1072 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 40, NO.

4, APRIL 2022

Next-Generation Multiple Access Based on

NOMA With Power Level Modulation
Xinyue Pei , Student Member, IEEE, Yingyang Chen , Member, IEEE,
Miaowen Wen , Senior Member, IEEE, Hua Yu , Member, IEEE, Erdal Panayirci , Life Fellow, IEEE,
and H. Vincent Poor , Life Fellow, IEEE

Abstract— To cope with the explosive traffic growth expected Index Terms— Bit error rate, next-generation multiple access
in next-generation wireless networks, it is necessary to design (NGMA), achievable rate, finite alphabet input, non-orthogonal
next-generation multiple access techniques that can provide multiple access (NOMA).
higher spectral efficiency as well as larger-scale connectivity.
As a promising candidate, power-domain non-orthogonal mul- I. I NTRODUCTION
tiple access (NOMA) has been widely studied. In conventional
power-domain NOMA, multiple users are multiplexed in the
same time and frequency band with different preset power
levels, which, however, may limit the spectral efficiency under
F UTURE wireless networks, such as beyond fifth-
generation (B5G) or sixth-generation (6G) networks, are
expected to support extremely high data rates and numerous
practical finite alphabet inputs. Inspired by the concept of
spatial modulation, we propose to solve this problem by encoding users or nodes with various applications and services [1].
extra information bits into the power levels, and exploiting However, the conventional orthogonal access (OMA) schemes
different signal constellations to help the receiver distinguish used in the previous wireless generations cannot meet these
between them. To convey this idea, termed power selection (PS)- unprecedented demands, limiting the improvement of the
NOMA, clearly, we consider a simple downlink two-user NOMA overall spectral efficiency (SE). Against the background, non-
system with finite input constellations. Assuming maximum-
likelihood detection, we derive closed-form approximate bit error orthogonal multiple access (NOMA) was proposed, which
rate (BER) expressions for both users. Moreover, the two-user allows numerous users to share the same resource (e.g.,
achievable rate region is also characterized. Simulation results a time/frequency resource block) and separate the users in
verify the analysis and show that the proposed PS-NOMA can power or code domains at the expense of additional receiver
outperform conventional NOMA in terms of BER and achievable complexity [2]–[4]. In power-domain NOMA, users are typi-
rate.
cally multiplexed with different power levels by using superpo-
Manuscript received July 27, 2021; revised November 13, 2021; accepted sition coding at the transmitter and are distinguished through
December 17, 2021. Date of publication January 14, 2022; date of current successive interference cancellation (SIC) at the receiver
version March 17, 2022. This work was supported in part by the National [5]–[7]. Compared with OMA, NOMA has higher significant
Natural Science Foundation of China under Grants U1809211, 61871190,
and 62001191; in part by the Natural Science Foundation of Guangdong system throughput and greater fairness [8]–[11]. Due to these
Province under Grants 2018B030306005 and 2021B1515120067; in part by advantages, NOMA has been recognized as a key technology
the Guangdong Provincial Key Laboratory of Short-Range Wireless Detection by the third-generation partnership project (3GPP) for future
and Communication, under Grant 2017B030314003; and in part by the Fun-
damental Research Funds for the Central Universities under Grant 21620351. wireless networks [12].
(Xinyue Pei and Yingyang Chen are co-first authors.) (Corresponding authors: It is worth noting that the majority of existing NOMA
Yingyang Chen; Miaowen Wen.) schemes assume Gaussian input signals [13]–[15]. Although
Xinyue Pei and Hua Yu are with the National Engineering Technology
Research Center for Mobile Ultrasonic Detection, South China Univer- Gaussian inputs can theoretically attain the channel capacity,
sity of Technology, Guangzhou 510640, China, and also with the Key the practical implementation faces many difficulties, e.g., very
Laboratory of Marine Environmental Survey Technology and Applica- large storage capacity, high computational complexity, and
tion, Ministry of Natural Resources, Guangzhou 510300, China (e-mail:
[email protected]; [email protected]). extremely long decoding delay [16]. Applying the results
Yingyang Chen is with the College of Information Science and Technology derived from the Gaussian inputs to the signals with finite
and the College of Cyber Security, Jinan University, Guangzhou 510632, alphabet inputs, e.g., pulse amplitude modulation (PAM), can
China (e-mail: [email protected]).
Miaowen Wen is with the National Engineering Technology Research Cen- result in a significant performance loss [17]. Attentive to
ter for Mobile Ultrasonic Detection, South China University of Technology, this, some researchers have considered finite alphabet inputs
Guangzhou 510640, China, and also with the Science and Technology on [18]–[21]. In particular, the authors of [18] studied NOMA-
Communication Networks Laboratory, Shijiazhuang, Hebei 050081, China
(e-mail: [email protected]). aided two-user Gaussian multiple access channels (MACs)
Erdal Panayirci is with the Department of Electrical and Electron- with finite complex input constellations, and designed a
ics Engineering, Kadir Has University, 34083 Istanbul, Turkey (e-mail: constellation rotation (CR) scheme for M -ary phase shift
[email protected]).
H. Vincent Poor is with the Department of Electrical and Computer keying (PSK) and M -PAM signals. In [19], the same authors
Engineering, Princeton University, Princeton, NJ 08544 USA (e-mail: proposed a novel power allocation scheme for the model
[email protected]). in [18], which can achieve similar performance to the CR
Color versions of one or more figures in this article are available at
https://doi.org/10.1109/JSAC.2022.3143240. scheme while reducing the decoding complexity for quadrature
Digital Object Identifier 10.1109/JSAC.2022.3143240 amplitude modulation (QAM) constellations. The authors
0733-8716 © 2022 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.

Authorized licensed use limited to: Mississippi State University Libraries. Downloaded on January 23,2025 at 00:07:31 UTC from IEEE Xplore. Restrictions apply.
PEI et al.: NEXT-GENERATION MULTIPLE ACCESS BASED ON NOMA WITH POWER LEVEL MODULATION 1073

