Mimo tutorial by-fuyun_ling

Topics in Digital Communications
Fundamentals of MIMO
Technology and Spatial
Multiplexing
© 2012 Fuyun Ling
fling@twinclouds.com
Fundamentals of MIMO
Technology and Spatial
Multiplexing
Fuyun Ling

Outline
• Introduction
• Efficiency of MIMO Data Transmission with Spatial
Multiplexing
• Basic Forms of MIMO Transmission and Reception
• MIMO Receivers – Complexity Reduction
• Examples of MIMO Systems
• Additional Discussion and Remarks
• Acknowledgement
© 2012 Fuyun Ling
• Introduction
• Efficiency of MIMO Data Transmission with Spatial
Multiplexing
• Basic Forms of MIMO Transmission and Reception
• MIMO Receivers – Complexity Reduction
• Examples of MIMO Systems
• Additional Discussion and Remarks
• Acknowledgement
2

Introduction
© 2012 Fuyun Ling
3

What is MIMO
• MIMO: Multiple Input/Multiple Output Systems
– Doesn’t have to be wireless
• Degenerated forms:
SIMO MISO SISO
• MIMO can provide multiple independent channels
(spatial multiplexing).
– SIMO and MISO only provide diversity over a single channel
© 2012 Fuyun Ling
• MIMO: Multiple Input/Multiple Output Systems
– Doesn’t have to be wireless
• Degenerated forms:
SIMO MISO SISO
• MIMO can provide multiple independent channels
(spatial multiplexing).
– SIMO and MISO only provide diversity over a single channel
Rev. 06/2012 4

History of MIMO
• Early Results
– A.R. Kaye and D.A. George (1970)
• First Description of MIMO Channels
– Branderburg and Wyner (1974)
• Capacity Analysis
– W. van Etten (1975, 1976)
• Receiver Structures
– Arogyaswami Paulraj and Thomas Kailath (1993, 1994)
• Concept of Spatial Multiplexing
– Telatar (1995)
• MIMO Channel Capacity
© 2012 Fuyun Ling
• Early Results
– A.R. Kaye and D.A. George (1970)
• First Description of MIMO Channels
– Branderburg and Wyner (1974)
• Capacity Analysis
– W. van Etten (1975, 1976)
• Receiver Structures
– Arogyaswami Paulraj and Thomas Kailath (1993, 1994)
• Concept of Spatial Multiplexing
– Telatar (1995)
• MIMO Channel Capacity
Rev. 06/2012 5

History of MIMO (cont.)
• Major Re-Introduction of MIMO Concepts and Systems
– Greg Raleigh and Gerard J. Foschini (1996)
• Described a practical system with “Layered Space-Time
Architecture” and detailed analysis
– Bell Labs announcement on BLAST MIMO technology (1998)
• First demonstrated spatial multiplexing in laboratory prototype
• Caught industry’s attention on the MIMO technology
• Started new development wave
• The title is a little vague to cause some misinterpretation
– Caused more interest due to the misunderstanding
– Nothing was wrong in the text
© 2012 Fuyun Ling
• Major Re-Introduction of MIMO Concepts and Systems
– Greg Raleigh and Gerard J. Foschini (1996)
• Described a practical system with “Layered Space-Time
Architecture” and detailed analysis
– Bell Labs announcement on BLAST MIMO technology (1998)
• First demonstrated spatial multiplexing in laboratory prototype
• Caught industry’s attention on the MIMO technology
• Started new development wave
• The title is a little vague to cause some misinterpretation
– Caused more interest due to the misunderstanding
– Nothing was wrong in the text
Rev. 06/2012 6

History of MIMO (cont.)
• Practical Development
– Iospan Wireless Inc. developed the first commercial system in
that used MIMO with OFDM (2001)
– Airgo Networks developed an IEEE 802.11n precursor
implementation based on MIMO-OFDM (2005)
– MIMO-OFDM in IEEE 802.11n standard (development from
2006)
– MIMO-OFDM in IEEE 802.16e (WiMax) standard (development
from 2006)
– MIMO-OFDM in 3GPP HSDPA, LTE standards (development
from 2008)
© 2012 Fuyun Ling
• Practical Development
– Iospan Wireless Inc. developed the first commercial system in
that used MIMO with OFDM (2001)
– Airgo Networks developed an IEEE 802.11n precursor
implementation based on MIMO-OFDM (2005)
– MIMO-OFDM in IEEE 802.11n standard (development from
2006)
– MIMO-OFDM in IEEE 802.16e (WiMax) standard (development
from 2006)
– MIMO-OFDM in 3GPP HSDPA, LTE standards (development
from 2008)
Rev. 06/2012 7

Key Features of MIMO
• Providing Spatial Diversity
– Not unique, SIMO and MISO can also provide spatial diversty
• SIMO – Receiver diversity
• MISO – Transmitter diversity
• Providing Spatial Multiplexing
– Unique to MIMO
– MIMO system can create N, up to min(Nt, Nr), independent
parallel channels
– Essential for achieving high spectrum efficiency (capacity >
1bit/Hz) at high SNR
– Spatial Multiplexing may not be advantageous in low SNR
environment
© 2012 Fuyun Ling
• Providing Spatial Diversity
– Not unique, SIMO and MISO can also provide spatial diversty
• SIMO – Receiver diversity
• MISO – Transmitter diversity
• Providing Spatial Multiplexing
– Unique to MIMO
– MIMO system can create N, up to min(Nt, Nr), independent
parallel channels
– Essential for achieving high spectrum efficiency (capacity >
1bit/Hz) at high SNR
– Spatial Multiplexing may not be advantageous in low SNR
environment
Rev. 06/2012 8

