1

Modulation and Coding Design for Simultaneous WirelessInformation and Power Transfer

Jie Hu, Yizhe Zhao, Kun Yang

Abstract—In order to satisfy the power-thirst ofthe IoT devices and thus extend their lifespan, theradio frequency (RF) signal aided wireless powertransfer (WPT) is exploited for remotely charging.Carefully coordinating both the WPT and wirelessinformation transfer (WIT) yields an emerging re-search trend in simultaneous wireless informationand power transfer (SWIPT). However, the SWIPTsystem designed by assuming Gaussian distributedinput signals may suffer from a substantial per-formance degradation in practice, when the finitealphabetical input is considered. In this article, wewill provide a design guide of the coding controlledSWIPT and study the modulation design in both thesingle-user and multi-user SWIPT systems. We hopethat this guide may push the SWIPT a step closerfrom theory to practice.

Index Terms—Simultaneous Wireless Informationand Power Transfer (SWIPT), Wireless InformationTransfer (WIT), Wireless Power Transfer (WPT),Coding, Modulation, Finite Alphabet

I. Introduction

In the imminent Internet of Things (IoT)era, massive number of sensors and IoT de-vices will be deployed for various applications.These miniature devices are normally pow-ered by embedded batteries. Frequent energy-consuming operations may quickly drain theirbatteries. Due to its high flexibility and lowinvestment on infrastructure, the carefully con-trolled wireless power transfer (WPT) relyingon the radio frequency (RF) signals [1] canbe invoked for remotely charging these low-power devices. Coordinating wireless informa-tion transfer (WIT) and WPT in the sameRF spectral band thus yields the research ofsimultaneous wireless information and powertransfer (SWIPT) [1]–[6].

With the aid of the SWIPT, the sensorsmay successfully recover the instruction infor-mation from the downlink RF signal emittedby the gateway, while harvesting a portionof the RF signal’s energy for powering their

All the authors are with the School of Information andCommunication Engineering, University of Electronic Sci-ence and Technology of China, Chengdu, 611731, China.Email: hujie@uestc.edu.cn, yzzhao@std.uestc.edu.cn, kun-yang@uestc.edu.cn.

The authors would like to acknowledge the financial supportof National Natural Science Foundation of China (NSFC) NO.U1705263.

Jie Hu would also like to acknowledge the financial supportof GF Innovative Research Program.

own operations [1]. Furthermore, the SWIPTis also suitable to be applied in the two-hop cooperative communication. The transmit-ter initially transmits the modulated RF signalsto the relays. With the aid of the harvest-then-cooperate and the harvest-store-cooperateprotocols, a portion of energy carried by themoddulated RF signals can be harvested bythe relay stations for powering their cooperativeinformation forwarding to the remote users [2].The benefit of the wireless powered relay istwo-fold: Firstly, no extra energy is consumedby the relay, which encourage the cooperationbetween users. Secondly, the deployment of therelay stations becomes more flexible, since theydo not have to be connected to the power grid.

Inspired by these promising applications,intense efforts have been invested in theSWIPT recently. Chen et al. [3] maximisedthe harvested power by optimally designing thetransmit covariance matrix in a point-to-pointmultiple-input-multiple-output (MIMO) aidedSWIPT system. Lv et al. [4] proposed an op-timal time-domain resource allocation schemeamong the multiple receivers for controlling theSWIPT in the downlink and for maximisingthe sum- and fair-throughput in the uplink.However, either as the constraint [3] or as theobjective [4], the attainable WIT throughputis always evaluated by the classic Shannon-Hartely channel capacity, which is achievedby assuming the infinite Gaussian distributedinput. By contrast, in any of the practical com-munication systems, only finite alphabet canbe transmitted due to the modulation schemeshaving limited constellation points and the cod-ing scheme having limited codewords. As aresult, the attainable WIT throughput can onlybe evaluated by the discrete-input-continuous-output mutual information. Therefore, Kim etal. [5] firstly studied the rate-energy trade-offin a SWIPT system with equi-probable finitealphabetic input, while Zhu et al. further ex-tended their study to a MIMO-SWIPT system[6]. However, the impact of the modulationdesign on the SWIPT performance is over-looked in [5], [6], especially when a practicalWPT receiver is invoked. In another hand, afterVarshney analysed the SWIPT from the infor-mation theoretical perspective in his seminalwork [7], surprisingly, few efforts have been in-vested in continually exploring the informationtheoretical limit and in designing the resultantcoding design for the SWIPT system.

