 Research
 Open Access
 Published:
Precoding design of NOMAenabled D2D communication system with low latency
EURASIP Journal on Wireless Communications and Networking volume 2019, Article number: 185 (2019)
Abstract
This letter investigated a devicetodevice (D2D) communication underlaying cellular system, where one wireless device (WD) can directly transmit information to two inner users under nonorthogonal multiple access (NOMA) protocol with low latency. Considering the user fairness and reliability, both users can adopt successive interference cancellation (SIC) to decode information by the given decoding order. Subject to predefined qualityofservice (QoS) requirement and the constraint of WD’s transmit power, the maximal rate of latter decoded information and the corresponding precoding vectors are obtained by applying the technique of semidefinite relaxation (SDR) and the Langrangian duality method. Then, the suboptimal scheme based on singular value decomposition (SVD) is also proposed with a lower computational complexity. Numerical simulation shows that with the design of proper precoding vectors D2D communication system assisted by NOMA has a better performance than orthogonal multiple access (OMA).
Introduction
Nonorthogonal multiple access (NOMA) is one of the promising multiple access to realize the challenging requirements of 5G [1, 2], such as massive connectivity, high data rate, and low latency. It has proved to be a viable solution for future dense networks and Internet of Things (IoT) devices. Unlike conventional orthogonal multiple access, NOMA uses the power domain to serve multiple users at different power levels at the same time, code, and frequency [3], in which superposition coding and successive interference cancellation (SIC) are employed [4]. Many various NOMA designs combined with multipleinput multipleoutput (MIMO) [5], cooperative relaying [6] and millimeterwave communications [7], have appeared in recent researches. In [8], the random opportunistic beamforming, which is a signal processing technique used in various wireless systems for directional communications [9], is first proposed for the MIMO NOMA systems, and the transmitter generated multiple beams and superposed multiple users within each beam. In [10], a beamforming design based on zeroforcing and user pairing scheme are proposed for the downlink multiuser NOMA system, assuming the perfect channel state information (CSI) is available at the transmitter. The integration of NOMA and multiuser beamforming thus has the potential to capture the benefits of both NOMA and beamforming.
Devicetodevice (D2D) communication makes it possible for users in proximity to communicate with each other directly rather than relying on base stations (BSs) [11], and thus, it is an available way for reliable and lowlatency communication. In [12] and [14], mode selection in underlay D2D networks is studied, while [13] investigates an efficient way of reusing the downlink resources for cellular and D2D mode communication. A step further from D2D pairs, [15] studies D2D groups that use NOMA as their transmission technique to serve multiple D2D receivers. Zhao et al. [16] consider the setting of an uplink singlecell cellular network communications. In order to further improve the outage performance of the NOMAweak user in a user pair and reduce cooperative delay, [17] focuses on fullduplex D2Daided cooperative NOMA.
Method
This paper considers the D2D communication underlaying cellular system in a multiple cellular networks, which takes advantages of NOMA and D2D systems to increase the available throughput of the wireless networks. The superposed signals are sent from the multiantenna wireless device (WD) to two singleantenna users under NOMA protocol. In particular, taking the user fairness into account, we study the case in which both users can adopt SIC to decode information by the given decoding order. To guarantee predefined qualityofservice (QoS) requirement and the constraint of WD’s transmit power, the maximal rate of latter decoded information and the corresponding precoding vectors are obtained by applying the technique of semidefinite relaxation and the Langrangian duality method. The suboptimal solution based on single value decomposition (SVD) is proposed with lower computational complexity. Then, simulations are also provided to verify the performance of the proposed NOMAenabled D2D schemes.
