You are on page 1of 24

Non-orthogonal Multiple Access

(NOMA) for Future Radio Access


Kenichi Higuchi
Tokyo University of Science

February 26, 2016

Kenichi Higuchi/TUS@Adachi_WS

Evolution of Radio Access (1)


Non-orthogonal multiple access (NOMA) based on DSCDMA for 3G
Rake diversity in multipath fading channel appropriate for low-rate
data (thus large spreading factor)
Large capacity via statistical multiplexing effect for circuit-switching
voice users

Orthogonal multiple access (OMA) based on OFDM for


3.9G and 4G
High robustness against multipath interference with simple receiver
structure (e.g. by using frequency-domain channel equalization)
Channel-aware packet scheduling in both frequency and time domain
multiuser diversity
Good affinity to the MIMO channel transmission (again assuming
relatively simple receiver structure)
February 26, 2016

Kenichi Higuchi/TUS@Adachi_WS

Evolution of Radio Access (2)


What about beyond 4G?
We think that NOMA with more advanced transceiver
can be a candidate multiple access scheme.
In theory, NOMA with successive interference cancellation
(SIC) achieves the multiuser capacity region.
Especially important for improving the performance of
cell-edge users with small system efficiency loss
OFDM-like signaling with superposition coding
3G

3.9G and 4G

Beyond 4G

NOMA

OMA

NOMA with advanced


transceiver?

February 26, 2016

Kenichi Higuchi/TUS@Adachi_WS

OMA vs. NOMA


OMA
User As
signal, xA

User Bs
signal, xB
Frequency

Narrow transmission bandwidth


per user
No inter-user (intra-cell)
interference

NOMA with SIC


User As signal, xA
User Bs signal, xB

Wide transmission bandwidth


per user
Frequency

SIC (successive interference cancellation) at the receiver


xA+xB+noise
xA With inter-user (intraDecode of xA
cell) interference
xB No inter-user (intraDecode of xB
xB+noise
cell) interference
February 26, 2016

Kenichi Higuchi/TUS@Adachi_WS

Study at Standardization Body: 3GPP


NOMA, namely downlink Multi-User Superposition
Transmission (MUST) is now being discussed at the
standardization body such as 3rd Generation Partnership
Project (3GPP) as a further evolution of LTE, i.e., LTE
Release 13.
3GPP RP-150496, New SI proposal: Study on downlink
multiuser superposition transmission for LTE, Mar. 2015.
In the following, we will briefly explain the principle of
downlink NOMA and our investigations, along with several
issues for implementing NOMA in real systems.
K. Higuchi and A. Benjebbour, Non-orthogonal multiple access (NOMA) with
successive interference cancellation for future radio access, IEICE Trans.
Commun. vol. E98-B, no. 3, pp. 403-414, Mar. 2015.
February 26, 2016

Kenichi Higuchi/TUS@Adachi_WS

Downlink
Base station

Terminals

February 26, 2016

Kenichi Higuchi/TUS@Adachi_WS

Principle of Downlink NOMA (1)


Base station (BS) transmitter performs superposition coding
for non-orthogonal user multiplexing.
Each users information is independently channel coded
and modulated and then added with other users signals.
K

x pu su
u 1

Tx power for user u

p1s1

p2 s2
p1

February 26, 2016

Coded modulation symbol of user u

x
p2

Kenichi Higuchi/TUS@Adachi_WS

Principle of Downlink NOMA (2)


The user terminal conducts SIC.
In the downlink, the decoding order of SIC should be in the
order of increasing channel gain, |hk|2/Nk.
Each user can remove interference from the user
whose channel condition is worse than that user.
Received signal at user k
y k hk x w k
hk u 1 pu su w k

yk

hk p1s1

Channel
coefficient of
user k

February 26, 2016

Noise + inter-cell
interference
(Power = Nk)

hk p2 s2
hk p3 s3

Kenichi Higuchi/TUS@Adachi_WS

Decode of s1

s1

Decode of s2

s2

Decode of s3

s3

Principle of Downlink NOMA (2)