January 25, 2019 DRAFT

Fig. 1: The tranceiver architecture with a SWIPT receiver.

In this article, we would like to highlight theimportance of both the coding and modulationdesign in the SWIPT system by introducingtheir theoretical fundamentals and by providingthe brief design guides. The main contributionsare summarised as follows:• The popular transceiver architecture of the

SWIPT is introduced in Section II.• A design guide of the coding controlled

SWIPT is provided by considering the bat-tery state at the receiver in Section III. Thecase studies of applying the unary codeand the run-length-limited (RLL) code inthe SWIPT is provided for illustrating thetrade-off between the attainable WIT andWPT performance.

• The impact of the wireless channels andthe hardware constraint on the practicalmodulation design in the SWIPT systemis studied for a single user scenario inSection IV, while the principle of the mod-ulation design in multi-user SWIPT sys-tem relying on the superposition symbolsis also introduced in Section V.

• Open problems concerning the modulationand coding design in the SWIPT systemare envisioned in Section VI.

II. Transceiver ArchitectureIn a SWIPT system, we may have SWIPT

users, which extract information and energyfrom the same RF signals. We may also havededicated WIT users and WPT users, whichreceives information and power requested, re-spectively by exploiting the broadcast nature ofthe wireless channel.

A. SWIPT TranscieverA typical SWIPT transmitter is illustrated in

the top half of Fig.1, which consists of the en-ergy source, the information source, the sourceand channel encoder, the digital modulator aswell as the transmit beamformer. The energysource powers the other functional modulesof the transmitter. The power allocated to thedigital modulator and the transmit beamformer

Fig. 2: Coding controlled SWIPT. A pair of codewords‘10101’ and ‘00100’ carry different amount of energy to theWPT receiver for charging its battery. All these codewordsare modulated by the OOK.

constitutes the actual transmit power carried bythe RF signal.

A typical SWIPT receiver is portrayed inthe bottom half of Fig.1, which is constitutedby the receive beamformer, the signal splitter,the WPT receiver and the WIT receiver. Ei-ther the power-splitter or the time-switcher isexploited for splitting the received RF signalinto two portions [8]. A portion of the receivedRF signal flows into the WIT receiver forthe demodulation and decoding. The recoveredinformation bits finally arrive at the informationdestination. The other portion of the receivedRF signal is converted by a rectifier to thedirect current (DC) and it is finally stored in thebattery. The rectifier and the battery constitutea typical WPT receiver.

The current research [1]–[6] mainly focus onthe design of the front ends of the transceiverby only considering the continuous signals,while largely overlooking the impact of thediscrete messages and symbols induced by thecoding and modulation of the SWIPT system,as illustrated in Fig.1.

B. Non-Linear RectifierAs portrayed in Fig.1, a simplified rectifier

consists of a diode and a low-pass-filter. Therectifier’s non-linearity exhibits in the follow-ing perspectives:• The circuit of the rectifier can only be

activated, when the input power of the RFsignal is higher than a threshold [9]. Thisactivation threshold is also regarded as thesensitivity of the rectifier.

• The output DC of the rectifier in Fig.1(c)can be formulated as a polynomial of theinput RF signal’s power [10], which indi-cates that a higher input power may resultsin a higher RF-DC conversion efficiency.

These characteristics of the rectifier should betaken into account in the coding and modula-tion design of the SWIPT system.

III. Coding Controlled SWIPTA. Fundamental

The landmark work [7] has firstly maximisedthe mutual information of the discrete-input-

2

discrete-output channel by optimising the dis-tribution of the discrete input messages, whileensuring that the energy carried by the outputmessages is higher than a pre-defined threshold.The energy carried by a binary codeword isjointly decided by the following factors:• The percentage of bit ‘1’ and that of bit ‘0’

in a binary codeword, which is regarded asthe structure of the codeword.