The rest of this letter is organized as follows. In Section 3, we introduce the system model and formulate the problem. In Section 4, we derive the optimal solution to this optimization problem. In Section 5, we propose the suboptimal solution based on SVD. In Section 6, we show simulation results that justify the performance of the proposed approaches. Our conclusions are included in Section 7.
Notations: Scalars are denoted by lowercase letters, vectors are denoted by boldface lowercase letters, and matrices are denoted by boldface uppercase letters. For a square matrix A,tr(A),rank(A), and A^{H} denote its trace, rank, and conjugate, respectively. A≥0 and A≤0 represent that A is a positive semidefinite matrix and a negative semidefinite matrix, respectively. ∥x∥ denotes the Euclidean norm of a complex vector x. E[·] denotes the statistical expectation. [ · ]^{+} means max(0, ·). \({\mathbb P}\left ({\, \cdot \,} \right)\) defines an outage probability event. The distribution of a circularly symmetric complex Gaussian (CSCG) random vector with mean vector x and covariance matrix Σ is denoted by CN(0,Σ), and ∼ stands for “distributed as”. \({{\mathbb {C}}^{x \times y}}\) denotes the space of x×y complex matrices.
System model and problem formulation
System model
In this paper, one D2D communication underlaying cellular system in a multiple cellular networks is considered, as illustrated in Fig. 1. One BS can serve a set of cellular users by WDs. Different cellular networks are allocated with orthogonal resource blocks, such as time, frequency, code, and space, in order to eliminate the inter interference between different cellular networks. Actually, WD is also regarded as a special user with relaying function and has a higher priority to decode its own message compared to s_{1} and s_{2}, which are transmitted to two ordinary users. So WD has direct links with two ordinary users, respectively, while no direct link between the BS and each ordinary user is assumed due to significant path loss [18]. We focus on a singlecell downlink transmission scenario, where a WD can receive the signals from the BS and then transmit the superposed signals using the wellknown amplifyandforward (AF) protocol [19] to corresponding users under NOMA protocol. The WD is equipped N antennas, and each user has a single antenna for the facility cost. Considering the complexity of the system, only two users are served at the same resource block. Note that it is of practical significance to choose two users to perform NOMA since NOMA systems are strongly interferencelimited [20]. It is often more appropriate to group two users together to perform NOMA with user pairing [21] to realize reliable and low latency communication. It is assumed that the users’ channel state information is perfectly available at the WD. We denote two types of information to both users by s_{1} and s_{2}, respectively. It is assumed that s_{1} and s_{2} are independent and identically distributed (i.i.d.) circularly symmetric complex Gaussian (CSCG) random signals with unit average power [22], i.e., E[s_{1}^{2}]=E[s_{2}^{2}]=1. Therefore, the complex baseband transmitted signal of WD can be expressed as:
where the \({\mathbf {w}_{1}},{\mathbf {w}_{2}} \in {{\mathbb {C}}^{N \times 1}}\) are the precoding vectors of s_{1} and s_{2}, respectively [23]. And the observations at two users are given by:
where \({\mathbf {h}_{1}}, {\mathbf {h}_{2}} \in {{\mathbb {C}}^{N \times 1}}\) denote the complex Gaussian channel vector of user 1 and user 2, respectively, which are independent and identically distributed (i.i.d) fading channels. n_{1} and n_{2} are additive Gaussian noise (AGN), satisfying n_{1},n_{2}∼CN(0,σ^{2}).
