Mimo BF&AA

Tutorial ─ MIMO Communications with Applications to (B)3G and 4G
Systems
Beamforming and Adaptive

Antennae
Markku Juntti and Juha Ylitalo
Contents
1. Introduction
2. Beamforming
3. Adaptive antennas
3. Optimal antenna weights
4. Adaptive antenna algorithms
5. Summary
References
MIMO Communications with Applications to (B)3G © J. Ylitalo & M. Juntti, University of Oulu, Dept. Electrical and
and 4G Systems ─ Beamforming & Adaptive Antennae Inform. Eng., Centre for Wireless Communications (CWC) 1
Beamforming: phased array
DOA = Direction of
• Beam steering = phasing the Arrival
antenna array elements
0
DOA
1 = 0 deg. M=8
2 = 30 deg.
DOA
-5
• Assumptions:
Array Gain [dB]

-10
– small angular spread (macro cell)
λ/2 ant. element separation -15
--> correlated antennas -20

--> Rxx ~diagonal
-25
• "Fixed spatial filter" -30
-50 0 50
– based on estimated DOA of desired Azimuth [deg]
user
– DOA tracking (slow vs. channel
tracking) Beams Butler
– fixed beams by e.g. Butler matrix 1-4 matrix
feasible
– interference suppressed by the beam
pattern
Beamforming: array response vector
θ
Uniform Linear Array :
antenna M separated by
(m-1)d from ref. antenna
d sinθ d= λ/2
θ
d Ref.
ant. M antenna
• Array response vector:

a(θ) = [1, e-j2πd sin(θ)/λ , e-j4πd sin(θ)/λ ... e-j2π(Μ−1)d sin(θ)/λ ]T
= [1 e-jπ sin(θ) e-j2π sin(θ) ... e-jπ(Μ−1) sin(θ) ] T
when d=λ/2 ; T denotes transpose

Beamforming: azimuth power spectrum
θ
d sinθ
θ
d Ref.
ant. M
antenna
• Power spectrum as a function of azimuth angle θ

:
P(θ) = [a(θ)H x(θ)]2 = a(θ)H {x(θ)x(θ)H} a(θ)
P(θ) = a(θ)H Rxx a(θ) H denotes here the

complex conjugate
DOA resolution is limited by the array transpose
aperture: ∆θ ~ 96deg/M with ULA
Adaptive antennas
" Optimal multi-channel filtering" Ant 1 Ant 2 Ant M
Rxx
• Optimisation criteria
- MMSE, ML, MV ...
• Optimal combining weights Criteria

- typically in the form: α Rxx-1 h* ... Algorithm
• Adaptive antenna algorithms Weights
- LMS, RLS, DMI, ...
Here we consider only optimal spatial combining

algorithms:
- Narrow-band assumption
- Wide-band assumed only in 2D Rake receiver sense
- Full-scale space-time equalisation will not be treated here
UE
Adaptive antenna algorithms: Introduction
UEINT
• Two basic cases:
- Beamforming with high correlated antennas
- Diversity reception with low correlated antennas
Interfering
• Beamforming can utilise directional w=h* UE
information
- allows simple DoA based algorithms: only Desired
DOA needed to be estimated UE
-optimal in Line-Of-Sight, AWGN case
- DoA based algorithms apply also to
transmission
- allows also optimum combining algorithms
Weight
Adaptation
• Diversity reception w=Rxx -1 h*
- optimum combining
- each antenna weight has to be separately
estimated +
- antenna weight proportional to SINR

Baseband processing
Optimisation criteria
• Assumptions:
- Signal and interference processes are uncorrelated --> different propagation paths
- Signal and interference processes are stationary --> during estimation of weights
- Signal and interference processes are additive --> linear radio channel
Performance of an adaptive algorithm depends on the

statistical character of the desired and interfering signal
in the temporal and spatial domain:
- fading (temporal correlation)
- correlation between antenna elements
Idealised or
perturbed
"Optimum combining algorithms can
suppress correlated interference at propagation
antenna array" conditions?
Optimisation criteria, cont'd
Optimisation of antenna combining weights based on

