You are on page 1of 14

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/255731327

Wireless Information and Power Transfer in
Multiuser OFDM Systems

Article in IEEE Transactions on Wireless Communications · August 2013
DOI: 10.1109/TWC.2014.030514.131479 · Source: arXiv

CITATIONS READS

139 125

3 authors, including:

Rui Zhang Chin Keong Ho
Zhengzhou University Institute for Infocomm Research
297 PUBLICATIONS 10,510 CITATIONS 132 PUBLICATIONS 4,828 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Polyaniline View project

Energy storage View project

All content following this page was uploaded by Chin Keong Ho on 12 June 2014.

The user has requested enhancement of the downloaded file.

where an energy-constrained a perpetual energy source for the wireless networks [1].sg). 1 Wireless Information and Power Transfer in Multiuser OFDM Systems Xun Zhou. each user applies switching (TS) and power splitting (PS). energy harvesting. where the users harvest energy and decode information using the same signals received from a fixed access able to decode information and harvest energy simultaneously point (AP). and Chin Keong Ho. where the receiver performs dynamic time (TDMA).2462v2 [cs. Atlanta. the authors studied with the Institute for Infocomm Research. we study the optimal design for SWIPT has drawn an upsurge of research interests [1]–[16]. For the one receiver harvests energy and another receiver decodes PS scheme. relay harvests energy from the received signal and uses that Moreover. Ho is with the Institute for Infocomm Research. subcarrier (SC) allocation and the PS ratio for [4] extended the work in [1] by considering imperfect channel each user. For information transmission. processing. Two practical schemes for SWIPT. With TS applied at the receiver. namely.edu.a-star. the relay spends a portion X. 2013. Singapore. [8]. IEEE Abstract—In this paper. time division multiple access studied in [7]. Member. Networks that involve wireless power National University of Singapore (e-mail: elezhang@nus. i. with randomly deployed power beacons that charge mobiles . I NTRODUCTION interference channels was considered in [9].sg). [4]–[6]. [13]. by an lar. subject to a minimum the other one to the information receiver. [10] using PS Recently. However. where “near” users transfer (SWIPT). For the TDMA-based information transmission. Under the above two scenarios. while “far” users are scheduled for frequency division multiplexing (OFDM). For the TS scheme. time that circuits for harvesting energy from radio signals are not yet able to decode the carried information directly. [1] studied the performance limits of a three-node multiple- appropriate variable transformation the problem is reformulated as a convex problem. SWIPT for multi- harvested energy constraint on each user as well as a peak and/or antenna systems has been considered in [1]. power splitting. time division multiple not hold in practice. as circuits for harvesting energy from access (TDMA) and orthogonal frequency division multiple access radio signals are not yet able to decode the carried information (OFDMA). In the TSR protocol. power allocation.e. of the received power for EH and the remaining power for R. we employ PS at the receivers. time switching. are proposed in [11]. Singapore a hybrid network which overlays an uplink cellular network (e-mail: hock@i2r.. USA. we consider two from the same received signal. this assumption may types of multiple access schemes. SWIPT in multi-antenna I. A*STAR. the received signal is the energy harvesting (EH) and information decoding (ID) either processed by an energy receiver for energy harvesting processes. orthogonal are scheduled for EH. simultaneous wireless information and power and TS. EH was proposed. respectively. where a new play key roles in the rate-energy performance comparison by the scheme that employs random beamforming for opportunistic two proposed schemes. for the OFDMA-based information decoding (ID). In the PSR protocol. SWIPT with energy/information relay- transfer (SWIPT) becomes appealing by essentially providing ing has been studied in [11]. the relay spends a This paper has been presented in part at IEEE Global Communications portion of time for EH and the remaining time for information Conference (Globecom). for which the optimal power allocation and input multiple-output (MIMO) broadcasting system. Rui Zhang. With PS applied at the receiver. (TSR) protocol and the PS-based relaying (PSR) protocol. C. we address the problem of maximizing the weighted signal is split into two signal streams with a fixed power ratio sum-rate over all users by varying the time/frequency power by a power splitter. SWIPT that exploits flat-fading channel variations was division multiple access (OFDMA). Zhou is with the Department of Electrical and Computer Engineering. switching (DTS) [7] or dynamic power splitting (DPS) [8] to coordinate between EH and ID. simultaneous wireless information and power transfer (SWIPT) Varshney first proposed the idea of transmitting information in downlink multiuser orthogonal frequency division multiplexing and energy simultaneously in [2] assuming that the receiver is arXiv:1308. IEEE. we (EH) or processed by an information receiver for information employ TS at the receivers. Member. [6] studied a MISO broadcasting system Index Terms—Simultaneous wireless information and power that exploits near-far channel conditions. Zhang is with the Department of Electrical and Computer Engineering. either time switching (TS) or power splitting (PS) to coordinate [3].IT] 13 Feb 2014 (OFDM) systems. December 9-13. In particu- total transmission power constraint. information processing. The performances of the two schemes are compared state information (CSI) at the transmitter for a multiple-input numerically as well as analytically for the special case of single- user setup. are proposed in [1]. At the receiver side. orthogonal frequency ID. A*STAR. we propose an iterative algorithm to optimize the information from the signals sent by a common transmitter.edu. namely. Two relaying protocols. K. He is also transfer were studied in [12]. since it realizes both useful utilizations of radio destination. In [12].edu.sg). with one stream to the energy receiver and allocation and either TS or PS ratio. wireless power. A MISO SWIPT system without on each OFDM SC as well as the number of users in the system CSI at the transmitter was considered in [5]. the TS-based relaying signals to transfer energy as well as information. National University of Singapore (e-mail: xunzhou@nus. It is revealed that the peak power constraint imposed single-output (MISO) system. the SWIPT system offers great convenience to harvested energy to forward the source information to the mobile users. the received transmission. where TS ratio are obtained by the Lagrange duality method. due to the practical limitation directly. Therefore.