of [20] developed a novel framework for a classical two- with bit allocation was studied, and its SE was analyzed.
transmitter two-receiver NOMA system over Z-channels with The authors of [33] mainly investigated the fundamental
QAM and max-min user fairness. Specifically, they formulated applicability of SM for multi-antenna channels and found
a max-min optimization problem to maximize the smaller intrinsic cooperation for constructing energy-efficient finite-
minimum Euclidean distance among the two resulting signal alphabet NOMA.
constellations at both receivers. Similarly, in [21], the authors In this paper, we consider a different way of integrating
considered a classical two-user MAC with NOMA and practi- NOMA with SM to improve the system SE by encoding extra
cal QAM constellations, aiming at maximizing the minimum information bits into the power levels. The proposed NOMA
Euclidean distance of the received sum constellation with a scheme with information-guided power selection (PS), referred
maximum likelihood (ML) detector by controlling transmitted to as PS-NOMA, shows its superiority, especially under finite-
powers and phases of users. alphabet inputs. The main contributions of this paper are
On the other hand, spatial modulation (SM) is regarded summarized as follows:
as another promising technique of improving SE for • We design a novel PS-NOMA scheme for the two-
next-generation wireless networks [22], [23]. Unlike con- user downlink channel, where the transmitter randomly
ventional multiple-input multiple-output (MIMO) techniques, chooses one out of preset power levels according to the
SM selects only one out of all available transmit antennas input bit stream. To make the power levels distinguishable
for each transmission, thus avoiding inter-channel interfer- by the receiver, we further design a specific signal con-
ence and the requirement of multiple radio frequency (RF) stellation as the signature for each power level. The best
chains. In SM systems, the transmitted information consists set of power levels is found by maximizing the minimum
of the index of an active antenna and a conventionally mod- Euclidean distance of the signal constellation.
ulated symbol. In [24] and [25], novel spatial constellation • The performance of PS-NOMA in terms of BER and
design (SCD) methods with generalized space shift keying achievable rate region is theoretically analyzed, where
(GSSK-SCD) signaling were proposed for physical layer the achievable rates for both users over Rayleigh fading
security (PLS) in multi-user (MU) MIMO communication channels are derived in closed-form. Since the exact
systems. In GSSK-SCD, the received spatial constellations BER expressions are intractable, approximate closed-
were optimized through a novel precoding scheme, which form BER expressions are derived for PAM constella-
minimizes the bit error rates (BERs) at legitimate users and tions, which well match with the simulation counterparts.
significantly worsens eavesdroppers’ BERs. In [26], a novel • Monte Carlo simulations are performed to verify the the-
SCD technique was proposed to enhance the PLS of optical oretical analysis. In consideration of comparison fairness,
GSSK, which can retain some benefits of SM. The perfor- the conventional NOMA scheme without PS is chosen as
mance of multiple-input single-output (MISO) and MIMO sys- a benchmark. Simulation results show that the proposed
tems is limited by spatial channel correlation. This limitation PS-NOMA can outperform the benchmark in terms of
becomes particularly severe in light fidelity (LiFi) systems, both BER and achievable rate region.
which use intensity modulation/direct detection (IM/DD) for The rest of this paper is organized as follows. Section II
data transmission, because light emitting diodes (LEDs) can describes the system model of PS-NOMA. Section III provides
only emit incoherent light. Intuitively, integrating NOMA with the achievable rate and BER analyses. Section IV analyzes the
SM (termed SM-NOMA) will further improve system SE numerical results. Finally, we conclude the work in Section V.
without increasing the power consumption and implementation Notation: The probability of an event and the probability
complexity compared with NOMA without SM [27]. There- density function (PDF) of a random variable are denoted by
fore, SM-NOMA has been exploited extensively in recent Pr(·) and p(·), respectively. Q(·), Γ(·), E{·}, and Var{·}
years, and capacity or BER analysis is the focus of many denote the Gaussian Q-function, gamma function, expectation,
works, such as [27]–[29]. In [27], the authors studied the BER and variance, respectively. | · | denotes the absolute value of
and capacity performance of a novel three-node cooperative a complex scalar, and · denotes the floor function. R{·}
relaying system using SM-aided NOMA. In [28], an iterative indicates the real part of a complex value. Superscript ∗ stands
algorithm for the spatial-domain design was proposed to maxi- for complex conjugates. I(·) denotes MI, and H(·|·) denotes
mize the instantaneous capacity of SM-NOMA. The authors of entropy. Finally, x ∼ Nc (μx , βx ) indicates that the random
[29] applied SM-NOMA in wireless vehicle-to-vehicle (V2V) variable x obeys a complex Gaussian distribution with mean
environments. They derived closed-form capacity expressions, μx and variance βx .
analyzed BER performance via Monte Carlo simulations, and
formulated a pair of power allocation optimization schemes for
the system. Notably, all the above mentioned works focused II. S YSTEM M ODEL
on Gaussian inputs, which do not apply to finite input constel-
lations. On the other hand, some researchers have considered A. Transmission Model
finite alphabet inputs in SM-NOMA systems [30]–[33]. The To begin with the introduction of PS-NOMA, let us recall
authors of [30], [31] proposed and studied SM-NOMA systems the conventional two-user downlink NOMA system first.
from the mutual information (MI) perspective. Since the MI In such a system, the base station (denoted by BS) transmits
lacks a closed-form formulation, they proposed a lower bound a superimposed signal to both users. Based on the NOMA
to quantify it. In [32], an SM aided cooperative NOMA scheme protocol, this signal consists of the data symbols desired

Authorized licensed use limited to: Mississippi State University Libraries. Downloaded on January 23,2025 at 00:07:31 UTC from IEEE Xplore. Restrictions apply.
1074 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 40, NO. 4, APRIL 2022

Fig. 1. System model with 2 users, where UA is the near user, and the other is the far user.

by both users, and the power allocated to the symbols. coded symbol
In conventional NOMA systems, information is conveyed only
sC = αA (l)sA + αB (l)sB (3)
by data symbols. However, the power allocation coefficients
contained in the superimposed signal can also be utilized to to UA and UB simultaneously, where sA ∈ SA and sB ∈ SB
carry information by appropriate design. are the data symbols intended for UA and UB , respectively.
To convey this idea, we consider a downlink PS-NOMA We assume that sA and sB are MA -ary PAM and MB -ary
network as illustrated in Fig. 1, which consists of a BS, a PAM symbols with E{|sA |2 } = E{|sB |2 } = Es = 1;
near user (denoted by UA ), a far user (denoted by UB ). All
the nodes are equipped with a single antenna. The channels of SA = {±dA , ±3dA , · · · , ±(MA − 1)dA } (4)
BS→ UA and BS→ UB are respectively denoted by hA and and
hB , which are assumed to follow Nc (0, βA ) and Nc (0, βB ).
Notably, it is assumed that all the nodes are synchronous SB = {±dB , ±3dB , · · · , ±(MB − 1)dB } (5)
in both time and frequency, and perfect channel state infor-
are the corresponding MA and  MB points constellations,
mation (CSI) is available to both users. Unlike conventional
respectively,
 where dA = 3/(MA2 − 1) and dB =
NOMA, we obtain an SE gain by selecting a power level (PL) 2
3/(MB − 1) respectively represent half of the minimum
from a set of N PLs at random at the transmitter where N is
distance between two adjacent points of the normalized
a power of two, so that the PL carries log2 (N ) bits in addition
MA -PAM and MB -PAM constellations. Denoting Mi =
to the amount of information carried by the transmitted data
{1, 2, · · · , Mi } (i ∈ {A, B}), the signal constellation of sC
symbols. Noticing that UA is the near user and UB is the far
can be expressed as
user, we have βA ≥ βB , and the power allocated to UA should
be smaller than that assigned to UB considering users’ fairness. SC = {sC |si = Si (ki ), PL = l, ki ∈ Mi , l ∈ N , i ∈ {A, B}},
In this manner, the transmit power matrix can be constructed (6)
as
⎡  ⎤ which is an irregular MA MB N -ary constellation. An example
pA (1)pt , (1 − pA (1))pt of MA = MB = N = 2 is shown in Fig. 2. The received
⎢ pA (2)pt , (1 − pA (2))pt ⎥
⎢ ⎥
P=⎢ .. .. ⎥, (1) signals at UA and UB can be given by
⎣ . . ⎦
  y A = h A sC + n A (7)
pA (N )pt , (1 − pA (N ))pt
where 0 < pA (l) < 0.5 denotes the power allocation coef- and
ficient for UA , l ∈ {1, . . . , N }, and pt denotes the transmit y B = h B sC + n B , (8)
power; the first and the second columns represent the power for
UA and UB , respectively, and different rows represent different respectively, where nA ∼ Nc (0, N0 ) (nB ∼ Nc (0, N0 ))
PLs. To ensure that the PLs are distinguishable, we delib- indicates the additive white Gaussian noise (AWGN) at UA
erately rotate each PL with a certain angle, which is given (UB ) with noise variance N0 .
by Θ = diag(exp(0), exp(jπ/N ), · · · , exp(j(N − 1)π/N )).
Therefore, the rotated power matrix can be derived as B. Signal Decoding
⎡ ⎤
αA (1), αB (1) The above subsection describes the transmission model of
⎢ αA (2), αB (2) ⎥
⎢ ⎥ PS-NOMA. Now we shift our focus to the decoding process.
G = ΘP = ⎢ .. .. ⎥, (2) To proceed, we need to first clarify which user desires the
⎣ . . ⎦
αA (N ), αB (N ) information carried by the PL. In this paper, we assume
 that the information of the PL is desired by UB , such that
where αA (l) = exp(j(l − 1)π/N ) pA (l)pt and αB (l) = the achievable rate of the far user can be further improved.
exp(j(l − 1)π/N ) (1 − pA (l))pt . In other scenarios of interest, it can also be used to improve
Suppose that the BS chooses the PL l ∈ N (N = the performance of UA or both users. To enhance the per-
{1, 2, · · · , N }). Subsequently, it conveys the superposition formance of both users, we can assume both users share the