Efficiency of MIMO Data
Transmission with Spatial
Multiplexing
© 2012 Fuyun Ling
Efficiency of MIMO Data
Transmission with Spatial
Multiplexing
Rev. 06/2012 9

Spectral Efficiency of Single Channel Transmission
• System Capacity from Information Theory
– Shannon’s unconstrained, bandwidth-limited channel capacity for
the ideal AWGN channel with arbitrary inputs is
C = log2(1 + SNR)
• C is in unit of bit/(second*Hz) per use (I.e. per channel)
• At low spectral efficiency, i.e., low SNR: C ≈ SNR/ln 2 (linear)
• At high spectral efficiency, i.e, C and SNR are large:
C ≈ log2(SNR) (~ 3 dB, or double SNR, per bit)
• Example:
– C = 1 => SNR = 1 (0 dB);
– C = 5 => SNR = 31 (14.9 dB);
– C = 10 => SNR = 1023 (30.1 dB);
– C = 15 => SNR = 32767 (45.2 dB: difficult to achieve!);
– Conclusion: difficult to achieve very high spectral efficiency
transmission with a single channel.
© 2012 Fuyun Ling
• System Capacity from Information Theory
– Shannon’s unconstrained, bandwidth-limited channel capacity for
the ideal AWGN channel with arbitrary inputs is
C = log2(1 + SNR)
• C is in unit of bit/(second*Hz) per use (I.e. per channel)
• At low spectral efficiency, i.e., low SNR: C ≈ SNR/ln 2 (linear)
• At high spectral efficiency, i.e, C and SNR are large:
C ≈ log2(SNR) (~ 3 dB, or double SNR, per bit)
• Example:
– C = 1 => SNR = 1 (0 dB);
– C = 5 => SNR = 31 (14.9 dB);
– C = 10 => SNR = 1023 (30.1 dB);
– C = 15 => SNR = 32767 (45.2 dB: difficult to achieve!);
– Conclusion: difficult to achieve very high spectral efficiency
transmission with a single channel.
Rev. 06/2012 10

MIMO for High Spectral Efficiency Transmission
• MIMO and Spatial Multiplexing
– In proper environments, a MIMO system can create multiple
independent parallel channels from Spatial Multiplexing
• The number (N) of such channels is ≤ min(NTx, NRx)
– In the ideal case, these channels have the same magnitudes
and SNRs
• To fairly compare with non-MIMO case, we assume total transmitted
power are the same (the same total interference power to other
cells) in both cases, i.e., each channel has an SNRMIMO = SNR/N.
– The total Spectral Efficiency is
• At low SNR
since
© 2012 Fuyun Ling
• MIMO and Spatial Multiplexing
– In proper environments, a MIMO system can create multiple
independent parallel channels from Spatial Multiplexing
• The number (N) of such channels is ≤ min(NTx, NRx)
– In the ideal case, these channels have the same magnitudes
and SNRs
• To fairly compare with non-MIMO case, we assume total transmitted
power are the same (the same total interference power to other
cells) in both cases, i.e., each channel has an SNRMIMO = SNR/N.
– The total Spectral Efficiency is
• At low SNR
since
Rev. 06/2012 11
1
2 2
0
log (1 ) log (1 / )
N
Total i
n
C SNR N SNR N


   
/
( 0), ln2
ln2
Total
N SNR N
SNR C SNR

  
ln(1 ) 0x x for x  

Spectral Efficiency with Spatial Multiplexing
(N = 1, 2, and 4)
-20
-10
0
10
20
30
40
50
60
70
0.01 0.1 1 10 100
SNR(dB)
# of bits per Hz
Requred SNR at different bits/Hz
N = 1
N = 2
N = 4
© 2012 Fuyun Ling
-20
-10
0
10
20
30
40
50
60
70
0.01 0.1 1 10 100
SNR(dB)
# of bits per Hz
Requred SNR at different bits/Hz
N = 1
N = 2
N = 4
Rev. 06/2012 12

Conclusions on Transmission Spectral
Efficiency using MIMO
• In high spectral efficiency region (high SNR), 3 dB SNR
increase is needed for the increase of each bit per Hz.
– It’s difficult to achieve high spectral efficiency if only a single
channel is used
• With spatial multiplexing, MIMO can create multiple
parallel channels to achieve high efficiency transmission
– It will be most efficient in a rich scattering environment, where
the magnitudes of these channels have small dispersion
• With constant transmitted power (constant interference),
MIMO does not improve the efficiency at low SNR
comparing to the single channel case
– MIMO is much more effective in high SNR regions
© 2012 Fuyun Ling
• In high spectral efficiency region (high SNR), 3 dB SNR
increase is needed for the increase of each bit per Hz.
– It’s difficult to achieve high spectral efficiency if only a single
channel is used
• With spatial multiplexing, MIMO can create multiple
parallel channels to achieve high efficiency transmission
– It will be most efficient in a rich scattering environment, where
the magnitudes of these channels have small dispersion
• With constant transmitted power (constant interference),
MIMO does not improve the efficiency at low SNR
comparing to the single channel case
– MIMO is much more effective in high SNR regions
Rev. 06/2012 13