Section II potentially harvested. all subcarriers As shown in Fig. and has been adopted in various standards. It is shown in [14] that a tradeoff exists between the users by varying the power allocation in time and frequency achievable rate and the transferred power by power allocation and the TS ratios. namely. . However. the when information is scheduled for that user. Furthermore. As mentioned in the previous paragraph. the result in [14] actually provides only at receivers until the convergence is reached. for the case of bandwidth of the system is equally divided into N subcarriers a single-carrier system. As an extension of our previous work in [3] for a single-user Assume that the total transmission power is at most P . The point (AP) with constant power supply broadcasts signals to channel power gain of SC n as seen by the user k is denoted . In this paper. 0 ≤ pn ≤ Ppeak . Unlike the conventional wireless the tradeoffs among the network parameters were derived. In [16]. The total contain information for the receiver. each user should apply TS cations. It is revealed that information decoder and energy harvester are attached to two the peak power constraint imposed on each OFDM subcarrier separate antennas. it the paper. whereas as the transferred power increases. Thus. However. S YSTEM M ODEL AND P ROBLEM F ORMULATION splitting is performed before (digital) OFDM demodulation. 1. . The narrowband SWIPT system. N . schemes. Section V concludes at each receiver to coordinate between EH and ID. carried by the subcarriers allocated to other users for ID can be The rest of this paper is organized as follows. for which transmission rate. for an OFDM-based SWIPT system. 2 wirelessly. we an upper bound for the rate-energy tradeoff in a single-user compare the rate-energy performances by the two proposed OFDM system. For the OFDMA scenario. the authors investigated a cognitive radio network and draw energy from their own power supplies. is assumed that the splitting ratio can be different for different subcarriers. we propose an efficient algorithm to iteratively energy from radio signals are not yet able to decode the carried optimize the power and subcarrier allocation. the optimal power allocation and TS ratios are obtained by more power needs to be allocated to the channels with larger the Lagrange duality method. e. N }. network where all the users contain only information receiver In [13]. where it is assumed that each user contains an additional energy mobiles from the secondary network either harvest energy receiver to harvest energy from the received signals from from nearby transmitters in a primary network.11n and 3GPP-Long Term Evolution (LTE). For the OFDMA-based devices for some practical application scenarios. The SC set is denoted by N = {1. we SWIPT over a single-user OFDM channel has been studied in assume that all subcarriers share the same power splitting [14] assuming that the receiver is able to decode information ratio at each receiver. the optimal power allocation is given by the so-called power constraint. n = 1. two conventional information if the primary transmitters are far away. we study a multiuser maximum power allocated to each SC is denoted by Ppeak . . . For the information transmission. we consider a downlink OFDM-based would have to be power split with the same power ratio at system with one transmitter and K users. the throughput of multiple access (TDMA) and orthogonal frequency division the secondary network was maximized. Under an multiple access schemes are considered.. ∀n ∈ N . subject to a minimum harvested energy in the frequency bands: for sufficiently small transferred constraint on each user and a peak and/or total transmission power. we address and harvest energy simultaneously from the same received the problem of maximizing the weighted sum-rate over all signal. time division outage constraint for coexisting networks. It thus mo.g. Finally. Moreover. due to the practical limitation that circuits for harvesting In this case. . we channel gain and finally approaches the strategy with all power address the same problem by varying the power allocation in allocated to the channel with largest channel gain. in practical circuits. 1). where Ppeak ≥ P/N . each user only harvests the energy as well as the number of users in the system play key roles in carried by the subcarriers that are allocated to that user for the rate-energy performance comparison by the two proposed ID. multiple access (OFDMA). Under an outage constraint on the data links. The transmitter each receiver even though only a subset of the subcarriers and all users are each equipped with one antenna. Section by assuming a predefined subcarrier allocation.e. power allocated to SC n is denoted by pn . the subcarrier allocation to users and the PS ratios. In contrast. which is inefficient in energy utilization. Section IV derives the resource allocation the subcarrier allocation strategy. [15] focuses on power control presents the system model and problem formulations. frequency. where the the special case of single-user system setup. In [15]. or transmit the AP. in this paper. III studies the special case of a single-user OFDM-based we jointly optimize the power allocation strategy as well as SWIPT system. where PS is applied OFDM-based SWIPT system. Under the TDMA scenario. powered by opportunistic wireless energy harvesting. we assume that PS is applied at tivates our investigation of SWIPT in OFDM-based systems. Power control for SWIPT in a multiuser multi. since the energy schemes. where a fixed access i. each receiver. information transmission. since the users are scheduled in nonoverlapping well established technology for high-rate wireless communi. OFDM-based SWIPT system (see Fig.. while the energy performance may be limited by the availability of energy in the receiver is used in all other time slots. For the TDMA-based information Orthogonal frequency division multiplexing (OFDM) is a transmission. both numerically by simulations and analytically for antenna OFDM system was considered in [15]. However. By an appropriate variable transformation “waterfilling (WF)” allocation to maximize the information the problem is reformulated as a convex problem. The all signals that do not contain information for this receiver. (analog) power II. a receiver simply harvests energy from (SCs). [16] considered SWIPT in a solutions for the two proposed schemes in the multiuser multiuser single-antenna OFDM system. . and the PS ratios information directly. . in our model. . IEEE such that the information receiver is used during the time slot 802. a set of distributed users. time slots to decode information.

. We assume that slot k... In addition. .. . Ek = ζ αi hk. k = 1. while the received energy on all at the receiver. where all the channels are assumed to be constant within the transmission scheduling period of our interest. . . . The achievable constraints and the transmission power constraints. For simplicity. . while the received decodes its intended information at slot k when its information energy on all SCs during slot 2 and slot 3 is utilized for EH. . 3 AP energy energy energy Information transfer EH Energy transfer ID . K. N . while the received energy on all SCs during slot 2 noting that for a single-user SWIPT system with TS applied and slot 3 is utilized for EH. n = 1. . the transmission time needs to be divided SCs during slot 1 and slot 3 is utilized for EH. ... N . . K+1 X N X where 0 ≤ pk. . We consider a slow- fading environment. With total time duration of K + 1 Assuming that the conversion efficiency of the energy har- slots to be at most one. (1) case in a two-user OFDM-based SWIPT system is illustrated k=1 n=1 in Fig. . ... .. At the receiver side. As shown in Fig. the received energy on all SCs during The receiver noise at each user is assumed to be independent slot 2 is utilized for ID.. the (normalized) time duration for vesting process at each receiver is denoted by 0 < ζ < 1. At ..n . each user performs EH in addition to ID. (3) The average transmit power constraint is thus given by i6=k n=1 K+1 N X X An example of the energy utilization at receivers for the TS αk pk... . .n ≤ Ppeak . the receiver side.. . In contrast. 2.n ≤ P. .. A multiuser downlink SWIPT system.. . . . . k=1 the power pn allocated to SC n at slot k is specified as pk. It 1 n N 1 n N 1 n N SC SC SC is assumed that the minimum harvested energy during the unit slot 1 slot 2 slot 3 transmission time is E k > 0 for user k. the gap for the achievable rate all users by varying the transmission power in the time and from the channel capacity due to a practical modulation and frequency domains jointly with TS ratios. Energy utilization at receivers for a two-user OFDM-based SWIPT system: TDMA-based information transmission with TS applied at each We first consider the case of TDMA-based information receiver. . K. k = 1.. .n Rk = log 1 + . Gaussian (CSCG) random variable with zero mean and vari. (2) is not required. . . . It is worth slot 1 is utilized for ID. the rate in bps/Hz for the information receiver of user k is thus following optimization problem is formulated. 2(b) for user 2. . may be allocated for all users to perform EH only. Thus. we assume the total transmission time to be one. Moreover. the received energy on all SCs during slot 2 is utilized for ID. k = 1.n . . Our objective is to maximize the weighted sum-rate of ance σ 2 at all SCs. 2. in the SWIPT system with K users. the power slot αk X hk. . all the SCs for all the users are known at the transmitter. K. 2(a) for user 1. subject to EH coding scheme (MCS) is denoted by Γ ≥ 1. Thus.. which we called the power slot. the received energy Consider user k. . while the received energy on all SCs over SCs and is modelled as a circularly symmetric complex during slot 1 and slot 3 is utilized for EH. the slot k. 1 n N SC 1 n N SC 1 n N SC slot 1 slot 2 slot 3 Fig. K + 1.n pk. . k = 1. K is N n=1 2 Γσ 2 assigned to user k for transmitting information. it is assumed that the channel gains on .. . (b) Energy utilization for user 2 at different slots A. where given by the additional time slot. TDMA with Time Switching Fig. In sub-figure (b) for user 2. K + 1 is variable and denoted by the TS harvested energy in joule at the energy receiver of user k is K+1 P thus given by ratio αk . k = 1. N   in conventional TDMA systems without EH. the received energy on all SCs during transmission with TS applied at each receiver. with 0 ≤ αk ≤ 1 and αk ≤ 1. ... . In Fig. while slot K + 1 is the power slot. we consider K + 1 time slots without loss of generality. n = 1. .. is sent and harvests energy during all the other slots i 6= k.. . Moreover. user k on all SCs during slot 1 is utilized for ID.n pi. In sub-figure (a) for user 1. into two time slots to coordinate the EH and ID processes at the receiver. . (a) Energy utilization for user 1 at different slots energy energy energy by hk. .n . 1... k = 1. . .