The user terminal conducts SIC.
User k can correctly decode user 1k1 signals and
remove these signal components from the received signal.
2

pk hk
(NOMA )
b/s/Hz
log2 1 K
Rk
2

p
h
N
i
k
k
u k 1

Received signal at user k


y k hk x w k
hk u 1 pu su w k

yk

hk p1s1

Channel
coefficient of
user k

February 26, 2016

Noise + inter-cell
interference
(Power = Nk)

hk p2 s2
hk p3 s3

Kenichi Higuchi/TUS@Adachi_WS

Decode of s1

s1

Decode of s2

s2

Decode of s3

s3

Capacity Region
Symmetric
channel

NOMA with SIC


OMA

R2 (b/s/Hz)
3.5

Capacity regions of
two access schemes
are identical.

ptotal|h1|2/N1 = 10 dB
ptotal|h2|2/N2 = 10 dB
0

R2 (b/s/Hz) 1
0.8
Asymmetric
0.6
channel
0.4
ptotal|h1|2/N1 = 20 dB
0.2
2
ptotal|h2| /N2 = 0 dB
0
0

3.5

R1 (b/s/Hz)

3
4
R1 (b/s/Hz)

Capacity region of NOMA is wider than that of OMA.


February 26, 2016

Kenichi Higuchi/TUS@Adachi_WS

10

System Efficiency and User Fairness


Generalized system throughput based on -proportional fair
K
Rk (t )1
U (t )
k 1 1
Parameter controls the tradeoff between the system
efficiency and user fairness.
System
efficiency

User
fairness

To maximize the system throughput, the following resource


allocation metric is to be maximized.
K

1
Weighted sum of user
R
W
P
t
,
;

throughput
R
t

(
1)
k 1 k

M (W , P;t )

As is increased for better user fairness, weight for throughput of


weak user increases.
February 26, 2016

Kenichi Higuchi/TUS@Adachi_WS

11

When NOMA is Effective?


MNOMA

0
R2 (b/s/Hz)

1
1
NOMA
R
t
R2NOMA t

R1(t 1)
R1(t 1)

0.8

OMA
MNOMA
R1 t R2 t
0

=0

>0

0.6
0.4
0.2 MOMA
0
0

1
1
OMA
OMA
R
t
R

t
R1(t 1) 1
R1(t 1) 2
1

3
4
R1 (b/s/Hz)

NOMA is more attractive than OMA when the user channels


are asymmetric and the system wants to take care of user
fairness.
February 26, 2016

Kenichi Higuchi/TUS@Adachi_WS

12

NOMA in Cellular Downlink


In cellular downlink, the channel conditions vary significantly
among users due to the near-far effect.
Strong

Weak

Inter-cell interference

When we want to improve the throughput of weak user


In OMA, bandwidth allocation to strong user is severely
limited.

NOMA with SIC


Wide transmission bandwidth for all users
Allocates large power to weak user
Although the weak user does not use SIC, the impact of interuser interference is small.
Strong user uses SIC to cancel out the interference from weak
user and small power allocation is enough due to strong channel.
February 26, 2016

Kenichi Higuchi/TUS@Adachi_WS

13

Example
1
0.9
Cumulative probability

0.8
0.7
0.6

Full-buffer traffic
model with 10 users
per cell

0.5
0.4

FTP traffic model 1


with = 0.8

0.3

OMA
NOMA

0.2
0.1
0 -1
10

February 26, 2016

10
10
User throughput (Mb/s)
Kenichi Higuchi/TUS@Adachi_WS

10

14

NOMA in MIMO Downlink


NOMA with SIC appropriate for MIMO downlink
Key issues in our mind
Use of SIC instead of dirty paper coding (DPC)
DPC is difficult to implement and sensitive to the error
in channel state information feedback.
Avoidance of increase in reference signaling overhead
In NOMA, the number of users multiplexed within a
same frequency is assumed to be beyond the number
of BS antennas.
If we assume the user-specific beamforming, the
number of reference signals in NOMA may exceed the
number of BS antennas.
February 26, 2016

Kenichi Higuchi/TUS@Adachi_WS

15

Proposed Method
Intra-beam superposition coding and SIC in NOMA with MIMO
BS transmitter: Intra-beam superposition coding
The number of beams is equal to that of BS antennas.
Within a beam, multiple-user signals are non-orthogonally
multiplexed based on superposition coding.
The number of reference signals is the same as in OMA
irrespective of the number of non-orthogonally multiplexed users.