• The mapping from the binary bits to themodulated symbols. For example, the bi-nary sequence ‘1111’ is mapped to thesymbol carrying the highest energy in the16-QAM.

In this section, we mainly focus on the de-sign of the codeword structure by adopting theon-off-keying (OOK) based modulation, whereonly the binary bit ‘1’ carries a single unit ofenergy, as portrayed in Fig.2.

There is an obvious trade-off between theWPT and the WIT performance. For example,if the channel input always generates an all-one binary sequence, the mutual informationis zero, although the maximum energy can betransferred. If the channel input is optimisedfor only maximising the mutual information,the energy carried by the channel output isa certain value, which might not satisfy theenergy request of the receiver.

The basic principle is to generate the code-words with a certain structure in order to sim-ulataneously satisfy the certain requirementsof both the WIT and WPT performance. Thefollowing coding schemes can be opted forreacing this design target [11]:• Compensation Energy Coding. Dummy bi-

nary bits are directly concatenated behindthe information bits in order to guaranteethat the resultant codeword has a certainpercentage of bit ‘1’. This coding ap-proach has the lowest complexity. How-ever, the dummy bits do not carry anyinformation, which may thus degrade theWIT performance.

• Inverse Source Coding. A classic sourceencoder takes non-equi-probable messagesto generate the binary sequence havingequi-probable binary bits. By contrast, aninverse source encoder takes equi-probablemessages to generate the binary sequencehaving a certain structure for satisfyingthe WPT requirement. However, the asyn-chronization between the encoder and thedecoder imposes difficulties in the efficientdecoding design.

• Constraint Coding. Some constraint cod-ing techniques have degrees of freedom tochange the codeword structure for satisfy-ing the WPT requirement. Since they donot include any dummy bits, the WIT per-formance may not suffer significant degra-

dation. Furthermore, the efficient symbol-level trellis can be adopted for decodingthe constraint code.

We will then introduce a pair of typical con-straint codes, namely the unary code and theRLL code.

B. Unary CodeThe unary encoder maps the j-th input bi-

nary sequence onto a j-bit codeword, whichhas a single bit ‘0’ at the end and all theother bits in front are ‘1’. For instance, the4-level unary encoder is capable of encodingfour different binary sequences {00, 01, 10, 11}.The first input sequence ‘00’ is encoded as acodeword ‘0’, while the fourth input sequence‘11’ is thus encoded as a codeword ‘1110’.Obviously, different input binary sequence maybe encoded as a codeword having differentpercentage of energy bit ‘1’. Therefore, theaverage percentage of energy bit 1 in unarycodewords can be adjusted by changing theoccurrence probabilities of the input binarysequences, which hence controls the SWIPTperformance of the codewords.

C. Run-Length-Limited CodeAnother constraint coding technique is the

RLL code [12]. A type-0 (d, k)-RLL code hasthe following constraints on a codeword:

• The runs of bit ‘0’ have a length of d atleast between successive bit ‘1’.

• The runs of bit ‘0’ have a length of k atmost between successive bit ‘1’.

The run-length of bit ‘0’ may be an ar-bitrary value between d and k. For in-stance, a type-0 (1, 3) RLL code is ca-pable of generating a binary bit sequenceof ‘10100010010001001 · · · ’, where the mini-mum run-length of bit ‘0’ is 1 and its maximumrun-length is 3. Obviously, the average percent-age of energy bit ‘1’ in type-0 RLL codewordis determined by the occurrence probabilitiesof the runs of bit ‘0’ having different lengths.For instance, if a type-0 (1, 3) RLL encoderincreases the occurrence probability of the runsof bit ‘0’ having a length of 1, the averagepercentage of energy bit ‘1’ can be thus in-creased. Therefore, by adjusting the occurrenceprobabilities of the runs of bit ‘0’ havingdifferent lengths, we may control the SWIPTperformance of the type-0 RLL codewords.For a type-1 RLL encoder, we should focuson adjusting the occurrence probabilities ofthe runs of bit ‘1’ having different lengthsin order to control its corresponding SWIPTperformance.