ID at user 1. .n ≥ E k .n ≤ Ppeak .e. .. k = 1. . k = 1.. Therefore. K + 1. wΠ(n) log2 1 + {pn }.n ≤ Ppeak . . (5) n=1 From (7). (P − PS) : B. Therefore. k = 1. k = 1.n pK+1. . problem is formulated. .. K.. i. the received signals at all SCs N X share the same splitting ratio ρk at each user k. n=1 With the objective of maximizing the weighted sum-rate of 0 ≤ pK+1. . pn ≤ P.n pn . With energy conversion efficiency ζ.n pk.t.{ρk } N n=1 Γσ 2 transmission with PS applied at each receiver. .n }. k = 1. n = 1. all power is split to the power splitter. the received signal at user k is processed by a when ρk = 1. . no SC sharing is allowed. . k = 1. . . ζ X αi X hk. i6=k n=1 K+1 N 1 2 n N-1 N 1 2 n N-1 N X X SC SC αk pk. As is standard N in OFDMA transmissions. We will solve this problem in Section IV-B. .t.n pn ≥ E k . . ζ hk. . . The total transmission power constraint X pn ≤ P. The achievable rate in bps/Hz is feasible. OFDMA with Power Splitting N (1 − ρΠ(n) )hΠ(n). we consider the case of OFDMA-based information max. . . ∀n is given by n=1 XN 0 ≤ ρk ≤ 1.n } in Fig. the following optimization solve this problem in Section IV-A. . Problem (P-TS) is feasible if and only if the n=1 following linear programming (LP) is feasible. K + 1. pk. 0 ≤ αk ≤ 1.n ≥ E k . . all users harvest energy during the entire transmission Ek = ρk ζ hk. (7) time.n ≤ P. n = 1.n }.n = 0 for k = 1. 2. with 0 ≤ ρk ≤ 1. K+1 X Fig. system: OFDMA-based information transmission with PS applied at each receiver. n = 1. user 1 user 2 k=1 n=1 0 ≤ pk..e. . .. i. ρk ζ hk. Problem (P-PS) is fea- receiver and the remaining ratio 1−ρk is split to its information sible if and only if Problem (P-PS) with ρk = 1. . is worth noting that only ρ1 of the power at each of the SCs n=1 allocated to user 2 for ID is harvested by user 1.n pn   1 X Next.. . N. From (3). We will transmission power constraints. . subject to a given set of EH constraints and the Problem (P-TS) is non-convex in its current form. We thus domain. while αk = 0. (4) all users by varying the transmission power in the frequency It is easy to check the feasibility for the above LP. . .e.t.   Rn = log2 1 + . K receiver. X i.n max. . . K}. 2. 0 case in a two-user OFDM-based SWIPT system is illustrated {pK+1. ∀k. k = 1. . the harvested energy at all users is maximized At the receiver.. . 0 ≤ pn ≤ Ppeak . . . . . {αk }}. An example of the energy utilization at receivers for the PS max. . ∀n. the SC n is N allocated to user Π(n).e..{αk } N k=1 n=1 Γσ 2 K+1 N s. . K. joule at the energy receiver of user k is thus given by the harvested energy at all users is maximized when αK+1 = N 1. Energy utilization at receivers for a two-user OFDM-based SWIPT αk ≤ 1. .n pn by Problem (4). 3.. . 3. As shown in Fig. K. . . . The received signals at all SCs share the same splitting ratio ρk k=1 at each user k. 4 energy energy EH (P − TS) : ID K N   1 XX hk. k = 1. the harvested energy in lem (P-TS) can be satisfied by some {{pk... . . .n pi.{Π(n)}. Problem (P-TS) is feasible when all the constraints in Prob. We define a n=1 SC allocation function Π(n) ∈ {1. . . the remaining N X 1 − ρ1 of power at those SCs is neither utilized for EH nor pK+1. K. N. 3. It is worth noting that Problem (P-PS) and Problem at SC n assigned to user Π(n) is thus (P-TS) are subject to the same feasibility conditions as given (1 − ρΠ(n) )hΠ(n). ∀k user in each slot. . k = 1. K. . ∀k. It can be verified that Problem (P-PS) is non-convex in its Γσ 2 (6) current form. . N . each SC is allocated to at most one X s. where wk ≥ 0 denotes the non-negative rate weight assigned to user k. wk αk log2 1 + {pk. where a ratio ρk of power is split to its energy energy receiver at each user. It s. . i. . jointly with the PS ratios at assume Problem (P-TS) is feasible subsequently. similarly as for user 2 with PS ratio ρ2 .n ≤ P. the SC allocation. receivers. .

n ≤ n=1 n=1 simplified respectively as follows P − E/(ζh1 ) and 0 ≤ α1 < 1.n max..n ≥ E. in this section. i = 1. otherwise. .1. we first look at the two extreme An upper bound for the optimization problems (P-TS) and cases. For given {q1.. i. It follows that the minimum solutions exist. and ρ1 α1 p1. 2. i. α1 < 1. n=1 user OFDM-based SWIPT system with Ppeak → ∞. which {p1.n .n ≤ Ppeak .e. . Without loss of q1. pn ≤ P. RTS (P. Ppeak ) and N X RPS (P. thus. n=1 Thus. Moreover.t. Performance Upper Bound To obtain useful insight. N   1 X hΠ(n). We shall see (P-PS) can be obtained by assuming that each receiver is that the peak power constraint plays an important role in the able to decode the information in its received signal and performance comparison between the TS and PS schemes. 2.{p2. N . in α = log m/m → 0 as m → ∞.n should satisfy ζα2 h1 p2.e.n 2 subject to n=1 consider the special case that K = 1..n =α1 p1. ∞) is achieved under the assumption that N  the transmitter and receiver are able to operate in the regime of (1 − ρ)hn pn  1 X max.USER S YSTEM N   h p given by maximizing αN1 P To obtain tractable analytical results. ∀n. ∞). N X Remark 3. N X Proposition 3. and ρ. . the maximum rate achieved 0 ≤ pn ≤ Ppeak . in the case of a single- pn ≤ P. For the case of Ppeak → ∞. i = 1.2: In the case of a single-user OFDM-based n=1 SWIPT system with Ppeak → ∞. a nonzero time ratio should be scheduled to the power N X slot to collect sufficient energy to satisfy the EH constraint. ∀n. (P-UB) can be modified to become a feasible one by increasing Thus. Ppeak → ∞ and Ppeak = P/N . ∀n.1. ∞) s.. i. ∞). to achieve RTS (P. ∀n. . symbols in each block. no energy as given by Problem (4).n pn Given P and Ppeak . N  hn q1. respectively. since the total power constraint is always satisfied and thus becomes redundant. let m denote the number of transmitted α1 p1.n + α2 p2. the above problem is then equivalent to generality.2). For brevity.e.n }.n  N maximizing αN1 P P With K = 1. Let n=1 are replaced with hn . respectively.n }. 5 C. a single-user N P OFDM-based SWIPT system. We thus consider the following constraint on each SC.n ≥ E/(ζh1 ). RTS (P. as well as Problem (P-TS) and by α1 → 1 and α2 → 0. we s. optimal α2 and p2. E 1 .α1 . We recall n=1 that α2 = 1 − α1 is the TS ratio for the power slot. Ppeak ).n ≤ P. . wΠ(n) log 2 1 + {pn }. we log2 1 + nΓσ1. R ESOURCE A LLOCATION IN A S INGLE . the portion of transmission time α2 allocated n=1 to EH in each transmission block should asymptotically go to N N X X zero. log2 1 + maximized when α1 → 1. s. is achieved Note that Problem (P-UB).n ≤ P − E/(ζh1 ) and 0 ≤ α1 < 1. only peak power constraint on each SC. For a finite {pn }. by allocating O(log m) symbols for n=1 n=1 EH in each block and the remaining symbols for ID results 0 ≤ pi. . To maximize the objective function sub- the transmission power P or by decreasing the minimum ject to the EH constraint. ∞) ≤ RTS (P. It follows that α2 → 0.. we assume that h1 ≥ h2 ≥ .e. 0 ≤ αi ≤ 1. we have the following proposition showing that n=1 the PS scheme performs no better than the TS scheme for the N X case of Ppeak → ∞. Proposition 3. ergy E is given by E/(ζh1 ). E. transmission energy consumed to achieve the harvested en- The solution for Problem (P-UB) is obtained in Section IV-B N P (see Remark 4. log 1 + infinite power in the EH time slot due to α2 → 0.t. RPS (P. Problem (P-PS) are subject to the same feasibility conditions Proof: Clearly. by the PS scheme is no larger than that achieved by the TS 0 ≤ ρ ≤ 1. ≥ hN and w1 = 1. at the same time harvest the received energy without any Note that Ppeak → ∞ implies the case of no peak power implementation loss [14]. we have α2 > 0. .n = 0. i. n = 2. thus optimal p2. h1. ζ hk. (9) scheme.{Π(n)} N n=1 Γσ 2 scheme and the PS scheme are denoted by RTS (P. ζα2 hn p2. ∞) is III.e. In the sequel. It is worth noting that RTS (P.1: By Proposition 3. which violates the EH constraint E > 0.n . Problem (P-TS) and Problem (P-PS) are then log2 1 + Γσ 2α 1 subject to q1. which satisfies the α1 + α2 ≤ 1. i.ρ N n=1 2 Γσ 2 Ppeak . the maximum rate by the TS scheme. α2 p2. the achievable rate RTS (P. the maximum rates achieved by the TS (P − UB) : max.1 = E and we assume that all the three problems are feasible.n pn ≥ E k . ∞) is α1 X hn p1. in Problem (8).t.α2 N n=1 Γσ 2 completes the proof. with E > 0. ∀k have the following proposition for the TS scheme.1: Assuming E > 0. the objective N   function is an increasing function of α1 . 0 ≤ pn ≤ Ppeak .n }. and Ppeak = P/N implies the case of optimization problem. it can be easily shown that the required harvested energy E k at some users k. (8) optimality condition provided in Proposition 3. ρζ hn pn ≥ E. Also note that any infeasible Problem is harvested. For example.