Authorized licensed use limited to: Mississippi State University Libraries. Downloaded on January 23,2025 at 00:07:31 UTC from IEEE Xplore. Restrictions apply.
PEI et al.: NEXT-GENERATION MULTIPLE ACCESS BASED ON NOMA WITH POWER LEVEL MODULATION 1075

Fig. 4. Constellation of 2-PAM signal sA .

Algorithm 1 Detection Algorithm

1: for UA do
2: Compare the received signal yA with the joint constel-
lation SB
R
:
SB
R
(î) = arg min |yA − hA SB
R
(i)|2 , (10)
i

where i ∈ {1, · · · , N MB }, and î is the estimate of i at

UA .
3: Determine active PL ˆl and the transmitted signal ŝB by
resolving SB
R
(î), where ŝB and ˆl are the estimates of sB
Fig. 2. Constellation SC (i) (i ∈ {1, · · · , 8}) of superimposed signal
and l at UA , respectively.
transmitted with 2-PAM signals sA as well as sB and N = 2. 4: Use SIC to decode sA , whose ML detector can be
expressed as
ŝA = arg min |yA −hA αB (l̂)ŝB −hA αA (l̂)sA |2 , (11)
sA

where ŝA is the estimate of sA at UA .

5: end for
6: for UB do
7: Compare the received signal yB with the joint constel-
lation SB
R
:
SB
R
(î) = arg min |yB − hB SB
R
(i)|2 , (12)
i

where i ∈ {1, · · · , N MB }, and î is the estimate of i at

UB .
8: Determine active PL ˆl and the transmitted signal ŝB by
resolving SB
R
(î), where ŝB and ˆl are the estimates of sB
Fig. 3. Constellations of sB with N = 2, where sB is a 2-PAM symbol, and and l at UB , respectively.
SBR denotes the union constellation of s R R
B and PL. SB (1), SB (3) : sB = 9: end for
−1, SB R (2), S R (4) : s = 1.
B B

information of the PL. For example, if there are two extra

equalization is close to SB R
(1), we can readily know the
bits carried by four PLs in total, we can allocate the first bit
estimated sB = −1 and l = 1. With these information, UA
(corresponding to the first two PLs) to UB , and the other bit
then subtracts the estimated sB and decodes its own signal,
to UA (corresponding to the last two PLs).
and this procedure can be seen as a common 2-PAM signal
Both users can invoke ML detectors to decode their signals.
decoding as shown in Fig. 4. The detailed detection algorithm
In conventional NOMA scheme, whose PLs are fixed, the
for both users is summarized in Algorithm 1.
users already know the values of PLs when decoding. While
in PS-NOMA, the PLs change over time and are unknown
at the users. Hence, the conventional ML detector cannot be III. P ERFORMANCE A NALYSIS
applied directly in PS-NOMA, and we should also decode PL In this section, we theoretically analyze the performance of
at both users. In our design, UA first decodes sB and l, while the proposed PS-NOMA scheme in terms of achievable rate
for UB , it detects sB and l by treating sA as noise. We herein and BER. Moreover, we discuss the optimal design of N and
use the joint constellation of sB and PL to help decoding. For consider the multi-user case.
example, in the case of N = 2 and 2-PAM sB as shown in
Fig. 3, a combined 4-ary constellation is formed, i.e.,
A. Achievable Rate Analysis
SB
R
(i) = {sB = ±1, l = 1, 2}, i ∈ {1, · · · , 4}. (9)
For convenience, we let {a, b, l} (a ∈ MA , b ∈ MB , l ∈
Explicitly, the information of sB and PL is contained in N ) denote (αA (l)SA (a) + αB (l)SB (b)). It is considered that
each point of the constellation. If the received signal after all PLs are activated equiprobably for PS-NOMA.

Authorized licensed use limited to: Mississippi State University Libraries. Downloaded on January 23,2025 at 00:07:31 UTC from IEEE Xplore. Restrictions apply.
1076 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 40, NO. 4, APRIL 2022

1) Achievable Rate Analysis of UB : It follows from the 2) Achievable Rate Analysis of UA : Different from the
chain rule that way at UB , UA detects its own signal after removing the
interference imposed by UB . Therefore, the achievable rate
I(sB , PL; yB ) = I(sB ; yB |PL) + I(PL; yB ). (17) for UA detecting sB is given by

Now, we deal with the calculations of I(sB ; yB |PL) and I(sB , PL; yA ) = I(sB ; yA |PL) + I(PL; yA ), (22)
I(PL; yB ) in the following. Specifically, we have where I(sB ; yA |PL) and I(PL; yA ) can be respectively derived
as (23) and (24), as shown at the bottom of the next page. With
I(sB ; yB |PL) = H(yB |PL) − H(yB |sB , PL), (18) the detected PL, the achievable rate for UA detecting sA , i.e.,
I(sA ; yA |PL), can be derived as (25), as shown at the bottom
where
of the next page.
N
1
H(yB |PL) = H(yB |PL = l) (19) B. BER Analysis
N
l=1
In this subsection, closed-form BER expressions are derived
and for UA and UB in PS-NOMA systems.
1) BER of UB : To obtain the BER of UB , we derive
MB N
1 the symbol error ratio (SER) of sB at UB first. Similar to
H(yB |sB , PL) = H(yB |sB = SB (kB ), PL= l). [34], [35], we utilize the conditional pairwise error probability
N MB
kB =1 l=1
(PEP), which indicates the probability of detecting sB as ŝB
(20)
conditioned on hB , namely
Here, H(yB |PL = l) and H(yB |sB = SB (kB ), PL = l) can Pr(sB → ŝB |hB )
be respectively given by (13) and (14), as shown at the bottom = Pr(|yB − hB αB (l)sB |2 > |yB − hB αB (l̂)ŝB |2 )
of the page. Substituting (13), (14), and (a)
= Pr(|hB xB |2 − |hB x̂B |2 − 2R{yB
∗
hB (xB − x̂B )} > 0)
p(yB |sA = SA (iA ), sB = SB (iB ), PL = l) (b)
= Pr(−|hB xB |2 − 2R{wB ∗
hB xB } > 0)
1 |yB − hB {iA , iB , l}|2 
= exp − (21) (c) |hB | (|xB | + 2R{x∗A xB })2
2 2
πN0 N0 = Q
2N0 |xB |2
into (18), I(sB ; yB |PL) can be given by (15), as shown at 
the bottom of the next page. Similarly, I(PL; yB ) can also be (d) |hB |2
= Q , (26)
derived as (16), as shown at the bottom of the page. 2N0equal


H(yB |PL = l) = p(yB |PL = l) log2 (p(yB |PL = l))dyB
MA MB N MA MB
kA =1 kB =1 l=1 log2 iA =1 iB =1 p(yB |sA = SA (iA ), sB = SB (iB ), PL = l)/MA MB
=− (13)
MA MB N
H(yB |sB = SB (kB ), PL = l)