Basic Forms of MIMO
Transmission and reception
© 2012 Fuyun Ling
Basic Forms of MIMO
Transmission and reception
Rev. 06/2012 14

High Level Description of a MIMO System
• Block Diagram of a MIMO System
• Matrix Representation
( ) Mod. Symbolsi k s
© 2012 Fuyun Ling
• Block Diagram of a MIMO System
• Matrix Representation
Rev. 06/2012 15
 0 1( ) ( ) ( ), ( ) ( ) ( )r r t t r t t
T
N N N N N N Nk k k k k k   y H x n x s s
( ) Mod. Symbolsi k s

Channel Matrix and Its Estimation
• To utilize the potential of MIMO, it is essential to
accurately know the channel matrix (or simply H).
• The channel matrix is usually estimated using known
symbols embedded in the transmitted data squence
– Called reference symbols or pilots
• Here we assume the elements of H are scalars
• The estimation is done in the receiver. Thus it is know
there.
• To fully utilize the spatial multiplexing, it is desirable to
feedback the parameters of H to the transmitter
– Require to using part of the reverse channel
r tN NH
© 2012 Fuyun Ling
• To utilize the potential of MIMO, it is essential to
accurately know the channel matrix (or simply H).
• The channel matrix is usually estimated using known
symbols embedded in the transmitted data squence
– Called reference symbols or pilots
• Here we assume the elements of H are scalars
• The estimation is done in the receiver. Thus it is know
there.
• To fully utilize the spatial multiplexing, it is desirable to
feedback the parameters of H to the transmitter
– Require to using part of the reverse channel
Rev. 06/2012 16

Channel Known to the Transmitter: Precoding
• Using Singular Value Decomposition, the matrix H can
be diagnalized by orthogonal matrices U and V such
that:
li‘s are the singular values of H
• Thus,
• By transmitting instead of
and pre-multiply the receive signal vector by
r t r r r t t tN N N N N N N N    H U V
1 0 0 0
0 0 0
0 0 0
0 0 0 0
0 0 0 0
r tN N N


 
 
  
 
 
 

© 2012 Fuyun Ling
• Using Singular Value Decomposition, the matrix H can
be diagnalized by orthogonal matrices U and V such
that:
li‘s are the singular values of H
• Thus,
• By transmitting instead of
and pre-multiply the receive signal vector by
Rev. 06/2012 17
1 0 0 0
0 0 0
0 0 0
0 0 0 0
0 0 0 0
r tN N N


 
 
  
 
 
 

( ) ( )t t t t
H
N N N Nk kx V x ( )tN kx
r r
H
N NU
( ) ( ) ( )r r r r t t t t rN N N N N N N N Nk k k    y U V x n

Precoding (cont.)
• The transformed received signal vector becomes:
• The i-the elements of : , i.e., N
parallel scalar channels!
• V is orthonormal – No change in noise statistics
• Water-filling weighting – Achieving channel total capacity
• Precoding require the transmitter has the knowledge of
the channel, which may or may not be easy to achieve
– Approximate knowledge of the channel also helps
• System capacity can be achieve by water-filling
weighting each channel (associated with li).
( ) ( ) ( )
( ) ( )
r r r r r r t t t t t t r r r
r t t r
H H H
N N N N N N N N N N N N N N N
N N N N
k k k
k k
     

  
  
y U U V V x U n
x n


( ) ( ) ( )i i i iy k s k n k  ( )rN ky
© 2012 Fuyun Ling
• The transformed received signal vector becomes:
• The i-the elements of : , i.e., N
parallel scalar channels!
• V is orthonormal – No change in noise statistics
• Water-filling weighting – Achieving channel total capacity
• Precoding require the transmitter has the knowledge of
the channel, which may or may not be easy to achieve
– Approximate knowledge of the channel also helps
• System capacity can be achieve by water-filling
weighting each channel (associated with li).
Rev. 06/2012 18

Linear MIMO Receivers
• Consider the basic MIMO system equation:
• We first consider the zero forcing form
– Premultiply y by HH we have
where is an Nt x Nt square matrix
– Premultiply by the (psudo)inverse of R, x is recovered as
– When n = 0,
• This is the zero forcing solution of Linear MIMO receiver
(indices dropped for clarity) y Hx n
H
  y H y Rx n

© 2012 Fuyun Ling
• Consider the basic MIMO system equation:
• We first consider the zero forcing form
– Premultiply y by HH we have
where is an Nt x Nt square matrix
– Premultiply by the (psudo)inverse of R, x is recovered as
– When n = 0,
• This is the zero forcing solution of Linear MIMO receiver
Rev. 06/2012 19
H
  y H y Rx n