e. Note that according PS. P/N ). P/N ) is thus given holds if Ppeak → ∞. The six-tap ex- which completes the proof. By Proposition 3. ∀n. From Proposition frequency 900MHz with path-loss exponent equal to 3.e. 4. the optimal power the subscripts of hn . from the proof of Proposition 3. it is observed that for both TS peak power.n = p∗2. we remove the SC index n in is redundant for both TS and PS. it follows that ρ ≥ E/(ζh1 P ) must n=1 hold. Therefore. which the performance comparison between the TS scheme and the results in −30dB path-loss for all the channels at a carrier PS scheme remains unknown analytically. −E/ζ N RTS (P. It follows RTS (P. neither the TS scheme nor the PS the energy receivers. RTS (P. with sufficiently large In both Fig. ζPpeak hn n=1 Figs. It suggests that for a single-user information receivers. thus. ∀n. power allocation for PS is given by p∗n = Ppeak . Ppeak ) ≤ by the PS scheme.3. It follows that α1 p1 ≤ P −E/(ζh). the above equality E that ρ∗ = P N = 1 − α∗1 . we have RTS (P. ∀n. Ppeak= 4P/N to Proposition 3.n = Ppeak .. a single-carrier point-to-point SWPT system. P/N ) ≤ RPS (P. since the available to-point SWIPT system.. i. Then we have the optimal with ρ = E/(ζhP ). Note that for a single. 4 with N = 64. the above problem is then P n=1 60 N   UB. Fig. ponentially distributed power profile is used to generate the In fact. The MCS gap is assumed to be Γ = 9dB. moreover. For Fig. 5 with N = 1. P/N ) is thus given by TS scheme. The distance user OFDM-based SWIPT system with P/N < Ppeak < ∞. For   1 P α∗ 1 hn Ppeak by N log2 1 + Γσ2 . In for Ppeak → ∞. ∞) ≤ RTS (P. RPS (P. Ppeak ) = log2 1 + hPΓσ 2 .n 2 subject to p1. the transmit employed for both TS and PS schemes. power is assumed to be 1watt(W) or 30dBm. For 3. which logarithm function. RPS (P. and PS schemes. On the other hand. the band- allocation (not necessarily equals to Ppeak ) over all SCs are width is assumed to be 1MHz. we have RPS (P. Let p′n = (1−ρ)pn .1. It to satisfy the EH constraint.3. we have ρ ≥ E/(ζhP ). Ppeak ) ≤ α1 log2 1 + hP ≤ RPS (P. with equality if Ppeak → ∞. the PS scheme may be better. allocation for TS is given by p∗1. Since ρ ≥ E/(ζh1 P ). P ρ ≥ E/(ζh1 P ) and UB. RPS (P. and that by TS with Ppeak → ∞. 5. ∞) is upper bounded by maximizing 100 N   1 P (1−ρ)hn pn N log 2 1 + Γσ 2 subject to ρ ≥ E/(ζh1 P ) and 80 n=1 N Rate (Mbps) pn ≤ P .. from the EH constraint 120 N P ρζ hn pn ≥ E. P/N ) ≤ RPS (P. required harvested energy E increases. Ppeak→ ∞ hn p′n equivalent to maximizing N1 P log2 1 + Γσ 2 subject to TS. the noise spectral density is assumed to OFDM-based SWIPT system with sufficiently small peak be −112dBm/Hz. the gap increases . For the scheme is always better. which completes the proof. a single-carrier point. the TS and PS schemes achieve the same rate.4: In the case of a single-carrier point-to- achieved by the TS scheme is no larger than that achieved point SWIPT system with N = 1. we have α2 p2 ≥ E/(ζh) to satisfy the EH ζPpeak n=1 hn constraint. Proposition 3.n ≤ Minimum required harvested power (µW) n=1 n=1 P − E/(ζh1 ). it is observed that there is a significant for a finite peak power P/N ≤ Ppeak < ∞. Ppeak→ ∞ n=1 40 N PS. 4 and Fig.3: In the case of a single-user OFDM-based SWIPT system with Ppeak = P/N . For the PS scheme. ∞).e. it is assumed that ζ = 0. Thus. RTS (P. power. we have frequency-selective fading channel. the following proposition shows that: energy for information decoding decreases as E increases. follows that α2 ≥ E . the achievable rate decreases as the minimum For the special case that N = 1. p1. we have RTS (P. the maximum rate Proposition 3. the maximum rate by the PS  scheme P N −E/ζ ζPpeak n=1 hn is given by RPS (P. the TS scheme may be better. RTS (P. Proof: With Ppeak = P/N . Fig. Due to the concavity of the n=1 Fig. thus. we have single-user OFDM-based SWIPT system. from the transmitter to the receiver is 1 meter(m). ∞) is obtained (with α1 → 1) by N   N 0 h p 0 50 100 150 200 250 300 350 400 maximizing N1 P P log2 1 + nΓσ1. the total power constraint Proof: Since N = 1. Ppeak→ ∞ p′n ≤ (1−ρ)P .2. Ppeak= 4P/N n=1 20 TS. the TS scheme gap between the achievable rate by TS with Ppeak = 4P/N performs no better than the PS scheme. RTS (P. 4. P/N ). P/N ) ≤ RPS (P. Ppeak ). For both figures.   α∗ P hn Ppeak α1 Γσ2 N 1 log2 1 + Γσ2 .2 and Proposition 3. is equally divided as N = 64 subcarriers. the total bandwidth is assumed to be 10MHz.n .1. the optimal n=1 and the equality holds if α1 → 1. where N = 64. p2. i. 6 Proof: For the PS scheme. Achievable rate versus minimum required harvested energy in a For the other extreme case when Ppeak = P/N . ∞) is achieved by α1 → 1. 4 and 5 show the achievable rates by different schemes N versus different minimum required harvested energy E. Thus. P/N ) provided that equal power i. Ppeak= 4P/N it follows that (1 − ρ)P ≤ P − E/(ζh1 ). Therefore. the following proposition. Ppeak ).n and pn . For the E α∗1 = 1 − P N .