= p(yB |sB = SB (kB ), PL = l) log2 (p(yB |sB = SB (kB ), PL = l))dyB
MA MB N MA
kA =1 kB =1 l=1 log2 iA =1 p(yB |sA = SA (iA ), sB = SB (kB ), PL = l)/MA
=− . (14)
MA MB N
1
I(sB ; yB |PL) = log2 (MB ) −
MA MB N
 MA MB 
MA MB 2
iB =1 exp(−|hB {kA , kB , l} − hB {iA , iB , l} + nB | /N0 )
N
iA =1
× E log2 MA . (15)
2
kA =1 kB =1 l=1 iA =1 exp(−|hB {kA , kB , l} − hB {iA , kB , l} + nB | /N0 )
I(PL; yB ) = H(yB ) − H(yB |PL)
1
= log2 (N ) +
MA MB N
 MA MB 
M M 2
iB =1 exp(−|hB {kA , kB , l} − hB {iA , iB , l} + nB | /N0 )
A N B
iA =1
× E log2 MA MB N .
2
kA =1 kB =1 l=1 iA =1 iB =1 ll=1 exp(−|hB {kA , kB , l} − hB {iA , iB , ll} + nB | /N0 )
(16)

Authorized licensed use limited to: Mississippi State University Libraries. Downloaded on January 23,2025 at 00:07:31 UTC from IEEE Xplore. Restrictions apply.
PEI et al.: NEXT-GENERATION MULTIPLE ACCESS BASED ON NOMA WITH POWER LEVEL MODULATION 1077

where we define in (a) xB  αB (l)sB and x̂B  αB (l̂)ŝB ; are correctly detected or not. Since we have N PLs, the overall
in (b) wB = hB αA (l)sA + nB = hB xA + nB , with xA  BER of UA can be evaluated as
αA (l)sA , and xB = xB − x̂B ; the detailed derivation of (c) N

1 P sB
is present in Appendix; (d) is obtained by defining equivalent PA ≈ (1 − PA ) sB
Pbit (l) + A ,
A
(31)
noise variance as N 2
l=1
N0 |xB |2 where PAsB
is the SER of sB at UA , and Pbit A
(l) denotes
N0equal  . (27)
(|xB | + 2R{x∗A xB })2
2 the BER of MA -ary PAM demodulation over Rayleigh fading
Since we model hB as a complex Gaussian random variable, channels with PL l selected. Since it is difficult to exactly
|hB |2 has a PDF calculate the BER when sB or PL is detected incorrectly at
UA , we use the estimate PAsB /2 instead. Similarly, to derive
1 x PAsB , we use the conditional PEP, Pr(sB → ŝB |hA ), which
p|hB |2 (x) = exp − . (28)
βB βB denotes the probability of detecting sB as ŝB conditioned on
With p|hB |2 (x), we can arrive at [36, Eq. (64)] hA [34], [35]
 +∞ Pr(sB → ŝB |hA )
Pr(sB → ŝB ) = Pr(sB → ŝB |hB )p|hB |2 (x)dx = Pr(|yA − hA αB (l)sB |2 > |yA − hA αB (l̂)ŝB |2 )
0
 
 +∞ |hA |2 (|xB |2 + 2R{x∗A xB })2
|hB |2 =Q
= Q p|hB |2 (x)dx 2N0 |xB |2
0 2N0equal 

1 βB |hA |2
= 1− , (29) =Q . (32)
2 4N0equal + βB 2N0equal

which can be averaged over all the possible values of x∗A to Definitions of xB , x̂B , xA , xB , and N0equal are given in (26)
consider all interference scenarios. Notably, xA herein should and (27). Having
bit
keep the same PL as xB . Let Pij,B denote the number of 1 x
error bits when symbol SB (i) is incorrectly detected as SB
R R
(j). p|hA |2 (x) =exp − , (33)
βA βA
Since sB and PL have log2 (N MB ) bits, the BER of sB and
we can arrive at [36, Eq. (64)]
PL detections at UB can be approximated as  +∞
1 Pr(sB → ŝB ) = Pr(sB → ŝB |hA )p|hA |2 (x)dx
PB ≈
N MB log2 (N MB ) 0

N MB 1 βA
× Pr(SB
R
(i) → SB
R bit
(j))Pij,B = 1− , (36)
. (30) 2 4N0 + βA
i=1 j=i
which can be averaged over all the possible values of x∗A
2) BER of UA : Recall the ML detection in (10) and (11). to consider all interference scenarios. Equally, the xA herein
UA first decodes PL as well as sB and extracts sB from the should keep the same PL as xB . Hence, PAsB is given by
received signal yA and then decodes sA from the residual N MB
1
signal. Obviously, the overall BER can be divided into two PAsB ≈ Pr(SB
R
(i) → SB
R
(j)). (37)
complementary parts, depending on whether sB and PL l ∈ N N MB i=1 j=i

1
I(sB ; yA |PL) = log2 (MB ) −
MA MB N
 MA MB 
MA MB 2
iB =1 exp(−|hA {kA , kB , l} − hA {iA , iB , l} + nA | /N0 )
N
iA =1
× E log2 MA (23)
2
kA =1 kB =1 l=1 iA =1 exp(−|hA {kA , kB , l} − hA {iA , kB , l} + nA | /N0 )
1
I(PL; yA ) = log2 (N ) +
MA MB N
 MA MB 
MA MB 2
iB =1 exp(−|hA {kA , kB , l} − hA {iA , iB , l} + nA | /N0 )
N
iA =1
× E log2 MA MB N .
2
kA =1 kB =1 l=1 iA =1 iB =1 ll=1 exp(−|hA {kA , kB , l} − hA {iA , iB , ll} + nA | /N0 )
(24)
1
I(sA ; yA |PL) = log2 (MA ) −
M N
A MA   
MA N
iA =1 exp |hA αA (l)(SA (kA ) − SA (iA )) + nA |2 /N0
× E log2 . (25)
exp (−|nA |2 /N0 )
kA =1 l=1

Authorized licensed use limited to: Mississippi State University Libraries. Downloaded on January 23,2025 at 00:07:31 UTC from IEEE Xplore. Restrictions apply.
1078 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 40, NO. 4, APRIL 2022