• Maximum Signal-to-Interference-and-Noise criterion (SINR)
• Least Mean Square criterion (LMS) --MMSE
• Maximum Likelihood criterion (ML)
• Minimum Variance criterion (MV)
It is interesting to note that all these criteria lead to similar

solution of optimal antenna array weight vector which can be
described as:
wopt= α Rxx-1 h*
in which α is a scalar scaling factor, wopt the optimal antenna weight vector
Rxx-1 is the spatial correlation matrix and h represents the array response
vector of the desired signal
MMSE criterion Received signal
x(t)= [x1(t) x2(t) … xM(t) ]T
- Widrow 1960 Ant 1 Ant 2 Ant M
Error signal:
ε(t) = d(t) - w* x(t)
Criteria Weight vector
Squared error: Algorithm w=[w1 … wM]T
ε(t)2 = d(t)2 -2d(t)w*x(t) + Weights
w*x(t)x*(t)w
Expected error value:

Σ
E{ε(t)2} = d(t)2 -2w*rxd(t) + w*Rxxw
-
Rxx = E { x x*} , is a MxM spatial correlation matrix
ε(t) + d(t)
rxd = E {d(t) x(t)}, is a Mx1 column vector (= h ,
ch.est.) error reference
MIMO Communications with Applications to (B)3G signal
© J. Ylitalo & M. Juntti, University of Oulu, Dept. Electricalsignal
and
d(t)2 =and
signal power
4G Systems ─ Beamforming & Adaptive Antennae Inform. Eng., Centre for Wireless Communications (CWC) 9
MMSE ... rxd
- selecting optimal w to minimise error: set Ant 1 Ant 2 Ant M

gradient of E{ε(t)2} with respect to w to zero
Rxx
∇w {<(ε2)>} = -2rxd(t) + 2Rxxw = 0

Weight
Criteria vector
Thus optimal weight vector is obtained
from: Algorithm w= Rxx-1 rxd
Weights
Rxxwopt = rxd and
wopt = Rxx-1 rxd Σ
-
Above equation is known as Wiener-Hopf
equation in matrix form. It is usually ε(t) + d(t)
referred as optimum Wiener solution
error reference
MIMO Communications with Applications to (B)3G © J. Ylitalo & M. Juntti, Universitysignal
of Oulu, Dept. Electrical andsignal
Optimal combining weights
• MMSE (Wiener S
−1 *
solution): wMMSE = R v = * −1
R −1 *
nn v
xx
1 + Sv Rnn v
• Max SINR: wSINR = α Rnn

−1 *
v
1 −1 *
• Maximum Likelihood: wML = * −1
Rnn v
v Rnnv
−1
Rnn 1
• Minimum variance: wMV = * −1
- Direction of desired signal 1 Rnn1
known
Note:
S= signal power, v* = array response vector of the desired signal,
Rnn= interference+noise covariance matrix, Rxx= total received here v = rxd
signal
MIMO (signal+interference+noise) covariance
Communications with Applications to (B)3G matrix,
© J. Ylitalo v=1University
& M. Juntti, for of Oulu, Dept. Electrical and
x(t) = s(t)v + n(t)
known
and 4Gsignal
Systemsweights in MV
─ Beamforming & Adaptive Antennae Inform. Eng., Centre for Wireless Communications (CWC) 11
Optimal combining weights
Linear matrix Scalar processor

Antennas filter
sSINR sML , sMV sMMSE
Rnn-1v* 1/N S/(S+N)
• Notations used above:

x(t) = s(t)v + n(t)
Note:
Rxx = E{x(t) x*(t)}
n(t) includes both
Rnn = E{n(t) n*(t)} interference and AWGN:
Rss = E{s(t) s*(t)} n(t)=I(t)+N0
s(t) = s(t)v v is the array response

vector corresponding to
Rxx = Rss +Rnn desired signal direction
Adaptive antenna algorithms
• Adaptive antenna algorithms solve the

optimal antenna weights for (rapidly) wopt= α Rnn-1 rxd*
changing radio channel state
• Different criteria (MMSE, max SINR,
ML, MV) lead to the same solution
For more details
• Task is to estimate Rnn and rxd and about matrix
inversion, see
find a simple and robust (direct or
iterative) way to invert Rnn S. Haykin: "Adaptive
filter theory",
• Rnn characteristics have an important Prentice Hall
role in the adaptation
- LMS
• Typical adaptive antenna algorithms:
- DMI
- RLS
Example: DMI algorithm
• Direct matrix inversion wopt= Rxx-1 rxd*
1 K
only estimates of Rnn and rxd are E{Rxx} = Rxx = ∑ x(k ) x * (k )
ˆ
K k =1
known:
1 K
E{rxd } = rˆxd = ∑ x(k )d * (k )
K k =1
• Thus we get wˆ opt = Rˆ nn

