Skip to content

Advertisement

  • Research
  • Open Access

A full-duplex SWIPT system with self-energy recycling to minimize energy consumption

EURASIP Journal on Wireless Communications and Networking20182018:253

https://doi.org/10.1186/s13638-018-1249-4

  • Received: 11 April 2018
  • Accepted: 18 September 2018
  • Published:

Abstract

In simultaneous wireless information and power transfer system (SWIPT), additional resource allocation for power transfer degrades system performance. Therefore, how to allocate power and time to data transmission and energy transfer is very important for the performance of time switching-based SWTPT system. In this paper, we propose a time switching-based full-duplex SWIPT system combined with self-energy recycling since self-interference harvesting is more efficient than simultaneous data transmission if the magnitude of self-interference is sufficiently large. Proposed system has additional self-energy recycling phases compared to the conventional time switching-based full-duplex SWIPT system. Considering the limitation of maximum transmission power, proposed system is optimized transmission time as well as power. For the validity of the proposed system, we show the bi-convexity of proposed system and propose an iterative algorithm for optimization of proposed system. In simulations, we compare proposed system with the half-duplex system without simultaneous data transmission phase and the full-duplex system without self-energy recycling phase. Simulation results show proposed system consume less energy than the conventional systems.

Keywords

  • SWIPT
  • Full duplex
  • Device-to-device communication
  • Self-energy recycling

1 Introduction

In recent years, simultaneous wireless information and power transfer (SWIPT) schemes are attracting in terms of enabling to carry information and energy simultaneously [1]. SWIPT schemes can be classified into two categories according to receiver types: time switching (TS) SWIPT scheme and power splitting (PS) SWIPT scheme. TS SWIPT scheme divides the downlink (DL) time into energy harvesting (EH) phase and information detection (ID) phase, and PS SWIPT scheme splits the received signal into EH signal and ID signal. Since the amount of harvested energy is also important as well as the data rate in SWIPT system, the trade-off between data rate and power transfer has been studied [1, 2]. Furthermore, various optimization problems such as power minimization, energy efficiency maximization, and weighted sum-rate maximization were studied in various system models [35].

SWIPT system combined with full duplex (FD) has advantages including the increase of spectral efficiency. In contrast to the conventional FD system where self-interference (SI) is harmful, SI can be beneficial in terms of energy source for harvesting in FD SWIPT system [610]. A number of studies using this advantage were performed for FD relay system where relay has no external power supply except for transmitted power from base station (BS) [7, 8, 10]. Also, the self-energy recycling (S-ER) scheme was proposed in point-to-point (P2P) FD system where PS-based EH is used at user equipment (UE) [9]. In general, the performance of PS SWIPT scheme is better than that of TS SWIPT scheme, because TS SWIPT scheme can be treated as a special case of PS SWIPT scheme with binary split power ratios [4, 11]. However, in contrast to TS receiver where a device requires only a simple switcher, PS receiver needs an RF signal splitter and its structure is relatively complicate [4, 12]. This means that TS-based SWIPT scheme is more applicable to UE than PS-based SWIPT scheme since the structure of UE needs to be simple.

TS-based SWIPT systems with S-ER were studied in [1315]. In [13], Zeng and Zhang proposed the beamformer design to maximize end-to-end throughput in multiple-input single-output (MISO) relay system. In the first phase, the source transmits information to relay, and the received signal at relay is amplified and forwarded to the destination during the second phase. At the same time, the source transfers the energy to relay, and relay harvests SI as well as the signal from the source node in the second phase. Ding et al. proposed the beamforming scheme of multiple-input multiple-output (MIMO) relay for rate maximization where relay harvests SI in the second phase [14]. In [15], Hwang et al. proposed the beamformers of BS and relay where relay has additional antenna to harvest SI.

The FD SWIPT system with S-ER is also helpful to device-to-device (D2D) communication for Internet of Thing (IoT) system—the communication between mobile phone and devices which cannot have sufficient battery. In the communication between smart phone and IoT devices such as bluetooth devices and smart watch, IoT devices are not connected with BS. Also, the short lifetime of IoT devices due to small size battery is a critical problem. In this environment, the device equipped with a large battery, such as a smartphone, can transfer energy-to-energy-deficient devices to prolong communication time.

Thus, we propose a TS-based FD SWIPT system for D2D communication to prolong the lifetime by minimizing energy consumption under the target rate constraint. In contrast to relay systems, D2D communication systems need to be considered as S-ER of both users since both users are not connected to power grid. Thus, proposed system has additional S-ER phases contrary to the conventional TS-based FD SWIPT system to employ SI efficiently. The entire time in the conventional TS-based FD SWIPT system consists of two phases if one user does not have enough energy to transmit data. In the first phase, one UE, i.e., UE1, transmits energy to another UE, i.e., UE2, and UE2 transmits data to UE1, and data transmissions in both directions are operated in the second phase. In the proposed system, however, each phase is divided into two parts for S-ER implementation. In these S-ER phases, UE2 does not transmit any signal and UE1 harvests SI. To minimize energy consumption, we should find time duration and power corresponding to each phase. Since the optimization problem for proposed system has a form of bi-convex, we propose an iterative scheme based on line search method to find a solution. In contrast to the alternative convex search (ACS) technique which can obtain a local minimum, the proposed scheme can obtain a better solution close to global optimum.

2 System model and optimization problem formulation

2.1 System model

As depicted in Fig. 1, we consider a D2D FD SWIPT system with the assumption that UE2 does not have energy for data transmission but UE1 has sufficient energy. In this scenario, UE1 should transfer energy to UE2 for data communication with UE2. However, since UE1 also does not connected to power grid, the total energy consumption of UE1 should be minimized. For this purpose, each user can harvest SI since harvesting SI can be more efficient than simultaneous data transmission if the magnitude of SI is sufficiently large. The proposed system model needs additional phases to harvest SI compared to the conventional TS-based FD SWIPT systems, and the operating protocol consists of four phases according to the mode of each user—energy transfer or data transmission mode of UE1 and only receive or data transmission mode of UE2. Also, it is assumed that all channels have quasi-static flat-fading characteristics, and the length of one channel block is normalized. Then, the relationship between time duration of each phase can be written as:
$$ {{\tau_{1}} + {\tau_{2}} + {\tau_{3}} + {\tau_{4}} = 1}, $$
(1)
Fig. 1
Fig. 1

The system model of the proposed system

and the received data signals at UE1 and UE2 can be, respectively, expressed as:
$$\begin{array}{*{20}l} {y_{1,2}} &= \sqrt {{P_{2,2}}} {h_{12}}{s_{2,2}} + \sqrt {{P_{1,2}}} {g_{1}}{s_{1,2}} + {n_{1}}, \end{array} $$
(2)
$$\begin{array}{*{20}l} {y_{2,3}} &= \sqrt {{P_{1,3}}} {h_{21}}{s_{1,3}} + {n_{2}}, \end{array} $$
(3)
$$\begin{array}{*{20}l} {y_{1,4}} &= \sqrt {{P_{2,4}}} {h_{12}}{s_{2,4}} + \sqrt {{P_{1,4}}} {g_{1}}{s_{1,4}} + {n_{1}}, \end{array} $$
(4)
$$\begin{array}{*{20}l} {y_{2,4}} &= \sqrt {{P_{1,4}}} {h_{21}}{s_{1,4}} + \sqrt {{P_{2,4}}} {g_{2}}{s_{2,4}} + {n_{2}}, \end{array} $$
(5)
where yk,i and Pk,i mean the received signal and the transmission power at UE k corresponding to τi, respectively; hij denotes the channel from UE j to UE i with hij(0,σ2); and gk denotes SI channel at UE k with a static amplitude and random phase characteristics, i.e., \({\left | {{g_{k}}} \right |^{2}} = \sigma _{SI}^{2}\) [7]. sk,i means the transmit symbol of UE k corresponding to τi, and nk is the additive white Gaussian noise at UE k with nk(0,N0). In energy transmission phases of UE1, i.e., τ1 and τ2, UE2 harvests the energy transmitted from UE1. At the same time, since SI can be used additional energy source during τ2, UE2 harvests SI without SI cancelation. During the phases of not receiving data from UE2, i.e., τ1 and τ3, UE1 also harvests SI without SI cancelation. Then, the harvested energies at UE1 and UE2 can be, respectively, written as:
$$\begin{array}{*{20}l} E_{1,1}^{\text{har}} &= {\tau_{1}}\eta {P_{1,1}}{\left| {{g_{1}}} \right|^{2}}, \end{array} $$
(6)
$$\begin{array}{*{20}l} E_{2,1}^{\text{har}} &= {\tau_{1}}\eta {P_{1,1}}{\left| {{h_{21}}} \right|^{2}}, \end{array} $$
(7)
$$\begin{array}{*{20}l} E_{2,2}^{\text{har}} &= {\tau_{2}}\eta \left({{P_{1,2}}{{\left| {{h_{21}}} \right|}^{2}} + {P_{2,2}}{{\left| {{g_{2}}} \right|}^{2}}} \right), \end{array} $$
(8)
$$\begin{array}{*{20}l} E_{1,3}^{\text{har}} &= \tau_{3} \eta {{P_{1,3}}{{{\left| {{g_{1}}} \right|}^{2}}}}, \end{array} $$
(9)

where \(E_{k,i}^{\text {har}}\) is the harvested energy at UE k during τi, and η means power conversion efficiency.