As for the calculation of PbitA

(l), we first derive the conditional
error probability for PL l under the mA -th bit (mA ∈
{1, . . . , log2 (MA )}) and the conditions of receive signal-to-
noise ratio (SNR) at UA as [37], i.e.,
αA (l)2 |hA |2
γA = , (38)
N0
and (34), as shown at the bottom of the page. By substituting
the alternative representation of the Gaussian Q function
in (34), i.e.,
 π
1 2 x2
Q(x) = exp − dψ, (39)
π 0 2 sin2 ψ
we can obtain
A
Pbit (l|mA , γA )
(1−2−mA )MA −1
2 mA −1)
 i(2 M 
= {(−1) A
MA i=0
Fig. 5. Constellations of sB with N = 4, where sB is a 2-PAM symbol,
and SBR denotes the union constellation of s
mA −1 B and PL.
i(2 ) 1
× (2mA −1 −  + )
M 2
 π/2
1 ((2i + 1)dA )2 γA C. Optimal Constellation Analysis
× exp − dψ }. (40)
π 0 sin2 ψ
   In this subsection, we discuss the optimal design of con-
Q(i|γA ) stellation in terms of the values of N . For convenience, let
By statistically averaging Q(i|γA ) over the PDF of γA , which dB,ij denote the Euclidean distance between points SB R
(i)
is given by and SB R
(j), where i, j ∈ {1, . . . , N MB }. According to the
geometric analysis, we can obtain the following.
N0 −N0 γA Lemma 1: The optimal BER performance is achieved at
pγA (γA ) = exp , (41)
(αA (l))2 βA (αA (l))2 βA N = 2.
Q(i) can be derived as Proof: Since the cases of N ≥ 2 all have similar
 π2  ∞ derivation processes, we can obtain pattern by analyzing a
1 ((2i + 1)dA )2 γA certain case. For ease of discussion, we consider the 2-PAM
Q(i) = exp −
π 0 0 sin2 ψ signal sA , 2-PAM signal sB , and N = 4 scenario as an
× pγA (γA )dγA dψ example. As indicated in Algorithm 1, N only affects the
1 − G(i) order of combined constellation. When N = 4, the combined
= , (42) constellation is shown as Fig. 5. Let dB,i,min denotes the
2
minimum Euclidean distance between SB R
(i) and other points.
where
Considering the symmetry of Fig. 5, we first take SB R
(1) for
(αA (l))2 βA ((2i + 1)dA )2 instance. From Fig. 5, we can obtain dB,1,min as
G(i) = . (43)
N0 + (αA (l))2 βA ((2i + 1)dA )2
dB,1,min = min{dB,12 , dB,13 , dB,16 , dB,18 }, (45)
Therefore, we have (35), as shown at the bottom of the page,
and where (46)–(48), as shown at the bottom of the next page, and
log2 (MA )
(49), as shown at the bottom of the next page, respectively.
1 In the similar manner, we can derive dB,2,min , and so on.
A
Pbit (l) = A
Pbit (l|mA ). (44)
log2 (MA ) mA =1
Therefore, the minimum Euclidean distance of Fig. 5, can be
derived as
Finally, by substituting (37) and (44) into (31), we can derive
the expression of PA . dB,min = min{dB,i,min, i ∈ {1, . . . , 8}}. (50)

(1−2−mA )MA −1   
2 i(2mA −1)/MA  mA −1 i(2mA −1 ) 1 
A
Pbit (l|mA , γA ) = (−1) 2 − +  Q (2i + 1)dA 2γA .
MA i=0
M 2
(34)
(1−2−mA )MA −1  
2 
i(2mA −1)
 mA −1 i(2mA −1 ) 1
A
Pbit (l|mA ) = (−1) MA
2 − +  Q(i) , (35)
MA i=0
M 2

Authorized licensed use limited to: Mississippi State University Libraries. Downloaded on January 23,2025 at 00:07:31 UTC from IEEE Xplore. Restrictions apply.
PEI et al.: NEXT-GENERATION MULTIPLE ACCESS BASED ON NOMA WITH POWER LEVEL MODULATION 1079

Since we have N ≥ 2, the angle π/N and 2π/N are values IV. N UMERICAL R ESULTS
between 0 and π. In other words, cos(π/N ) is monotonously In this section, we illustrate the BER and achievable rate
increasing with respect to N . Evidently, to achieve optimal performance of the proposed scheme through Monte Carlo
BER performance, N should be equal to the minimum value, simulation and numerical results. Since the power coefficients
which proves Lemma 1.  of PS-NOMA are randomly varying, from the fairness perspec-
tive, we set the counterpart without PS (denoted by NOMA)
D. Multi-User Case as a regular NOMA scheme. We consider a single-cell with
1 km radius in which users are uniformly distributed within
In this subsection, we consider the multi-user scenario.
the cell. We set the transmit power pt of BS to 1 W. In all
Since it needs specific design for multi-user case, we just give
simulations, the channels are assumed to be Rayleigh fading
the general thinking here. We herein also use the information
and perfectly known to both users. Without loss of generality,
of PL to enhance the performance of the farthest user. Notably,
we can respectively set the average powers of hA , hB as
the core idea of design is the same as two-user case.
βA = 10, βB = 1 by choosing the appropriate distances
We assume that there are U users (denoted by U1 , . . . , UU )
between users and BS. For brevity, we will refer to “PS-
to be served and N PLs to be selected. The channels BS→
NOMA (MA -PAM, MB -PAM, N )” as the PS-NOMA scheme
Ui (i ∈ {1, . . . , U}) is denoted by hi , whose variance is
where the BS has N PLs and maps the information bits for UA
βi . Without loss of generality, we sort the average powers
and UB into MA -PAM and MB -PAM, respectively. Similarly,
of channels as β1 > . . . > βU . On the transmitter side,
“NOMA (MA -PAM, MB -PAM)” denotes the NOMA scheme
we can also use rotation to design the transmitted constellation.
with MA -PAM and MB -PAM signals intended for UA and
Specifically, the rotated power matrix can be written as
UB , respectively. In the following figures, the power matrices
⎡ ⎤
α1 (1), . . . , αU (1) of “PS-NOMA (2-PAM, 2-PAM, N = 2)” and “NOMA (2-
⎢ α1 (2), . . . , αU (2) ⎥ PAM, 4-PAM)” are respectively set by
⎢ ⎥  √ √ 
G=⎢ .. .. .. ⎥, (51)
⎣ ⎦ 0.2, 0.8
. . . G= √ √ (54)
α1 (N ), αU (N )
..., 0.1 exp(jπ/2), 0.9 exp(jπ/2)
 √ √
where αi (l) = exp(j(l − 1)π/N ) pi (l), l ∈ N = and G = [ 0.2, 0.8] according to Lemma 1. Besides,
{1, . . . , N }, pi (l) is the power for Ui in the l-th PL. where we take SNR = pt Es /N0 as the horizontal axis of these
si ∈ Si with figures.

Si = {±di , ±3di , · · · , ±(Mi − 1)di } (52) A. BER Performance

beingthe corresponding Mi -PAM constellation, in which In this subsection, the BER performance of PS-NOMA and
di = 3/(Mi2 − 1) represents half of the minimum distance NOMA is compared, assuming that UA and UB in all consid-
between two adjacent points of the normalized constellation. ered schemes employ ML detection. Since the comparison of
Consequently, the irregular M1 × . . . × MU N -ary signal OMA and NOMA has been widely argued in previous works,
constellation of sC can be expressed as the BER curves of OMA counterparts are not presented.
In Fig. 6, we compare the BER performance of “PS-NOMA
SC = {sC |si = Si (ki ), PL = l, ki ∈ Mi , l ∈ N , (2-PAM, 2-PAM, N = 2)” and “NOMA (2-PAM, 4-PAM)”
Mi = {1, 2, · · · , Mi }, i ∈ {1, . . . , U}}. (53) with their theoretical BER results presented. Given fairness,
we assume that the maximum spectral efficiency for both
As for the receiver side, Ui decodes the data symbol for Uj schemes are 3 bits per second per Hertz (bps/Hz), and both
(i < j < U ) and the chosen PL sequentially through joint schemes have the same constellation order. It can be seen
constellation like Fig. 3. For UU , it just has to decode its own from Fig. 6 that the theoretical curves approximately match
data symbol and PL. Specific decoding process is similar to with their simulation counterparts. Besides, one can find that
Algorithm 1. the BER performance of UA is superior to that of UB in

dB,12 = 2dB |αB (1)|, (46)

 π
dB,13 = |dB αB (1)|2 + |dB αB (2)|2 − 2d2B |αB (1)αB (2)| cos , (47)
N
2π
dB,16 = |dB αB (1)|2 + |dB αB (3)|2 − 2d2B |αB (1)αB (3)| cos , (48)
N