Problem formulation
First, by taking the fairness of users and the decoding order into consideration, it is assumed that both users can use SIC to decode s_{2} first, then subtract it from the observation before s_{1} is decoded. The signaltointerferenceplusnoise ratio (SINR) of two users to decode s_{1} and s_{2} are respectively given by:
Then, according to the decoding order, s_{2} has a higher priority to be decoded. Actually, the cognitive radio concept is used here [24]. So the achievable rate of s_{2} is dependent on the minimal SINR of s_{2} which is decoded by each user, and its achievable rate can be expressed as:
In fact, s_{1}, which is intend for user 1, is inevitably decoded by user 2, and the achievable rate of s_{1} is considered to subtract the interception by user 2. So the achievable rate of s_{1} is written by [25]:
It can be verified that there is an effective rate of s_{1} only if the channel condition of user 1 is no worse than that of use 2.
Next, we will discuss the transmit power of WD. Suppose that the energy for receiving and amplifying the information in WD is supplied by extra power, and the transmit power P of WD is almost used to forward two types of information. So the transmit power must be satisfied [26]:
At last, in this paper, we aim to maximize the achievable rate of s_{1} subject to the predefined QoS of user 2 and the given transmit power of the WD. The optimization problem can be formulated as:
where \(\phantom {\dot {i}\!}{\gamma _{M}}{\mathrm { = }}{2^{2{R_{M}}  1}}\), and R_{M} is the target rate of s_{2} to satisfy the corresponding requirement of QoS.
Note that due to the nonconvex nature of (9a) and (9b), problem (9) is undoubtedly nonconvex in its current form. In the following subsection, we will find the optimal solution based on the analysis and transformation of problem (9).
The optimal solution
In order to solve the above nonconvex problem, we consider the nontrivial case of the problem (9), in which the objective function is positive and can be rewritten as:
It is obvious that with the same constraints, \({\text {lo}}{{\mathrm {g}}_{\mathrm {2}}}\frac {{{\sigma ^{\mathrm {2}}}{\mathrm { + }}\mathbf {h}_{\mathrm {1}}^{H}{\mathbf {w}_{\mathrm {1}}}{{\mathrm {}}^{\mathrm {2}}}}}{{{\sigma ^{\mathrm {2}}}{\mathrm { + }}\mathbf {h}_{\mathrm {2}}^{H}{\mathbf {w}_{\mathrm {1}}}{{\mathrm {}}^{\mathrm {2}}}}}\) and \(\frac {{{\sigma ^{\mathrm {2}}}{\mathrm { + }}\mathbf {h}_{\mathrm {1}}^{H}{\mathbf {w}_{\mathrm {1}}}{{\mathrm {}}^{\mathrm {2}}}}}{{{\sigma ^{\mathrm {2}}}{\mathrm { + }}\mathbf {h}_{\mathrm {2}}^{H}{\mathbf {w}_{\mathrm {1}}}{{\mathrm {}}^{\mathrm {2}}}}}\) have the same optimal solution.
Meanwhile, (9b) is divided into two constraints about the SINR of s_{2}. So the problem (9) has the same optimal solution with the following problem:
The semidefinite relaxation (SDR) [27] is applied to obtain the optimal solution of (9). Define \({\mathbf {W}_{1}} = {\mathbf {w}_{1}}\mathbf {w}_{1}^{H},{\mathbf {W}_{2}} = {\mathbf {w}_{2}}\mathbf {w}_{2}^{H},{\mathbf {H}_{1}} = {\mathbf {h}_{1}}\mathbf {h}_{1}^{H},{\mathbf {H}_{2}} = {\mathbf {h}_{2}}\mathbf {h}_{2}^{H}\), and ignore the constraint of rank(W_{1})=rank(W_{2})=1, we can obtain:
The problem (12) we proposed is a fractional programming obviously. The Dinkelbach method is widely adopted in solving the fractional programming. Thus, we use this method to transform and solve the problem. With a continuous auxiliary variable t [28], we reformulate the objective function in (12a) as:
where \(\left \{\! \left. \Psi \right \frac {tr({\mathbf {H}_{1}}{\mathbf {W}_{2}})}{tr({\mathbf {H}_{1}}{\mathbf {W}_{1}}{\mathrm {)}} + {\sigma ^{2}}}\! \ge \!{\gamma _{M}},\frac {{tr({\mathbf {H}_{2}}{\mathbf {W}_{2}})}}{{tr({\mathbf {H}_{2}}{\mathbf {W}_{1}}) + {\sigma ^{2}}}} \ge {\gamma _{M}},tr({\mathbf {W}_{1}}) + tr({\mathbf {W}_{2}}) \le P {\vphantom {\frac {tr({\mathbf {H}_{1}}{\mathbf {W}_{2}})}{tr({\mathbf {H}_{1}}{\mathbf {W}_{1}}{\mathrm {)}} + {\sigma ^{2}}}}}\right \}\) is the feasible set in (12). We have the following lemma, whose proof can refer to [29]:
Lemma 1
F(t) is a strictly decreasing and continuous function, and it has a unique zero solution, which is denoted as t^{∗}. Then, the optimal value of the objective function in (12a) is t^{∗}.
From Lemma 1, we can see that the optimal solution can be obtained by solving (12) if we know t^{∗} in advance. Though t^{∗} is unknown at first, Lemma 1 tells us that the Dinkelbach method by round search [28] can be used to find the root of F(t) efficiently. In each iteration, the nonnegative variable t will update its value and finally reach the optimal denoting as t^{∗}.
Therefore, in the following, we will optimize (12) for a given t at first. For convenience, we rewrite (13) into the following form:
Meanwhile, problem (14) is a convex semidefinite problem (SDP) and can be efficiently solved by convex optimization solvers, e.g., CVX [30].
Proposition 1
The optimal solution to problem (14) satisfies rank(W1∗)=rank(W2∗)=1.
Proof
Obviously, the problem (14) is a separate SDP with three generalized constraints. According to [31], the optimal solution (W1∗,W2∗) to (14) always satisfies rank^{2}(W1∗)+rank^{2}(W2∗)≤3. Here, we consider the nontrivial case where W1∗≠0 and W2∗≠0, then rank(W1∗)=rank(W2∗)=1 can be obtained. Proposition 1 is proved, and this implies that the SDR here is tight. □
Let α_{1},α_{2}, and α_{3} denote the dual variables associated with constraints (14b), (14c), and (14d), respectively. The dual problem of (14) is expanded as:
where
and d_{min} is the value of problem (15).
Proposition 2
The optimal dual solution α3∗ to problem (15) satisfies α3∗>0.
Proof
We show that the optimal solution α3∗>0 by contradiction. Assume that \(\alpha _{3}^{*} = 0\), since the Lagrangian dual variables are all nonnegative. The matrix B^{∗} is negative semidefinite, i.e., α1∗H_{1}+α2∗H_{2}≤0. Moreover, (14) is convex and satisfies Slater’s condition, and the duality gap between (14) and (15) is zero. Thus F(t)=(1−t)σ^{2}. According to Lemma 1, the optimal solution to problem (14) is the same with the problem (12) when F(t^{∗})=0. So t^{∗}=1, and the maximal rate of R_{1} is log2t^{∗}=0. It is not reasonable, and then α3∗>0 must be true. Proposition 2 is proved. □
With the optimal dual solution \(\left (\alpha _{1}^ *,\alpha _{2}^ *,\alpha _{3}^ * \right)\) and optimal value dmin∗ obtained by solving problem (15), we can derive A^{∗} and B^{∗}, respectively, by substituting \(\left (\alpha _{1}^ *,\alpha _{2}^ *,\alpha _{3}^ * \right)\) into (16) and (17). Moreover, the complementary slackness condition of (15b) yields to A^{∗}W1∗=0,B^{∗}W2∗=0. Since rank(W1∗)=1 and rank(W2∗)=1, we have rank(A^{∗})=N−1 and rank(B^{∗})=N−1. Let u_{1} and u_{2} be the basis of the null space of A^{∗} and B^{∗}, respectively, and define \({\hat {\mathbf {W}}_{1}} = {\mathbf {u}_{1}}\mathbf {u}_{1}^{H},{\hat {\mathbf {W}}_{2}} = {\mathbf {u}_{2}}\mathbf {u}_{2}^{H}\), then we have:
where τ_{1} and τ_{2} are the power allocation coefficients for transmitting information s_{1} and s_{2}, respectively. And
Thus, the optimal precoding vectors are w1∗=τ_{1}u_{1},w2∗=τ_{2}u_{2} with given t.
Note that 2N complex variables are to be optimized for problem (12), while only three real variables for problem (15). So problem (15) has the lower computational complexity than problem (12). Actually, the complexity reduction is significant as the number of antennas at WD increases. Detailed steps of proposed optimal algorithm are summarized as Algorithm 1.
The SVDbased suboptimal solution
As described in the previous section, we can derive the optimal solution to problem (11) by using fractional programming and solving dual problem. But the round search for finding optimal t^{∗} reduces the feasibility of the optimal solution to a certain extent in practice. In this section, we propose a suboptimal solution based on SVD to further reduce the computational complexity.