3

0 0.1 0.2 0.3 0.4Reqiured Information Rate (bit/channel use)

10-3

10-2

10-1

100

Batte

ry O

verf

low

Pro

babi

lity

p of

2-level Unary Code(0,1)-RLL code(1,2)-RLL code(2,3)-RLL code

0 0.1 0.2 0.3 0.4Reqiured Information Rate(bit/channel use)

10-3

10-2

10-1

100

Batte

ry U

derf

low

Pro

babi

lity

p uf

Fig. 3: The battery overflow/underflow probability versusthe minimum required information rate.

D. Battery-Aware DesignSince the battery has a finite capacity, the

energy carried by a codeword should not betoo high in order to avoid the energy wasteinduced by the battery overflow. By contrast,the energy carried by a codeword should not betoo low in order to avoid the energy shortageinduced by the battery underflow, which mayimpair the routine operation of the receiver. Theenergy stored in the battery of the receiver canbe modelled by a discrete queuing process.

By considering the routine energy consump-tion of the receiver, when a type-0 (or type-1)RLL encoder is adopted for SWIPT, the energyqueue is renewed for the duration of a completerun of bit ‘0’ (or bit ‘1’). The code design thenaims for minimising either the battery over-flow probability or its underflow probabilityby optimising the run-length constraints andthe occurrence probabilities of the runs of bit‘0’ (or bit ‘1’) having different lengths, whilesatisfying the minimum requirement of the in-formation transmission. By contrast, when theunary encoder is adopted, the energy queue isrenewed for the duration of a unary codeword.The code design then aims for optimising thelevel parameter and the occurrence probabili-ties of the input binary sequences for achievinga satisfactory SWIPT performance.

The SWIPT performance of the RLL codeand the unary code is investigated in Fig.3. Weassume that the information decoding may notconsume any energy carried by the receivedsignal, which is then exploited for chargingthe battery. For each renewal interval, thereceiver consumes a single energy unit witha probability of 0.5. The maximum capacityof the battery is 2 energy unit. A classic Zchannel is invoked, where bit ‘0’ can alwaysbe successfully received, while bit ‘1’ may beerroneously received as ‘0’ with a probabilityof 0.2 and it may be correctly received witha probability of 0.8. Observe from Fig.3 thattype-0 (0, 1)-RLL code has the highest densityof energy bit ‘1’. Therefore, it may achieve

Fig. 4: The input constellations at the WPT receiver. All theconstellations have the same average power. The activationthreshold of the WPT receiver is denoted as Pth.

the highest battery overflow probability andthe lowest battery underflow probability. Bycontrast, 2-level unary code has the highestbattery underflow probability, since it has thelowest density of energy bit ‘1’.

IV. SWIPT WithModulation: Single-User

A. FundamentalIn a practical communication system, the

coded information is then modulated by a spe-cific symbol from a finite alphabet. Therefore,the mutual information of the M-QAM finallyconverges to a constant, as the transmit powercontinuously increases. For example, as weincrease the transmit power of the modulatedRF signal, the mutual information of the 16-QAM converges to 4 bit/channel use, while thatof the 256-QAM converges to 8 bit/channeluse. Hence, more transmit power should beallocated for the WPT purpose, when the WITperformance converges.

Different modulation schemes exhibit di-verse WPT performance, when the non-linearity of the rectifier is considered. Weexemplify the received constellations of 16-phase-shift-keying (16-PSK), 16-QAM and16-pulse-amplitude-modulation (16-PAM) inFig.4. If the rectifier can only be activated bythe received power higher than its activationthreshold, which is illustrated by the red cir-cle/rectangle in Fig.4(a), we observe that allthe symbols of 16-PSK have been filtered bythe rectifier. As a result, the energy carried bythe 16-PSK symbols cannot be harvested bythe rectifier. In the case of 16-QAM, although12 symbols are filtered by the rectifier, westill have 4 symbols capable of delivering theenergy to the WPT receiver. Furthermore, 16-PAM performs best in terms of the WPT, sinceit still has 8 symbols capable of deliveringthe energy. If the rectifier threshold is lower

4

than the average power of these modulationschemes, 16-PSK may have the best WPTperformance. Furthermore, a higher order mod-ulation scheme has a better WPT performance,since it has more symbols carrying higher en-ergy. Note that when we consider the non-linearRF-DC conversion efficiency of Section II-B,the modulation schemes may exhibit a similartrend in terms of the WPT performance, sincea higher input power may result in a higherRF-DC conversion efficiency.