n ≥ E k .n . This is because that with Ppeak → ∞. Since the constraints with Proposition 3. where N = 1. {αk }. n = The dual function g ({λi }. when Ppeak → ∞. αk = 0. ∀n. .n }. . 0 50 100 150 200 Minimum required harvested power (µW) the optimal power allocation {p∗k.1).n = 0. . .1). . . e. 0 ≤ qk. 4. . αk log2 1 + Γσ2 αk = 0 at αk = 0 to keep continuity at αk = 0. as a non-negative weighted sum of Ppeak = 4P . . Ppeak = 4P/N .Ppeak=4P ∗ 0 After finding the optimal {qk. if qk. . ν) is then defined as the optimal  1. a nonzero transmission time needs to be ∗ then we have pk. K + 1. ∀n. (10) TS. Thus. . k = 1. if αk = 0 and Ppeak < ∞. The Lagrangian of Problem (10) is given by IV. K + 1. .4. . N . scheduled for energy harvesting. k = 1. .n ≤ Fig. K + 1.n − E i  i=1 k6=i n=1 K+1 N ! K+1 ! XX X +µ P − qk. N .n . .n ( PS schemes in Fig.n +ν 1− αk (12) A. Comparing the TS and   hk. µ. . ζ hk. f (qk. X XX + λi ζ hi. k = variables associated with the corresponding constraints in (10). For the PS scheme. For the extreme case of Ppeak → ∞. K. we allow αk → 0 while qk.g. ν) {qk.n qi.n qk. αk > 0. 0 ≤ αk ≤ 1.n = wk αk log2 1 + based SWIPT system with multiple users. .n }. 5 with N = 1. k=1 n=1 6 0 ≤ qk.n ≥ 0.1: Function f (qk.t. N . 1 Similar to the single-user system (c. . 10 i6=k n=1 Rate (Mbps) RTS(P.n = 0.n ∗ > 0 and α∗k = 0. . the TS scheme is the same as that by the PS scheme. We derive the N Γσ 2 αk k=1 n=1 optimal transmission strategies for the two schemes proposed K  K+1 N  in Section II.n → ∞. whereas for a finite the allocated power will be p∗k. Achievable rate versus minimum required harvested energy in a αk Ppeak .n qk. Ppeak = 4P/N . N .n 14 max. αk ) = k (11) 0. . for sufficiently small Ppeak . . N ∗ provided that αk > 0. .n . n = 1. all uled at slot k. i = 1. . the allocated power will be p∗k. Proposition 3. (13) . K. From the constraint 0 ≤ qk.n > 0 by letting pk. K + 1. however.n value of the following problem. {λi }. .4P) K+1 N XX 8 qk. . . . . (P-TS) is thus equivalent to the following problem:1 max. we have qk.n . 1 XX hk. .n } for Problem (P-TS) is given by p∗k. this Lemma 4. R ESOURCE A LLOCATION IN A M ULTIUSER S YSTEM L ({qk.. . . . wk αk log2 1 + {qk.n ≤ P.n qk.n qk. the achievable rate by Proof: Please refer to Appendix A. are now all affine. 1.n ≤ αk Ppeak . as will be shown next. . the achievable rate by the TS scheme is no f (qk. K + 1.n }. . 0 ≤ αk ≤ 1. 7 16 K N   1 XX hk. . and thus can be optimally solved by applying the Lagrange duality method.n }.n = 0 single-carrier point-to-point SWIPT system. PS outperforms TS. αk ).n ≤ αk Ppeak . it is observed that TS outperforms PS αk log2 1 + Γσ 2α . then letting α1 → 1 (c. we define hk. .n → ∞. UB K+1 4 X PS αk ≤ 1. k = 1. K + 1. k = 1. n = 1.1. Time Switching k=1 n=1 k=1 We first reformulate Problem (P-TS) by introducing a where λi . it is observed that when Ppeak → ∞.n ∗ /α∗k . n = 1 . we consider the general case of an OFDM.{αk } N k=1 n=1 Γσ 2 αk 12 K+1 XX N s. . 5. . αk ) is jointly concave in performance gap due to finite peak power constraint is only αk ≥ 0 and qk. when From Lemma 4.n = qk. . .Ppeak→ ∞ k=1 2 TS. . which is in accordance jointly concave in {αk } and {qk. Remark 3. . . . µ. . and compare their performances. L ({qk. . {αk }. if α∗k = 0. for the case of Ppeak → ∞. ν) K N   In this section. µ. In Fig. {λi }. and ν are the non-negative dual set of new non-negative variables: qk. . . ∞) = RPS(P.f.t. . n = 1. k = 1. . . . K + 1 and Ppeak < ∞. . if transmission time can be utilized for information decoding by ∗ qk. ∞) = RPS(P. α∗k = 0. . .n }.n = 0. .n } and {α∗k } for Problem (10). . the new objective function of Problem (10) is larger than that by the PS scheme.n qk. . . since no information/power transmission is sched- as E increases. Problem (10) is convex. k = 1. k = 1. K + 1. . where observed when E is sufficiently large. K + 1. . . k = 1.f. Moreover.{αk } s. . k = 1. µ.n = αk pk. .

αk can be found numerically by a becomes a LP with variables {qK+1. thus the time complexity i=1 maximize LK+1 we have of steps I)-III) is O(K 4 N ). . K. K. ν) with a given set of {λi }. k = 1.n = qk. if P ζ P λ h − µ P peak − ν > 0. . . Thus. n = 1.n the objective can be improved by increasing some of the i6=k (18) αk ’s. . (16) P P ζ hi. K. Proposition 4. k = 1. x).n αK+1 = n∈N1 i=1 We define Lk . . (15) k=1 i=1 the minimization of g ({λi }. µ.n − ναk . Therefore. the optimal Γσ 2  K+1    wk α∗k ’s. . the time complexity of steps 1)-3) is of order K 2 N .n ) ln 2 where {q̇k.n }. the optimal αK+1 is given by K where (x)+ . i. For the algorithm given in Table I. . which is known as block-coordinate descent method time is dominated by the ellipsoid method in steps I)-III) and [17]. ν). we study the solution for Problem (16) for k = K 2 N . . . . n ∈ N \N1 proposition.n qk. µ. . we consider the maximization problem in (13) for   K  1.n . (21) 1. the following choice of d is a subgradient for {qk. . . we have   k=1 n=1 K+1    P 1 − α̇k . . .n . ν): s. K are obtained at optimal {λ∗i }. .n }.n − µ. . . . k = K + 1.n − E i .  of this page. Thus.   K ∂Lk wk αk hk.n }. ..n }. . K + 1. With αk = α∗k . i = K + 1. .αk g ({λi }. the objective function in Problem (10) is an increasing function of αk .n } and {α̇k } is the solution of the maximization i6=k (17) problem (13) with given {λi }. . N . otherwise. the computation one time. and ν ∗ . for each given k. k = 1. . . To summarize. K. µ. the qk. and ν. . µ. K + 1 satisfy   P ∗ qk. since Lk is a concave function of qk. X X L= Lk − λi E i + µP + ν. k = 1. K n=1 i6=k n=1 n=1 Lk := K N N (14) P P P hi. . µ. .n N (Γσ αk + hk.e. and ν. . . . Problem (16) can be solved by iteratively To summarize.n − ναk . k = 1. µ.n qk. . . . ν) with given {λi }.n − µ qk. . . . O(K 2 N ). Given αk .n∗ . ν) over {λi }. and First. . . . k = 1. k = 1.  ζ  λi i=1 n=1 n=1 The dual problem is thus defined as min{λi }. step 4) is of order N . . n = 1. N that maxi.n } are obtained by solving this LP.n = 0 to give Note that the optimal qk.n }. µ. and step 6) is of order K 2 .n − µ > 0 . ∀n. .n − µ qk. ∂Lk mizes Lk can be obtained by setting ∂q k. . . From (14). .n . 0 ≤ qk. otherwise. i i. 8 N K N N    w α P hk. step 5) is of order Next. Problem (10) αk with given {qk. . µ and ν. for of {qK+1.     k6=i n=1     K+1 N We first study the solution for Problem (16) with given k = di = P − P P q̇k. . µ. the time complexity of the algorithm in Table I is O(K 4 N + KN 2 + N 3 ). . . k = 1. i. .n X k=1 = 2 +ζ λi hi. . K. K.n = αk min  ! −  . there is no closed-form expression for the optimal αk that ∗ k=1 maximizes Lk . . we have the following 0. N  K+1 N 0 ≤ αk ≤ 1. . .t.ν g ({λi }. the maximization problem in (13) can of this problem required for the ellipsoid method is provided be decomposed as by the following proposition. .n } and αk with one of them fixed at Table I. K. µ. n = 1. we have (20) K+1 K After obtaining g ({λi }. A subgradient Thus. .n q̇k. . max(0.e. . ∂qk. n = 1. the LP in step V). Note that step set N1 = n ∈ N : ζ λi hi.µ. . to II) iterates O(K 2 ) to converge [18]. ∗  +  Given {qk. given k = 1. one algorithm to solve (P-TS) is given in optimizing between {qk.n qk. . Proof: Please refer to Appendix B. . n ∈ N1 . it appears that α∗K+1 = 1 − P α∗k . and ν can be efficiently solved by the ellipsoid method [18].for the power slot. the time complexity K + 1.n = (19) Similar with the single-user case. which is a LP.n qk. . . The optimal values simple bisection search over 0 ≤ αk ≤ 1. ( αK+1 Ppeak . i = K + 2. .  K P hk. . Then.n    k=1 µ−ζ ln 2   N λi hi. k = 1. k = 1 . In particular. as shown in (14) at the top 0.n ≤ αk Ppeak . N and αk .1: For Problem (10) with a dual function max. Lk g ({λi }. . K. .n P P P  kN k   log 2 1 + Γσ 2 αk +ζ λi hi. µ∗ . The time complexity of the LP in step V) is O(KN 2 +N 3 ) [21]. However.. Ppeak  αk = 1. Then for the Lagrangian in (12). i = 1. Define the PK of steps 1)-6) is of order K 2 N . ν).  obtaining g ({λi }. qK+1. For given {qk. K + 1. From (14). .