When N≥2, the SVDbased precoding scheme can be used to eliminate the interference caused by s_{1} at the WD by restricting precoding vector w_{1} to satisfy \(\mathbf {h}_{2}^{H}{\mathbf {w}_{1}} = 0\) [32], which simplifies the precoding vector design. With SVDbased precoding vector, user 2 cannot decode the information s_{1}. It implies that the precoding vector w_{1} must lie in the null space of h_{2}. Let the SVD of h_{2} be expressed as \(\mathbf {h}_{2}^{H} = {\mathbf {u\Lambda }}{\mathbf {v}^{H}}{\mathrm { = }}{\mathbf {u\Lambda }}{[{\mathbf {v}_{1}}{\mathrm { }}{\mathbf {v}_{0}}]^{H}}\), where \(\mathbf {u} \in {\mathbb {C}^{1 \times 1}},\mathbf {v} \in {\mathbb {C}^{N \times N}}\) are orthogonal left and right singular vectors of h_{2}, respectively, and \(\mathbf {\Lambda } \in {\mathbb {C}^{1 \times N}}\) contains one positive singular value of h_{2}. \({\mathbf {v}_{0}} \in {\mathbb {C}^{N \times (N  1)}}\), which satisfies \(\mathbf {v}_{0}^{H}{\mathbf {v}_{0}} = {\mathbf {I}_{N  1}}\), is the last N−1 columns of v and forms an orthogonal basis for the null space of \(\mathbf {h}_{2}^{H}\). The SVDbased precoding vector w_{1} can be expressed as \({\mathbf {w}_{1}} = {\mathbf {v}_{0}}{\tilde {\mathbf {w}}_{1}}\), where \({\tilde {\mathbf {w}}_{1}} \) denotes the new vector to be designed, and the corresponding precoding vector of s_{2} to be designed is \({\tilde {\mathbf {w}}_{2}}\). It is obvious to observe that in order for the SVDbased solution to be feasible, we must have N≥2. Problem (11) is consequently formulated as:
Define \({\tilde {\mathbf {w}}_{1}} = {\tilde {\mathbf {w}}_{1}}\tilde {\mathbf {w}}_{1}^{H}, {\tilde {\mathbf {w}}_{2}} = {\tilde {\mathbf {w}}_{2}}\tilde {\mathbf {w}}_{2}^{H}\), \({\tilde {\mathbf {H}}_{1}} = \mathbf {v}_{0}^{H}{\mathbf {h}_{1}}\mathbf {h}_{1}^{H}{\mathbf {v}_{0}}\), we have the SVDbased SDP:
Obviously, the achieved optimal solution also satisfies the rankone constraint. Let \({\tilde {\alpha }_{1}}, {\tilde {\alpha }_{2}}\), and \({\tilde {\alpha }_{3}}\) denote dual variables, and its dual problem is given by:
where
Different from problem (12), problem (20) is convex and the dual gap between (21) and (22) is also zero. In the same way as the previous section, we can also solve the problem (20) by its Lagrangian dual problem (22) for complexity reduction. With the SVDbased solution \((\tilde {\alpha }_{1}^ *,\tilde {\alpha }_{2}^ *,\tilde {\alpha }_{3}^ *)\) achieved by problem (22), we can derive \({\tilde {\mathbf {A}}^ * }\) and \({\tilde {\mathbf {B}}^ * }\) according to (23) and (24). Let \({\tilde {\mathbf {u}}_{1}}\) and \({\tilde {\mathbf {u}}_{2}}\) be the basis of the null space of \({\tilde {\mathbf {A}}^ * }\) and \({\tilde {\mathbf {B}}^ * }\), respectively, and define \({\hat {\mathbf {W}}_{1}}{\mathrm { = }}{\tilde {\mathbf {u}}_{1}}\tilde {\mathbf {u}}_{1}^{H},{\hat {\mathbf {W}}_{2}} = {\tilde {\mathbf {u}}_{2}}\tilde {\mathbf {u}}_{2}^{H}\). Similar to (18), we have the precoding vectors based on SVD as \(\tilde {\mathbf {w}}_{1}^ * {\mathrm { = }}{\tilde \tau _{1}}{\tilde {\mathbf {u}}_{1}}, \tilde {\mathbf {w}}_{2}^ * {\mathrm { = }}{\tilde \tau _{2}}{\tilde {\mathbf {u}}_{2}}\), where
And the detailed steps of proposed SVDbased suboptimal algorithm are presented as Algorithm 2. Compared with Algorithm 1, the proposed SVDbased suboptimal scheme in Algorithm 2 further reduces the computational complexity without the Dinkelbach method.
Simulation results and discussions
In this section, we numerically evaluate the performance of the proposed optimal and suboptimal schemes. The simulation parameters are listed in Table 1. It is assumed that the channels from WD to two ordinary users are deep fading channels and the information signal attenuations are 55 dB and 60 dB, respectively, corresponding to an identical distance of 15 m and 20 m. We set the parameter in the Dinkelbach method ε=10^{−4} and the power of noise σ^{2}=−50 dBm. The channels h_{1} and h_{2} are assumed to be quasistatic flat Rayleigh fading, and each element of them follows an independent complex Gaussian distribution CN(0,1). All the simulation results are averaged over 1000 channel realizations. The optimal scheme and SVDbased scheme in this section, respectively, mean the optimal precoding vector scheme and the SVDbased precoding vector scheme of WD.