Furthermore, the ‘adverse’ effect of the wire-less channel on the WIT may actually improvethe WPT performance. If the scattering of achannel becomes more severe, the multi-pathpropagation may have a chance to construc-tively strengthen the signal at the receiver. As aresult, the WPT performance can be improved.Furthermore, additional interference is also ca-pable of improving the WPT performance, al-though it may impair the WIT performance.

B. MIMO aided ModulationThe beamformer/precoder design of the

MIMO-SWIPT system has been studied byconsidering the discrete-input-continuous-output mutual information with finite alphabet[13], where the spatial multiplexing gain isexploited for realising the SWIPT.

Furthermore, the implementation of multipleantennas is capable of facilitating the modu-lation in the spatial dimension. A specific an-tenna (or a subset of antennas) can be activatedfor transmitting a specific information symbolby exploiting the information driven antenna-switching mechanism, which is regarded as thespatial-modulation (SM) or space-shift-keying(SSK). The difference of the channel responseimpulse is relied upon for identifying the trans-mit antenna at the receiver for the demodula-tion. The SM/SSK may substantially reduce thenumber of RF chains in order to increase theenergy efficiency.

Since the SM/SSK system relies on the ac-tivation of limited number of antennas for theWIT, the rest of idle antennas can be exploitedfor gleaning the energy of the ambient RFsignals and recycling the energy transmitted bythe activated antennas. As a result, the energyefficiency of the SM/SSK system can be furtherincreased.

Furthermore, when we activate a transmitantenna for transmitting a specific informationsymbol, we may simultaneously activate anadditional antenna for the WPT. Since the acti-vation of the additional antenna may certainlydeteriorate the WIT performance, the WPToriented antenna activation scheme should becarefully designed. For example, the WPTchannel should have a huge difference responseimpulse with the WIT channel. As a result, the

receiver can identify the WIT channel for thedemodulation by firstly cancelling the interfer-ence from the WPT channel.

C. Hardware ConstraintIn order to improve the WPT performance,

a high-order modulation scheme has to beadopted. However, it may impose great chal-lenges on both the transmitter and the receiver:• A high-order modulation scheme normally

has a high peak-to-average-power-ratio(PAPR). For example, given the same av-erage power, the PAPR of 256-QAM is 25

17times higher than 16-QAM. The transmit-ter thus requires a power amplifier havinga very large linear region in order to avoidthe energy leakage. As a result, the charac-teristic of the practical power amplifier hasto be considered in the modulation design.

• For the demodulation of the high ordermodulated symbol, the receiver requiresaccurate channel state information (CSI)for carrying out the coherent detection. Inorder to avoid the energy consumption inthe CSI acquisition, the non-coherent de-tection based differential modulation [14]can be adopted by the IoT devices in theSWIPT system. However, the differentialmodulation may sacrifice both the WITand WPT performance to some extent,since high order differential modulation isstill a technical blank in the literature.

Furthermore, when we have to pack a largenumber of antennas in a limited area, the fol-lowing pair of hardware constraints has to beconsidered:• Uncorrelated spacing among the large

number of antennas cannot be guaranteedin a practical system. The modulated sym-bols can be further constructively com-bined by exploiting the antenna correla-tion both at the transmitter and at thereceiver. By considering the non-linearityof the rectifier, high antenna correlationmay result in a high WPT performance.The antenna correlation should also beconsidered in the SM/SSK-MIMO aidedSWIPT system.

• As the number of antennas becomes large,it is impossible for feeding each antennawith a single RF chain. As a result, wemay process the modulated symbols se-quentially in the digital domain and inthe analog domain, before they finallytransmitted by antennas. By optimisingthis hybrid beamformer, we may max-imise the discrete-input-continuous-outputmutual information, while ensuring therequired WPT performance.

