OFDM-MIMO implementation

in Line Of Sight
microwave/millimeter wave
link
Baruch Cyzs
[email protected]
Introduction
Implementation of OFDM-MIMO in line of
sight microwave link
Description of hardware prototype of mm
wave PTP microwave that employs
OFDM-MIMO.
Important Implementation issues in
microwave link that employs OFSM-MIMO
The MIMO Spatial multiplexing
implementation
The MIMO spatial implementation exploits
random independent and identical distributed
(iid) channel.
The orthogonality of the channel is usually
achieved by existing of reach scattering.
Spatial multiplexing suffer degradation in its
performance if significant direct path (LOS)
exists in the Rician channel.
LOS microwave link cannot implement MIMO
since it relies mainly on strong LOS component
How can MIMO implemented in
LOS Microwave
For 3 decades LOS microwave links use polar
multiplexing by transmitting via orthogonal
polarizations.
Witcom in 2001 has initiated new activity of
implementing geometric spatial multiplexing
project.
Prior to project kickoff Witcom has initiated
extensive outdoor field test to evaluate MIMO
performance in 5.8GHz in Tel Aviv.
Test results has shown low rank (mostly
singular) channel even in near/non line of sight.
The results has driven Witcom to seek solution
in the geometric spatial multiplexing.
MIMO SM field test in 5.8GHz
The Spatial Multiplexing principles
in LOS link
The receiver antenna array is located in the near
field zone of the transmitter antenna array.
The spherical wave front from each transmitter
antenna causes phase difference at the receiver
antennas.
LOS Spatial Multiplexing exploits this phase
difference to grant orthogonality to the link
channel for implementation of spatial
multiplexing.


The antenna array approach
As opposed to polar multiplexing in spatial
multiplexing the number of SM channels
can be greater than 2.
The LOS microwave link multiplexing
employs antenna arrays at both sides (no
need to be equal number of elements)
The array antenna spacing is the key
factor for achieving orthogonality.
The Near Field multiplexing
The receiving array is located in the near
field of the transmitting array.
Since the wave front is not planar there is
phase gradient upon the receiving array.
If the phase gradient is set to certain
predetermined value the link channel
becomes orthogonal.
Geometry orthogonalization
R

R

A|
Linear antenna array requirement full
rank condition
d
R

R

d
R

R

Phase difference between R and R:
360/(2*n) in optimal orthogonal condition
n antennas
1 4
2
opt
nR
d
n

= +
n
R
=
The asymmetric case
d
t
R

R

R

R

n antennas
n
R
d d
r t

=
d
t
d
r
d
r
Optimal antenna spacing versus link
distance and frequency
Singular values of dual array acts as
virtual channel gain
3 dB gain
optimal
Antenna spacing
The optimal orthogonal case
characteristics
Low sensitivity to antenna position.
No sensitivity to transversal shifts.
It is possible to work in suboptimal spacing
by employing adaptive modulation.
Antenna constellation can be linear or
regular polygon the same antenna spacing
rule holds.
h11
The channel Measurement
TX
TERMINAL
RX
TERMINAL
X1
X2
X3
y1
y2
y3
( ) * ( ) y t H x t = ( ) * ( ) y t H x t = * ( ) y H x t =
Measuring H matrix by a training/pilot
sequence and calculating beam formers terms
for the channel separation
Inherent diversity gain
Apart of spatial multiplexing Beam formers exhibits
inherent diversity gain over SISO channel
The gain depends on
N
t
transmitters
N
r,
receivers
N
c
active sub-channels, for inherent system
gain:


10*[log( ) log( ) log( )] g Nt Nr Nc = +
N
c
N
c
N
t
N
r
Singular Value Decomposition
y
1
2
3
o
o
o
1
2
3
o
o
o
z
z U x V y U
z x V U y
z Hx y
domain f requncy
t z t x t h t y
domain time
H H H
H
+ E =
+ E =
+ =
+ =

) ( ) ( * ) ( ) (

X
The de-multiplexing process
3
2
1
3
2
1
3
2
1
3
2
1
'
'
'
'
'
'
'
'
'
z
z
z
x
x
x
y
y
y
+ =
o
o
o
Noise statistics has not changed (unitary rotation)
Singular values represent virtual gain
Graphical presentation of SVD
Encoding
&
Modulation
V
+
+
Z
1

