since !z!k :=! ! hj!xj +
j6=k
! w , we have Kzk := 2
In r +
j6=k
Pj hj h⇤j

• Hence the achievable rate can be computed

• We can further improve the performance of the MMSE

receiver by combining with SIC

30
 Linear  MMSE:  Performance
8.3 Receiver architectures

Performance (the R 1
to the capacity) of MMSE
C 0.9
of MMSE Matched filter
compared to the 0.8 Decorrelator
r bank and to the
bank. MMSE
0.7
ter than both, 0.6
re range of SNR. R
is i.i.d. Rayleigh C88 0.5
= 8. 0.4
0.3
0.2
0.1
0
–30 –20 –10 0 10 20 30
SNR (dB)

• Low SNR: MMSE → MF

C
• High SNR: MMSE → ZF
Analogous to what we did in Section 8.3.2 for the decorrelator, we can now
upgrade the basic bank of linear MMSE receivers by allowing successive
cancellation of streams as well, as depicted in Figure 8.16. What is the
performance improvement in using the MMSE–SIC receiver? Figure 8.17
31
 r
MMSE receivers with successive cancellation and equal power allocation
MMSE-­‐SIC
achieves the capacity of the i.i.d. Rayleigh fading channel.

MMSE Decode Stream 1

receiver 1 stream 1

Subtract MMSE Decode Stream 2

stream 1 receiver 2 stream 2

Subtract MMSE Decode Stream 3

y[m] stream 1, 2 stream 3
receiver 3

6 MMSE–SIC: a
near MMSE receivers,
mating one of the
ta streams, with
uccessively cancelled Subtract MMSE Decode Stream nt
eceived vector at stream receiver nt stream nt
1, 2, ... , nt –1
.

P
SINRk,MMSE = Pk h⇤k Kzk1 hk Kzk := 2
In r + Pj hj h⇤j
j>k

32
 MMSE-­‐SIC  Achieves  MIMO  Capacity
MIMO II: capacity and multiplexing architectures

The MMSE–SIC R 1.1

eves the capacity
C 1
channel when MMSE–SIC
. Rayleigh. 0.9 MMSE
Matched filter
0.8 Decorrelator
0.7
R 0.6
C88
0.5
0.4
0.3
0.2
0.1
0
–30 –20 –10 0 10 20 30
SNR (dB)

• In fact for any channel matrix H and Kx, we have

Innfact, the MMSE–SIC receiver is optimal in a much
P t ⇣ stronger sense:⇤ it⌘
achieveslog
the (1
best+possible
SINRsum rate (8.2)
) = oflog
thedet
transceiver
I architecture
+ HKx H in
k,MMSE nr 2
k=1 8.1 for any given H. That is, if the MMSE–SIC receiver is used for
Section
demodulating the streams and the SINR and rate for stream k are SINRk and
log!1 + SINRk " respectively, then the rates sum up to
33
 Optimality  of  MMSE-­‐SIC
• Sending independent streams at individual Tx antennas
⟹ can be viewed as a multi-user uplink channel
• Hence, the optimality of SIC is straightforward

• At each stage, MMSE filter gives a sufficient statistics for

decoding the data from the vector signal
- This is because the capacity achieving code for each data stream
looks like Gaussian
- Hence the aggregate noise + interference is still Gaussian
• Hence, MMSE is information lossless at each stage

34
 Summary  of  Linear  Filters

y v v*y
X
y = h k xk + h j xj + w
j>k

• Matched filter: vMF = hk

• ⇤
Zero-forcing filter: vZF = (Qk Qk ) hk
- The rows of! Qk form an orthonormal basis of the null space of
{hj | j > k}

• v
MMSE filter: MMSE = K 1
zk h k
P
- Here Kzk := 2
In r +
j>k
Pj hj h⇤j

35
Slow  Fading  MIMO  Channel

36
 Optimal  Outage  Performance
• Outage probability:
- For a given input covariance
⇣ Kx, the outage
⌘ event is the event
HKx H⇤
log det Inr + 2 <R
- To find the optimal outage probability, need to find Kx such that
the corresponding outage probability is minimized:
⇢ ✓ ⇤
◆
HKx H
pout (R) = min Pr log det Inr + 2
<R
Kx :Tr(Kx )P

• With i.i.d. Rayleigh fading, a good upper bound of pout is

obtained with uniform power allocation:
⇢ ✓ ◆
SNR
pi.i.d.
out (R) = Pr log det Inr + HH⇤ <R
nt

37
Diversity  Order  of  i.i.d.  Rayleigh  MIMO
• A simple repetition scheme:
- Use one Tx antenna at a time to send the same symbol x
- Successively on the different nt antennas over nt time slots
- Equivalent channel: a SIMO channel with nt×nr outputs
- Diversity order nt×nr is achieved