2.2 Optimization problem for the proposed system

The system energy consumption is the same with the total energy consumption of UE1. Thus, the system energy consumption, ET, can be written as:
$$ {} {E_{T}} \,=\, {\tau_{1}}\left({1 - \eta \sigma_{\text{SI}}^{2}} \right){P_{1,1}} + {\tau_{2}}{P_{1,2}} + {\tau_{3}}\left({1 - \eta \sigma_{\text{SI}}^{2}} \right){P_{1,3}} + {\tau_{4}}{P_{1,4}}. $$
(10)
Since total energy consumption of UE2 should not be larger than the sum of the energies harvested from UE1 and SI, the relationship between the powers of UE1 and UE2 can be expressed as:
$$ {{} \begin{aligned} \eta {\left| {{h_{21}}} \right|^{2}}\left({{\tau_{1}}{P_{1,1}} + {\tau_{2}}{P_{1,2}}} \right) \ge {\tau_{2}}\left({1 - \eta \sigma_{\text{SI}}^{2}} \right){P_{2,2}} + {\mkern 1mu} {\tau_{4}}{P_{2,4}}. \end{aligned}} $$
(11)
Assuming that SI is imperfectly canceled in data receiving phase [16], the achievable rate of each user can be, respectively, written as:
$$ {\begin{aligned} {R_{1}} &= {\tau_{2}}{\log_{2}}\left({1 + \frac{{\left| {{h_{12}}} \right|}}{{{N_{0}} + \sigma_{\text{RSI}}^{2}}}{P_{2,2}}} \right) \\&\qquad+ {\tau_{4}}{\log_{2}}\left({1 + \frac{{\left| {{h_{12}}} \right|}}{{{N_{0}} + \sigma_{\text{RSI}}^{2}}}{P_{2,4}}} \right), \end{aligned}} $$
(12)
$$ {\begin{aligned} {R_{2}} &= {\tau_{3}}{\log_{2}}\left({1 + \frac{{\left| {{h_{21}}} \right|}}{{{N_{0}}}}{P_{1,3}}} \right) \\&\qquad+ {\tau_{4}}{\log_{2}}\left({1 + \frac{{\left| {{h_{21}}} \right|}}{{{N_{0}} + \sigma_{\text{RSI}}^{2}}}{P_{1,4}}} \right), \end{aligned}} $$
(13)
where \(\sigma _{\text {RSI}}^{2}\) represents the variance of residual self-interference. The residual self-interference model can be divided into two cases: the complicated case and the optimistic case [1722]. In the complicated case, the power of residual self-interference is increased with the transmission power [17, 19]. In the optimistic case, the power of residual self-interference has constant value regardless of the transmission power [18, 2022]. We assume that the power of residual self-interference is constant as the optimistic case. Instead, we set the power of residual self-interference to upper bound of the power of residual self-interference, i.e., \(\sigma _{\text {RSI}}^{2} = \sigma _{\text {SIC}}^{2}\sigma _{SI}^{2}P_{\text {max}}\) where Pmax is the maximum transmission power of UE. Since the data transmission of UE2 does not exist during τ3, there is no residual self-interference during τ3. Then, the optimization problem to minimize the system energy consumption can be formulated as follows:
$$\begin{array}{*{20}l} &\mathop{\min}\limits_{{\textbf{P}_{1}},{\boldsymbol{\tau}}} {E_{T}} \end{array} $$
(14a)
$$\begin{array}{*{20}l} &{s.t.}\qquad {\tau_{1}} + {\tau_{2}} + {\tau_{3}} + {\tau_{4}} = 1, \end{array} $$
(14b)
$$\begin{array}{*{20}l} &R_{k} \ge {C_{k}}, \end{array} $$
(14c)
$$\begin{array}{*{20}l} &\eta {\left| {{h_{D}}} \right|^{2}}\left({\left({1 - \sigma_{\text{SI}}^{2}} \right){\tau_{1}}{P_{1,1}} + {\tau_{2}}{P_{1,2}}} \right) \qquad\qquad \\ &\qquad\qquad\qquad\ge {\tau_{2}}\left({1 - \eta \sigma_{\text{SI}}^{2}} \right){P_{2,2}} + {\tau_{4}}{P_{2,4}}, \end{array} $$
(14d)
$$\begin{array}{*{20}l} &{\boldsymbol{\tau}} \ge {\bf{0}}, \end{array} $$
(14e)
$$\begin{array}{*{20}l} &0 \le {P_{k,i}} \le {P_{\max }}, \end{array} $$
(14f)
where Ck means a target rate of UE k. For convenience, we set P1=[P1,1,P1,2,P1,3,P1,4]T, P2=[0,P2,2,0,P2,4]T, and τ=[τ1,τ2,τ3,τ4]T. The optimization problem is a form of linear programming to τ when the powers of both users are fixed, and the objective function and all constraint functions are convex about P when τ is fixed. Therefore, the optimization problem is a bi-convex problem [23]. However, ACS technique which can be used to solve a bi-convex optimization problem can obtain a local minimum. Thus, we propose an iterative scheme to derive the value close to the global optimum of the proposed system in the next section, and the algorithm of the iterative scheme is given in Table 1.
Table 1

Algorithm of the proposed iterative scheme for the optimization of the proposed system

1) set initial point

If \({P_{max}} - \frac {{\sigma _{SI}^{2}\left ({{N_{0}} + \sigma _{RSI}^{2}} \right)}}{2} \ge 0\)

Let \(\textbf {P}_{1}^{\left (1 \right)} = {\left [ {\begin {array}{*{20}{c}} {0,}&{{P_{\max }},}&0&{\left ({{2^{\frac {{{C_{2}}}}{{{1-\tau _{min}}}}}} - 1} \right)\frac {{{N_{0}+\sigma _{RSI}^{2}}}}{{{{\left | {{h_{21}}} \right |}^{2}}}}} \end {array}} \right ]^{T}}\)

\(\textbf {P}_{2}^{\left (1 \right)} \,=\, {\left [ {\begin {array}{*{20}{c}}{0,}&{\frac {{\eta {{\left | {{h_{21}}} \right |}^{2}}}}{{1 - \eta \sigma _{SI}^{2}}}{\alpha _{2}^{\left (1 \right)}} \,-\, \frac {{{N_{0} + \sigma _{RSI}^{2}}}}{{{{\left | {{h_{12}}} \right |}^{2}}}},}&{0,}{\eta {{\left | {{h_{21}}} \right |}^{2}}{\alpha _{2}^{\left (1 \right)}} \,-\, \frac {{{N_{0} + \sigma _{RSI}^{2}}}}{{{{\left | {{h_{12}}} \right |}^{2}}}}}\end {array}} \right ]^{T}}\)

\({{\boldsymbol {\tau }}^{\left (1 \right)}} = {\left [ {\begin {array}{*{20}{c}}{0,}&{{\tau _{\min }},}&{0,}&{1 - {\tau _{\min }}}\end {array}} \right ]^{T}}\)

\({\alpha _{1}^{\left (1 \right)}} = \frac {{{2^{\frac {{{C_{2}}}}{{{\tau _{3}} + {\tau _{4}}}}}}{{\left \{ {\left ({1 - \eta \sigma _{SI}^{2}} \right){N_{0}}} \right \}}^{\frac {{{\tau _{3}}}}{{{\tau _{3}} + {\tau _{4}}}}}}{{\left ({{N_{0}} + \sigma _{RSI}^{2}} \right)}^{\frac {{{\tau _{4}}}}{{{\tau _{3}} + {\tau _{4}}}}}}}}{{{{\left | {{h_{21}}} \right |}^{2}}}}\)

\(\alpha _{2}^{\left (1 \right)} = \left ({{P_{\max }} - \frac {{\sigma _{SI}^{2}\left ({{N_{0}} + \sigma _{RSI}^{2}} \right)}}{{{{\left | {{h_{21}}} \right |}^{2}}{{\left | {{h_{12}}} \right |}^{2}}}}} \right){\tau _{\min }} + \frac {{{ {{N_{0}} + \sigma _{RSI}^{2}} }}}{{\eta {{\left | {{h_{21}}} \right |}^{2}}{{\left | {{h_{12}}} \right |}^{2}}}}\)

\({\tau _{\min }} = \frac {{AW\left ({\left ({\log 2} \right){e^{y\log 2}}} \right) - BC\log 2}}{{AC\log 2}}\)

else

Let \(\textbf {P}_{1}^{\left (1 \right)} = {\left [ {\begin {array}{*{20}{c}}{0,} & {{P_{\max }},} & {0,} & {\left ({{2^{\frac {{{C_{2}}}}{{{1-\tau _{min}}}}}} - 1} \right)\frac {{{N_{0} + \sigma _{RSI}^{2}}}}{{{{\left | {{h_{21}}} \right |}^{2}}}}}\end {array}} \right ]^{T}}\)

\(\textbf {P}_{2}^{\left (1 \right)} = {\left [ {\begin {array}{*{20}{c}}{0,}&{\frac {{\eta {{\left | {{h_{21}}} \right |}^{2}}}}{{1 - \eta \sigma _{SI}^{2}}}{P_{\max }},}&{0,}&0\end {array}} \right ]^{T}}\)

\({{\boldsymbol {\tau }}^{\left (1 \right)}} = {\left [ {\begin {array}{*{20}{c}}{0,} & {{\tau _{\min }},} &{0,} & {1 - {\tau _{\min }}}\end {array}} \right ]^{T}}\)

\({\alpha _{1}^{\left (1 \right)}} = \frac {{{2^{\frac {{{C_{2}}}}{{{\tau _{3}} + {\tau _{4}}}}}}{{\left \{ {\left ({1 - \eta \sigma _{SI}^{2}} \right){N_{0}}} \right \}}^{\frac {{{\tau _{3}}}}{{{\tau _{3}} + {\tau _{4}}}}}}{{\left ({{N_{0}} + \sigma _{RSI}^{2}} \right)}^{\frac {{{\tau _{4}}}}{{{\tau _{3}} + {\tau _{4}}}}}}}}{{{{\left | {{h_{21}}} \right |}^{2}}}}\)