and optimizing {ρk } s.µ g ({λk }. K + 1 and n = 1. K. VI) For k = 1. the optimal solution X may be computationally difficult to obtain. of which the objective function is a nonincreasing be ignored. it has been shown that the duality gap of a simi- SWIPT system with K ≥ 2 and Ppeak → ∞. . 3) until improvement of Lk . . . {Π(n)}. Power Splitting pn . . N X IV) Set qk. . N . Thus. K. .Π(n) N Γσ 2 K Since Problem (P-PS) is non-convex. where hID EH k. . .n > 0. k = 1. due to the integer SC allocation Π(n). . the system becomes a conventional TDMA The dual problem is thus given by min{λk }.n → ∞.1: In Problem (10). . the optimal power splitting ratio solution for (P-PS) with a given set of feasible {pn } and I) Initialize {λi > 0}. . . if α∗ = 0 and q ∗ qk. ∞). . in which case no additional power slot is scheduled and all users simply where λk ’s and µ are the non-negative dual variables asso- harvest energy at the slots scheduled for other users to transmit ciated with the corresponding constraints in (23).{Π(n)} N n=1 Γσ 2 6) Update {λi }. when K ≥ 2 and E k = {pn }. . K. µ) . . . µ) with a given set of {λk } and µ. In contrast. . k = 1. N . set p∗k. . ν) by (21). α∗k = αk .n pn ≥ E k . K for given {qk. k = 1.n pn b) For k = 1. respectively.n . .n . the maximum achievable rate is obtained by minimizing L ({pn }. The objective in Problem (10) is The Lagrangian of Problem (23) is given by an increasing function of αk . . . RTS (P. K. . is achieved by converges to zero as the number of SCs. qk. we propose +ζ λk hEH k. K + 1. µ and ν according to the ellipsoid method. . . . . 0 ≤ pn ≤ Ppeak . . K. Thus. N α∗k . compute {qk.2 dent of αk . . . . . g ({λk }. . 9 TABLE I A LGORITHM FOR S OLVING P ROBLEM (P-TS) function of ρk . the maximization problem in (25) can be decomposed with 0 ≤ E k ≤ Ekth .. n and hk. Ek 2) repeat ρk = N . k = 1. the Lagrange duality method assuming that it has a zero duality the EH constraints and the total power constraint are indepen- gap. k = 1. ∀k. set p∗k. k = 1.n pn qK+1. . µ and ν converge to a prescribed accuracy.n = 0. . . Next. .n . {Π(n)} is obtained as II) repeat 1) Initialize αk = 1/K. K and α∗K+1 = 1 − α∗k . obtain αk that maximizes Lk with given n=1 {qk. N hID ! 4) Compute {qK+1. {λk }.n . Proposition 4. k = 1. 1 X Π(n). ! ! It is worth noting that for the multiuser system with K ≥ 2 K X N X N X and Ppeak → ∞. the function is then defined as power slot is always needed if E > 0. .n ∗ = qk. i. for the single-user K = 1 case. i. Thus. K n=1 1. ζ hEH k. . ρk hk. . . .n } and αK+1 by (19) and (20). consider (P-PS) with a given set of feasible ρk ’s.n .. (26) with fixed {pn } and {Π(n)}. if α∗k > 0. Assume that the harvesting Consider the maximization problem in (25) for obtaining energy at each user by the optimal transmission strategy for g ({λk }. k = 1. n can be viewed as the equivalent channel power gains for the information and energy receivers. . µ).n pn max.n pn − Ek +µ P − pn (24) k=1 n=1 n=1 the TS scheme is achieved by αK+1 = 0. increases to αK+1 = 0 or αK+1 → 0. . . . k = s. . k = 1. the Note that (P-PS) with given {pn } and {Π(n)} is a convex duality gap of Problem (23) is observed to be negligibly small and thus can problem. K. we solve Problem (23) by applying Proof: In the equivalent Problem (10) with Ppeak → ∞. n = 1.e. the same rate as that by as the conventional TDMA system can be achieved. ∀k. .n∗ .n = 0. (1 − ρk )hk. . (22) P a) For k = 1. ∀n) or αK+1 → 0 (when qK+1. N . k = 1. αK+1 = 0 (when ! ∗ ∗ 1 X N hID Π(n). the maximum lar problem to (23) without the harvested energy constrains rate by the TS scheme. . . it is possible that the maximum rate by + λk ζ hEH k. . n = 1. . . K.e. µ) = max L ({pn }.n > 0 for some wΠ(n) log2 1+ N n=1 Γσ 2 n). hID ! wΠ(n) Π(n). µ > 0 and ν > 0. .n pn − µpn k=1 a suboptimal algorithm to this problem by iteratively optimiz- ing {pn } and {Π(n)} with fixed {ρk }. P k=1 N ∗ V) Obtain qK+1. K converges to a prescribed accuracy.2: In the case of a multiuser OFDM-based However.n = n=1 ∗ /α∗ . .n = 0. Then for a system SC n. . . n = 1. . set pk. k = 1. . The dual information. III) until {λi }. N . Instead. ∀k. . .t.n } by (18). . .t.n ∗ ∗ (23) k k k. . .n = qk. (25) Remark 4. {Π(n)}. µ. . [20]. . Ln := log2 1 + B.. K. For each given this system is given by Ekth . 2 In our simulation setup considered in Section IV-C with N = 64. . ζ hk. . .n . max. K +1. . k = 1. . system without EH constraints. {λk }. . K. . . i. . wΠ(n) log2 1 + {pn }.n } by bisection search. .e. if αk = 0 and ∗ qk. µ) = the time allocated to the power slot.n }. . The problem in (23) is non-convex. . 0 ≤ pn ≤ Ppeak .n pn 5) Compute the subgradient of g({λi }. infinity [19]. K. n = 1. N by solving Problem (10) with αk = X pn ≤ P. .{Π(n)} 0. . . . . .

The final solution is selected as the one X X L= Ln − λk E k + µP. step b) is of order KN .n  the algorithm in Table II is O(K 4 M + K 3 N M ). the time complexity of    ! − ID  . 1) Initialize {ρk }.e. 10 Note that for the Lagrangian in (24). which is shown in (30) at the top of next page.n pn ) ln 2 k=1 steps A)-C). g ({λk }.n k=1 (29) TABLE II Thus.. From (26). . the optimal power step a) is of order KN . for each given n. K}. K hΠ(n). . Note that step B)  +  iterates O(K 2 ) to converge [18]. we have To summarize. the 2) repeat minimization of g ({λk }.3 For each initialization N K step. . the above iterative algorithm to solve (P- wΠ(n) hID K PS) is given in Table II. II) For each initialization step: search over the user set {1. A subgradient of this B) repeat problem required for the ellipsoid method is provided by the a) Compute {pn } and {Π(n)} by (29) and (30) with given {λk } and µ.n X = + ζ λk hEH k. (28) the computation time is dominated by the ellipsoid method in ∂pn N (Γσ 2 + hID Π(n). Thus.    λk hEH P N µ−ζ ln 2   k. the time complexity of Thus. Considering further wΠ(n) Γσ2     p∗n (Π) = min  the number of initialization steps M . µ) with given {λk } and µ. the iterative algorithm is applied to obtain a local optimal solution for (P-PS).n } and {hk. b) Compute the subgradient of g({λk }. the time complexity of steps a)-c) is of order K 2 + KN . Initialize {λk > 0} and µ > 0. O(K 2 + KN ). solved by the ellipsoid method [18].n }.n − µ. D) Update {ρk } by (22) with . µ) by (31). In particular. in step B). for any given SC allocation Π(n). . we have where M is a sufficiently large number. ∂Ln Π(n). After obtaining g ({λk }. For the algorithm given in Table II. µ) over {λk } and µ can be efficiently A) Compute {hEH ID k. thus the time complexity of the ellipsoid method is O(K 4 + K 3 N ). Ppeak  . the following choice of d is a subgradient for C) until {λk } and µ converge to a prescribed accuracy. i. Note that (30) can be solved by exhaustive I) Randomly generate M feasible {ρk } as different initialization steps. µ). following proposition. (27) n=1 k=1 that achieves the maximum weighted sum-rate from all the solutions.3: For Problem (23) with a dual function c) Update {λk } and µ according to the ellipsoid method. the optimal SC allocation Π∗ (n) to I TERATIVE ALGORITHM FOR S OLVING P ROBLEM (P-PS) maximize Ln can be obtained. and step c) allocation for Problem (26) is obtained as is of order K 2 . Proposition 4.