Z
4

Decoding
&
Demodulation
U

U
x
x
y
y
Diversity
gain
Carrier
separation
o
1

o
2

Precoding is needed for diversity gain
Basic Block diagram - dual
antenna arrays
V
21

V
12

V
22

V
11

U
11

U
21

U
12

U
22

x
1

x
2

y
1

y
2

H
11

H
22

H
21

H
12

x
1

x
2

y
2

y
1

Tx Beam former
Diversity Gain
Rx Beam former
Separation Channel
Capacity discussion - theory
Theoretical capacity
2 options:
Transmitter knows channel state:

1
log( )
n
i
W
C
=
=
=

total
0
i
P
N

Where satisfies
1
( )
i
=

Water filling algorithm

1
log(1 )
n
i
C
n
=

= +

Transmitter does not know channel state:

Capacity discussion in real life
Real modem has maximum throughput
so there exists maximum bound of
throughput for higher SNR values.
It transmitter knows the channel it can
set the throughput accordingly in the
modulator.
This is in fact real-life water filling.
The dual mode QR-SVD weight
computation
In order to decouple beam former update instances in
both sides of the link coefficients was set by dual
algorithm.
Precoding V coefficient in transmitter updated in slow
manner (sigma-beam sterring effect) on diversity gain
by SVD calculation exploiting slow return channel.
Receiver beam former is calculated by QR
decomposition in fast manner update locally at the
receiver (delta-null steering effect):

z U z
Q R QR U
QR HV
H
R
T
=
= =
=
1 1
) (
The QR-SVD characteristics
If preceding V is set by SVD result R becomes
diagonal with singular values at its diagonal.
In the case of the the optimal orthogonal spacing
R becomes diagonal and V is not needed.
Off diagonal elements energy of R (upper
triangle) proportional to the non unitary noise
enhancement.
The spatial multiplexing
implementation
Witcom has built in the first half of the decade a
prototype system that utilizes the LOS mm wave
MIMO technology. The project was called
TeraWave.
This system was tested with successful results
for 9 months in France Telecom site.
Unfortunately due to marketing reasons the
program has discontinued.
Terawave outline architecture
TeraWave general specification
Frequency: 23GHZ
Bandwidth 28MHz.
Capacity: STM-4 (622MBS).
4 parallel channels 155MBS each employ
polar+spatial multiplexing.
Modulation: OFDM 46 subcarriers/symbol up to
128QAM.
Full pilot symbol every 16 OFDM symbol.
Coding: Turbo Product Code
Outline: Full digital IDU connected via fiber to dual
ODUS direct mounted to dish antenna.
DSP calculates SVD/QR coefficients in zero forcing
fashion.
TeraWave gallery
Test site in France
The Spatial/polar system
Beam forming for multiple spatial channels separation
OFDM optimized modulation for spatial system
Smart mux for payload delivery over multiple spatial
channels

encoder
encoder
encoder
IFFT
IFFT
IFFT
Modulator
beam
formaer
Modulator
Modulator
OFDM framing
Demodul ator
Demodul ator
Demodul ator FFT
FFT
FFT
beam
formaer
decoder
decoder
decoder
OFDM synchronizer
D
e
m
u
x
M
u
x
Data in
Data out
Spatial architecture
Transmitter Reciver
M
A
C

f
r
a
m
e
r
M
u
x
F
E
C

f
r
a
m
e
r
ARQ
memory
management
return
channel
SDH
Ethernat
Payload
QoS
B
e
a
m

f
o
r
m
e
r
TPC IFFT
Up
converter
IF / RF
OFDM
f ramer
spatial
channels
TPC
TPC
adptive modulation control
code rate QAM
Up
converter
IF / RF
clock LO ref erence
Up
converter
IF / RF
Fiber
channels
IFFT
OFDM
f ramer
IFFT
OFDM
f ramer
from receiver
TeraWave transmission system
ODUs
IDU
M
A
C

f
r
a
m
e
r
M
u
x
F
E
C

f
r
a
m
e
r
ARQ
memory
management
return
channel
SDH
Ethernat
Payload
B
e
a
m

f
o
r
m
e
r
TPC FFT
Down
converter
IF / RF
OFDM
f ramer
spatial
channels
TPC
TPC
adptive modulation control
code rate QAM
Down
converter
IF / RF
clock LO ref erence
Down
converter
IF / RF
Fiber
channels
channel
estimator
FFT
OFDM
f ramer
FFT
OFDM
f ramer
to transmit
TeraWave receiving system
ODUs
IDU

Resource Allocation

Constant throughput mode in (TDM radio):
Assign modulation modes so as to maximize gain margin (dB
above minimum S/N required for reliable communication).