\(\alpha _{2}^{\left (1 \right)} = {P_{\max }} + \frac {{\left ({N_{0} + \sigma _{RSI}^{2}} \right)\left ({1 - \eta \sigma _{SI}^{2}} \right)}}{{\eta {{\left | {{h_{21}}} \right |}^{2}}{{\left | {{h_{12}}} \right |}^{2}}}} \)

\({\tau _{\min }} = \frac {{{C_{1}}}}{{{{\log }_{2}}\left ({1 + \frac {{{{\left | {{h_{12}}} \right |}^{2}}}}{{{N_{0} + \sigma _{RSI}^{2}}}}{P_{2,2}}} \right)}}\)

2) for k=1:kmax

i) Find ET(τ(k)), pk, and βk

ii) Let τ(k+1)=τ(k)+βkpk

iii) If |τ(k+1)τ(k)|<ε

Algorithm terminated. The power and time are set to \(\textbf {P}_{1}^{\left ({k} \right)}\), \(\textbf {P}_{2}^{\left ({k} \right)}\),

and τ.

else

Find \(\textbf {P}_{1}^{\left ({k + 1} \right)}\) and \(\textbf {P}_{2}^{\left ({k + 1} \right)}\)

3 Power and time optimization to minimize energy consumption of the proposed system

3.1 Optimal power allocation with fixed τ

Since P1,1 and P1,2 only affect R1, and P1,3 and P1,4 only affect R2, the optimization problem can be divided into two parts: UE1 target rate part and UE2 target rate part. Then, the optimization problem for UE1 target rate can be written as:
$$\begin{array}{*{20}l} {} &\mathop {\min }\limits_{{\textbf{P}_{1}},{\textbf{P}_{2}}} {\tau_{1}}\left({1 - \eta \sigma_{\text{SI}}^{2}} \right){P_{1,1}} + {\tau_{2}}{P_{1,2}} \end{array} $$
(15a)
$$\begin{array}{*{20}l} &{} {s.t.}\qquad {\tau_{1}}{P_{1,1}} \,+\, {\tau_{2}}{P_{1,2}} \!\ge\! {\tau_{2}}\frac{{\left({1 - \eta \sigma_{\text{SI}}^{2}} \right)}}{{\eta {{\left| {{h_{21}}} \right|}^{2}}}}{P_{2,2}} \,+\, {\tau_{4}}\frac{1}{{\eta {{\left| {{h_{21}}} \right|}^{2}}}}{P_{2,4}}, \end{array} $$
(15b)
$$\begin{array}{*{20}l} &{} \quad{\tau_{2}}{\log_{2}}\left({1 \,+\, \frac{{{P_{2,2}}{{\left| {{h_{12}}} \right|}^{2}}}}{{{N_{0} \,+\, \sigma_{\text{RSI}}^{2}}}}} \right) \,+\, {\tau_{4}}{\log_{2}}\left({1 + \frac{{{P_{2,4}}{{\left| {{h_{12}}} \right|}^{2}}}}{{{N_{0} \,+\, \sigma_{\text{RSI}}^{2}}}}} \right) \!\ge\! {C_{1}}. \end{array} $$
(15c)
Since UE1 only needs to transmit energy as much as the required energy of UE2 to satisfy C1, we can divide this optimization problem into two stages. First, the minimum energy consumption of UE2 is derived to satisfy C1 under the assumption that the energy harvested from UE1 is more than the minimum energy consumption of UE2. Next, the minimum transfer energy of UE1 is obtained based on the minimum energy consumption of UE2. Then, the optimization problem of the first stage can be expressed as:
$$\begin{array}{*{20}l} &\mathop {\min }\limits_{{\textbf{P}_{2}}} {\tau_{2}}\frac{{\left({1 - \eta \sigma_{\text{SI}}^{2}} \right)}}{{\eta {{\left| {{h_{21}}} \right|}^{2}}}}{P_{2,2}} + {\tau_{4}}\frac{1}{{\eta {{\left| {{h_{21}}} \right|}^{2}}}}{P_{2,4}} \end{array} $$
(16a)
$$\begin{array}{*{20}l} &{} {s.t.}\qquad{\tau_{2}}{\log_{2}}\left({1 + \frac{{{P_{2,2}}{{\left| {{h_{12}}} \right|}^{2}}}}{{{N_{0} + \sigma_{\text{RSI}}^{2}}}}} \right) \\&\qquad+ {\tau_{4}}{\log_{2}}\left({1 + \frac{{{P_{2,4}}{{\left| {{h_{12}}} \right|}^{2}}}}{{{N_{0} + \sigma_{\text{RSI}}^{2}}}}} \right) \ge {C_{1}}. \end{array} $$
(16b)
If the objective function and all constraints are convex and the optimization problem has one possible set to satisfy the KKT conditions, then the feasible set is a global minimizer [24]. According to the KKT conditions, we can obtain following equations:
$$\begin{array}{*{20}l} {} {\tau_{2}}\frac{{\left({1 - \eta \sigma_{\text{SI}}^{2}} \right)}}{{\eta {{\left| {{h_{21}}} \right|}^{2}}}} &\,-\, {\mu_{1}}\left({{\tau_{2}}\frac{{{{\left| {{h_{12}}} \right|}^{2}}}}{{\left({{N_{0}} + \sigma_{\text{RSI}}^{2} + {{\left| {{h_{12}}} \right|}^{2}}{P_{2,2}}} \right)\log 2}}} \right) \,=\, 0, \end{array} $$
(17a)
$$\begin{array}{*{20}l} {}{\tau_{4}}\frac{1}{{\eta {{\left| {{h_{21}}} \right|}^{2}}}} &\,-\, {\mu_{1}}\left({{\tau_{4}}\frac{{{{\left| {{h_{12}}} \right|}^{2}}}}{{\left({{N_{0}} + \sigma_{\text{RSI}}^{2} \,+\, {{\left| {{h_{12}}} \right|}^{2}}{P_{2,4}}} \right)\log 2}}} \right) = 0, \end{array} $$
(17b)
$$\begin{array}{*{20}l} &{\mu_{1}}\left({C_{1}} - {\tau_{2}}{{\log }_{2}}\left({1 + \frac{{{P_{2,2}}{{\left| {{h_{12}}} \right|}^{2}}}}{{{N_{0} + \sigma_{\text{RSI}}^{2}}}}} \right)\right. \\ & \left.- {\tau_{4}}{{\log }_{2}}\left({1 + \frac{{{P_{2,4}}{{\left| {{h_{12}}} \right|}^{2}}}}{{{N_{0} + \sigma_{\text{RSI}}^{2}}}}} \right) \right) = 0, \end{array} $$
(17c)
$$\begin{array}{*{20}l} & {\mu_{1}} \ge 0 \end{array} $$
(17d)
$$\begin{array}{*{20}l} &{}{{C_{1}} \,-\, {\tau_{2}}{{\log }_{2}}\left({1\! +\! \frac{{{P_{2,2}}{{\left| {{h_{12}}} \right|}^{2}}}}{{{N_{0} \,+\, \sigma_{\text{RSI}}^{2}}}}} \right) \!\,-\, {\tau_{4}}{{\log }_{2}}\left(\! {1 \,+\, \frac{{{P_{2,4}}{{\left| {{h_{12}}} \right|}^{2}}}}{{{N_{0} \,+\, \sigma_{\text{RSI}}^{2}}}}} \!\right)} \!\le\! 0, \end{array} $$
(17e)
where μ1≠0 since (17a) and (17b) are not satisfied when μ1=0. Then, P2,2 and P2,4 can be expressed in terms of μ1:
$$\begin{array}{*{20}l} {P^{*}_{2,2}} &= \frac{{\eta {{\left| {{h_{21}}} \right|}^{2}}}}{{1 - \eta \sigma_{\text{SI}}^{2}}}{\alpha_{2}} - \frac{{{N_{0} + \sigma_{\text{RSI}}^{2}}}}{{{{\left| {{h_{12}}} \right|}^{2}}}}, \end{array} $$
(18)
$$\begin{array}{*{20}l} {P^{*}_{2,4}} &= \eta {\left| {{h_{21}}} \right|^{2}}{\alpha_{2}} - \frac{{{N_{0} + \sigma_{\text{RSI}}^{2}}}}{{{{\left| {{h_{12}}} \right|}^{2}}}}, \end{array} $$
(19)
where \({\alpha _{2}} = \frac {\mu _{1}^{*} }{{\log 2}}\) and \(\mu _{1}^{*}\) can be obtained by (17c), (18), and (19) as follow:
$$ \mu_{1}^{*} = {2^{\frac{{{C_{1}}}}{{{\tau_{2}} + {\tau_{4}}}}}}{\left({1 - \eta \sigma_{\text{SI}}^{2}} \right)^{\frac{{{\tau_{2}}}}{{{\tau_{2}} + {\tau_{4}}}}}}\left({\frac{{\left({N_{0} + \sigma_{\text{RSI}}^{2}} \right) \log 2}}{{\eta {{\left| {{h_{21}}} \right|}^{2}}{{\left| {{h_{12}}} \right|}^{2}}}}} \right). $$
(20)
Since the minimum energy consumption of UE2 is obtained under the condition of (15c), (15a) can be solved by only considering (15b). To minimize the energy consumption of UE1, it is efficient that required energy of UE2 is harvested during τ1 since τ1 is more efficient than τ2 due to S-ER. However, since the maximum transmission power is limited, all required energy of UE2 may not be harvested during τ1. In this case, additional energy harvesting should be done during τ2. Thus, the transmission powers of UE1 during τ1 and τ2 can be, respectively, expressed as:
$$ {P}_{1,1}^{*} = \min \left({\frac{{{{\tau}_{2}}\left({1 - \eta {\sigma}_{\text{SI}}^{2}} \right){P}_{2,2}^{*} + {{\tau}_{4}}{P}_{2,4}^{*}}}{{{{\tau}_{1}}\eta {\left| {{{h}_{21}}} \right|{~}^{2}}}},{{P}_{\max}}} \right), $$
(21)
$$ {P}_{1,2}^{*} = \frac{1}{\tau}_{2}\left({{\tau}_{2}}\frac{\left({1 - \eta {\sigma}_{\text{SI}}^{2}} \right)}{{\eta {\left| {h}_{21} \right|{~}^{2}}}}{P}_{2,2}^{*} + {{\tau}_{4}}\frac{1}{{\eta {\left| {{h_{21}}} \right|{~}^{2}}}}{P}_{2,4}^{*} - {\tau}_{1}{P}_{1,1}^{*} \right). $$
(22)
Then, the optimization problem for UE2 target rate can be written as:
$$\begin{array}{*{20}l} &\mathop {\min }\limits_{{\textbf{P}_{1}}} {\tau_{3}}\left({1 - \eta \sigma_{\text{SI}}^{2}} \right){P_{1,3}} + {\tau_{4}}{P_{1,4}} \end{array} $$
(23a)
$$\begin{array}{*{20}l} &{s.t.}\quad {\tau_{3}}{{\log }_{2}}\left({1 \,+\, \frac{{{P_{1,3}}{{\left| {{h_{21}}} \right|}^{2}}}}{{{N_{0}}}}} \right)\\&\quad \,+\, {\tau_{4}}{{\log }_{2}}\left({1 \,+\, \frac{{{P_{1,4}}{{\left| {{h_{21}}} \right|}^{2}}}}{{{N_{0} + \sigma_{\text{RSI}}^{2}}}}} \right) \ge {C_{2}}, \end{array} $$
(23b)
using the KKT conditions, P1,3 and P1,4 can be expressed as:
$$\begin{array}{*{20}l} {P^{*}_{1,3}} &= \frac{1}{{1 - \eta \sigma_{\text{SI}}^{2}}}{\alpha_{1}} - \frac{{{N_{0}}}}{{{{\left| {{h_{21}}} \right|}^{2}}}}, \end{array} $$
(24)
$$\begin{array}{*{20}l} {P^{*}_{1,4}} &= {\alpha_{1}} - \frac{{{N_{0} + \sigma_{\text{RSI}}^{2}}}}{{{{\left| {{h_{21}}} \right|}^{2}}}}, \end{array} $$
(25)
where \({\alpha _{1}} = \frac {{{2^{\frac {{{C_{2}}}}{{{\tau _{3}} + {\tau _{4}}}}}}{{\left \{ {\left ({1 - \eta \sigma _{\text {SI}}^{2}} \right){N_{0}}} \right \}}^{\frac {{{\tau _{3}}}}{{{\tau _{3}} + {\tau _{4}}}}}}{{\left ({{N_{0}} + \sigma _{\text {RSI}}^{2}} \right)}^{\frac {{{\tau _{4}}}}{{{\tau _{3}} + {\tau _{4}}}}}}}}{{{{\left | {{h_{21}}} \right |}^{2}}}}\). Therefore, if τ has fixed value, the optimal power allocation to minimize energy consumption can be written as:
$$\begin{array}{*{20}l} \textbf{P}_{1}^{*} &= \left[ {\begin{array}{c} {\min \left({\frac{{{\tau_{2}}\left({1 - \eta \sigma_{\text{SI}}^{2}} \right)P_{2,2}^{*} + {\tau_{4}}P_{2,4}^{*}}}{{{\tau_{1}}\eta {{\left| {{h_{21}}} \right|}^{2}}}},{P_{\max }}} \right)}\\ {\frac{1}{{{\tau_{2}}}}\left({{\tau_{2}}\frac{{\left({1 - \eta \sigma_{\text{SI}}^{2}} \right)}}{{\eta {{\left| {{h_{21}}} \right|}^{2}}}}P_{2,2}^{*} + {\tau_{4}}\frac{1}{{\eta {{\left| {{h_{21}}} \right|}^{2}}}}P_{2,4}^{*} - {\tau_{1}}P_{1,1}^{*}} \right)}\\ {\frac{1}{{1 - \eta \sigma_{\text{SI}}^{2}}}{\alpha_{1}} - \frac{{{N_{0}}}}{{{{\left| {{h_{21}}} \right|}^{2}}}}}\\ {{\alpha_{1}} - \frac{{{N_{0}} + \sigma_{\text{RSI}}^{2}}}{{{{\left| {{h_{21}}} \right|}^{2}}}}} \end{array}} \right], \end{array} $$
(26)
$$ {} \textbf{P}_{2}^{*} \,=\,\! {\left[\!\! {0,{\mkern 1mu} \frac{{\eta {{\left| {{h_{21}}} \right|}^{2}}}}{{1 - \eta \sigma_{\text{SI}}^{2}}}{\alpha_{2}} \,-\,\! \frac{{{N_{0}} \,+\, \sigma_{\text{RSI}}^{2}}}{{{{\left| {{h_{12}}} \right|}^{2}}}},{\mkern 1mu}0,{\mkern 1mu} \eta {{\left| {{h_{21}}} \right|}^{2}}{\alpha_{2}} \,-\, \frac{{{N_{0}} \,+\, \sigma_{\text{RSI}}^{2}}}{{{{\left| {{h_{12}}} \right|}^{2}}}}}\!\! \right]^{T}}. $$
(27)