H
R H H
y

1 1
ˆ H 
  x R H y x R n

ˆ x x

Linear MIMO Receivers (cont.)
• Derivation of MMSE MIMO receiver
– Consider again
– To determine coefficient matrix A, such that, the estimate of x,
minimize the norm of the error vector, i.e., minimize
– A can be determined by taking derivative of ewith respect to A
or using orthogonal principle, i.e.:
– Thus A satisfies:
where we assume , and the noise is white.
– Finally
 y Hx n
ˆMMSE x Ay
2
| | | |H
E e E         Ay x
© 2012 Fuyun Ling
• Derivation of MMSE MIMO receiver
– Consider again
– To determine coefficient matrix A, such that, the estimate of x,
minimize the norm of the error vector, i.e., minimize
– A can be determined by taking derivative of ewith respect to A
or using orthogonal principle, i.e.:
– Thus A satisfies:
where we assume , and the noise is white.
– Finally
Rev. 06/2012 20
  0H H H H
E E E E                  ey Ay x y A yy xy
    
1 12H H H H
E E 
 
        A xy yy H HH I
H
E    xx I H
E    nx 0
   
12 2
ˆ , ( ) ~H H H H
MMSE r rN N 

    x AH y H HH I y HH I

Linear MIMO Receivers (cont.)
• Discussions of zero forcing (ZF) and MMSE receivers
– ZF receiver generate an unbiased estimate of Tx symbols
– It’s main problem is noise enhancement
– MMSE receiver provides better SNR at output
– The estimate is biased
• Not a serious problem for PSK Tx symbols
• Need bias removal for QAM signals
– It can be show that MMSE receiver converges to as ZF receiver
when SNR goes to infinity, i.e.,
– Linear receivers do not provide as good performance as non-
linear receivers
© 2012 Fuyun Ling
• Discussions of zero forcing (ZF) and MMSE receivers
– ZF receiver generate an unbiased estimate of Tx symbols
– It’s main problem is noise enhancement
– MMSE receiver provides better SNR at output
– The estimate is biased
• Not a serious problem for PSK Tx symbols
• Need bias removal for QAM signals
– It can be show that MMSE receiver converges to as ZF receiver
when SNR goes to infinity, i.e.,
– Linear receivers do not provide as good performance as non-
linear receivers
Rev. 06/2012 21
  2
12 1
0
lim ( )H H H


 

 H HH I H H H

Maximum Likelihood (ML) Estimator
• Recall:
• The ML estimator determines , which minimizes
where , i.e., its elements are Tx symbols
• This is a constraint LS estimation problem
– It is nonlinear and its complexity is (NP)-hard
– In the receivers, , i.e., its elements are complex numbers
• The performance of the ML estimator is much better than
the linear receivers. However, it is also much more
complex than the linear receivers
– Complexity on order of KNt (K – symbol order)
 y Hx n
ˆMLy Hx
ˆMLx
ˆ tN
ML x s
© 2012 Fuyun Ling
• Recall:
• The ML estimator determines , which minimizes
where , i.e., its elements are Tx symbols
• This is a constraint LS estimation problem
– It is nonlinear and its complexity is (NP)-hard
– In the receivers, , i.e., its elements are complex numbers
• The performance of the ML estimator is much better than
the linear receivers. However, it is also much more
complex than the linear receivers
– Complexity on order of KNt (K – symbol order)
Rev. 06/2012 22
ˆ tN
ML x 

QRD ML Estimator
• Prerequisite: QR decomposition (often used in MIMO)
– Assuming H has a rank of Nt, we have: H = QR
where Q is an Nr x Nt orthnormal matrix, R is an Nt x Nt upper
triangular matrix.
• Solution:
– Since Q is orthnormal, we have:
– Can be viewed as an Nt layer system
© 2012 Fuyun Ling
• Prerequisite: QR decomposition (often used in MIMO)
– Assuming H has a rank of Nt, we have: H = QR
where Q is an Nr x Nt orthnormal matrix, R is an Nt x Nt upper
triangular matrix.
• Solution:
– Since Q is orthnormal, we have:
– Can be viewed as an Nt layer system
Rev. 06/2012 23
2 22 2
2
00 01 0 10 0
2 1 11 1 1 1
1 11 1
ˆ ˆ ˆ ˆ( )
ˆ
ˆ0
ˆ
ˆ0 0
t
t
t tt t
H H
ML ML ML ML
N
N
ML
N NN N
r r ry s
y r r s
y sr


  
      
    
    
            
            
y Hx y QRx Q Q y Rx Q y Rx
y Rx

 

    
 

QRD ML Estimator (cont.)
– Each layer has a metric function:
– The objective of the estimator is to find a sequence of that
minimizes the overall metric
– Tree search can be used to find the ML solution
2 2
0 1 1 1 1 1 1 2 1 2 2 2 2 1 2 1
ˆ ˆ ˆ ˆ ˆ ˆ( ) , ( , )t t t t t t t t t t t t t tN N N N N N N N N N N N N Ng s y r s g s s y r s r s                  
2
1 0 1 1 0 00 0 01 1 0 1 1
ˆ ˆ ˆ ˆ ˆ ˆ( , , )t t t tN N N Ng s s s y r s r s r s        
0 1 1 2 1 1 0 1 1
ˆ ˆ ˆ ˆ ˆ ˆ( ) ( , ) ( , , )t t t t tN N N N Ng s g s s g s s s      
ˆ{ }is
© 2012 Fuyun Ling
– Each layer has a metric function:
– The objective of the estimator is to find a sequence of that
minimizes the overall metric
– Tree search can be used to find the ML solution
Rev. 06/2012 24
0 1 1 2 1 1 0 1 1
ˆ ˆ ˆ ˆ ˆ ˆ( ) ( , ) ( , , )t t t t tN N N N Ng s g s s g s s s      