 π
dB,18 = |dB αB (1)|2 + |dB αB (4)|2 − 2d2B |αB (1)αB (4)| cos , (49)
N

Authorized licensed use limited to: Mississippi State University Libraries. Downloaded on January 23,2025 at 00:07:31 UTC from IEEE Xplore. Restrictions apply.
1080 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 40, NO. 4, APRIL 2022

Fig. 6. BER comparison between PS-NOMA and NOMA. Fig. 9. Achievable rate and MI performance of PS-NOMA.

than those of NOMA. This phenomenon tells us that applying

PS can improve performance without increasing constellation
complexity.
Figures 7, 8 respectively investigate two different scenarios
where the power matrices of the compared counterparts
 √ √ 
0.4, 0.6
G= √ √ (55)
0.1 exp(jπ/2), 0.9 exp(jπ/2)
for case 1,
 √ √ 
0.4, 0.6
G= √ √ (56)
0.3 exp(jπ/2), 0.7 exp(jπ/2)
for case 2, respectively. We first discuss the BER performance
of UB . Concretely, in both cases, the minimum Euclidean
distances of the constellations are smaller than those of
Fig. 7. BER comparison between PS-NOMA schemes with different PL the benchmark. Hence they show worse BER performances.
values: Case 1. Moreover, one can observe that the BER performance of UB
changes more significantly than that of UA . The illustration
of this phenomenon is shown in the following. From (10)
and (11), we know that false detection of sB and PL may
result in the wrong sB , or wrong PL, or both wrong sB and
PL. The first detection process (10) is related to both sB
and PL, and it is directly related to the correctness of (11),
so the power of UB has a more significant influence than UA .
Notably, we assume that MA = MB in this figure. Therefore
this conclusion stands. In wording, we can consistently achieve
better BER performance through designing power matrices
when N = 2 and the design of lower-order matrices are much
simpler than higher-order ones.

B. Achievable Rate Performance

In this subsection, we first evaluate the achievable rate
and MI performance of the proposed PS-NOMA, and then
Fig. 8. BER comparison between PS-NOMA schemes with different PL compare the achievable rate of PS-NOMA with the NOMA
values: Case 2. counterparts.
In Fig. 9, we show the achievable rate and MI performance
both PS-NOMA and NOMA, which comes from the fact of PS-NOMA. As seen from the figure, all the curves grow
βA > βB . Although all the curves have the same constellation steadily as the SNR increases at low-to-medium SNR while
order as 8, the performances of PS-NOMA are always better achieving floors at high SNR. Obviously, the rates of UA

Authorized licensed use limited to: Mississippi State University Libraries. Downloaded on January 23,2025 at 00:07:31 UTC from IEEE Xplore. Restrictions apply.
PEI et al.: NEXT-GENERATION MULTIPLE ACCESS BASED ON NOMA WITH POWER LEVEL MODULATION 1081

are equal. It is worth noting that the performance improvement

of UA is more obvious than that of UB .
Fig. 11 depicts the simulation results of the achievable rate
region when the transmitted power is fixed at the √ BS. For
NOMA,
√ √ we set
√ transmitted power matrix as G = [ 0.1 →
0.5, 0.9 → 0.5], while for PS-NOMA, we have
G  √ √ √ √ 
0.1 → 0.5,
√ 0.9 →√ 0.5 √
= .
exp(jπ/2)×(0 → 0.4), exp(jπ/2)×( 1 → 0.6)
(57)
It can be seen from Fig. 11 that the PS-NOMA method
outperforms NOMA scheme in terms of achievable rate region.
Furthermore, the interval between curves will be largest when
we choose the power coefficients appropriately. Notably, this
Fig. 10. Achievable rate comparison between PS-NOMA and NOMA.
result comes at the expense of different power allocation,
which implies that a larger achievable rate region can be
achieved when adopting appropriate power allocation matrix.

V. C ONCLUSION
In this paper, we have proposed a novel two-user NOMA
scheme using PS, in which the ordinary PAM symbol carries
the bits of users. In addition, the BS randomly chooses a
PL from the power matrix preset for each transmission to
carry more information. For PS-NOMA, approximate BER
expressions have been derived in closed form for both users
over Rayleigh flat fading channels. The rates and MI under
finite input have also been developed. Computer simulations
have verified the performance analysis and shown that the
proposed scheme achieves better BER performance than con-
ventional NOMA without increasing constellation complex-
ity. Moreover, we observe that the proposed PS-NOMA can
achieve better BER performance with the minimum number of
PLs, which reduces the design difficulty of the power matrix.
On the other hand, the proposed scheme also shows superior
Fig. 11. Achievable rate region comparison between PS-NOMA and NOMA. achievable rate performance and larger rate region compared to
that of the NOMA counterpart. This work focuses only on the
two-user scenario. We herein only give the original design for
obtained from (25), UB obtained from (17), and UB observed
multi-user case, and we leave the specific more-user extension
at UA obtained from (22) saturate to log2 (MA ), log2 (N MB ),
for future study.
and log2 (N MB ), respectively. On the other hand, the MI
curves I(PL; yA ) and I(PL; yB ) respectively denote the PL
A PPENDIX
gains at UA and UB and are respectively generated from (24)
P ROOF OF (c)
and (16). Evidently, both curves saturate at 1 bps/Hz since the
input entropy of the power-domain is log2 (N ). In other words, To process step (c) of (26), we use the method introduced in
all the gains gleaned from the power-domain are assigned to [35]. Let D  −|hB xB |2 − 2R{wB ∗
hB xB }. With wB =
UB . Moreover, since the channel quality of UA is much better hB xA + nB , D can be rewritten as
than that of UB , the achievable rate or MI obtained at UA is D = −|hB xB |2 −|hB |2 2R{x∗A xB }−2R{n∗B hB xB }.
always higher than that at UB , which guarantees the success
of SIC. (A.1)
Subsequently, in Fig. 10, we present the curves of Obviously, D is Gaussian distributed with
PS-NOMA versus NOMA. Compared to NOMA, PS-NOMA
provides achievable rate gains for both UA and UB , and E{D} = −|hB xB |2 − |hB |2 2R{x∗A xB } (A.2)
obtains the sum achievable rate enhancement in turn. This and
system performance improvement is more pronounced at low
SNR, while the system gap narrows with increasing SNR. Var{D} = Var{n∗B hB xB } = 2N0 |hB xB |2 . (A.3)

Since PS-NOMA and NOMA have the same constellation Therefore, we have Pr(D > 0) = Q(−E{D}/ Var{D}),
order, at high SNR, the achievable rate performances of them which can be easily simplified as the result of (c).