3.2 Deriving minimum energy transmission time and initial point

Since maximum transmission power is limited, τmin should be guaranteed to satisfy C2. For finding τmin, we assume that τ1 and τ3 are 0 to secure maximum data transmission time, and P1,2 takes Pmax since Pmax is available maximum transmission power of UE1. Then, (11) can be modified with the assumption that τ1=τ3=0 and P1,1=P1,3=0 as:
$$ \eta {\left| {{h_{21}}} \right|^{2}}{\tau_{2}}{P_{\text{max}}} = {\tau_{2}}\left({1 - \eta \sigma_{\text{SI}}^{2}} \right){P_{2,2}} + (1 - {\tau_{2}}){P_{2,4}}, $$
(28)
and α2 can be written in terms of τ2 by using (27):
$$ \alpha_{2}^{\left(1 \right)} \,=\, \left({{P_{\max }} - \frac{{\sigma_{\text{SI}}^{2}\left({{N_{0}} + \sigma_{\text{RSI}}^{2}} \right)}}{{{{\left| {{h_{21}}} \right|}^{2}}{{\left| {{h_{12}}} \right|}^{2}}}}} \right){\tau_{\min }} + \frac{{{ {{N_{0}} + \sigma_{\text{RSI}}^{2}} }}}{{\eta {{\left| {{h_{21}}} \right|}^{2}}{{\left| {{h_{12}}} \right|}^{2}}}}. $$
(29)

The initial point can be derived as two cases according to the value of A where \(A = {{P_{\max }} - \frac {{{\sigma _{\text {SI}}^{2}}\left ({{N_{0}} + \sigma _{\text {RSI}}^{2}} \right)}}{{{{\left | {{h_{21}}} \right |}^{2}}{{\left | {{h_{12}}} \right |}^{2}}}}}\), because A<0 means that P2,4 obtained by (27) is negative value.

3.2.1 Case I: A≥0

We can find τmin by using (17d), (27), and (29):
$$ {\tau_{\min }} = \frac{{AW\left({\left({\log 2} \right){e^{y\log 2}}} \right) - BC\log 2}}{{AC\log 2}}, $$
(30)
where \(B = \frac {{{N_{0} + \sigma _{\text {RSI}}^{2}}}}{{\eta {{\left | {{h_{21}}} \right |}^{2}}{{\left | {{h_{12}}} \right |}^{2}}}}\), \( C = {\log _{2}}\left ({\frac {1}{{1 - \eta {\sigma _{\text {SI}}^{2}}}}} \right)\), \(D = {\log _{2}}\left ({\frac {{\eta {{\left | {{h_{21}}} \right |}^{2}}{{\left | {{h_{12}}} \right |}^{2}}}}{{{N_{0} + \sigma _{\text {RSI}}^{2}}}}} \right) - {C_{1}}\), \(y = {\log _{2}}\frac {C}{A} + \frac {{BC}}{A} - D\), and W(∙) means Lambert W function. Then, we can derive the initial point based on τmin as follows:
$$\begin{array}{*{20}l} {}{{\boldsymbol{\tau }}^{\left(1 \right)}} &= {\left[ {\begin{array}{*{20}{c}} {0,}&{{\tau_{\min }},}&{0,}&{1 - {\tau_{\min }}} \end{array}} \right]^{T}}, \end{array} $$
(31)
$$\begin{array}{*{20}l} {}\textbf{P}_{1}^{\left(1 \right)} &= {\left[ {\begin{array}{*{20}{c}} {0,}&{{P_{\max }},}&0&{\left({{2^{\frac{{{C_{2}}}}{{{1-\tau_{\text{min}}}}}}} - 1} \right)\frac{{{N_{0}+\sigma_{\text{RSI}}^{2}}}}{{{{\left| {{h_{21}}} \right|}^{2}}}}} \end{array}} \right]^{T}}, \end{array} $$
(32)
$$ {{}\begin{aligned} \textbf{P}_{2}^{\left(1 \right)} &\!\,=\, \!{\left[ {\begin{array}{*{20}{c}}{\!\!0,\!\!}&{\frac{{\eta {{\left| {{h_{21}}} \right|}^{2}}}}{{1 - \eta \sigma_{\text{SI}}^{2}}}{\alpha_{2}^{\left(1 \right)}} \!\,-\, \frac{{{N_{0} + \sigma_{\text{RSI}}^{2}}}}{{{{\left| {{h_{12}}} \right|}^{2}}}},} {0,}{\eta {{\left| {{h_{21}}} \right|}^{2}}{\alpha_{2}^{\left(1 \right)}} \!\,-\, \frac{{{N_{0} + \sigma_{\text{RSI}}^{2}}}}{{{{\left| {{h_{12}}} \right|}^{2}}}}}\end{array}} \!\!\right]^{T}}\!, \end{aligned}} $$
(33)
$$\begin{array}{*{20}l} {}{\alpha_{1}^{\left(1 \right)}} &= \frac{{{2^{\frac{{{C_{2}}}}{{{\tau_{3}} + {\tau_{4}}}}}}{{\left\{ {\left({1 - \eta \sigma_{\text{SI}}^{2}} \right){N_{0}}} \right\}}^{\frac{{{\tau_{3}}}}{{{\tau_{3}} + {\tau_{4}}}}}}{{\left({{N_{0}} + \sigma_{\text{RSI}}^{2}} \right)}^{\frac{{{\tau_{4}}}}{{{\tau_{3}} + {\tau_{4}}}}}}}}{{{{\left| {{h_{21}}} \right|}^{2}}}}, \end{array} $$
(34)
$$\begin{array}{*{20}l}{} \alpha_{2}^{\left(1 \right)} &\,=\, \left(\!{{P_{\max }} \,-\, \frac{{\sigma_{\text{SI}}^{2}\left({{N_{0}} + \sigma_{\text{RSI}}^{2}} \right)}}{{{{\left| {{h_{21}}} \right|}^{2}}{{\left| {{h_{12}}} \right|}^{2}}}}} \!\right)\!{\tau_{\min }} + \frac{{{ {{N_{0}} + \sigma_{\text{RSI}}^{2}} }}}{{\eta {{\left| {{h_{21}}} \right|}^{2}}{{\left| {{h_{12}}} \right|}^{2}}}}. \end{array} $$
(35)