QRD ML Estimator (cont.)
• QRD Complexity increases exponentially with the length
of Tx data vector and the Tx data symbol size
– e.g. for Nt = 4 and QAM64 (K=64), 64+642+643+644 = 17,043,520
metrics need to be computed.
– Practically impossible for large alphabet size (K) and long vector
• The result is the ML estimate of the Tx data vector.
• In a coded system it is necessary to generate soft bit
metrics in order to achieve better system performance
– Generating soft bit metrics (LLRs) for decoding:
• The values in exp() were calculated above
• It’s very computationally intense ~ 2xlog2(K)xKNt
© 2012 Fuyun Ling
• QRD Complexity increases exponentially with the length
of Tx data vector and the Tx data symbol size
– e.g. for Nt = 4 and QAM64 (K=64), 64+642+643+644 = 17,043,520
metrics need to be computed.
– Practically impossible for large alphabet size (K) and long vector
• The result is the ML estimate of the Tx data vector.
• In a coded system it is necessary to generate soft bit
metrics in order to achieve better system performance
– Generating soft bit metrics (LLRs) for decoding:
• The values in exp() were calculated above
• It’s very computationally intense ~ 2xlog2(K)xKNt
Rev. 06/2012 25
   2 2
2 21 1
: ( ) 1 : ( ) 1
ln exp ln exp
k ks s
LLR  
 
   
        
   
 x x x x
y Hx y Hx

MIMO Receivers – Complexity
Reduction
© 2012 Fuyun Ling
MIMO Receivers – Complexity
Reduction
Rev. 06/2012 26

General Discussion
• MIMO linear receivers (ZF and MMSE) are simple to
implement. However, their performance are relatively
poor comparing to ML receiver
• Practical MIMO receivers has been developed to bridge
the gap between these two extremes including:
– Zero-Forcing receiver with decision feedback
– QRD ML receiver with Sphere Detector (QRD SD)
– M-algorithm based QRC ML receiver (QRD-M)
– Lattice reduction
• There are many variations of these implementations
• The simplified QRD methods are direct consequences of
corresponding results from investigations of
simplification of MLSE equalizers in 1970-80s
© 2012 Fuyun Ling
• MIMO linear receivers (ZF and MMSE) are simple to
implement. However, their performance are relatively
poor comparing to ML receiver
• Practical MIMO receivers has been developed to bridge
the gap between these two extremes including:
– Zero-Forcing receiver with decision feedback
– QRD ML receiver with Sphere Detector (QRD SD)
– M-algorithm based QRC ML receiver (QRD-M)
– Lattice reduction
• There are many variations of these implementations
• The simplified QRD methods are direct consequences of
corresponding results from investigations of
simplification of MLSE equalizers in 1970-80s
Rev. 06/2012 27

Zero-Forcing with Decision Feedback (ZF-DF)
• Zero-Forcing receiver can also be expressed in the QRD
form, namely
i.e., , to be solved using backward substitution
– A decision is made in each step of backward substitution;
– The decision is used in next step
H
       y Hx n QRx n Q y y Rx n
1
ˆ 
x R y
© 2012 Fuyun Ling
• Zero-Forcing receiver can also be expressed in the QRD
form, namely
i.e., , to be solved using backward substitution
– A decision is made in each step of backward substitution;
– The decision is used in next step
Rev. 06/2012 28

QRD-M
• In the ZF-DF, one decision is made in each stage (one
survivor path)
• The performance can be improved if more than one (M)
survivor paths with smallest cumulative metrics are
saved
• The M survivors are extended to MK paths (K = symbol
order) to next stage.
• The M paths with smallest cumulative metrics are saved
• Repeat the above two steps until detection is completed
• QRD-M has fixed computational complexity
• Performance depends on M
– The larger M, the better performance, but the higher complexity
© 2012 Fuyun Ling
• In the ZF-DF, one decision is made in each stage (one
survivor path)
• The performance can be improved if more than one (M)
survivor paths with smallest cumulative metrics are
saved
• The M survivors are extended to MK paths (K = symbol
order) to next stage.
• The M paths with smallest cumulative metrics are saved
• QRD-M has fixed computational complexity
• Performance depends on M
– The larger M, the better performance, but the higher complexity
Rev. 06/2012 29

QRD-SD (Sphere Detector)
• In QRD-SD, the paths with cumulative metrics smaller
than a predetermined threshold are saved at each stage
• The survivors are extended to PK paths (P – number of
survived paths, K – symbol order) to next stage.
• Again, the paths with cumulative metrics smaller than a
predetermined threshold are saved
(The end nodes of the survived paths are in a sphere
around the received signal points)
• The computational complexity of QRD-SD is not fixed
• The performance and complexity depend on the threshold
– The higher threshold, the better performance but also the higher
complexity
© 2012 Fuyun Ling
• In QRD-SD, the paths with cumulative metrics smaller
than a predetermined threshold are saved at each stage
• The survivors are extended to PK paths (P – number of
survived paths, K – symbol order) to next stage.
• Again, the paths with cumulative metrics smaller than a
predetermined threshold are saved
(The end nodes of the survived paths are in a sphere
around the received signal points)
• The computational complexity of QRD-SD is not fixed
• The performance and complexity depend on the threshold
– The higher threshold, the better performance but also the higher
complexity
Rev. 06/2012 30