Authorized licensed use limited to: Mississippi State University Libraries. Downloaded on January 23,2025 at 00:07:31 UTC from IEEE Xplore. Restrictions apply.
1082 IEEE JOURNAL ON SELECTED AREAS IN COMMUNICATIONS, VOL. 40, NO. 4, APRIL 2022

R EFERENCES [25] N. Su, E. Panayirci, A. Yesilkaya, M. Koca, H. Haas, and H. V. Poor,

“Multiuser physical layer security in visible light communications
[1] “Cellular networks for massive IoT,” Ericsson, Stockholm, Sweden, systems employing MIMO-NOMA technique,” IEEE Trans. Commun.,
White Paper, Jan. 2016. [Online]. Available: https://www.ericsson.com/ vol. 69, no. 4, pp. 2585–2598, Jan. 2021.
en/reports-and-papers/white-papers/cellular-networks-for-massive-iot– [26] E. Panayirci, A. Yesilkaya, T. Cogalan, H. V. Poor, and H. Haas,
enabling-low-power-wide-area-applications “Physical-layer security with optical generalized space shift keying,”
[2] L. Dai, B. Wang, Y. Yuan, S. Han, C.-L. I, and Z. Wang, “Non- IEEE Trans. Commun., vol. 68, no. 5, pp. 3042–3056, May 2020.
orthogonal multiple access for 5G: Solutions, challenges, opportunities, [27] Q. Li, M. Wen, E. Basar, H. V. Poor, and F. Chen, “Spatial modulation-
and future research trends,” IEEE Commun. Mag., vol. 53, no. 9, aided cooperative NOMA: Performance analysis and comparative study,”
pp. 74–81, Sep. 2015. IEEE J. Sel. Areas Commun., vol. 13, no. 3, pp. 715–728, Jun. 2019.
[3] Y. Liu, Z. Ding, M. Elkashlan, and H. V. Poor, “Cooperative non- [28] C. Liu, M. Ma, Y. Yang, and B. Jiao, “Optimal spatial-domain design
orthogonal multiple access with simultaneous wireless information for spatial modulation capacity maximization,” IEEE Commun. Lett.,
and power transfer,” IEEE J. Sel. Areas Commun., vol. 34, no. 4, vol. 20, no. 6, pp. 1092–1095, Jun. 2016.
pp. 938–953, Apr. 2016. [29] Y. Chen, L. Wang, Y. Ai, B. Jiao, and L. Hanzo, “Performance analysis
[4] Y. Chen, L. Wang, and B. Jiao, “Cooperative multicast non-orthogonal of NOMA-SM in vehicle-to-vehicle massive MIMO channels,” IEEE
multiple access in cognitive radio,” in Proc. IEEE Int. Conf. Commun. J. Sel. Areas Commun., vol. 35, no. 12, pp. 2653–2666, Dec. 2017.
(ICC), May 2017, pp. 1–6. [30] X. Wang, J. Wang, L. He, Z. Tang, and J. Song, “On the achievable
[5] M. Vaezi, Z. Ding, and H. V. Poor, Multiple Access Techniques for 5G spectral efficiency of spatial modulation aided downlink non-orthogonal
Wireless Networks and Beyond. Cham, Switzerland: Springer, 2019. multiple access,” IEEE Commun. Lett., vol. 21, no. 9, pp. 1937–1940,
[6] X. Pei, H. Yu, X. Wang, Y. Chen, M. Wen, and Y.-C. Wu, “NOMA- Sep. 2017.
based pervasive edge computing: Secure power allocation for IoV,” IEEE [31] X. Wang, J. Wang, L. He, and J. Song, “Spectral efficiency analysis for
Trans. Ind. Informat., vol. 17, no. 7, pp. 5021–5030, Jul. 2021. downlink NOMA aided spatial modulation with finite alphabet inputs,”
[7] L. Wei, Y. Chen, and B. Jiao, “Physical layer security performance analy- IEEE Trans. Veh. Technol., vol. 66, no. 11, pp. 10562–10566, Nov. 2017.
sis and optimization for FD-NOMA V2V systems,” China Commun., [32] L. Pan and J. Zheng, “Spatial modulation aided cooperative NOMA:
vol. 17, no. 11, pp. 29–41, Nov. 2020. Implementation and achievable rate analysis,” in Proc. IEEE Int. Conf.
[8] Y. Chen, M. Wen, L. Wang, W. Liu, and L. Hanzo, “SINR-outage Commun. (ICC), Shanghai, China, May 2019, pp. 1–6.
minimization of robust beamforming for the non-orthogonal wireless [33] Y.-Y. Zhang and J.-K. Zhang, “Fundamental applicability of spatial
downlink,” IEEE Trans. Commun., vol. 68, no. 11, pp. 7247–7257, modulation: high-SNR limitation and low-SNR advantage,” IEEE J. Sel.
Nov. 2020. Areas Commun., vol. 37, no. 9, pp. 2165–2178, Sep. 2019.
[9] Z. Ding et al., “Application of non-orthogonal multiple access in LTE [34] L. Bariah, A. Al-Dweik, and S. Muhaidat, “On the performance of non-
and 5G networks,” IEEE Commun. Mag., vol. 55, no. 2, pp. 185–191, orthogonal multiple access systems with imperfect successive interfer-
Feb. 2017. ence cancellation,” in Proc. IEEE Int. Conf. Commun. Workshops (ICC),
[10] S. M. R. Islam, N. Avazov, O. A. Dobre, and K.-S. Kwak, “Power- May 2018, pp. 1–6.
domain non-orthogonal multiple access (NOMA) in 5G systems: Poten- [35] E. Başar, U. Aygölü, E. Panayırcı, and H. V. Poor, “Orthogonal
tials and challenges,” IEEE Commun. Surveys Tuts., vol. 19, no. 2, frequency division multiplexing with index modulation,” IEEE Trans.
pp. 721–742, 2nd Quart., 2017. Signal Process., vol. 61, no. 22, pp. 5536–5549, Nov. 2013.
[11] A. Arafa, E. Panayirci, and H. V. Poor, “Relay-aided secure broadcasting [36] M.-S. Alouini and A. J. Goldsmith, “A unified approach for calculating
for visible light communications,” IEEE Trans. Commun., vol. 67, no. 6, error rates of linearly modulated signals over generalized fading chan-
pp. 4227–4239, Jun. 2019. nels,” IEEE Trans. Commun., vol. 47, no. 9, pp. 1324–1334, Sep. 1999.
[12] Technical Specification Group Radio Access Network; Study on Non- [37] K. Cho and D. Yoon, “On the general BER expression of one- and two-
Orthogonal Multiple Access (NOMA) for NR, Standard TR 38.812, dimensional amplitude modulations,” IEEE Trans. Commun., vol. 50,
Release 16, (V16.0.0), 3GPP, Rep., 3rd Generation Partnership Project, no. 7, pp. 1074–1080, Jul. 2002.
Sophia Antipolis, France, Dec. 2018.
[13] D. Slepian and J. K. Wolf, “A coding theorem for multiple access
channels with correlated sources,” Bell Syst. Tech. J., vol. 52, no. 7,
pp. 1037–1076, 1973.
[14] R. S. Cheng and S. Verdú, “Gaussian multiaccess channels with ISI:
Capacity region and multiuser water-filling,” IEEE Trans. Inf. Theory, Xinyue Pei (Student Member, IEEE) received the
vol. 39, no. 3, pp. 773–785, May 1993. B.S. degree from the South China University of
[15] G. Caire and S. Shamai (Shitz), “On the achievable throughput of a Technology, Guangzhou, China, in 2018, where she
multiantenna Gaussian broadcast channel,” IEEE Trans. Inf. Theory, is currently pursuing the Ph.D. degree. Her main
vol. 49, no. 7, pp. 1691–1706, Jul. 2003. research interests include non-orthogonal multiple
[16] T. M. Cover and J. A. Thomas, Elements of Information Theory, 2nd ed. access, massive MIMO, physical layer security, the
Hoboken, NJ, USA: Wiley, 2006. Internet of Things, and caching.
[17] A. Lozano, A. M. Tulino, and S. Verdu, “Optimum power allocation
for parallel Gaussian channels with arbitrary input distributions,” IEEE
Trans. Inf. Theory, vol. 52, no. 7, pp. 3033–3051, Jul. 2006.
[18] J. Harshan and B. S. Rajan, “On two-user Gaussian multiple access
channels with finite input constellations,” IEEE Trans. Inf. Theory,
vol. 57, no. 3, pp. 1299–1327, Mar. 2011.
[19] J. Harshan and B. S. Rajan, “A novel power allocation scheme for
two-user GMAC with finite input constellations,” IEEE Trans. Wireless
Commun., vol. 12, no. 2, pp. 818–827, Feb. 2013. Yingyang Chen (Member, IEEE) received the
[20] Z. Dong, H. Chen, J.-K. Zhang, and L. Huang, “On non-orthogonal
B.Eng. degree in electronic engineering from
multiple access with finite-alphabet inputs in Z-channels,” IEEE J. Sel.
the Yingcai Honors College, University of Elec-
Areas Commun., vol. 35, no. 12, pp. 2829–2845, Dec. 2017.
[21] Z. Dong, H. Chen, J. K. Zhang, L. Huang, and B. Vucetic, “Uplink tronic Science and Technology of China (UESTC),
non-orthogonal multiple access with finite-alphabet inputs,” IEEE Trans. Chengdu, China, in 2014, and the Ph.D. degree in
Wireless Commun., vol. 17, no. 9, pp. 5743–5758, Sep. 2018. signal and information processing from Peking Uni-
[22] Y. Yang and B. Jiao, “Information-guided channel-hopping for high versity, Beijing, China, in 2019. From March 2018 to
data rate wireless communication,” IEEE Commun. Lett., vol. 12, no. 4, September 2018, she worked as a Visiting Student
pp. 225–227, Apr. 2008. with the Next Generation Wireless Group, University
[23] M. Wen et al., “A survey on spatial modulation in emerging wireless of Southampton, supervised by Prof. Lajos Hanzo.
systems: Research progresses and applications,” IEEE J. Sel. Areas She is currently an Assistant Professor with Jinan
Commun., vol. 37, no. 9, pp. 1949–1972, Sep. 2019. University, Guangzhou, China. Her current research interests include perfor-
[24] N. Su, E. Panayirci, M. Koca, and H. V. Poor, “Spatial constellation mance analysis and optimization in wireless communication systems, cross-
design-based generalized space shift keying for physical layer security layer resource management, and array signal processing. She is serving as an
of multi-user MIMO communication systems,” IEEE Wireless Commun. Editor for the IEEE C OMMUNICATIONS L ETTERS , and a Guest Editor for
Lett., vol. 10, no. 8, pp. 1785–1789, Aug. 2021. the IET C OMMUNICATIONS .