3.2.2 Case II: A<0

A<0 means the optimal value of P2,4 from (27) is negative, but all power values should not be smaller than 0. Thus, we take P2,4=0, and all harvested energy is used in τ2. Then, P2,2 can be obtained by (28):
$$ {P_{2,2}} = \frac{{\eta {{\left| {{h_{21}}} \right|}^{2}}}}{{1 - \eta \sigma_{\text{SI}}^{2}}}{P_{\max }}, $$
(36)
and τmin can be also derived based on (17d),
$$ {\tau_{\min }} = \frac{{{C_{1}}}}{{{{\log }_{2}}\left({1 + \frac{{{{\left| {{h_{12}}} \right|}^{2}}}}{{{N_{0} + \sigma_{\text{RSI}}^{2}}}}{P_{2,2}}} \right)}}. $$
(37)
Then, the initial point of this case can be expressed as:
$$\begin{array}{*{20}l} {{\boldsymbol{\tau }}^{\left(1 \right)}} &= {\left[ {\begin{array}{*{20}{c}} {0,}&{{\tau_{\min }},}&{0,}&{1 - {\tau_{\min }}} \end{array}} \right]^{T}}, \end{array} $$
(38)
$$\begin{array}{*{20}l} \textbf{P}_{1}^{\left(1 \right)} &= {\left[ {\begin{array}{*{20}{c}}{0,} & {{P_{\max }},} & {0,} & {\left({{2^{\frac{{{C_{2}}}}{{{1-\tau_{\text{min}}}}}}} - 1} \right)\frac{{{N_{0} + \sigma_{\text{RSI}}^{2}}}}{{{{\left| {{h_{21}}} \right|}^{2}}}}}\end{array}} \right]^{T}}, \end{array} $$
(39)
$$\begin{array}{*{20}l} \textbf{P}_{2}^{\left(1 \right)} &= {\left[ {\begin{array}{*{20}{c}} {0,}&{\frac{{\eta {{\left| {{h_{21}}} \right|}^{2}}}}{{1 - \eta \sigma_{\text{SI}}^{2}}}{P_{\max }},}&{0,}&0 \end{array}} \right]^{T}}, \end{array} $$
(40)
$$\begin{array}{*{20}l} {\alpha_{1}^{\left(1 \right)}} &= \frac{{{2^{\frac{{{C_{2}}}}{{{\tau_{3}} + {\tau_{4}}}}}}{{\left\{ {\left({1 - \eta \sigma_{\text{SI}}^{2}} \right){N_{0}}} \right\}}^{\frac{{{\tau_{3}}}}{{{\tau_{3}} + {\tau_{4}}}}}}{{\left({{N_{0}} + \sigma_{\text{RSI}}^{2}} \right)}^{\frac{{{\tau_{4}}}}{{{\tau_{3}} + {\tau_{4}}}}}}}}{{{{\left| {{h_{21}}} \right|}^{2}}}}, \end{array} $$
(41)
$$\begin{array}{*{20}l} \alpha_{2}^{\left(1 \right)} &= {P_{\max }} + \frac{{\left({N_{0} + \sigma_{\text{RSI}}^{2}} \right)\left({1 - \eta \sigma_{\text{SI}}^{2}} \right)}}{{\eta {{\left| {{h_{21}}} \right|}^{2}}{{\left| {{h_{12}}} \right|}^{2}}}}. \end{array} $$
(42)

3.3 Iterative scheme based on line search method

As mentioned before, since ACS technique obtains a local minimum, we propose an iterative scheme to derive a value close to global optimum. In proposed scheme, τ is updated iteratively based on the line search method using gradient of ET about τ [24]. Then, the kth τ can be expressed as:
$$ {{\boldsymbol{\tau }}^{\left({k + 1} \right)}} = {{\boldsymbol{\tau }}^{\left(k \right)}} + {\beta_{k}}{\textbf{p}_{k}}, $$
(43)
where τ(k) means updated τ at the kth iteration, pk denotes a search direction vector, and βk is the magnitude of Δτ(k)=τ(k+1)τ(k). To minimize ET, τ should be varied from the direction of the maximum partial derivative to that of the minimum partial derivative, because the maximum partial derivative means that the increase of ET is minimum when the duration of the corresponding τ is decreasing. Thus, pk can be set as follows:
$$ {\textbf{p}_{k}} = \left\{ {\begin{array}{*{20}{l}} 1&{{\mkern 1mu} {\kern 1pt} {i^{*}} = \mathop {\min }\limits_{i} \frac{{\partial {E_{T}}}}{{\partial {\tau_{i}}}}}\\ { - 1}&{{\kern 1pt} {j^{*}} = \mathop {\max }\limits_{j} \frac{{\partial {E_{T}}}}{{\partial {\tau_{j}}}}}\\ 0&\text{otherwise} \end{array}} \right.. $$
(44)
Without loss of generality, P1,1 can be fixed to Pmax before deriving the gradient of ET. Since data transmission is not performed during τ1 and the relationship between power and time is linear, minimizing τ1 is efficient in the aspect that it maximizes the available data transmission time. Thus, the gradient of ET over τ1 can be obtained by the condition that the required energy to satisfy C1 should not be varied:
$$ \frac{{\partial {E_{T}}}}{{\partial {\tau_{1}}}} = \left({1 - \eta \sigma_{\text{SI}}^{2}} \right){P_{\max }} + {\tau_{2}}\frac{{\partial {P_{1,2}}}}{{\partial {\tau_{1}}}}, $$
(45)
where the partial derivatives of P1,3 and P1,4 can be ignored since these do not affect to C1. Then, the partial derivative of P1,2 can be written using (11):
$$ \frac{{\partial {P_{1,2}}}}{{\partial {\tau_{1}}}} = - \frac{{{P_{\max }}}}{{{\tau_{2}}}}. $$
(46)
Since the magnitude of P1,1 is fixed to Pmax, P1,2 can be represented by the function of the power of UE2 using (11). Then, the gradient of ET about τ can be expressed using (10) and (11):
$$ {{} \begin{aligned} \nabla {E_{T}} &= \left[ - \eta \sigma_{\text{SI}}^{2}{P_{\max }},\,\,\,\frac{{1 - \eta \sigma_{\text{SI}}^{2}}}{{\eta {{\left| {{h_{21}}} \right|}^{2}}}}{P_{2,2}} + {\tau_{2}}\frac{{1 - \eta \sigma_{\text{SI}}^{2}}}{{\eta {{\left| {{h_{21}}} \right|}^{2}}}}\frac{{\partial {P_{2,2}}}}{{\partial {\tau_{2}}}}, \right.\\ &\left({1 - \eta \sigma_{\text{SI}}^{2}} \right){P_{1,3}} + {\tau_{3}}\left({1 - \eta \sigma_{\text{SI}}^{2}} \right)\frac{{\partial {P_{1,3}}}}{{\partial {\tau_{3}}}},\\ &\left. {\frac{1}{{\eta {{\left| {{h_{21}}} \right|}^{2}}}}{P_{2,4}} + {P_{1,4}} + {\tau_{4}}\left({\frac{1}{{\eta {{\left| {{h_{21}}} \right|}^{2}}}}\frac{{\partial {P_{2,4}}}}{{\partial {\tau_{4}}}} + \frac{{\partial {P_{1,4}}}}{{\partial {\tau_{4}}}}} \right)} \right]. \end{aligned}} $$
(47)
To obtain each partial derivative of (47), we assume that the rates of both UEs are not varied by τ:
$$ \frac{{\partial {R_{l}}}}{{\partial {\tau_{i}}}} = 0,\,\,\,\,l = 1,2. $$
(48)
This means that if τi is varied, the power corresponding to τi is also varied to satisfy target rate. By using (12), (13), (26), (27), and (48), ET can be expressed in terms of α1 and α2.
$$ {{}\begin{aligned} \nabla {E_{T}} &= \left[ { - \eta \sigma_{\text{SI}}^{2}{P_{\max }},} \right.\\ &{\alpha_{2}} - \frac{{\left({{N_{0}} + \sigma_{\text{RSI}}^{2}} \right)\left({1 - \eta \sigma_{\text{SI}}^{2}} \right)}}{{\eta {{\left| {{h_{21}}} \right|}^{2}}{{\left| {{h_{12}}} \right|}^{2}}}} - {\alpha_{2}}\log \left({\frac{{\eta {{\left| {{h_{21}}} \right|}^{2}}{{\left| {{h_{12}}} \right|}^{2}}}}{{\left({{N_{0}} + \sigma_{\text{RSI}}^{2}} \right)\left({1 - \eta \sigma_{\text{SI}}^{2}} \right)}}{\alpha_{2}}} \right),\\ &{\alpha_{1}} - \frac{{{N_{0}}\left({1 - \eta \sigma_{\text{SI}}^{2}} \right)}}{{{{\left| {{h_{21}}} \right|}^{2}}}} - {\alpha_{1}}\log \left({\frac{{{{\left| {{h_{21}}} \right|}^{2}}}}{{{N_{0}}\left({1 - \eta \sigma_{\text{SI}}^{2}} \right)}}{\alpha_{1}}} \right),\\ &{\alpha_{1}} - \frac{{{N_{0} + \sigma_{\text{RSI}}^{2}}}}{{{{\left| {{h_{21}}} \right|}^{2}}}} - {\alpha_{1}}\log \left({\frac{{{{\left| {{h_{21}}} \right|}^{2}}}}{{{N_{0} + \sigma_{\text{RSI}}^{2}}}}{\alpha_{1}}} \right) + \\ &\left.{{\alpha_{2}} - \frac{{{N_{0} + \sigma_{\text{RSI}}^{2}}}}{{\eta {{\left| {{h_{21}}} \right|}^{2}}{{\left| {{h_{12}}} \right|}^{2}}}} - {\alpha_{2}}\log \left({\frac{{\eta {{\left| {{h_{21}}} \right|}^{2}}{{\left| {{h_{12}}} \right|}^{2}}}}{{{N_{0} + \sigma_{\text{RSI}}^{2}}}}{\alpha_{2}}} \right)} \right] \end{aligned}} $$
(49)
Since the sum of the phases involved in energy transfer, i.e., τ1+τ2, should not be smaller than τmin, pk cannot be updated in the direction in which τ1+τ2 becomes smaller than τmin. Therefore, the direction finding procedure for updating pk can be expressed as Table 2.
Table 2