• nt×nr is also the maximum diversity order that can be

extracted from the MIMO system (easy to check)

38
 V-­‐BLAST  is  Suboptimal
• In V-BLAST, essentially each data stream is sent on one
Tx antenna only
• ⟹ diversity order of each data stream ≤ nr
• A more detailed look:
- Outage event of the channel:
⇣ ⌘
P
nt
HKx H ⇤ P
nt
log (1 + SINRk,MMSE ) = log det Inr + 2 <R= Rk
k=1 k=1
- For V-BLAST, on the other hand, outage occurs whenever one
data stream k cannot be successfully decoded:
log (1 + SINRk,MMSE ) < Rk

39
 8.5.2 Coding across transmit antennas:
a channel D-BLAST
with a (random) ratio SI
signal-to-noisechannels
viewed a
D-­‐BLAST:  Coding  over  Tx  Antennas
viewed as a Antennachannel
parallel 1: Receive
with n sub-channe
and get b
Antenna 1: t
Significant
no coding improvement
acrossAntenna
these of V-BLASToutage
2:sub-channels: noto
has
in codin
com
there
Sectio
Antenna 2: Antenna 1:
transmit antennas.
of these sub-channels How do we
is Receive improve
in a deep fade and cann of
the these
archite
• Example: 2Receive
Tx antennas
(a)
more clearly
Antenna 2:
how to proceed, one
(a)
stream using that sub-channel. On the other hand,
Suppress
can stream
draw an
channels
u
an
- Breaks each codeword into and the
channels, parallel
two blocks
a channel
fading
we can average over
withoutage
a (random)
channel. the
(b)
Receive
Inrandomness
V-BLAST,
signal-to-noise and
the
of th
SIb
getthe
ratio
Receive and get better performance. From our From discu
Antenna 1: viewed as a parallel
Antenna channel
1: with
in Section 5.4.4, we know reliable communication
Antenna 1: n t in Sectio
sub-channe
condition
Antenna 1:
Antenna 2: no coding across these2:
Antenna
Antenna 2:sub-channels: outage there
universal
Antenna 2: !n t
of these sub-channels is in a log"1 deep +
(c) fadeSINRand cann
universal
# > R%
Receive Suppress k
Suppress stream using that sub-channel. On the other Howev hand,
(a) k=1 (b)
(b) Cancel
channels, we can
From the decomposition average over the randomness
(8.88), we see SINR From of
that 2 ),the
thiswt
Receive and get better Antenna
outage 1:
Antenna 1: Antennaperformance.
1: From received
our discu
condition
condition of the original
Antenna 2:
MIMO channel as well
Antenna
Antenna 2:
1: in Section codes
universal 5.4.4,Antenna
we know
for 2:
the reliablechannel
parallel communication
universal
the
cantwo bes
(c)
Antenna 2: (c)
universal codes for the original !nt
MIMO universal
egg
channel. prob
Receive
log"1
(d) + SINRk #Howev> R%
However, there is a channel
problem here. To two and get
obtain theth
• Each codeword Suppress
Cancel
(b)
is faced with a parallel
SINR2 ), we are Figure
assuming
8.18 How
Cancel
k=1
that the
D-BLAST
with
first SINR
is to
stream ), isw
2stag
sub-channels,
Antenna 1: with SINR 1 and
From
receivedthe SINR
decomposition
signal
2
works.
is
respectively
Antenna 1: (8.88), we see that this
(a) A soft estimate
cancelled off. of
However, received
mitto anten
code
Antenna 1: condition of the Antenna 2:
original MIMO channelantennas
as wellss
Antenna 2:
the two streams block
shouldA of the
be first codeword
jointly decoded. the two
There
Antenna 2:
universal codes for
(layer) the
obtainedparallel
without
Receive
egg problem: without decoding(d)the first stream, channel can
egg be
prob
Let on u
Receive
(c) interference. (b) A soft MMSE
(d) universal codes
and get the second for the
stream original
in the MIMO
first place.channel.
and
(Figure
The get
keyt
8
40estimate of block B is obtained
 Parallel  Channel  Conversion
• D-BLAST convert the MIMO channel into a parallel
channel for each data stream
- The sub-channels have SINR’s {SINR1,MMSE, SINR2,MMSE,...}

• Hence, the outage event becomes

⇣ ⌘
P
nt
HKx H ⇤
log (1 + SINRk,MMSE ) = log det Inr + 2 <R
k=1
- The same as in the optimal scenario
• D-BLAST attains the optimal outage performance

41