In the present thesis, the concept for beyond 3G mobile radio systems is described. A service area concept is introduced in order to combat the performance limiting interferences present in the cellular mobile communication systems, with each service area consisting of a various simultaneously active mobile terminals, a number of fixed access points and a central unit per- forming signal processing. About uplink transmission, the main characteristic of this service area system is that with the aid of joint detection of the transmit signals from the mobile ter- minals performed at uplink transmission, all interferences between the simultaneously active mobile terminals using the same bandwidth is drastically reduced. Moreover the use of OFDM subcarrierwise in the described service area based system allows for intersymbol interference free communication and for simple equalization in the frequency domain.
Through this subcarrierwise equalization the service area based system is equivalent with a number of smaller parallel systems, a fact that affects in a reduced computational complexity in the case of optimum multiuser operating with the maximum likehood principle and subop- timum linear detector zero-forcing. In this thesis the parallel interference cancellation detector is introduced, according to which the multi access interference is iteratively reconstructed and subtracted from the received signal. Parallel interference cancellation detector is compared in terms of performance with suboptimum linear detector, due to the reduced computational com- plexity.
Using standardized COST 207 channel models, the performance of parallel interference can- cellation detector compared with suboptimum linear detector has been investigated for a frozen channel, with the same snapshot using the same parameters of the channel, as well as for a number of system loads. A fact that can be observed in simulations results is that parallel in- terference cancellation detectors could achieve the same performance with a reduction of the complexity as suboptimum linear detector zero-forcing in the case of no estimate refinement and with estimate refinement by hard quantization depending on the load system. With esti- mate refinement by soft quantization the performance of the parallel interference cancellation detector is improved, having better performance than zero-forcing detector in cases with nor- mal load. Moreover with the improvement raised in this thesis, in this normal system load, the performance is more improved. On the other hand in the case of full load system, the PIC detector can not substract all the multi access interference producing error flow, this thing is not too important taking in account that the fully loaded system case should not be never present.
Parallel Interference Cancellation in beyond 3G multi-user and multi-antenna OFDM systems
1. David Research Group for RF Communications 2003-5-26 1
Parallel interference cancellation
in beyond 3G multi-user and
multi-antenna OFDM systems
David Sabater Dinter
University of Kaiserslautern
dinter@rhrk.uni-kl.de
Supervisor: A. Sklavos
2. David Research Group for RF Communications 2003-5-26 2
summary
• service area based system in the uplink
• transmission model
• subcarrierwise investigation
• optimum and suboptimum linear detection
• parallel interference cancellation
• PIC with improved estimate refinement
• simulation results
• conclusions
3. David Research Group for RF Communications 2003-5-26 3
uplink transmission in a service area
MT
AP
CU
AP
AP
MT
MT
( )1
ˆd
( )2
ˆd
( )
ˆ K
d
( )1
d
( )2
d
( )K
d
4. David Research Group for RF Communications 2003-5-26 4
( ) ( )
( ) ( )
( ) ( )B B
1,1 ,1
1,2 ,2
1, , B
(1)
(2)
( )B
(1)
(2)
( )
(1)
(2)
( )
K
K
K K K K KK
•= +
H H
H H
H H
d
d
d
n
n
n
e
e
e
% %L
% %L
M O M
% %L
M
%
%
M
%
%
%
M
%
transmission model
• Additive noise vector at AP :
• Received signal vector at AP :
Bk
Bk
B B B F( ) ( ,1) ( , ) T
( )k k k N
n n=n% % %K
B B B F( ) ( ,1) ( , ) T
( )k k k N
e e=e% % %K
B F FK N KN× F 1KN ×B F 1K N × B F 1K N ×
• data symbol vector sent by MT k: F( , )(k) ( ,1) T
( )k Nk
d d=d K
5. David Research Group for RF Communications 2003-5-26 5
subcarrierwise investigation
( )
( )
( )F
1
2
N
÷
÷
= ÷
÷
÷ ÷
H 0 0
0 H 0
H
0 0 H
% L
% L%
M M O M
%L
( ) ( ) ( )
( ) ( ) ( )
( ) ( ) ( )B B B
1,1 2,1 ,1
1,2 2,2 ,2
1, 1, ,
K
K
K K K K
÷
÷
= ÷
÷
÷
H H H
H H H
H
H H H
% % %L
% % %L%
M M O M
% % %L
B F F block diagonal matrixK N xKN
•conversion of totalsystem to smaller parallel systems
•significant effort reduction in
•linear ZF, MMSE
•non linear MLVE
FN
B F F matrixsparseK N xKN
6. David Research Group for RF Communications 2003-5-26 6
( )
( )
( )
( ) ( ) ( )
F
F
F FF F
2
all data vectors equiprobable and Gauss noise
maximum likehood vector estimator (MLVE)
ˆ arg min
n
n
K
n nn n
∈
→
= −
d
d
d e H d
g
%%
D
optimum non-linear and
suboptimum linear detector
( )
( )
( ) ( )
( ){ }F
F F Fˆ arg max |n
n n n
K
P
∈
=
d
d d e%
D
F F F
( ) ( ) ( )ˆ .