Channels with gain margin below a threshold are turned off if
throughput can be maintained with fewer channels.

Variable throughput mode (in packet radio):
When all channels have minimum gain margin, reduce
throughput in order to maintain gain margin.

Beamformer Implementation
SVD loop
[V] [U']
SVD
/QRcalculat
or
Actual channel
[H(t)]
[U']
[V]
[E]
Insert
pilots
Extract
pilots
Reference
Pilot
Generator
X
Virtual Channel
DATA DATA
Pilot
Generator
Return
Channel
CPE Phase
Normalization
x
Random phase
|(t)
Estimated
virtual channel
transfer
function
Implementation issues in mm wave
OFDM-MIMO
Significant dynamic issues have been found due to the
large aperture of the antenna array and interruptions close
to the antennas that caused too rapid changes in
differential channel.
Differential phase noise due to separate LOs in ODUs.
High common phase noise due to large PLL factor in
23GHz.
Larger back off due of OFDM compared to SC.
128QAM require for low implementation degradation CINR
of above 35dB.
PTP microwave require 99.999% availability that reflect
BER=10
-12
which permits less than 5 minutes outage per
year!.
Interpolation filters degrade symbol+CP periodity - a fact
that increased noise in higher BB frequency. Remedy:
Over sample OFDM, better recover of symbol timing
phase.
To grant TeraWave signal processing
means to combat the known OFDM
drawbacks in order to eliminate most of
the inferiorities compared to single carrier
system.

The challenge was met!
The TeraWave challenge
MIMO-OFDM Phase noise error
discussion
In SISO OFDM channel error :
Close to LO carrier common phase error.
Far from LO carrier Inter carrier Interference
Spatial OFDM error suffers from uncorrelated noise:
Common phase error cause CPE error in each modem
that can be corrected by CPE compensation.
Differential Phase error cause uncorrected cross-talk
between sub channels that cannot be compensated by
conventional CPE.
Channel Doppler causes mainly differential phase
error

Correlated phase noise -
analysis
Terawave has high OFDM symbol rate. Most of phase
noise is CPE type.
In Terawave there are no pilots in OFDM symbol. Full
training symbol is transmitted every 12-36 OFDM
symbols.
Channel model acquired after fading average of the pilot
reference.
Channel model update rate 300Hz.
Safe acquisition and tracking for 128QAM requires
integrated RMS phase error of less than 3.
Stringent requirement for MM wave receiver with
synthesizer with integrated RMS phase noise from
300Hz of less than 3 in conventional phase recovery .
Solution decision directed
CPE
Correcting common phase error after equalizing
among all the sub-carrier in the OFDM symbol after
slicing each sub-carrier.
Can be done in both forward (correcting actual data
advantage over SC modem) and backward feed.
Prone to slice error due to AWGN, channel cross-
talk ICI phase noise, channel behavior and non
linearity processing gain depends on number of
carriers.
In Terawave simulation showes 30KHz dual order
loop bandwidth. Practical allowed integrated RMS
phase noise from 3KHz - 3
The spatial phase error noise -
calculation

1
1
1

1
'
, 1
' ( ) ( )
' ( )
' ( )
: ( )
'
t r
j j
r t
r t
r t
r t
r t
H U e U V e V
if
H U I j U V I j V
H I j U U V V
H I j U U V V
Define E error matrix
H I jE
U U V V
u u

= E E
u u <<
= E + u E + u
= + E u E+ E u
= + u + u

= +
E = E u E+ u

1
1
1

1
'
, 1
' ( ) ( )
' ( )
' ( )
: ( )
'
t r
j j
r t
r t
r t
r t
r t
H U e U V e V
if
H U I j U V I j V
H I j U U V V
H I j U U V V
Define E error matrix
H I jE
U U V V
u u