QRD Estimator Examples
• QRD-ML
• ZF-DF
© 2012 Fuyun Ling
• QRD-ML
• ZF-DF
Rev. 06/2012 31

QRD Estimator Examples (cont.)
• QRD-M (M = 2)
• QRD-SD (T = 5)
© 2012 Fuyun Ling
• QRD-M (M = 2)
• QRD-SD (T = 5)
Rev. 06/2012 32

MLD Complexity Reduction for Small Nt
• In [LA07] a method to reduce MLD complexity for small
Nt was described, it is summarized as follows:
(1) Decompose H and x as:
(2) We rewrite:
(3) Determine MNt-1 that minimize for each
distinctive as:
a)
b) Find the closest constellation point to by
slicing as the estimate of s1.
c) Find the smallest one in the MNt-1 metrics with the
corresponding and
– The complexity is proportional to MNt-1 instead of MNt
   1,
t
s H h H x s 
22
1s   y Hx y Hs h 
1,
ˆs s
2
1s y Hs h 
s
© 2012 Fuyun Ling
• In [LA07] a method to reduce MLD complexity for small
Nt was described, it is summarized as follows:
(1) Decompose H and x as:
(2) We rewrite:
(3) Determine MNt-1 that minimize for each
distinctive as:
a)
b) Find the closest constellation point to by
slicing as the estimate of s1.
c) Find the smallest one in the MNt-1 metrics with the
corresponding and
– The complexity is proportional to MNt-1 instead of MNt
Rev. 06/2012 33
2
1s y Hs h 
s
 1 1
1 1( ) ( ) ( )H H H H
s s 
    h h h y Hs h h h h y Hs  
1
( ) ( )H H
h h h y Hs 
1
ˆs
1
ˆs s

MLD Complexity Reduction for Small Nt (cont.)
• This approach is most effective for small Nt and large
constellation size, e.g, two antenna with 64QAM
• Bit metrics can be computed similar to full MLD
• The result is optimal (no approximation)
• It was originally proposed in
[AL07] Y. Lomnitz and D. Andelman, “Efficient maximum
likelihood detector for MIMO systems with small number
of streams,” Electronics Letters, pp. 1212-1214, Vol. 43,
No. 22, Oct. 2007.
© 2012 Fuyun Ling
• This approach is most effective for small Nt and large
constellation size, e.g, two antenna with 64QAM
• Bit metrics can be computed similar to full MLD
• The result is optimal (no approximation)
• It was originally proposed in
[AL07] Y. Lomnitz and D. Andelman, “Efficient maximum
likelihood detector for MIMO systems with small number
of streams,” Electronics Letters, pp. 1212-1214, Vol. 43,
No. 22, Oct. 2007.
Rev. 06/2012 34

Soft Decoding Bit Metrics (LLRs)
• Generating soft bit metrics (LLRs) for decoding coded data,
has the similar form as the QRD-ML case:
– where the sk’s are only taking from the survived paths
• Further simplification: Dual-Max or Max-Log approximation
– Also applicable for the QRD-ML case
– Good approximation at high SNR
• In both cases, if sk(x) = 1, or sk(x) = −1 does not exist, it will
take the smallest possible value.
   2 2
2 21 1
: ( ) 1 : ( ) 1
ln exp ln exp
k ks s
LLR  
 
   
        
   
 x x x x
y Hx y Hx
© 2012 Fuyun Ling
• Generating soft bit metrics (LLRs) for decoding coded data,
has the similar form as the QRD-ML case:
– where the sk’s are only taking from the survived paths
• Further simplification: Dual-Max or Max-Log approximation
– Also applicable for the QRD-ML case
– Good approximation at high SNR
• In both cases, if sk(x) = 1, or sk(x) = −1 does not exist, it will
take the smallest possible value.
Rev. 06/2012 35
   2 2
2 21 1
( ) 1 ( ) 1
max max
k ks s
LLR   
     
x x
y Hx y Hx

BLAST and Its Variations
• BLAST stands for Bell Labs Layered Space Time
• It was first implemented in the lab environment
demonstrating the advantage of MIMO (1996 – 1998)
• Later designs were based the same concept with
improvements
• D-BLAST (D is for Diagonal) was first developed before
1998
• V-BLAST (V is for Vertical) was soon developed
afterwards (1998) to simplify implementation
– Used in most practical designs
• Modified D-BLAST to take advantage of both
© 2012 Fuyun Ling
• BLAST stands for Bell Labs Layered Space Time
• It was first implemented in the lab environment
demonstrating the advantage of MIMO (1996 – 1998)
• Later designs were based the same concept with
improvements
• D-BLAST (D is for Diagonal) was first developed before
1998
• V-BLAST (V is for Vertical) was soon developed
afterwards (1998) to simplify implementation
– Used in most practical designs
• Modified D-BLAST to take advantage of both
Rev. 06/2012 37

BLAST (cont.)
• D-BLAST
– Layered design with the number of layers equal to the number of
antennas
– Each layer coded independently
– Advantages: Fully utilize spatial and time diversity
– Disadvantages: Multiple decoders, increased complexity
© 2012 Fuyun Ling
• D-BLAST
– Layered design with the number of layers equal to the number of
antennas
– Each layer coded independently
– Advantages: Fully utilize spatial and time diversity
– Disadvantages: Multiple decoders, increased complexity
Rev. 06/2012 38