−1
rˆxd
if Rnn can be estimated (α =1)

or if only Rxx can be estimated wˆ opt = Rˆ xx
−1
rˆxd
DMI algorithm
• Example: GSM burst 1 K
E{Rxx } = Rˆ xx = ∑ x(k ) x * (k )
K k =1
58 bit 26 bit ( d ) 58 bit
1 K
Data Training sequence Data E{rxd } = rˆxd = ∑ x(k )d * (k )
K k =1
• Procedure
– calculate E{Rxx} over the whole burst (K=148) wˆ opt = Rˆ xx
−1
rˆxd
– calculate E{rxd} over the known symbols (K=26)
( actually E{rxd} = E{h}= estimate of the impulse
response)
• E{Rxx} and E{rxd} are calculated for each burst
• E{Rxx} and E{rxd} can be averaged over several
bursts
snr K +2− M
• If rxd is known and Rnn can be estimated: E { } ≅
snropt K +1
=> K > 3M (roughly)
K=no of independent samples, M= no
Inform. Eng., Centre for Wireless Communications (CWC) 15
of
and 4G Systems ─ Beamforming & Adaptive Antennae
antennas
Summary
• Simple beamforming (beampointing) is simple and

robust and applies well to FDD systems
• Adaptive antenna algorithms can be applied to
interference suppression
• Adaptive antenna algorithms/ optimum combining
is required for good MIMO performance: ”space-
time equalisation”
References
• Monzingo and Miller, Introduction to Adaptive Arrays”, Wiley,

1980.
• Winters, J. ”Optimum Combining in Digital Mobile Radio with
Cochannel Interference”, Selected Areas in Communications,
IEEE Journal on ,Volume: 2 , Issue: 4 , Jul 1984, Pages:528 -
539
• J. C. Liberti. and T. S. Rappaport, Smart Antennas for Wireless
Communications: IS-95 and Third Generation CDMA
Applications, Prentice Hall, Upper Saddle River, NJ, 1999.
• Ylitalo, J.; Tiirola, E.; ” Performance evaluation of different
antenna array approaches for 3G CDMA uplink”, Vehicular
Technology Conference Proceedings, 2000. VTC 2000-Spring
Tokyo. 2000 IEEE 51st ,Volume: 2 , 15-18 May 2000, Pages:
883 - 887 vol.2

Mimo BF&AA

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Mimo BF&AA

Uploaded by

Copyright:

Available Formats

Tutorial ─ MIMO Communications with Applications to (B)3G and 4G

Beamforming and Adaptive

Array Gain [dB]

--> correlated antennas -20

• Array response vector:

= [1 e-jπ sin(θ) e-j2π sin(θ) ... e-jπ(Μ−1) sin(θ) ] T

when d=λ/2 ; T denotes transpose

• Power spectrum as a function of azimuth angle θ

P(θ) = a(θ)H Rxx a(θ) H denotes here the

" Optimal multi-channel filtering" Ant 1 Ant 2 Ant M

• Optimal combining weights Criteria

• Adaptive antenna algorithms Weights

- LMS, RLS, DMI, ...

Here we consider only optimal spatial combining

- antenna weight proportional to SINR

Performance of an adaptive algorithm depends on the

Optimisation of antenna combining weights based on

It is interesting to note that all these criteria lead to similar

Expected error value:

- selecting optimal w to minimise error: set Ant 1 Ant 2 Ant M

∇w {<(ε2)>} = -2rxd(t) + 2Rxxw = 0

wopt = Rxx-1 rxd Σ

• Max SINR: wSINR = α Rnn

Linear matrix Scalar processor

• Notations used above:

s(t) = s(t)v v is the array response

• Adaptive antenna algorithms solve the

• Direct matrix inversion wopt= Rxx-1 rxd*

• Thus we get wˆ opt = Rˆ nn

if Rnn can be estimated (α =1)

• Simple beamforming (beampointing) is simple and

• Monzingo and Miller, Introduction to Adaptive Arrays”, Wiley,

You might also like