Figure 2 shows the maximal rate performance of different schemes versus the transmit power P under the number of antennas at WD N = 4 and R_{M} = 4 bps/Hz. Without loss of generality, time division multiple access (TDMA) is used to a representative of orthogonal multiple access (OMA), in which WD only serves single user in one time slot with joint power allocation among two time slots for two users. It can be observed that the proposed optimal and suboptimal scheme outperform traditional TDMA in terms of maximal rate of R_{1}, and this performance advantage is more obvious in the high transmit power region, though in TDMA scheme the s_{1} is not interrupted by s_{2}. And it is noted that the proposed SVDbased suboptimal scheme only has a slight performance loss compared to the optimal scheme.
Figure 3 compares the maximal rate of R_{1} versus the number of antennas at WD for different schemes under the transmit power of WD P = 25 dBm, when the target rate of s_{2}, i.e., R_{M}, is 4 bps/Hz and 6 bps/Hz, respectively. It is obviously noted that the maximal rate of R_{1} is enhanced as the number of antennas grows. However, the growth trend gradually becomes slow. Besides, the gap between proposed optimal scheme and SVDbased schemes in terms of the maximal rate of R_{1} is reducing with the increasing R_{M}.
In Fig. 4, the maximal rate region of R_{1} versus R_{M} is characterized for different schemes with the number of antennas at WD N = 4, when the transmit power of WD is 25 dBm and 20 dBm, respectively. It is also noted that the optimal scheme achieves better rate regions than the SVDbased scheme. Furthermore, the higher target rate of s_{2} requires, the smaller gap between the optimal and suboptimal scheme is. Meanwhile, the impact of transmission power at WD on the achieved rate regions for different schemes is also shown in Fig. 4. We can find that under certain transmit power of WD, when R_{M} is large, the rate of R_{1} may be zero. The reason is that all power should be allocated to precoding vector w_{2} to first satisfy the target rate demand of s_{2}. So the precoding vector w_{1} has little effect on the system performance no matter it is designed in optimal solution or suboptimal solution. And the rate of R_{1} becomes zero almost at the same value of target rate R_{M} for the optimal and SVDbased schemes.
Finally, Fig. 5 presents the outage performance of R_{1} when R_{M} varies from 2 to 11 bps/Hz with the transmit power of WD P = 25 dBm and the number of antennas at WD N = 4. Given the transmit power of WD P, the number of antennas N, and the target rate R_{M}, the outage probability is \({p_{\text {out}}}\left ({P,\,{R_{M}},\,N} \right) \buildrel \Delta \over = {\mathbb P}\left ({{R_{1}} = 0} \right)\). Especially, we set P = 25 dBm and N = 4. It is observed that the proposed optimal and suboptimal solutions achieve a similar performance to each other, and our proposed two schemes significantly decrease the outage probability of rate of R_{1} compared with the TDMA scheme.
Conclusion
In this paper, an optimization problem of precoding vectors for twouser D2D communication underlaying system enabled by NOMA is investigated. Given the target rate R_{M} and the transmit power of WD P, the maximal rate of R_{1} and corresponding optimal precoding vectors have been obtained. Then, the suboptimal solution based on SVD is proposed for complexity reduction. Finally, simulation results are provided to show the proposed optimal and suboptimal precoding algorithms can outperform OMA scheme, such as TDMA.
Availability of data and materials
Data sharing is not applicable to this article as no datasets were generated or analyzed during the current study.
Abbreviations
 AF:

Amplifyandforward
 AGN:

Additive Gaussian noise
 BS:

Base station
 CSI:

Channel state information
 CVX:

Matlab software for disciplined convex programming
 D2D:

Devicetodevice
 IoT:

Internet of Things
 MIMO:

Multipleinput multipleoutput
 NOMA:

Nonorthogonal multiple access
 OMA:

Orthogonal multiple access
 QoS:

Quality of service
 SDP:

Semidefinite problem
 SDP:

Semidefinite relaxation
 SIC:

Successive interference cancellation
 SINR:

Signaltointerferenceplusnoiseratio
 SVD:

Singular value decomposition
 TDMA:

Time division multiple access
 WD:

Wireless device
References
 1
Z. Ding, et al., Application of nonorthogonal multiple access in LTE and 5G networks. IEEE Commun. Mag.55(2), 185–191 (2017).
 2
R Hu, Y Qian, An energy efficient and spectrum efficient wireless heterogeneous network framework for 5G systems. IEEE Commun. Mag.52(5), 94–101 (2014).
 3
Y Saito, Y Kishiyama, A Benjebbour, T Nakamura, ALK Higuchi, in Proc. IEEE 77th Vehicular Technology Conference (VTC Spring). Nonorthogonal multiple access (NOMA) for cellular future radio access. (IEEE, Dresden, 2013), 1–5.
 4
K Higuchi, A Benjebbour, Nonorthogonal multiple access (NOMA) with successive interference cancellation. IEICE Trans. Commum. E98B(3), 403–414 (2015).
 5
Z Ding, F Adachi, H Poor.The application of MIMO to nonorthogonal multiple access. IEEE Trans. Wirel. Commum. Lett. 15(1), 537–552 (2016).
 6
D Wan, M Wen, F Ji, Y Liu, Y Huang, Cooperative NOMA systems with partial channel state information over Nakagami m fading channels. IEEE Trans. Commun.66(3), 947–958 (2018).
 7
B Wang, L Dai, M Xiao, Millimeter Wave NOMA. Encyclopedia of Wireless Networks (Springer, Cham, 2018).
 8
A Benjebbour, Y Saito, Y Kishiyama, A Li, A Harada, T Nakamura, in Proc. 2013 International Symposium on Intelligent Signal Processing and Communication Systems (ISPACS). Concept and practical considerations of nonorthogonal multiple access (NOMA) for future radio access. (IEEE, Naha, 2013), 770–774.
 9
W Wu, B Wang, Robust secrecy beamforming for wireless information and power transfer in multiuser MISO communication system. EURASIP J Wireless Com Network. 2015(1), 161 (2015).
 10
B Kim, S Lim, H Kim, S Suh, J Kwun, S Choi, C Lee, S Lee, D Hong, in Proc. MILCOM 20132013 IEEE Military Communications Conference. Nonorthogonal multiple access in a downlink multiuser beamforming system (IEEESan Diego, 2013), pp. 1278–1283.
 11
H ElSawy, E Hossain, M Alouini, Analytical modeling of mode selection and power control for underlay D2D communication in cellular networks. IEEE Trans. Commun.62(11), 4147–4161 (2014).
 12
K Doppler, C Yu, C Ribeiro, P Janis, in Proc. 2010 IEEE Wireless Communication and Networking Conference. Mode selection for devicetodevice communication underlaying an LTEadvanced network. (IEEE, Sydney, 2010), 1–6.
 13
K Akkarajitsakul, P Phunchongharn, E Hossain, VK Bhargava, in Proc 2012 IEEE International Conference on Communication Systems (ICCS). Mode selection for energyefficient D2D communications in LTEadvanced networks: A coalitional game approach (IEEESingapore, 2012), pp. 488–492.
 14
D Zhu, J Wang, A Swindlehurst, C Zhao, Downlink resource reuse for devicetodevice communications underlaying cellular networks. IEEE Signal Process. Lett.21(5), 531–534 (2014).
 15
J Zhao, Y Liu, K Chai, Y Chen, M Elkashlan, J AlonsoZarate, in Proc. IEEE Global Commun. Conf. (GLOBECOM). NOMAbased D2D communications: towards 5G. (IEEE, Washington, 2016), 1–6.
 16
J Zhao, Y Liu, K Chai, Y Chen, M Elkashlan, Joint subchannel and power allocation for NOMA enhanced D2D communications. IEEE Trans. Commun.65(11), 5081–5094 (2017).
 17
Z Zhang, Z Ma, M Xiao, Z Ding, P Fan, Fullduplex devicetodeviceaided cooperative nonorthogonal multiple access. IEEE Trans. Veh. Technol.66(5), 4467–4471 (2017).
 18
