Tensor-Based Multiuser Detection

TELFOR-2013
November 26-28, 2013

Tensor-Based Multiuser Detection
and
Intra-Cell Interference Mitigation
in LTE PUCCH
www.huawei.com

Vladimir Lyashev | Ivan Oseledets | Delai Zheng
Huawei Technologies
Russian Research Center, Moscow
(vladimir.lyashev@huawei.com)

Huawei Technologies Co., Ltd.

All right reserved

Slide 1

Why does PUCCH important?
“40% Of YouTube Traffic Now Mobile,
Up From 25% In 2012, 6% In 2011.”

“PUCCH occupies too much bandwidth and
is used not in the most efficient way.”
Field-test scenario:
• eRAN7, 10 MHz
• 60 connected UEs
• 30 UEs constantly downloading
large files (i.e. video streaming)

eNodB allocates
• 10RB for PUCCH,
• 3 UE per RB in average

“VoLTE dramatically
increases PUCCH usage.”
•
•
•
•

8 million VoLTE users worldwide
VoLTE will take off in 2015-2016 worldwide
over 10-20 MHz spectrum - hundreds users
46% MBB providers required VoLTE during 1 year

All right reserved

AMR calls/1MHz
GSM

8

UMTS

12

HSPA

24

VoLTE

50

Slide 2

LTE Uplink: resources


All right reserved

Slide 3

PUCCH Allocation and SRS Signal

SRS bandwidth is multiplied
by 4RB: 4, 8, 12, …


All right reserved

Slide 4

Intra-cell Interference in LTE PUCCH
Up to:
• 36UE per 1RB in Format 1x
• 12UE per 1RB in Format 2x
Separation by CAZAC sequence

In practice, time-alignment of the
signals at the eNodeB receiver is
not perfect.

All right reserved

Slide 5

Timing Error: Main Reasons

 limited resolution and measurement errors
 propagation time change due to UE movement

 oscillator drift
 abrupt change of the multipath channel

 misdetection of the Timing Advance
(Initial or Update) command


All right reserved

Slide 6

Abrupt changes in channel delay profile
can’t be compensated by TA commands alone!
200ms-1s

200ms-1s

200ms-1s

TA
command

abs(timing
error)


Timing
correction at
UE

Timing
correction at
UE

Path birthdeath

All right reserved

Timing
correction at
UE
Path birthdeath

Slide 7

Field-test measurements: scenario
 20 km/h speed
 600 m length difference
 720kHz (6RB) SRS signal
generation
// blue line

1

 14.4 MHz measurement signal
// red line

2

2
1


All right reserved

Slide 8

Field test measurements: results


All right reserved

Slide 9

Mathematical Model
1,
𝐏𝑗 𝐏 𝑞 =
0,
𝐻

CAZAC property for ideal sync.:

𝑗 = 𝑞;
𝑗 ≠ 𝑞.

Q

Y j  P H j T j P j X j   P H q Tq Pq X q   P I
 
 
q 1
power loss
q j
  


H
j

H
j

Measurement #

Measurement results

NUE = 6, TAerror = 0 μs
Desired user
Interference
23.7
0.11
12.52
0.67
13.37
0.95
9.83
0.65
7.5
0.26
10.6
0.51

2

H
j

intra  cell interference

SIR
23 dB
22 dB
11 dB
12 dB
15 dB
13 dB

NUE = 6, TAerror = 1.56 μs
Desired user
Interference
14.41
4.12
11.23
2.63
16.91
1.01
2.42
3.15
7.97
2.71
5.11
4.21

All right reserved

Slide 10

SIR
5.5 dB
6.5 dB
12 dB
-1.2 dB
4.5 dB
0.8 dB

Mathematical Model and Its Approximation
Q

Y (n, l , k )   P(q, k , l ) H (q, n, k , l ) X (q, l )T (q, k )  E (n, l , k )
q 1

B-rank channel approximation:
B

H ( q , n, k , l )   W (  , q , n ) S (  , k )
 1

 Rank-2 model basically gives a very good fit to the
experimental channel H(q, n, l, k), usually of a fit of order 95%.
 The rank-1 model also look promising, and can approximate
70% of the energy.

All right reserved

Slide 11

Rank-1 (B=1) Model Approximation
Mathematical Notation in
Slice Form
Yl (n, k )  Y (n, l , k )
l th slice for 3D  tensor (received signal)

Pl (q, k )  P(q, l , k )


T (k )  T (k ) S (k )



Yl  WX l Pl T  El

All right reserved

Slide 12

Joint
Algorithm
ALS-1

l th slice for 3D  tensor (reference sequence)

Joint Detection
^
Update T

Update W



ˆ
WX l Pl  T IT  Yl

Update X



ˆ
W  X I Xl Pl  Yl

ˆ
W X l  W I  Yl

Iteration++

Receive Signal

Y

Simple Channel
Estimation & MRC
with equalizing

H0
W0

QPSK-symbols
demapping &
decoding

XMRC
X0

Set as initial guess
for ALS iterations

X
ALS-1 iterations

^
T0= I12x12


All right reserved

Slide 13

Output CQI bits

Quality decoding

Joint Detection with Quality Control

TErr = 3 TErr = 0

Pilots
FER in MRC:
FER in ALS-1:
MRC & ALS-1 have the same error frames:
ALS-1 males mistake (MRC not):
FER in MRC:
FER in ALS-1:
MRC & ALS-1 have the same error frames:
ALS-1 males mistake (MRC not):


128 / 12 000
225 / 12 000
114 / 12 000
111 / 12 000
407 / 12 000
180 / 12 000
107 / 12 000
73 / 12 000

All right reserved

Slide 14

Simulation Parameters
Parameter
LTE PUCCH format
Bandwidth
CQI
Modulation type
Number of Rx antennas
Number of Tx antennas per user
Number of users
Cyclic shift (CS) interval for RS

Value
format 2
1.4 MHz
7 bits
QPSK
4
1
6
π/3

Power of desired user (CS=0)
Power for UE with CS=1,3,5
Power for UE with CS=2,4
Timing error (uniform distribution)
Propagation channel
Number of simulated sub-frames

All right reserved

0 dB
3 dB
0 dB
-1.56 … 1.56 us
ETU70
20 000
Slide 15

Tensor-Based Multiuser Detection

More Related Content

What's hot (20)

Viewers also liked (13)

Similar to Tensor-Based Multiuser Detection (20)

Recently uploaded (20)

Tensor-Based Multiuser Detection