UE receiver: Intra-beam SIC


Inter-beam interference is suppressed by the linear spatial
filtering.
Then, intra-beam interference due to superposition coding
is removed by SIC.
Effective use of SIC as in the SISO or SIMO downlink
February 26, 2016

Kenichi Higuchi/TUS@Adachi_WS

16

Pictorial Example
Spatial
filtering

Signal to User2
Signal to User1

User2 signal
decoding

User 2

Beam 1
User 1

BS

Spatial
filtering

SIC of User
2 signal

User1 signal
decoding

Spatial
filtering

SIC of User
4 signal

User3 signal
decoding

User 3

Beam 2
Signal to User4
Signal to User3

User 4

Large

Spatial
filtering

User4 signal
decoding
Small

Equivalent normalized channel gain, g


February 26, 2016

Kenichi Higuchi/TUS@Adachi_WS

17

Throughput Performance
700

1.5 times

40
Sum throughput (Mb/s)

Cell-edge user throughput (kb/s)

50

30

20
NOMA
OMA

10

Num. of BS antennas = 1
Num. of BS antennas = 2
0

10

20
30
40
Number of users per cell

February 26, 2016

50

NOMA
OMA

600
500
400
300
200
100
0

Num. of BS antennas = 1
Num. of BS antennas = 2
10

Kenichi Higuchi/TUS@Adachi_WS

20
30
40
Number of users per cell

50
18

Issues and recent investigations for


implementing NOMA in real systems

February 26, 2016

Kenichi Higuchi/TUS@Adachi_WS

19

Generation of Superimposed Signal (1)


Two approaches under investigation at 3GPP
SOMA (semi-orthogonal multiple access)
3GPP R1-151425, Multiuser superposition schemes, Apr. 2015.
3GPP R1-151848, Candidate schemes for superposition transmission, Apr. 2015.

Each users information is QAM modulated and summed


among users with appropriate power allocations so that
the resultant signal still forms higher-order QAM signal.
E.g. QPSK + QPSK = 16QAM, 16QAM + QPSK = 64QAM

Near user modulation


symbol (QPSK)
February 26, 2016

Far user modulation


symbol (QPSK)
Kenichi Higuchi/TUS@Adachi_WS

Superimposed symbol
(16QAM)
20

Generation of Superimposed Signal (2)


SOMA (semi-orthogonal multiple access) contd
No need for defining new modulation symbols
Good performance even when SIC is not applied
Question: Do we need Gray mapping after superposition?
RA-CEMA (rate-adaptive constellation expansion multiple
access)
A. G. Perotti and B. M. Popovi, Non-orthogonal multiple access for degraded broadcast
channels: RA-CEMA, in Proc. IEEE WCNC2015, New Orleans, USA, Mar. 2015.

Superimposed signal forms 2mQAM similar to SOMA.


Mapping of coded bits of multiple users on m bits at
2mQAM modulation is adaptively controlled depending on
the channel conditions of respective users.
February 26, 2016

Kenichi Higuchi/TUS@Adachi_WS

21

Generation of Superimposed Signal (3)


RA-CEMA (rate-adaptive constellation expansion multiple
access) contd
No need for defining new modulation symbols similar to
SOMA
By changing the bit mapping of multiple non-orthogonally
multiplexed users based on channel conditions, more
detailed rate control is achieved.
Q

Far user = 1 bit


Near user = 3 bits

February 26, 2016

0111

0011

1011

1111

0110

0010

1010

1110

0100

0000

1000

1100

0101

0001

1001

1101

Kenichi Higuchi/TUS@Adachi_WS

22

Others
Resource allocation
In addition to the time/frequency block allocation in OMA,
the power allocation needs to be appropriately conducted.
Frequency block-distributed codeword mapping in LTEAdvanced should also be taken into account.
Control signaling
Transport format (modulation scheme, code rate, power,
etc.) of other users should be informed for SIC process.
Reference signaling
Additional format of reference signal appropriate for
NOMA may be defined.
Hybrid ARQ in NOMA

February 26, 2016

Kenichi Higuchi/TUS@Adachi_WS

23

Conclusion
We briefly explained the principle of downlink NOMA and our
investigations, along with several issues for implementing
NOMA in real systems.

Thank you very much for your kind attention!

February 26, 2016

Kenichi Higuchi/TUS@Adachi_WS

24