= E E
u u <<
= E + u E + u
= + E u E+ E u
= + u + u

= +
E = E u E+ u

1
1
1

1
'
, 1
' ( ) ( )
' ( )
' ( )
: ( )
'
t r
j j
r t
r t
r t
r t
r t
H U e U V e V
if
H U I j U V I j V
H I j U U V V
H I j U U V V
Define E error matrix
H I jE
U U V V
u u

= E E
u u <<
= E + u E + u
= + E u E+ E u
= + u + u

= +
E = E u E+ u
The spatial phase error noise
2X2 case
2 2
11 2 2 11 1 2 11 1 2
2 2
22 2 2 12 1 2 12 1 2
* *
2
12 11 21 1 2 11 21 1 2
1
* *
1
21 21 11 1 2 21 11 1 2
2
( ) ( ) (1 )( )
( ) ( ) (1 )( )
( ) ( )
( ) ( )
t r t t r r
t r t t r r
t t r r
t t r r
e v v
e v v
e v v u u
e v v u u
| | | | | |
| | | | | |
o
| | | |
o
o
| | | |
o
= + + +
= + + +
= +
= +
Common
CPE error
Diff. CPE error
Diff, x-talk
error
Spatial phase error -
requirement
Differential spatial phase noise causes CPE error and
leakage from other spatial channel (noise like).
Without treatment algorithm the error budget force RMS
integrated phase error requirement of less than 0.7
degree tough solution in MM wave.
Alternatives:
To use common RF Lo (main contributor) for all ODUs.
Implication on deployment.
To use ultra quite separate RF LO, with basic high
frequency (low phase error multiplication).
To add to decision directed algorithm that mitigate
differential phase error mitigation the CPE leakage
compensation.


Spatial phase noise
(CPE+leakage) mitigation

xd1

CPE
calc.
Antenna 1
FFT
Antenna 2
FFT
decision


y
cpe1

y
cpe2

y
1
y
2

cpe2

xd2

+
+
u
*
11

u
*
12

u
*
21

u
*
22

CPE
decision

cpe1

2 1
12 12
1 2
* * 2
12 11 21 1 2 11 21 1 2 1
1
* * 1
21 21 11 1 2 21 11 1 2 2
2
1 21
1 1 *
2 11 21
2
2 2
1
( )
( )
( ) ( )
( ) ( )
( )
(
CPE
CPE
CPE
t t r r
xd
t t r r
xd
xd new old xd
xd new old xd
y CPE y
x decision y
y x
e e
x x
e v v u u
e v v u u
e
u u
c
c c
o
| | | |
o
o
| | | |
o
o

o
o

o


=
=
=
= =
= + +
= + +
=
=
*
12
*
11 21
)
e
u u
y
cpe2

1
x
2
x
Differential Phase and amplitude
change due to wind
Solution to dynamics
Phase tracking loop in receivers according
to master transmitter to avoid differential
phase error.
Differential amplitude correction between
DSP calculation.
Phase loop for high dynamics
|
|
|
The amplitude correction
for high dynamics
u
12
h
21
h
12
h
22
o
12
u
22
u
21
h
11
+
+
+
+
x
1
x
2
u
11
x
2
x
1
X
X
X
X
X
X
X
X
X
X
X
o
11
X
X
o
22
X
X
o
21
|
2
X
X
|
1
2 11 11 21 21 21 22 1 21 21 12 11 11 11 1
) ( ) ( x u h u h x u h u h x o o o o + + + =
2 12 12 21 22 22 22 1 12 12 11 22 22 12 2
) ( ) ( x u h u h x u h u h x o o o o + + + =
leakage
Leakage path





Leakage path
Gain path
Gain path
Phase noise and hit resilience
Phase
rotator
Phase error
measure
Feedback
loop
control
dela
y
Feed
forward
loop
control
Phase
rotator
Phase
rotator
dela
y
Phase
rotator
H
V
H error
V error
input
<2
filter
Accumulato
r
counter
Divider
Accumulato
r
counter
M
U
X
>-2
filter
Mediu
m
Ring
filter
M
UX
comparator
To phase
rotator
input
The constellation before and
after
After Correction Before Correction
PAPR reduction in MIMO
Each FEC Block is interleaved among 4
channels.
Novel approach of multiplying output with
unitary matrix .
Rotation is selected according to minimum
peak to average.
2-3 dB gain in this approach.