BLAST (cont.)
• V-BLAST
– Can be implemented with one encoder per antenna or with only one
encoder and one decoder
– Encoded data assigned to multiple antennas and over time
– Advantages: Simplified implementation, good performance in AWGN
– Disadvantages: Lower time diversity gain
– Modified D-BLAST: Assign the encoded data (single encoder)
diagonally and interleaved in time
© 2012 Fuyun Ling
• V-BLAST
– Can be implemented with one encoder per antenna or with only one
encoder and one decoder
– Encoded data assigned to multiple antennas and over time
– Advantages: Simplified implementation, good performance in AWGN
– Disadvantages: Lower time diversity gain
– Modified D-BLAST: Assign the encoded data (single encoder)
diagonally and interleaved in time
Rev. 06/2012 39

MIMO in OFDM Based Systems
• MIMO is most widely used in OFDM Systems
– MIMO is most suitable for single path (flat fading) channels
– OFDM can be viewed as a collection of multiple parallel single
path channels
• Each subcarrier is a single path system
• Examples of OFDM Systems with MIMO
– 802.11n/ac
– LTE
– WiMax
© 2012 Fuyun Ling
• MIMO is most widely used in OFDM Systems
– MIMO is most suitable for single path (flat fading) channels
– OFDM can be viewed as a collection of multiple parallel single
path channels
• Each subcarrier is a single path system
• Examples of OFDM Systems with MIMO
– 802.11n/ac
– LTE
– WiMax
Rev. 06/2012 40

MIMO in OFDM Systems (cont.)
• 802.11n/ac
– Spatial multiplexing with 2, 3, or 4 streams
• In 802.11, # of streams is the rank of spatial multiplexing
• 2x is mandatory, 3x and 4x are optional
• 11ac can have up to 8 streams
– Modulation: BPSK, QPSK, 16QAM, and 64 QAM
– FFT size: 64 (20 MHz), 128 (40MHz)
– Transmitter procedure:
• Single coded bit sequence mapped into multiple spatial (bit)
streams in round-robin fashion
• Each bit stream goes through interleaving, mod. symbol
mapping, space-time coding and iFFT blocks to yield OFDM
signal for transmission
– No special layering processing performed
© 2012 Fuyun Ling
• 802.11n/ac
– Spatial multiplexing with 2, 3, or 4 streams
• In 802.11, # of streams is the rank of spatial multiplexing
• 2x is mandatory, 3x and 4x are optional
• 11ac can have up to 8 streams
– Modulation: BPSK, QPSK, 16QAM, and 64 QAM
– FFT size: 64 (20 MHz), 128 (40MHz)
– Transmitter procedure:
• Single coded bit sequence mapped into multiple spatial (bit)
streams in round-robin fashion
• Each bit stream goes through interleaving, mod. symbol
mapping, space-time coding and iFFT blocks to yield OFDM
signal for transmission
– No special layering processing performed
Rev. 06/2012 41

• MIMO in LTE Release 8
– There are 7 downlink “MIMO” modes in LTE Rel. 8
– Only Modes 3 and 4 can support spatial multiplexing
– The maximum number of Layers, i.e., the rank of the MIMO
spatial multiplexing, is 4
– The data is organized as resource blocks, which consist of
multiple OFDM symbols and multiple subcarriers in each of the
OFDM symbols
– The block of coded bits of each code word are scrambled and
mapped to modulation symbols. These modulation symbols of
one or two codewords are mapped to one or more layers.
• The modulation symbols in each layer are mapped to the
resource elements – sequentially in frequency bins from one
OFDM symbol to next.
© 2012 Fuyun Ling
• MIMO in LTE Release 8
– There are 7 downlink “MIMO” modes in LTE Rel. 8
– Only Modes 3 and 4 can support spatial multiplexing
– The maximum number of Layers, i.e., the rank of the MIMO
spatial multiplexing, is 4
– The data is organized as resource blocks, which consist of
multiple OFDM symbols and multiple subcarriers in each of the
OFDM symbols
– The block of coded bits of each code word are scrambled and
mapped to modulation symbols. These modulation symbols of
one or two codewords are mapped to one or more layers.
• The modulation symbols in each layer are mapped to the
resource elements – sequentially in frequency bins from one
OFDM symbol to next.
Rev. 06/2012 42

• MIMO in LTE Release 8 (cont.)
– Precoding in LTE
• To save feedback information on reverse link, code-books
are used for close-loop precoding
• LTE Codebooks:
– TDD-LTE could use optimal precoding due to the link reciprocity
Codebook
index
Number of layers 
1 2
0
1
2
3 -
© 2012 Fuyun Ling
• MIMO in LTE Release 8 (cont.)
– Precoding in LTE
• To save feedback information on reverse link, code-books
are used for close-loop precoding
• LTE Codebooks:
– TDD-LTE could use optimal precoding due to the link reciprocity
© 2012 Fuyun Ling Rev. 06/2012 43
Codebook
index
Number of layers
1 2
0 