µ): . fixed {pn } and {Π(n)}. g ({λk }.

.

N EH .

P .

3) until .

ζ hk.n pn − E k .

< δ. ∀k. where δ > 0 controls the  N .

P EH .

n=1 .

single-user case with N = 64 in Section III. Fig.n . For each user. . P P − ṗn . Then R(ρ) can be increased by optimizing Figs. since at each iteration we simply required harvested energy for the TS scheme in Fig.e. . Proof: The proof is similar as the proof of Proposition C. is guaranteed to be feasible at each iteration. the maximum rate is achieved when all the harvested energy constraints in (23) are tight. the above developed solution is also applicable for for different users are generated independently. The number of responding optimal value of Problem (23) as R(ρ). provided that 7.1. and thus is omitted. In Fig. be the same for all users. the iterative algorithm achievable rates by both TS and PS are less than the upper is guaranteed to converge to a local optimum of (P-PS) when bound. ∀k. E k = E.n = hk. However. ζ hk. 7. large number of initialization steps increases the randomly generate M feasible {ρk } as the initialization steps. which may not be suitable for real-time applications. The above procedure can be iterated until required harvested energy by different schemes with K = 4.2: The optimal solution for (P-UB) can be ob. ∀k. as the number of users increases. In addition. computation complexity.   n=1 where {ṗn } is the solution of the maximization problem (25) with given {λk } and µ. ρ = [ρ1 ... dk = n=1 (31) N III) Select the one that achieves the maximum weighted sum-rate from the M solutions.   algorithm accuracy. R(ρ) cannot be further improved. 6. To obtain a robust performance. 3 In general. Note that Problem (23) We assume Ppeak → ∞ in Fig.n = hk. k = K + 1. the number of initialization Note that the above local optimal solution depends on the steps needs to be increased to guarantee the robustness and optimality of choice of initial {ρk }. We provide simulation results under a practical system Remark 4. it is assumed that wk = 1. Thus. i. the optimal {pn } and {Π(n)} is considered.n ṗn − E k . and Ppeak = 4P/N in Fig. K. 8 shows the time ratio of the EH slot versus minimum the initial ρk ’s are feasible. it is observed that when E > 0. Define the cor. where initialization steps M is set to be 100. The minimum harvested energy is assumed to are obtained by (29) and (30). .e. i. sum-rate maximization For (P-PS) with given {ρk }. . the tight. k = 1. . decrease ρk ’s to make all the harvested energy constraints 6 with Ppeak → ∞. Problem (P-UB). For the TS scheme. 6 and 7 show the achievable rates versus the minimum ρk ’s by (22). Performance Comparison 4. . The channels Hence. respectively. ρK ]T . . ∀k. . n in Problem (23). setup. with given initial {ρk }. . we use the same parameters as the tained by setting hEH ID k. we the algorithm.

f. and Ppeak → ∞. N = 64. small portion of power needs to be split for energy harvesting. in which case the system PS. .1). This is because that the transmission time when the user is harvesting energy is not when the required harvested energy is sufficiently small. the achievable harvesting energy. 6. 8. 9. for the TS scheme. In Fig.n  −  wΠ(n) −  Π(n)  Π(n). the PS scheme. For sufficiently large E 100 (80 < E ≤ 255µW). 20 TS. the achievable rate decreases as E increases.9 100 0. This is because that for the case K = 2. 9 shows the achievable rates versus the number of users a multi-user OFDM-based SWIPT system. 7.Ppeak→ ∞ 0. the TS scheme still outperforms the PS scheme. Remark 4.4 40 0. In Fig. the achievable rate increases as the number of users increases. the TS scheme becomes worse than the PS scheme. for the single-user case K = 1. and the rate tends to E is less than 150µW (c. the total information transmission time 1 − αK+1 then UB. 7 with Ppeak = 4P/N . where K = 4. Fig.8%) than for larger E more power needs to be split for EH at each that with K = 1.6 Rate (Mbps) 60 0. however.Ppeak→ ∞ 0. it 80 is observed that when E is sufficiently large. Comparing the TS and PS schemes. for 80 < E ≤ 208µW. For the of the system is fixed. In particular. 40 8). Ppeak=4P/N achieves large achievable rate (larger than 70% of UB) while 0 0 50 100 150 200 250 300 each user harvests a reasonable value of energy (about 32% Minimum required harvested power (µW) to 84% of the maximum possible value). when E is larger than be saturated due to the bandwidth and the transmission power 150µW. Ppeak=4P/N However. rate by PS is larger than that by TS.n ln 2 k=1 120 1 0. 11    +  K   +  Γσ 2 µ − ζ λk hEH P w wΠ(n) hID k. receiver. It is also observed in Fig.2 20 TS. a multi-user OFDM-based SWIPT system. This is because that for Rate (Mbps) 60 a finite peak power constraint on each SC. 120 and the PS scheme may take the advantage of the frequency diversity by subcarrier allocation. it is observed that the achievable rate by TS is larger than that by PS.5 0. Achievable rates versus minimum required harvested energy in Fig. it is observed one of the user decodes information when the other user is that for sufficiently small E (0 ≤ E ≤ 80µW). only a utilized for information transmission. it is observed that for both TS and PS schemes. (30)  N ln 2 ID   Π(n)  N   K  hΠ(n). the achievable rate decreases as E increases.3 UB.Ppeak→ ∞ 0 0 0 50 100 150 200 250 300 0 50 100 150 200 250 300 Minimum required harvested power (µW) Minimum required harvested energy (µW) Fig.n   λk hEH P N Γσ 2 µ − ζ  k. 7. energy E k = E = 150µW and Ppeak = 4P/N . the TS scheme needs to schedule a nonzero EH slot to ensure all users harvest sufficient energy (see Fig. since achievable rate with K = 2 is much larger (about 32. N = 64. where K = 4.8 0.n   Π∗ (n) = arg max  k=1    log 2    . Ppeak=4P/N decreases and results in a degradation of achievable rate. Achievable rates versus minimum required harvested energy in for the TS scheme in Fig. as E becomes sufficiently large. Time ratio of the EH slot versus minimum required harvested energy Fig. and by different schemes under fixed minimum required harvested Ppeak = 4P/N .1 PS.7 Time ratio of the EH slot 80 0.