The direction finding procedure for updating pk

If

\(\tau _{1}^{\left (k \right)} + \tau _{2}^{\left (k \right)} = {\tau _{\min }},{j^{*}} = \mathop {\max }\limits _{j} \frac {{\partial {E_{T}}}}{{\partial {\tau _{j}}}},\,{\mathrm {where\, }}{j^{*}} \in \left \{ {1,2} \right \}\)

 

\({i^{*}} = \mathop {\min }\limits _{i} \frac {{\partial {E_{T}}}}{{\partial {\tau _{i}}}},{\text {where}}\,i \in \left \{ {1,2} \right \}\)

else

 
 

\({j^{*}} = \mathop {\max }\limits _{j} \frac {{\partial {E_{T}}}}{{\partial {\tau _{j}}}},\,{\mathrm {where\, }}j \in \left [ {1,4} \right ]\)

 

\({i^{*}} = \mathop {\min }\limits _{i} \frac {{\partial {E_{T}}}}{{\partial {\tau _{i}}}},{\mathrm {where\, }}i \in \left [ {1,4} \right ]\)

end

 
The proposed scheme updates τ considering the variation of P1 and P2 contrary to ACS. ET can be considered as the difference between the local optimums corresponding to adjacent τ values. Thus, the energy consumption of kth updated τ can be considered as the lowest local optimum among adjacent local optimums at the kth iteration. Also, the saturated point of the proposed iterative scheme can be explained by using the Hessian matrix of the energy consumption:
$$ H = \left[ {\begin{array}{*{20}{c}} {\frac{{{\partial^{2}}{E_{T}}}}{{\partial \tau_{1}^{2}}}}&0&0&0\\ 0&{\frac{{{\partial^{2}}{E_{T}}}}{{\partial \tau_{2}^{2}}}}&0&0\\ 0&0&{\frac{{{\partial^{2}}{E_{T}}}}{{\partial \tau_{3}^{2}}}}&0\\ 0&0&0&{\frac{{{\partial^{2}}{E_{T}}}}{{\partial \tau_{4}^{2}}}} \end{array}} \right], $$
(50)
and the second-order partial derivatives are given by:
$$ \frac{{{\partial^{2}}{E_{T}}}}{{\partial \tau_{1}^{2}}} = 0, $$
(51)
$$\begin{array}{*{20}l} \frac{{{\partial^{2}}{E_{T}}}}{{\partial \tau_{2}^{2}}} &= \frac{{{\alpha_{2}}}}{{{\tau_{2}}}}{\left\{ {\log \left({\frac{{\eta {{\left| {{h_{21}}} \right|}^{2}}{{\left| {{h_{12}}} \right|}^{2}}}}{{{{\left({{N_{0}} + \sigma_{\text{RSI}}^{2}} \right)}}\left({1 - \eta \sigma_{\text{SI}}^{2}} \right)}}{\alpha_{2}}} \right)} \right\}^{2}}, \end{array} $$
(52)
$$\begin{array}{*{20}l} \frac{{{\partial^{2}}{E_{T}}}}{{\partial \tau_{3}^{2}}} &= \frac{{{\alpha_{1}}}}{{{\tau_{3}}}}{\left\{ {\log \left({\frac{{{{\left| {{h_{21}}} \right|}^{2}}}}{{{N_{0}}\left({1 - \eta \sigma_{\text{SI}}^{2}} \right)}}{\alpha_{1}}} \right)} \right\}^{2}}, \end{array} $$
(53)
$$\begin{array}{*{20}l} \frac{{{\partial^{2}}{E_{T}}}}{{\partial \tau_{4}^{2}}} &= \frac{{{\alpha_{1}}}}{{{\tau_{4}}}}{\left\{ {\log \left({\frac{{{{\left| {{h_{21}}} \right|}^{2}}}}{{{\left({{N_{0}} + \sigma_{\text{RSI}}^{2}} \right)}}}{\alpha_{1}}} \right)} \right\}^{2}}\\&\quad+ \frac{{{\alpha_{2}}}}{{{\tau_{4}}}}{\left\{ {\log \left({\frac{{\eta {{\left| {{h_{21}}} \right|}^{2}}{{\left| {{h_{12}}} \right|}^{2}}}}{{{\left({{N_{0}} + \sigma_{\text{RSI}}^{2}} \right)}}}{\alpha_{2}}} \right)} \right\}^{2}}. \end{array} $$
(54)

Since α1>0, α2>0, and τi≥0, each second-partial derivative is not less than 0. This means that Hessian matrix about τ is positive semi-definite even if P1 and P2 are variables about τ. Therefore, the energy consumption according to local optimum sets which are obtained by the proposed iterative scheme is convex about τ, the solution of the proposed scheme can saturate to the value close to global optimum.

Next, βk should be derived for updating τ. βk decides the convergence gap ε and the number of iterations up to saturation. ET can be modified the approximated objective function as Taylor series.
$$\begin{array}{*{20}l} &{E_{T}}\left(\tau \right) \approx \\ &{E_{T}}\left({{\tau^{\left(k \right)}}} \right) + \nabla {E_{T}}\left({{\tau^{\left(k \right)}}} \right)\left({\tau - {\tau^{\left(k \right)}}} \right) \\&\quad+ \frac{1}{2}{\left({\tau - {\tau^{\left(k \right)}}} \right)^{T}}{\nabla^{2}}{E_{T}}\left({{\tau^{\left(k \right)}}} \right)\left({\tau - {\tau^{\left(k \right)}}} \right), \end{array} $$
(55)
where 2ET(τ(k)) is Hessian matrix of ET at τ(k) and the higher order terms are ignored. Then, ET(τ(k+1)) is:
$$ \begin{aligned} {E_{T}}\left({{\tau^{\left(k \right)}} + {\beta_{k}}{\textbf{p}_{k}}} \right) &\approx {E_{T}}\left({{\tau^{\left(k \right)}}} \right) + \nabla {E_{T}}\left({{\tau^{\left(k \right)}}} \right){\beta_{k}}{\textbf{p}_{k}} \\&\quad+ \frac{1}{2}{\beta_{k}}\textbf{p}_{k}^{T}{\nabla^{2}}{E_{T}}\left({{\tau^{\left(k \right)}}} \right){\beta_{k}}{\textbf{p}_{k}}. \end{aligned} $$
(56)
Since pk is the direction toward to decreasing the total energy consumption and ET(τ(k+1))≤ET(τ(k)),
$$ \begin{aligned} \nabla {E_{T,\min }}\left({{\tau^{\left(k \right)}}} \right){\beta_{k}} &- \nabla {E_{T,\max }}\left({{\tau^{\left(k \right)}}} \right){\beta_{k}} \\&+ \frac{1}{2}\beta_{k}^{2}\left({{\nabla^{2}}{E_{T,\min }} + {\nabla^{2}} {E_{T,\max }}} \right) \le 0, \end{aligned} $$
(57)
where the direction of maximum partial derivative is − 1, that of minimum partial derivative is 1, and others are 0. ET,min(τ(k)) and 2ET,min(τ(k)) mean partial and second partial derivatives at τ(k) corresponding to i, respectively, and ET,max(τ(k)) and 2ET,max(τ(k)) denote partial and second partial derivatives at τ(k) corresponding to j, respectively. Then the scaling factor βk can be expressed as:
$$ {\beta_{k}} \le \frac{{2\left({\nabla {E_{T,\max }}\left({{\tau^{\left(k \right)}}} \right) - \nabla {E_{T,\min }}\left({{\tau^{\left(k \right)}}} \right)} \right)}}{{{\nabla^{2}}{E_{T,\max }}\left({{\tau^{\left(k \right)}}} \right) + {\nabla^{2}}{E_{T,\min }}\left({{\tau^{\left(k \right)}}} \right)}}. $$
(58)
Since all components of τ cannot be smaller than 0, the upper bound of βk should be defined as:
$$ {} {\beta_{k}} \,=\, \min \left({\gamma \times {\tau_{{j^{*}}}},\frac{{2\left({\nabla {E_{T,\max }}\left({{\tau^{\left(k \right)}}} \right) - \nabla {E_{T,\min }}\left({{\tau^{\left(k \right)}}} \right)} \right)}}{{{\nabla^{2}}{E_{T,\max }}\left({{\tau^{\left(k \right)}}} \right) + {\nabla^{2}}{E_{T,\min }}\left({{\tau^{\left(k \right)}}} \right)}}} \right), $$
(59)