1
1
2
1






10
01
2
1
1 





1
1
2
1






11
11
2
1
2 





j
1
2
1






 jj
11
2
1
3 





 j
1
2
1
-

Further Discussion and Remarks
• The most important feature of MIMO technology is that it
provides a practical mean to achieve spatial multiplexing,
i.e. creates multiple transmission channels.
• With multiple transmission channel, the bits per channel
is reduced for the same throughput
– With fewer bits per channel, the system will keep in the “linear” channel
capacity region
– The SNR per channel is reduced when keeping a constant total
transmission power
– The advantage of statistical multiplexing is mainly in the high SNR/high
spectral efficiency region
– At low SNR region, there’s not much gains
• Practical MIMO system efficiency depends on if the
channel is “rich scattering”
© 2012 Fuyun Ling
• The most important feature of MIMO technology is that it
provides a practical mean to achieve spatial multiplexing,
i.e. creates multiple transmission channels.
• With multiple transmission channel, the bits per channel
is reduced for the same throughput
– With fewer bits per channel, the system will keep in the “linear” channel
capacity region
– The SNR per channel is reduced when keeping a constant total
transmission power
– The advantage of statistical multiplexing is mainly in the high SNR/high
spectral efficiency region
– At low SNR region, there’s not much gains
• Practical MIMO system efficiency depends on if the
channel is “rich scattering”
Rev. 06/2012 45

Further Discussion and Remarks (cont.)
• When the number of transmitter and/or receiver antenna
is more than the created channel, a MIMO system can
also have diversity gain
– Diversity gain can be obtained by different means and not
unique to MIMO
– MIMO can most effectively utilize the diversity gain together with
obtaining statistical multiplexing as shown by Tse et. al.
• MIMO can also be used in multiuser communications
(Multiuser MIMO) with the same principle
– Communication paths are from Tx antenna arrays to antennas
on multiple receivers
– Multiple channels can be created as long as the channels from
the Tx antennas to multiple Rx antennas are not correlated
© 2012 Fuyun Ling
• When the number of transmitter and/or receiver antenna
is more than the created channel, a MIMO system can
also have diversity gain
– Diversity gain can be obtained by different means and not
unique to MIMO
– MIMO can most effectively utilize the diversity gain together with
obtaining statistical multiplexing as shown by Tse et. al.
• MIMO can also be used in multiuser communications
(Multiuser MIMO) with the same principle
– Communication paths are from Tx antenna arrays to antennas
on multiple receivers
– Multiple channels can be created as long as the channels from
the Tx antennas to multiple Rx antennas are not correlated
Rev. 06/2012 46

• The MIMO transmitter can be implemented by simply
distributing the coded data bits/symbols to multiple
antennas
– No special Tx data channel “layering” arrangement are
necessary if so desired
• Precoding in a MIMO system can achieve channel
capacity
– The transmitter will need to know the channel state information
(CSI)
– The channel at the opposite direction is used to sent the CSI
information and thus reduce its capacity, unless it is for a TDD
system
© 2012 Fuyun Ling
• The MIMO transmitter can be implemented by simply
distributing the coded data bits/symbols to multiple
antennas
– No special Tx data channel “layering” arrangement are
necessary if so desired
• Precoding in a MIMO system can achieve channel
capacity
– The transmitter will need to know the channel state information
(CSI)
– The channel at the opposite direction is used to sent the CSI
information and thus reduce its capacity, unless it is for a TDD
system
Rev. 06/2012 47

• Various receiver structures are discussed
– Linear receivers are easier to implement but the gaps between
their performances and the optimum are relatively large
– Maximum Likelihood (ML) receiver can achieve close to optimal
performance
• Brute force implementation of ML detector is practical
impossible in practice except for simple cases
• Simplified, mostly suboptimal, algorithms have been adopted
from corresponding simplified method previously developed
for maximum likelihood equalization and other types of
receivers
• QRD is most suitable for ML and approximate ML detectors
– Lattice reduction methods was developed for improving linear
receiver performance but not widely used today, due to its
complexity and performance.
© 2012 Fuyun Ling
• Various receiver structures are discussed
– Linear receivers are easier to implement but the gaps between
their performances and the optimum are relatively large
– Maximum Likelihood (ML) receiver can achieve close to optimal
performance
• Brute force implementation of ML detector is practical
impossible in practice except for simple cases
• Simplified, mostly suboptimal, algorithms have been adopted
from corresponding simplified method previously developed
for maximum likelihood equalization and other types of
receivers
• QRD is most suitable for ML and approximate ML detectors
– Lattice reduction methods was developed for improving linear
receiver performance but not widely used today, due to its
complexity and performance.
Rev. 06/2012 48

Acknowledgement
• This presentation is based many references that I have
found. Because there were so many people contributed
to the MIMO technology, It would be impossible to list all
of the authors who contributed to the each individual
parts discussed in the presentation. Any and all of the
materials presented here should be credited to
respective authors and I would like to express my
sincere thanks to all of the contributors together here.
© 2012 Fuyun Ling
• This presentation is based many references that I have
found. Because there were so many people contributed
to the MIMO technology, It would be impossible to list all
of the authors who contributed to the each individual
parts discussed in the presentation. Any and all of the
materials presented here should be credited to
respective authors and I would like to express my
sincere thanks to all of the contributors together here.
Rev. 06/2012 49

Mimo tutorial by-fuyun_ling

More Related Content

What's hot (20)

Viewers also liked (18)

Similar to Mimo tutorial by-fuyun_ling (20)

More from Fuyun Ling (10)

Recently uploaded (20)

Mimo tutorial by-fuyun_ling