Authorized licensed use limited to: Mississippi State University Libraries. Downloaded on January 23,2025 at 00:07:31 UTC from IEEE Xplore. Restrictions apply.
PEI et al.: NEXT-GENERATION MULTIPLE ACCESS BASED ON NOMA WITH POWER LEVEL MODULATION 1083

Miaowen Wen (Senior Member, IEEE) received Erdal Panayirci (Life Fellow, IEEE) received the
the Ph.D. degree from Peking University, Beijing, Ph.D. degree in electrical engineering and sys-
China, in 2014. tem science from Michigan State University, USA,
From January 2019 to January 2021, he was in 1971. He is currently a Professor with the
with The University of Hong Kong, Hong Kong, Department of Electrical and Electronics Engineer-
as a Post-Doctoral Research Fellow. He is currently ing, Kadir Has University, Istanbul, Turkey, and
a Professor with the South China University of a Visiting Research Collaborator with the Depart-
Technology, Guangzhou, China. He has published ment of Electrical Engineering, Princeton University,
two books and more than 130 journal articles. His Princeton, NJ, USA. He spent the academic years
research interests include a variety of topics in the 2008–2009 and 2017–2018 with the Department of
areas of wireless and molecular communications. Electrical Engineering, Princeton University. He has
He was a recipient of the IEEE Asia–Pacific (AP) Outstanding Young published extensively in leading scientific journals and international confer-
Researcher Award in 2020, and four best paper awards from the IEEE ences and coauthored the book Principles of Integrated Maritime Surveillance
ITST’12, the IEEE ITSC’14, the IEEE ICNC’16, and the IEEE ICCT’19. Systems (Kluwer Academic, 2000). His research interests include communica-
He was the winner in data bakeoff competition (Molecular MIMO) at tion theory, synchronization, advanced signal processing techniques, and their
IEEE Communication Theory Workshop (CTW) 2019, Selfoss, Iceland. applications to wireless electrical, underwater, and optical communications.
He served as a Guest Editor for the IEEE J OURNAL ON S ELECTED A REAS He is a member of the National Academy of Sciences. He has served
IN C OMMUNICATIONS (Special Issue on Spatial Modulation for Emerging as a member of the IEEE Fellow Committee from 2005 to 2008 and
Wireless Systems) and for the IEEE J OURNAL OF S ELECTED T OPICS IN from 2018 to 2020, and as a member of the IEEE GLOBECOM/ICC
S IGNAL P ROCESSING (Special Issue on Index Modulation for Future Wireless Management and Strategy Standing Committee from 2017 to 2020. He is
Networks: A Signal Processing Perspective and Special Issue on Advanced currently a member of the IEEE ComSoc Awards Standing Committee,
Signal Processing for Local and Private 5G Networks). He is currently serving between 2022–2024. He served as the Technical Program Co-Chair, the
as an Editor for the IEEE T RANSACTIONS ON C OMMUNICATIONS, the General Co-Chair, and the Technical Program Chair for several IEEE ICC,
IEEE T RANSACTIONS ON M OLECULAR , B IOLOGICAL , AND M ULTI -S CALE IEEE PIMRC, and IEEE WCNC conferences. He was an Editor of the IEEE
C OMMUNICATIONS, and the IEEE C OMMUNICATIONS L ETTERS . T RANSACTIONS ON C OMMUNICATIONS in synchronization and equalization
from 1995 to 2000.

H. Vincent Poor (Life Fellow, IEEE) received the

Hua Yu (Member, IEEE) received the B.S. degree in Ph.D. degree in EECS from Princeton University in
mathematics from Southwest University, Chongqing, 1977. From 1977 to 1990, he was on the faculty
China, in 1995, and the Ph.D. degree in commu- of the University of Illinois at Urbana–Champaign.
nication and information system from the South Since 1990, he has been on the faculty at Princeton,
China University of Technology, Guangzhou, China, where he is currently the Michael Henry Strater
in 2004. He was a Visiting Scholar with the University Professor. From 2006 to 2016, he served
School of Marine Science and Policy, University of as the Dean of Princeton’s School of Engineering
Delaware, from 2012 to 2013. He is currently a Pro- and Applied Science. He has also held visiting
fessor with the School of Electronic and Information appointments at several other universities, includ-
Engineering, South China University of Technology, ing most recently at Berkeley and Cambridge. His
and an Adjunct Researcher with the Key Laboratory research interests are in the areas of information theory, machine learning and
of Marine Environmental Survey Technology and Application, Ministry of network science, and their applications in wireless networks, energy systems,
Natural Resources. He is also the Director of the Department of Underwater and related fields. Among his publications in these areas is the forthcoming
Communications and the National Engineering Technology Research Center book Machine Learning and Wireless Communications (Cambridge University
for Mobile Ultrasonic Detection. His main research interests are in the areas Press). Dr. Poor is a member of the National Academy of Engineering and
of wireless communications and underwater acoustic communications and the National Academy of Sciences, and a foreign member of the Chinese
networks. He was a co-recipient of the Best Paper Award from IEEE ICNC Academy of Sciences, The Royal Society, and other national and international
2016 and the Best Demo Award from ACM WUWNET 2018. academies. He received the IEEE Alexander Graham Bell Medal in 2017.

Authorized licensed use limited to: Mississippi State University Libraries. Downloaded on January 23,2025 at 00:07:31 UTC from IEEE Xplore. Restrictions apply.