C Li, H Yang, F Sun, J Cioffi, L Yang, Multiuser overhearing for cooperative twoway multiantenna relays. IEEE Trans. Veh. Technol.65(5), 3796–3802 (2016).
 19
C Li, S Zhang, P Liu, F Sun, J Cioffi, L Yang, Overhearing protocol design exploiting intercell interference in cooperative green networks. IEEE Trans. Veh. Technol.65(1), 441–446 (2016).
 20
Z Ding, X Lei, G Karagiannidis, R Schober, J Yuan, V Bhargava, A survey on nonorthogonal multiple access for 5G networks: research challenges and future trends. IEEE J. Select. Areas Commun.35(10), 2181–2195 (2017).
 21
Z Ding, P Fan, H Poor, Impact of user pairing on 5G nonorthogonal multiple access downlink transmissions. IEEE Trans. Veh. Technol.65(8), 6010–6023 (2016).
 22
C Li, P Liu, C Zou, F Sun, J Cioffi, L Yang, Spectralefficient cellular communications with coexistent one and twohop transmissions. IEEE Trans. Veh. Technol.65(8), 6010–6023 (2016).
 23
C Li, H Yang, F Sun, J Cioffi, L Yang, Adaptive overhearing in twoway multiantenna relay channels. IEEE Signal Process. Lett.23(1), 117–120 (2016).
 24
F Zhou, Y Wu, R Hu, Y Wang, KK Wong, Energyefficient NOMA heterogeneous cloud radio access networks. IEEE Netw.32(2), 152–160 (2017).
 25
W Wu, F Zhou, Pei Li, P Deng, B Wang, VCM Leung, in Proc IEEE Inter. Conf. Commun. (ICC). Energyefficient secure NOMAenabled mobile edge computing networks (IEEEShanghai, 2019).
 26
P Gover, A Sahai, in Proc. IEEE International Symposium on Information Theory. Shannon meets Tesla: wireless information and power transfer. (IEEE, Austin, 2010), 2363–2367.
 27
Z Luo, W Ma, AM So, Y Ye, S Zhang, Semidefinite relaxation of quadratic optimization problems. IEEE Signal Process. Mag.27(3), 20–34 (2010).
 28
H Zhang, Y Huang, S Li, L Yang, Energyefficient precoder design for MIMO wiretap channels. IEEE Commun. Lett.18(9), 1559–1562 (2014).
 29
J Xu, L Qiu, Energy efficiency optimization for MIMO broadcast channels. IEEE Trans. Wirel. Commun.12(2), 690–701 (2013).
 30
M Grant, S Boyd, CVX: Matlab software for disciplined convex programming, version 2.0 beta (2013). http://cvxr.com/cvx.
 31
Y Huang, D Palomar, Rankconstrained separable semidefinite programming with applications to optimal beamforming. IEEE Trans. Signal Proc.58(2), 664–678 (2010).
 32
C Li, F Sun, J Cioffi, L Yang, Energy efficient MIMO relay transmissions via joint power allocations. IEEE Trans. Circ. Syst.61(7), 531–535 (2014).
Acknowledgements
Not applicable.
Funding
This paper was supported in part by the National Natural Science Foundation of China under Grant 61271232, the Natural Science Foundation of Jiangsu Province of China under BK20180757, the Project of Educational Commission of Jiangsu Province of China under 18KJB510028, the Introducing Talent Research StartUp Fund of Nanjing University of Posts and Telecommunications under NY218100, and the Postgraduate Research & Practice Innovation Program of Jiangsu Province under Grant CXZZ13_0487.
Author information
Affiliations
Contributions
PD is the main author of the current paper. PD contributed to the development of the ideas, design of the study, theory, result analysis, and article writing. WW contributed to the development of the ideas, design of the study, theory, and article writing. PD and PL conceived and designed the experiments. XS and PL performed the experiments. BW undertook revision works of the paper. All authors read and approved the final manuscript.
Corresponding author
Correspondence to Ping Deng.
Ethics declarations
Competing interests
The authors declare that they have no competing interests.
Additional information
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
About this article
Received
Accepted
Published
DOI
Keywords
 Nonorthogonal multiple access (NOMA)
 Low latency
 Semidefinite relaxation
 Langrangian duality
 Singular value decomposition (SVD)