where γ×τj is an upper bound to prevent βk from becoming larger than τj. The proposed optimization scheme can obtain a value close to global optimum, since the Hessian matrix of ET about τ is positive semi-definite.

4 Simulation results

In this section, we discuss the pros and cons of the proposed system compared with the conventional SWIPT systems. We set the path loss exponent as 2, N0=−70 [dBm], Pmax=23 [dBm], and η=0.5 [22, 25, 26]. We set ε=10−2, γ=0.1, and kmax=1000 for the iterative scheme.

Figure 2 shows the convergence of the proposed scheme compared to ACS where \(\sigma _{\text {SIC}}^{2} = -100\,[\!\text {dB}]\). The values of the exhaustive search are derived by comparing all local minimums when τ is fixed. In exhaustive search, the difference of each fixed τ between adjacent local minimums is set 0.05. This means that the difference between the global optimum point and the point of the exhaustive search is smaller than 0.05, i.e., |τoptτexh|≤0.05, where τopt is the global optimum point of τ and τexh denotes the nearest point to τopt among local minimums of the exhaustive search. In ACS scheme, linear programming is used to derived local minimum when P1 and P2 are fixed, and the KKT conditions are used to find local minimum when τ is fixed. It is shown that the performance of the proposed iterative optimization scheme is almost the same as the performance of the exhaustive search. In contrast to the proposed scheme, the results of ACS cannot obtain the value similar to exhaustive search since ACS technique cannot assure global optimum. Also, we can see that the performance of ACS depends on the initial points. The performance of ACS with the proposed initial point is similar to that of the proposed scheme in low \(\sigma ^{2}_{\text {SI}}\) region because the efficiency of S-ER is trivial in low \(\sigma ^{2}_{\text {SI}}\) region. This means that the initial point obtained in Section 3.2 is close to the optimal value since simultaneous transmission is more efficient than S-ER. However, the optimal point is far from the initial point obtained by Section 3.2 when \(\sigma ^{2}_{\text {SI}}\) is large since the performance difference between the exhaustive search and ACS with the proposed initial point becomes large for larger \(\sigma ^{2}_{\text {SI}}\).
Fig. 2
Fig. 2

Convergence of the proposed scheme compared to ACS. C1=C2=1 [bps/Hz], D=10 [m]

Figure 3 shows the minimum energy consumption of various systems according to \(\sigma _{\text {SI}}^{2}\). Half-duplex (HD) SWIPT system can be considered as a system with τ4=0, P1,2=0, and \(\sigma _{\text {RSI}}^{2}=0\) in Fig 1. Also, FD SWIPT system can be treated as a system considering only τ2 and τ4 in Fig 1. It can be seen that the proposed system and FD system have almost the same energy consumption while HD system has the worst performance in low \(\sigma _{\text {SI}}^{2}\) region, because the effectivenesses of S-ER and residual self-interference are trivial. This means that most of the transmission time is used for simultaneous data transmission in the proposed system, and the time allocation of the proposed system is almost the same with FD system. Therefore, the energy consumption of FD system is similar to that of the proposed system in low \(\sigma _{\text {SI}}^{2}\) region. In the case of HD system, however, simultaneous data transmission time is not assigned. This causes an increase of transmit power to achieve target rates of both users; thus, the energy consumption of HD system has the largest value. However, the performance gap between FD system and HD system is decreased as \(\sigma _{\text {SI}}^{2}\) becomes larger. It is because the efficiency of S-ER increases as \(\sigma _{\text {SI}}^{2}\) becomes larger in HD system, and the performance loss due to residual self-interference occurs in FD system. In HD system, the reduction rate of the energy consumption is the largest since S-ER is operated in all transmission phase. On the other hand, the reduction rate of the energy consumption of FD system is the smallest because S-ER is only implemented during one phase. Also, the effect of residual self-interference cannot be ignored. Thus, the energy consumption of HD system is less than that of FD system if the magnitude of \(\sigma _{\text {SI}}^{2}\) is larger than − 5 [ dB] and − 6 [ dB] when \(\sigma _{\text {SIC}}^{2} = -100\,[\text {dB}]\) and \(\sigma _{\text {SIC}}^{2} = -70\,[\!\text {dB}]\), respectively. Similarly, the performance gap between the proposed system and FD system increases as the magnitude of \(\sigma _{\text {SI}}^{2}\) grows. In the proposed system, most of the transmission time is assigned to simultaneous data transmission phase when the magnitude of \(\sigma _{\text {SI}}^{2}\) is small since the efficiency of simultaneous data transmission is higher than that of S-ER. However, time allocation to S-ER phase is increased as the efficiency of S-ER becomes higher, which makes the energy consumption of the proposed system more efficient than that of FD system. In the case of \(\sigma _{\text {SIC}}^{2} = -70\,[\text {dB}]\), the performance of the proposed system is lower than that of HD system because the power of the residual self-interference is fixed to the upper bound level. Ideally, the time allocation to τ1 is increased, and most of the harvesting energy of UE2 is transferred during τ1 as \(\sigma _{\text {SI}}^{2}\) become larger. This means that the transmission power of UE1 during τ2 becomes smaller. However, since the power of the residual self-interference is not varied keeping upper bound value, the performance of the proposed system is lower than that of HD system in high \(\sigma _{\text {SI}}^{2}\) region if sufficient self-interference cancelation is not conducted.
Fig. 3
Fig. 3

Total energy consumptions of various systems according to \(\sigma _{\text {SI}}^{2}\). C1=C2=1 [bps/Hz], D=10 [m]

Next, we discuss the effect of Pmax, and the simulation result is shown in Fig. 4. It can be seen that the energy consumption of HD system is decreased as Pmax becomes larger because HD system has the phase where only energy transmission is operated. Since the length of this phase is reduced as Pmax becomes larger, available data transmission time can be increased, and it decreases the required energy for data transmission. Therefore, the energy consumptions of HD system decreases as Pmax becomes larger. This tendency is also shown in the proposed system with σSIC=−100 [dB]. However, the energy consumption of FD system where σSIC=−100 [dB] does not change according to Pmax because FD system does not have an energy transfer only phase. Also, the length of the transmission time corresponding to each phase is dependent to data transmission rather than energy transfer. Thus, decreasing energy transmission time by increasing Pmax does not affect the variation of the length of each phase in FD system. In the case of \(\sigma _{\text {SIC}}^{2} = -70\,[\text {dB}]\), both energy consumption of the proposed system and FD system are increased as Pmax becomes larger. Since the variance of the residual self-interference is affected by Pmax, the performance degradation occurs in both system if sufficient self-interference cancelation is not conducted.
Fig. 4
Fig. 4

Total energy consumptions of various systems according to Pmax. C1=C2=1 [bps/Hz], \(\sigma _{\text {SI}}^{2} = -4\,[\text {dB}]\), and D=10 [m]

Figure 5 shows the total energy consumption according to target rate where \(\sigma _{\text {SI}}^{2} = -2\,[\text {dB}]\). It can be seen that the increase of the energy consumption according to target rate is largest in HD system, because simultaneous data transmission phase does not exist in HD system. Thus, the energy consumption of HD system is less than that of FD system when target rate is 2.5 [bps/Hz], but the energy consumption of HD system is greater than that of FD system when target rate is larger than 4 [bps/Hz]. In low target rate region, S-ER phase is more efficient than simultaneous transmission phase since \(\sigma _{\text {SI}}^{2}\) is large. However, simultaneous data transmission is more efficient than S-ER because energy consumption increases exponentially as target rate become higher. Thus, the performance reversal between HD and FD systems occurs. On the other hand, the proposed system has the lowest energy consumption in all target rate region, because the proposed system can efficiently allocate the time using both S-ER phase and simultaneous data transmission phase.
Fig. 5
Fig. 5

Total energy consumptions of various systems according to target rate. \(\sigma _{\text {SI}}^{2} = -2\,[\text {dB}]\), D=5 [m]

5 Conclusions

In this paper, we propose the TS-based FD SWIPT system in D2D communication system to minimize total energy consumption where S-ER is implemented at both users. Harvesting SI is more efficient than simultaneous data transmission when the magnitude of SI is sufficiently large. Thus, S-ER operating phases are combined in the proposed system, and this enables efficient time and power allocation. To minimize total energy consumption, we optimize transmission time and power corresponding to each phase under target rate constraints. Also, in contrast to the previous power minimization problems, we limit maximum transmission power of users to reflect practical limitation. First, we show the optimization problem of the proposed system is bi-convex problem, and the optimal power allocation scheme is proposed when the transmission time of each duration is fixed. Though ACS can be applied to solve the conventional bi-convex problem, it can give only a local optimum solution. Thus, we proposed the iterative algorithm to derive a better solution based on the line search method. Numerical results showed the proposed system gives better performance compared with the conventional SWIPT systems. However, since the proposed system is limited to single D2D pair, it needs to expand to general multiple D2D pairs system by considering the different interference patterns from each D2D pair.