5

Fig. 5: Conventional modulation and constellation rotationbased modulation for the SWIPT. The symbols requested bythe WIT user pair are denoted by the stars. The Euclideandistance between the superposition symbol and the originis the square root of the actual transmit power.

V. SWIPT WithModulation: Multi-UserA. Superposition Symbol

In order to further improve the spectral ef-ficiency, the symbols requested by multipleusers may be superimposed with each otherin the same resource block. For example,in the power-domain non-orthogonal-multiple-access (NOMA), the symbols destined to dif-ferent users are differentiated by their transmitpower. Then, the superposition symbol canbe demodulated by exploiting the successive-interference-cancellation (SIC). In the sparse-code-multiple-access (SCMA), a symbol is de-composed for modulating onto different sub-carriers. The reconstructed superposition sym-bol on a specific sub-carrier can be demod-ulated by the message-passing-algorithm. Inthe network coded cooperative network, thesymbols of different users are superimposedat the hub, which is then broadcast to theusers and demodulated. The energy carried bythe superposition symbols can be harvestedby WPT users due to the broadcast nature ofwireless channels.

In all the above-mentioned scenarios, weshould constructively superimpose the symbolsdestined to different WIT users in order toachieve the required WPT performance of theWPT users in the SWIPT system. In the follow-ing example, we consider a pair of dedicatedWIT users and a single WPT user.

B. Constellation RotationIn the conventional signal superposition, as

exemplified in the top half of Fig.5, the sym-bols requested by this WIT user pair are de-structively combined. The resultant superposi-tion symbol suffers from a substantial energyloss, when compared to the original symbols.The WPT performance is thus significantlydegraded. If we rotate the WIT users’ con-stellation for a certain angle, as exemplified in

the bottom half of Fig.5, the energy carried bythe superposition symbol can be increased inorder to satisfy the WPT requirement. How-ever, the constellation rotation may result inthe reduction of the minimum Euclidean dis-tance between adjacent superposition symbols,as portrayed in Fig.5, which may deterioratethe WIT performance. An optimal constellationrotation angle should be chosen in order toachieve a balance between the WPT and theWIT performance [15].

The optimal scheme is to design the constel-lation rotation angles for every pair of symbolsrequested by WIT User A and B. However,this may impose unaffordable control overheadon the system, since the WIT users have toacquire the knowledge of the rotation anglesfor demodulating the superposition symbol. Inorder to reduce the control overhead, the con-stellation rotation angles can be designed for apair of symbol blocks requested by WIT UsersA and B, respectively, which are constitutedby the same length of symbols. As a result,all the symbols in the block requested byWIT user A are rotated by an identical angle,while those in the block requested by WITuser B are rotated by another identical angle.After the rotation, the symbols in these blocksare superimposed pairwise. Therefore, a newsymbol block consisting of the superpositionsymbols is generated and broadcasted in thewireless channel. The constellation rotation an-gles should be optimised in order to maximisethe total power carried by the superpositionsymbol block, while satisfying a specific WITconstraint. For example, if the maximum like-lihood based multi-user detector is invokedby the WIT users, the minimum Euclideandistance of the superposition constellation afterthe optimal rotation should be higher than apredefined threshold.

C. ComparisonWe demonstrate in Fig.6 that our

constellation-rotation aided modulation designoutperforms its conventional counterpart interms of the WPT performance in most cases,where the classic AWGN channels having thepath loss of 30 dB are invoked for all the users.Moreover, the average transmit power for WITUser A’s and WIT User B’s transmissionsare 1 Watt and 0.1 Watt, respectively. Thesymbol duration is 10−6 second. Specifically,when the activation threshold Pth of therectifier is 1.2 mW, our constellation rotationaided 4-QAM (QPSK) performs worse thanthe conventional approach. This is becausethe constellation rotation may produce morelow-power symbols, which cannot activatethe rectifier. Hence, the WPT performance isdegraded. Furthermore, when we increase the