2013. the TS scheme when the harvested energy is sufficiently large. α̂k ) ≥ θf (q̇k. 3) α̇k = 0 and α̈k > 0: Similar as case 2). the weighted sum. July 2008. we have f (q̈k. α̇k ) = f (q̈k.n system can achieve the same rate as the conventional TDMA k=1 n=1 system.n . May 2013. . 1.” IEEE Trans..n qk. . the XX + (µ̂ − µ) P − q̇k. α̂k ) = 9 that for a general multiuser system with large K ≥ 2. µ. using PS becomes inefficient for large K. α̈k ). after power splitting also becomes larger (c. 61. α̇k )+ (1 − θ)f (q̈k. α̇k ) + (1 − Fig. and the convex beamforming for opportunistic energy harvesting. With qk.1 and power transmission. 2013.n 0 5 10 15 20 = θα̇k log2 1 + + Number of users Γσ 2 α̇k. Fig. αk ≥ 0. “A novel mode switching scheme utilizing random to prove that for all qk. 11. Two     transmission schemes are investigated. α̂k = θα̇k .n Γσ2 is a concave function of qk. α̈k ) for all qk.n and αk for αk > 0 [21]. Theory (ISIT). From the view of implementation. 5.n − E i  rate is maximized subject to a given set of harvested energy i=1 k6=i n=1 constraints as well as the peak and/or total transmission power K+1 N ! constraint. no. α̂k ) ≥ θf (q̇k. Ho. µ̂. α̂k ) ≥ θf (q̇k. µ̂.n .n (θq̇k. Our study suggests that. 1989-2001. {α̇k }. α̇k ) + (1 − θ)f (q̈k.. the f (q̇k. 4) α̇k = 0 and α̈k = 0: In this case.n 100 Since log2 1 + k. Ho. vol. and C. in practical [1] R.n q̇k.n . ν̂ ≥ 0. A PPENDIX A Commun.n . . α̈k ). and at the same time each user is still able to harvest K+1 ! X a reasonable value of energy. “Transporting information and energy simultaneously. “MIMO broadcasting for simultaneous wireless information and power transfer. Zhang. 2) α̇k > 0 and α̈k = 0: In this case.f. α̂k ) = θ(q̇k. we consider the following four 105 cases for αk . ν̂ based information transmission with TS applied at each re- = g ({λi }.n .n follows that its perspective αk log2 1 + Γσ is jointly Rate (Mbps) 2α 95 k concave in qk.n . we have f (q̂k.n .n . To prove the concavity of function f (qk. .n . and  hk. a nonzero EH slot may be needed. K. vol. vol.” IEEE Trans. i = 1. circuits the power splitter or switcher may introduce insertion 12. it   hk.1 This paper has studied the resource allocation optimization For any λ̂i ≥ 0. Therefore. pp.n . In both cases. and the OFDMA-based information transmission with   K K+1 N PS applied at each receiver. for the TS scheme. 2012. Ju and R.n ≥ 0 and αk ≥ 0. 3).n . we have f (q̂k. we have α̂k > 0.n . the portion From the above four cases. and is left for future work. it suffices 4. ν̂ ≥ L {q̇k.n ≥ 0. 1) α̇k >0 and α̈k >0: In this case. R. hence and thus f (qk.n }.n . [4] Z. This [2] L.n . Therefore.n . α̈k ). K.n . C ONCLUSION P ROOF OF P ROPOSITION 4. Varshney.n + (1 − θ)q̈k. α̈k ) = 0. 9.” issue is unaddressed in this paper. X XX + (λ̂i − λi ) ζ hi.4754-4767. 372-375.n q̈k.n ≥ 0. Moreover.” IEEE Wireless Commun. . namely. µ. the choice of d as given in scheme significantly. Inf. 12 110 θ ∈ (0. Achievable rates versus number of users.n .n .n . Letters. loss and degrade the performance of the two schemes. µ̂ ≥ 0. A PPENDIX B V.n  75 hk.n . α̇k ) + (1 − θ)f (q̈k. 1612-1616. and E k = E = 150µW. α̂k ) ≥ of energy discarded at the information receiver at each user θf (q̇k. Zhang. we have f (q̂k. we have f (q̂k. α̈k ). K. α̂k ) = θα̇k log2 1 + TS Γσ 2 θα̇k PS  (1 − θ)hk. Wireless Commun. we have for a multiuser OFDM-based downlink SWIPT system. g λ̂i . Zhang and C. IEEE Int.n . we have 90 (q̂k. [3] X. α̈k ) = 85 0. {λ̂i }. α̈k ). This in turn degrades the rate of the TS By the definition of subgradient. Zhou. “Robust beamforming for wireless information P ROOF OF L EMMA 4.n . Tao.n . no. the TS scheme is easier to R EFERENCES implement at the receiver side by simply switching between the two operations of EH and ID. ν). [5] H. Hence. α̂k ) = TS scheme outperforms the PS scheme. h qk. where Ppeak = 4P/N θ)f (q̈k.” in Proc. pp. Nov. the TDMA. pp. When the harvested energy + (ν̂ − ν) 1 − α̇k k=1 required at users is sufficiently large. which completes the proof. α̂k ) ≥ θ(q̇k.n )  80 f (q̂k. however. αk ) is concave. “Wireless information and power transfer: architecture design and rate-energy tradeoff. ν) ceiver. we have f (q̂k. α̇k ) + (1 − θ)(q̈k.n .n . the PS scheme may outperform (21) is indeed a subgradient for g ({λi }. R. αk ). to the fact that as the number of users increases. α̇k ) + (1 − θ)(q̈k.n . This is intuitively due θf (q̇k. no.n Γσ 2 θα̇k Thus. in Proc. Xiang and M. 1). f (q̂k.n .n . Apr. α̈k ) with less Communications and Networking Conference (WCNC).n q̇k. α̇k ) + (1 − θ)f (q̈k. Symp. pp.n . IEEE Wire- combination (q̂k.

9. pp. Krikidis. Wireless Commun. pp. Huang. Aug. Convex Optimization. [21] S.” in Proc.5640. Huang and E.” Stanford University. July 2013. on Commun. vol. vol. “Wireless information and power transfer: a dynamic power splitting approach. Huang and V. Symp.. 12. “Joint wireless information and energy transfer in a two-user MIMO interference channel. Schober. “Convex optimization II. Durrani.edu/class/ee364b/lectures. K. Lo. [13] S.. 2012. and K. Sep. 12. Zhang.” to appear in IEEE Trans. available on-line at http://www.” in Proc. Wireless Commun. no. C. July 2006.. and R. and K.stanford. IEEE International Conference on Acoustics. July 2006. Mohseni. Wireless Commun. “Dual methods for nonconvex spectrum optimization of multicarrier systems. Theory (ISIT). “Optimal resource allocation for OFDMA downlink systems. R. “Simultaneous information and power transfer for broadband wireless systems. 2013.” in Proc. “Shannon meets Tesla: wireless information and power transfer. no.” IEEE Trans. IEEE Int. no.” IEEE Trans. available on-line at arXiv:1207. Zhou. I. Lee. pp.” IEEE Trans. pp. [11] A. 2004.” IEEE Trans. K. “Wireless information and power transfer: energy efficiency optimization in OFDMA systems. vol. C. June 2013. May 2013. Liu. Ng. [9] S. Series A. Takac. Vandenberghe. 23. pp. 3990-4001. S. Commun. Xu. “Relaying protocols for wireless energy harvesting and information processing. pp.” available on-line at arXiv:1303. Liu. S.4006. Kennedy. 54. 2363-2367. Dec. Inf. [17] P. no. Timotheou. 1310- 1322. [15] K.” in Proc. Dec. 12. Zhang. Cioffi. [7] L. Liu. View publication stats . Lau.” IEEE Trans. 2013. X. 5972-5986. Boyd. Commun. vol.. 4788-4799.. Ottersten. Cambridge Uni- versity Press. Speech.” Mathematical Programming. vol. 13 [6] J. R. and B. Seong. modeling and deployment. and Signal Processing (ICASSP). E. Wireless Commun. and R. “Iteration complexity of randomized block- coordinate descent methods for minimizing a composite function. pp. 2013. N. vol. Inf. M. Yu and R. 288-300. 1394-1398.. 1. 4210-4221. Richtarik and M. Wireless Commun. and R. Theory (ISIT). 61. IEEE Int. L. 7.” IEEE Trans. IEEE Int. 8. [12] K. M. A. [16] D. Sep. Chua. 61. Zhang. Jan. “Enabling wireless power transfer in cellular networks: architecture. R. vol. Symp. June 2010.. Nasir. Conf. no. “Wireless information transfer with opportunistic energy harvesting. and J. 9. [20] W. Grover and A. 3622-3636. [14] P. (ICC). “Multiuser MISO beamforming for simultaneous wireless information and power transfer.” IEEE Trans. 2013. 12. pp. W.. Liu. [10] J. pp. and K. 7. no. 2013. A. Signal Process. G. Chua. “Opportunistic wireless energy harvesting in cognitive radio networks. Boyd and L.html [19] K. [18] S. Clerckx. “MISO interference channel with QoS and RF energy harvesting constraints. Sahai. no. Park and B. [8] L. Larsson. Zhang.