n n n
=d D e% %
( )F F F F
1( ) ( )*T ( ) ( )*Tn n n n−
=D H H H% % % %
• Optimum multiuser detector,
• suboptimum linear detection
• Example: ZF criterion
F
F F
( )( ) ( )
2
min
nn n
−e H d%%
7. David Research Group for RF Communications 2003-5-26 7
parallel interference cancellation
( )( )
( )
1
*T
*T
diag
diag
−
=
=
F H H
R H H
% % %
% % %
*T
H%
e% r%
F%
R%
$ ( )pd
$
( )ˆ 1p −d
-bank
of MF
( )ˆ pu
iterative MUD
estimaterefinement
andFECdecoding
• Forward matrix:
• Feedback matrix:
8. David Research Group for RF Communications 2003-5-26 8
no estimate refinement
F
, , (non zero) eigenvalues ofKNλ λ1 FR% %K
• no estimate refinement,
• PIC convergent if
• convergence value
ˆˆ ˆ=d d
( ) 1ρ ≤FR% %
$d FECdemod
ˆd
ˆu
$ˆd
estimate refinement and
FEC decoding
spectral radius
( ) { }F
FR max , , KNρ λ λ1=% % K
ZF
ˆ ˆ( )=∞d d
9. David Research Group for RF Communications 2003-5-26 9
spectral radius example
2,...,8K =
{ }P Rρ ≤
divergenceconvergence
•
• exp. Channel
snapshot
B 8K =
R
10. David Research Group for RF Communications 2003-5-26 10
estimate refinement by hard quantization
$d FECdemod
ˆd
ˆu
$ˆd
estimate refinement and
FEC decoding
• exploit knowledge of discrete
• quantization of to the modulation constellation
F( , )k n
d ∈D
$d
{ }F
2ˆˆ ˆarg min ( )
KN
p
∈
= −
d
d d d
D
D
11. David Research Group for RF Communications 2003-5-26 11
estimate refinement by soft quantization
$d FECdemod ˆu
$ˆd
estimate refinement and FEC decoding
$$ ( )
( )
2
min E
k
k
m md d
−
$$ ( )k
md must satisfy
estim.
2
dσ
2
d
ˆ2 σ
( )tanh 2•
( )sign •
mod
( ) ( )
{ }ˆk k
m mL d d
12. David Research Group for RF Communications 2003-5-26 12
-10 -5 0 5 10 15 20
10
-3
10
-2
10
-1
10
0
simulation results
F( )
1n
ρ >
•
•
• subc.
• exp. channel
• no quant.
4K =
B 4K =
( )10 b 010log / /dBE N
bP
AWGN ZF
PICMF
13. David Research Group for RF Communications 2003-5-26 13
-10 -5 0 5 10 15 20
10
-3
10
-2
10
-1
10
0
simulation results
F( )
1n
ρ >
( )10 b 010log / /dBE N
bP
•
•
• subc.
• exp. channel
• hard quant.
4K =
B 4K =
AWGN
ZF
PICMF
14. David Research Group for RF Communications 2003-5-26 14
-10 -5 0 5 10 15 20
10
-3
10
-2
10
-1
10
0
simulation results
F( )
1n
ρ >
( )10 b 010log / /dBE N
bP
•
•
• subc.
• exp. channel
• soft quant.
4K =
B 4K =
AWGN
ZF
PICMF
15. David Research Group for RF Communications 2003-5-26 15
-10 -5 0 5 10 15 20
10
-3
10
-2
10
-1
10
0
simulation results
bP
( )10 b 010log / /dBE N
•
•
• subc.
• exp. channel
• no quant.
2K =
B 4K =
F( )
1n
ρ ;
AWGN
PIC
(even iterations)
PIC
(odd iterations)
ZF
16. David Research Group for RF Communications 2003-5-26 16
PIC with improved estimate refinement
*T
H%
e% r%
F%
$ ( )1d
bank
of MF
iterative MUD
first iterationsecond iterationthird iteration
*T
H%
e% r%
F%
R%
$ ( )2d
-bank
of MF
iterative MUD
demod
( )ˆ 2u
$
( )ˆ 1d
*T
H%
e% r%
F%
R%
$ ( )3d
-bank
of MF
iterative MUD
demod
( )ˆ 3u
$
( )ˆ 2d
hard Q
or
soft Q
• principle: input MAI-free at quantization process
• starting estimate refinement at the third iteration, errors
introduced by quantization method can be reduced
$d
17. David Research Group for RF Communications 2003-5-26 17
-10 -5 0 5 10 15 20
10
-3
10
-2
10
-1
10
0
simulation results
( )10 b 010log / /dBE N
bP
•
•
• subc.
• exp. channel
• hard quant.
• hard mod.
quant.
2K =
B 4K =
F( )
1n
ρ >
AWGN
MF
ZF
PIC
PIC mod
18. David Research Group for RF Communications 2003-5-26 18
-10 -5 0 5 10 15 20
10
-3
10
-2
10
-1
10
0
simulation results
( )10 b 010log / /dBE N
bP
•
•
• subc.
• exp. Channel
• soft quant.
• soft mod.
quant.
2K =
B 4K =
F( )
1n
ρ >
AWGN
MF
ZF
PICPIC mod
19. David Research Group for RF Communications 2003-5-26 19
-10 -5 0 5 10 15 20
10
-3
10
-2
10
-1
10
0
simulation results
( )10 b 010log / /dBE N
bP
•
•
•
• exp. Channel
• hard quant.
• hard mod.
quant.
3K =
B 6K =
F 32N =
AWGN
MF
ZF
PIC
PIC mod
20. David Research Group for RF Communications 2003-5-26 20
-10 -5 0 5 10 15 20
10
-3
10
-2
10
-1
10
0
simulation results
( )10 b 010log / /dBE N
bP
•
•
•
• exp. Channel
• soft quant.
• soft mod.
quant.
3K =
B 6K =
F 32N =
AWGN
MF
ZF
PIC mod
PIC
21. David Research Group for RF Communications 2003-5-26 21
conclusions
• PIC is a flexible JD scheme
• PIC is not always convergent
• performance improvement with modified
estimate refinement
• more investigation towards PIC necessary