6

1 2 5 10 20 25 50 100

The Size of the Data Block (L) ( 104)

1

2

3

4

5

6

7

8

9

10

11

Tota

l Har

vest

ed E

nerg

y (J

)

10-4

4QAM with CR16QAM with CR64QAM with CR

4QAM without CR16QAM without CR64QAM without CR

Pth=0.6mW

Pth=1.2mW

Fig. 6: A case study of the constellation rotation (CR) aidedmodulation desgin in a SWIPT system having a pair of WITusers and a single WPT user.

size of the symbol block, observe from Fig.6that the WPT performance of the constellation-rotation aided scheme reduces. This remindsus that we should also strike a trade-offbetween the control signalling overheadand the attainable system performance. Theperformance gap between the different-ordermodulation schemes can be explained in asimilar way as we have done in Section IV-A.

VI. Future Challenges and Conclusions

The following open problems still need ourfurther investigation:

Concatenated Code: A concatenated encoderconsisting of a source encoder, channel encoderand an energy encoder should be carefullydesigned, while a powerful iterative decoder isalso required for processing the sophisticatedconcatenated codewords.

Coded Modulation: The bit-to-symbol map-ping from the binary bits to the modulatedsymbols has to be designed by jointly consider-ing the codeword structure and the modulationcharacteristic in order to satisfy both the WITand WPT requirements.

Adaptive Modulation: In order to exploit thedistinctive WPT and WIT features of a spe-cific modulation scheme, we should design anadaptive modulation scheme by considering thewireless channel characteristics, the non-linearrectifier and the diverse SWIPT requirements.

This article aims for introducing the funda-mental of the coding controlled SWIPT and themodulation design of the signal-user and multi-user SWITP system and for inspiring moreendeavour invested in this promising topic.

References

[1] G. Pan and et al, “Performance Analysis and Optimiza-tion for SWIPT Wireless Sensor Networks,” IEEE Trans.on Comm., vol. 65, no. 5, May 2017, pp. 2291–2302.

[2] H. Chen and et al., “Harvest-Then-Cooperate: Wireless-Powered Cooperative Communications,” IEEE Trans. onSign. Proc., vol. 63, no. 7, April 2015, pp. 1700–1711.

[3] Z. Chen and et al., “Joint Transceiver Optimization ofMIMO SWIPT Systems for Harvested Power Maximiza-tion,” IEEE Sign. Proc. Lett., vol. 24, no. 10, Oct. 2017,pp. 1557–1561.

[4] K. Lv and et al., “Throughput Maximization and FairnessAssurance in Data and Energy Integrated CommunicationNetworks,” IEEE Internet of Things Journal, vol. 5, no. 2,April 2018, pp. 636–644.

[5] I.-M. Kim and et al., “Wireless Information andPower Transfer: Rate-Energy Tradeoff for Equi-ProbableArbitrary-Shaped Discrete Inputs,” IEEE Trans. on Wire-less Comm., vol. 15, no. 6, June 2016, pp. 4393–4407.

[6] X. Zhu and et al., “Precoder Design for SimultaneousWireless Information and Power Transfer Systems WithFinite-Alphabet Inputs,” IEEE Trans. on Vehi. Tech.,vol. 66, no. 10, Oct. 2017, pp. 9085–9097.

[7] L. R. Varshney, “Transporting Information and EnergySimultaneously,” in 2008 IEEE ISIT, July 2008, pp.1612–1616.

[8] J. Hu and et al, “Integrated data and energy communi-cation network: A comprehensive survey,” IEEE Comm.Surveys Tutorials, vol. 20, no. 4, Fourthquarter 2018, pp.3169–3219.

[9] S. D. Assimonis and et al., “Sensitive and EfficientRF Harvesting Supply for Batteryless Backscatter Sen-sor Networks,” IEEE Trans. on Micr. Theo. and Tech.,vol. 64, no. 4, April 2016, pp. 1327–1338.

[10] B. Clerckx and et al., “Waveform Design for WirelessPower Transfer,” IEEE Trans. on Sign. Proc., vol. 64,no. 23, Dec. 2016, pp. 6313–6328.