6 Methods/Experimental

The purpose of this study is to minimize the system consumption energy. The system consists of two FD users where imperfect self-interference cancelation is assumed. The channels between each user are assumed as Rayleigh fading channel, and the self-interference channel of each user is assumed as random variable with static amplitude and random phase. The simulation results of the conventional schemes and exhaustive search are derived by comparing all local minimum when τ is fixed, and each local minimum is derived by using optimal power allocation scheme which is introduced in Section 3.1. HD SWIPT system can be considered as the proposed system with τ4=0, P1,2=0, and \(\sigma _{\text {RSI}}^{2}=0\). Also, FD SWIPT system can be treated as a system set to τ1=0 and τ3=0 in the proposed system model. In the process of finding each local minimum, the difference of each fixed τ between adjacent local minimums is set to 0.05. This means that the difference between the global optimum point and the point of the exhaustive search is smaller than 0.05, i.e., |τoptτexh|≤0.05. Since the feasible set may not exist due to the limitation of maximum transmission power, we take the system parameters which make the feasibility of all schemes close to 1. Also, if any of the entire systems has no feasible set, the results of its instantaneous channel model are excluded from Monte Carlo simulations.

Abbreviations

ACS: 

Alternative convex search

DL: 

Downlink

D2D: 

Device-to-device

EH: 

Energy harvesting

FD: 

Full duplex

HD: 

Half duplex

ID: 

Information detection

IoT: 

Internet of things

MIMO: 

Multiple-input multiple-output

MISO: 

Multiple-input single-output

PS: 

Power splitting

P2P: 

Point-to-point

S-ER: 

Self-energy recycling

SWIPT: 

Simultaneous wireless information and power transfer

TS: 

Time switching

UE: 

User equipment

Declarations

Funding

This study was supported by the National Research Foundation of Korea (NRF-2015R1D1A1A01058644).

Availability of data and materials

Data sharing is not applicable to this article as no datasets were generated or analyzed during the current study.

Authors’ contributions

HN has contributed 70% of the work, and the rest is contributed by CL. Both authors read and approved the final manuscript.

Competing interests

The authors declare that they have no competing interests.

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

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.

Authors’ Affiliations

(1)
The Department of Electrical and Electronic Engineering, Yonsei University, C418, Seoul, 03722, South Korea
(2)
The Department of Electrical and Electronic Engineering, Yonsei University, C527, Seoul, 03722, South Korea

References

  1. L.R. Varshney, in IEEE International Symposium on Information Theory - 573 Proceedings. Transporting information and energy simultaneously (IEEEToronto, 2008), pp. 1612–1616.Google Scholar
  2. L. Liu, R. Zhang, K.C. Chua, Wireless information transfer with opportunistic energy harvesting. IEEE Trans. Wirel. Commun.12(1), 288–300 (2013).View ArticleGoogle Scholar
  3. D.W.K. Ng, E.S. Lo, R. Schober, Wireless information and power transfer: energy efficiency optimization in OFDMA systems. IEEE Trans. Wirel. Commun.12(12), 6352–6370 (2013).View ArticleGoogle Scholar
  4. Q. Shi, L. Liu, W. Xu, R. Zhang, Joint transmit beamforming and receive power splitting for MISO SWIPT systems. IEEE Trans. Wirel. Commun.13(6), 3269–3280 (2014).View ArticleGoogle Scholar
  5. X. Zhou, R. Zhang, C.K. Ho, Wireless information and power transfer in multiuser OFDM systems. IEEE Trans. Wirel. Commun.13(4), 2282–2294 (2014).View ArticleGoogle Scholar
  6. M. Maso, C.F. Liu, C.H. Lee, T.Q.S. Quek, L.S. Cardoso, Energy-recycling full-duplex radios for next-generation networks. IEEE J. Sel. Areas Commun.33(12), 2948–2962 (2015).View ArticleGoogle Scholar
  7. M. Mohammadi, H.A. Suraweera, G. Zheng, C. Zhong, I. Krikidis, in IEEE Workshop on Signal Processing Advances in Wireless Communications, SPAWC, 2015-Augus. Full-duplex MIMO relaying powered by wireless energy transfer (IEEEStockholm, 2015), pp. 296–300.Google Scholar
  8. Y. Zeng, R. Zhang, Full-duplex wireless-powered relay with self-energy recycling. IEEE Wirel. Commun. Lett.4(2), 201–204 (2015).View ArticleGoogle Scholar
  9. Z. Hu, C. Yuan, F. Zhu, F. Gao, Weighted sum transmit power minimization for full-duplex system with SWIPT and self-energy recycling. IEEE Access. 4:, 4874–4881 (2016).View ArticleGoogle Scholar
  10. D. Wang, R. Zhang, X. Cheng, L. Yang, Capacity-enhancing full-duplex relay networks based on power-splitting (PS-)SWIPT. IEEE Trans. Veh. Technol.66(6), 5446–5450 (2017).Google Scholar
  11. X. Zhou, R. Zhang, C.K. Ho, Wireless information and power transfer: architecture design and rate-energy tradeoff. IEEE Trans. Commun.61(11), 4754–4767 (2013).View ArticleGoogle Scholar
  12. Y. Wu, Y. Liu, Q. Xue, S. Li, C. Yu, Analytical design method of multiway dual-band planar power dividers with arbitrary power division. IEEE Trans. Microw. Theory Tech.58(12), 3832–3841 (2010).View ArticleGoogle Scholar
  13. Y. Zeng, R. Zhang, Full-duplex wireless-powered relay with self-energy recycling. IEEE Wirel. Commun. Lett.4(2), 201–204 (2015).View ArticleGoogle Scholar
  14. Z. Ding, S. Hu, Q. Ni, Beamforming optimisation in energy harvesting cooperative full-duplex networks with self-energy recycling protocol. IET Commun.10(7), 848–853 (2016).View ArticleGoogle Scholar
  15. D. Hwang, K.C. Hwang, D.I. Kim, T.J. Lee, Self-energy recycling for RF powered multi-antenna relay channels. IEEE Trans. Wirel. Commun.16(2), 812–824 (2017).View ArticleGoogle Scholar
  16. Y. Liao, K. Bian, L. Song, Z. Han, Full-duplex MAC protocol design and analysis. IEEE Commun. Lett.19(7), 1185–1188 (2015).View ArticleGoogle Scholar
  17. T. Riihonen, S. Werner, R. Wichman, Mitigation of loopback self-interference in full-duplex MIMO relays. IEEE Trans. Signal Process.59(12), 5983–5993 (2011).MathSciNetView ArticleGoogle Scholar
  18. M. Jain, J.I. Choi, T. Kim, D. Bharadia, S. Seth, K. Srinivasan, P. Levis, S. Katti, P. Sinha, in Proceedings of the 17th Annual International Conference on Mobile Computing and Networking. Practical, real-time, full duplex wireless, MobiCom ’11 (ACMNew York, 2011), pp. 301–312.Google Scholar
  19. D.W.K. Ng, E.S. Lo, R. Schober, Dynamic resource allocation in MIMO-OFDMA systems with full-duplex and hybrid relaying. IEEE Trans. Commun.60(5), 1291–1304 (2012).View ArticleGoogle Scholar
  20. D. Bharadia, E. McMilin, S. Katti, Full duplex radios. SIGCOMM Comput. Commun. Rev.43(4), 375–386 (2013).View ArticleGoogle Scholar
  21. L.J. Rodríguez, N.H. Tran, Performance of full-duplex AF relaying in the presence of residual self-interference. IEEE J. Sel. Areas Commun.32(9), 1752–1764 (2014).View ArticleGoogle Scholar
  22. Y. Chang, H. Chen, F. Zhao, Energy efficiency maximization of full-duplex two-way relay-assisted device-to-device communications underlaying cellular networks. EURASIP J. Wirel. Commun. Netw.2016(1), 222–224 (2016).View ArticleGoogle Scholar
  23. J. Gorski, F. Pfeuffer, K. Klamroth, Biconvex sets and optimization with biconvex functions: a survey and extensions. Math. Methods Oper. Res.66(3), 373–407 (2007).MathSciNetView ArticleGoogle Scholar
  24. E.K.P. Chong, S.H. Zak, An Introduction to Optimization, 3rd (Wiley-Interscience, New Jersey, 2008).View ArticleGoogle Scholar
  25. S. Ali, N. Rajatheva, M. Latva-aho, in 2014 European Conference on Networks and Communications (EuCNC). Full duplex device-to-device communication in cellular networks (IEEEBologna, 2014), pp. 1–5.Google Scholar
  26. Z. Zhou, M. Dong, K. Ota, J. Wu, T. Sato, Energy efficiency and spectral efficiency tradeoff in device-to-device (D2D) communications. IEEE Wirel. Commun. Lett.3(5), 485–488 (2014).View ArticleGoogle Scholar

Copyright

© The Author(s) 2018

Advertisement