[11] Z. Babar and et al., “Unary-Coded Dimming Control Im-proves ON-OFF Keying Visible Light Communication,”IEEE Trans. on Comm., vol. 66, no. 1, Jan. 2018, pp.255–264.

[12] A. M. Fouladgar and et al., “Constrained Codes forJoint Energy and Information Transfer,” IEEE Trans. onComm., vol. 62, no. 6, June 2014, pp. 2121–2131.

[13] Z. Ding and et al., “Application of Smart AntennaTechnologies in Simultaneous Wireless Information andPower Transfer,” IEEE Comm. Mag., vol. 53, no. 4, April2015, pp. 86–93.

[14] G. Kaddoum and et al., “Design of Simultaneous Wire-less Information and Power Transfer Scheme for ShortReference DCSK Communication Systems,” IEEE Trans.on Comm., vol. 65, no. 1, Jan. 2017, pp. 431–443.

[15] Y. Zhao and et al., “Constellation rotation aided modula-tion design for the multi-user swipt-noma,” in 2018 IEEEICC, May 2018, pp. 1–6.

Jie Hu [S’11, M’16] (hujie@uestc.edu.cn) received the Ph.D.degree from the Faculty of Physical Sciences and Engineering,University of Southampton, U.K., in 2015. Since March 2016,he has been working with the School of Information andCommunication Engineering, University of Electronic Scienceand Technology of China (UESTC), China, as an AssociateProfessor. His research interests include energy harvesting,wireless power transfer, social networking and their applica-tions in wireless communication and networking.

Yizhe Zhao [S’16] (yzzhao@std.uestc.edu.cn) received his BSdegree from the School of Communication Engineering inXidian University, in 2014, and MSc degree from the Schoolof Communication and Information Engineering in Universityof Electronic Science and Technology of China (UESTC),in 2017. He is currently a PhD candidate in the School ofInformation and Communication Engineering, UESTC. Hisresearch interests include wireless networks as well as wirelessinformation and energy transfer.

Kun Yang [M’00, SM’07] (kunyang@uestc.edu.cn) receivedhis PhD from the Department of Electronic & ElectricalEngineering of University College London, UK. He is currentlyan affiliated professor at UESTC, China. His research interestsinclude wireless networks, future Internet technology and net-work virtualization, mobile cloud computing and networking.

7

Fig. 1: The tranceiver architecture with a SWIPT receiver.

8

Fig. 2: Coding controlled SWIPT. A pair of codewords ‘10101’ and ‘00100’ carry differentamount of energy to the WPT receiver for charging its battery. All these codewords are modulatedby the OOK.

9

0 0.1 0.2 0.3 0.4Reqiured Information Rate (bit/channel use)

10-3

10-2

10-1

100

Batte

ry O

verf

low

Pro

babi

lity

p of

2-level Unary Code(0,1)-RLL code(1,2)-RLL code(2,3)-RLL code

0 0.1 0.2 0.3 0.4Reqiured Information Rate(bit/channel use)

10-3

10-2

10-1

100

Batte

ry U

derf

low

Pro

babi

lity

p uf

Fig. 3: The battery overflow/underflow probability versus the minimum required informationrate.

10

Fig. 4: The input constellations at the WPT receiver. All the constellations have the same averagepower. The activation threshold of the WPT receiver is denoted as Pth.

11

Fig. 5: Conventional modulation and constellation rotation based modulation for the SWIPT.The symbols requested by the WIT user pair are denoted by the stars. The Euclidean distancebetween the superposition symbol and the origin is the square root of the actual transmit power.

12

1 2 5 10 20 25 50 100

The Size of the Data Block (L) ( 104)

1

2

3

4

5

6

7

8

9

10

11

Tota

l Har

vest

ed E

nerg

y (J

)

10-4

4QAM with CR16QAM with CR64QAM with CR

4QAM without CR16QAM without CR64QAM without CR

Pth=0.6mW

Pth=1.2mW

Fig. 6: A case study of the constellation rotation (CR) aided modulation desgin in a SWIPTsystem having a pair of WIT users and a single WPT user.

13