Modulation Design for Wireless Information and

Power Transfer with Nonlinear Energy Harvester

Modeling
Ekaterina Bayguzina and Bruno Clerckx
Department of Electrical and Electronic Engineering, Imperial College London, London, U.K.
Email: {ekaterina.bayguzina08, b.clerckx}@imperial.ac.uk
arXiv:1802.06512v1 [cs.IT] 19 Feb 2018

Abstract—Far-field wireless power transfer (WPT) and simul- Moreover, most of the existing works in the SWIPT literature
taneous wireless information and power transfer (SWIPT) have are based on the linear model of the energy harvester (EH) [5]-
become increasingly important in radio frequency (RF) and com- [9]. Yet, the linear EH model is limited in terms of accuracy,
munication communities recently. The problem of modulation
design for SWIPT has however been scarcely addressed. In this and systematic waveform design utilizing the nonlinear model
paper, a modulation scheme based on asymmetric phase-shift allows to achieve a higher ECE (and overall end-to-end
keying (PSK) is considered, which improves the SWIPT rate- power transfer efficiency) in a practical rectenna circuit with
energy tradeoff region significantly. The nonlinear rectifier model, the use of multiple sinewaves [10]-[13]. Leveraging [10],
which accurately models the energy harvester, is adopted for eval- SWIPT signal design accounting for the rectifier nonlinearity
uating the output direct current (DC) power at the receiver. The
harvested DC power is maximized under an average power con- was studied in [14]. It was concluded in [14] that the
straint at the transmitter and a constraint on the rate of informa- rectifier nonlinearity radically changes the design of SWIPT.
tion transmitted via a multi-carrier signal over a flat fading chan- Indeed “it favours a different waveform, modulation, input
nel. As a consequence of the rectifier nonlinearity, this work high- distribution and transceiver architecture as well as a different
lights that asymmetric PSK modulation provides benefits over use of the RF spectrum. Exploiting the rectifier nonlinearity
conventional symmetric PSK modulation in SWIPT and opens
the door to systematic modulation design tailored for SWIPT. in the SWIPT design also makes a more efficient use of the
Index Terms—SWIPT, nonlinear model, rectenna, energy har- resources by enabling enlarged rate-energy regions compared
vesting, modulation. to those obtained by ignoring the nonlinearity in the system
design.” Motivated by those observations, the optimal input
I. I NTRODUCTION
distribution of SWIPT subject to nonlinear power constraints
Radio frequency (RF) wireless power transfer (WPT) has was studied in [15]-[16]. Remarkably, it was shown that the
become more important recently with the advent of mainly capacity of an additive white Gaussian noise (AWGN) channel
battery free wireless sensor networks and Internet of Things. under transmit average power and nonlinear delivered power
At the same time, new wireless communications systems constraints is the same as the capacity of an AWGN channel
enabling simultaneous wireless information and power transfer under an average power constraint. In other words, the
(SWIPT) are widely investigated in the research community. capacity of an AWGN channel is independent of the value of
With regard to practical modulation schemes, their effect the delivered power constraint. The capacity can be arbitrarily
on the energy conversion efficiency (ECE) as compared to approached by using time sharing between distributions
continuous wave (CW) transmission has been considered by with high amount of information, e.g. circularly symmetric
the RF community [1]-[4]. In [1], FSK modulated signals are complex Gaussian (CSCG) inputs, and distributions with high
shown to be detrimental in terms of harvested power due to the amount of power reminiscent of flash signaling exhibiting a
mixing operation of the diode-based rectifier. In [2], amplitude low probability of high amplitude signals.
and phase variations introduced by QPSK and 16QAM are In this paper, motivated by the above observations, we
demonstrated to decrease the ECE as compared to the CW. design modulation for SWIPT with nonlinear energy harvester
However, it is established in [3] that these modulation schemes based on finite constellations. We study modulation of infor-
can improve the ECE at low input powers. In [4], it is shown mation symbols onto the multi-carrier energy-carrying wave-
that the ECE, achieved via multitone and modulated signals, form, resulting in a unified SWIPT waveform. We adapt PSK
depends on the circuit optimization, mainly on the matching modulation to SWIPT requirements and propose asymmetric
network and the load. Accordingly, such signals can enhance PSK modulation, which allows to gain the maximum output
the ECE for certain load values and input power levels in energy while maintaining the required information rate.
relation to a CW signal with the same average power. Notation: E{·} refers to the DC component of a signal.
However, modulation schemes designed specifically for E{·} refers to the expectation over statistical randomness.
the purpose of SWIPT have scarcely been addressed [5]. V ar{·} refers to the variance of a random variable. U [−a, a]
This work was supported in part by the EPSRC of U.K. under Grants denotes the uniform distribution over the interval [−a, a]. The
EP/K502856/1, EP/L504786/1 and EP/P003885/1. convolution operator is denoted by ∗.
 illustrated in Fig. 2 (left), a lossless antenna can be modelled
as a voltage source vs (t) followed by a series resistance Rs
and a parallel reactance Xs . The rectifier is modelled as a
resistance Rin in parallel with a reactance Xin . Assuming
perfect matching (Rin = Rs , Xin = −Xs ), all the available
RF power Pin,av is transferred
  to the2rectifier and absorbed by
Fig. 1: System model with the receiver simultaneously decod-

Rin , so that Pin,av = E E |vin
√ (t)| /Rs Since Pin,av =
.
ing information and harvesting energy. E E |y(t)|2 , vin (t) = y(t) Rs .
 

vd C. Rectifier and Diode Nonlinear Model

vin
vout Consider a rectifier composed of a single series diode
Rs id iout followed by a low-pass filter with a load as in Fig. 2
Xs Xin Rin C
(right). Denote the voltage drop across the diode as vd (t) =
vs
~
vin
~ RL
vin (t) − vout (t) where vin (t) is the input voltage to the
diode and vout (t) is the output voltage across the load re-
Antenna and Rectifier non-linear low-pass sistor. A tractable behavioural diode model is obtained by
matching network device filter and load
Taylor series expansion of the diode characteristic equation
vd (t)
Fig. 2: Antenna equivalent circuit (left) and a single series id (t) = is (e nvt − 1) (with is the reverse bias saturation
diode rectifier (right). current, vt the thermal voltage, and n the diode ideality
factor) around
P∞ a quiescent operating point vd =a a, namely
i
id (t) = i=0 ki (vd (t) − a) , where k0 = is (e nvt − 1) and
II. SWIPT S YSTEM M ODEL a
nv
e t
ki = is i!(nv i , i = 1, . . . , ∞. Assume a steady-state response
We study the system model illustrated in Fig. 1, in which t)

a single-input single-output (SISO) point-to-point SWIPT is and an ideal low pass filter, such that vout (t) is at a constant
considered and the receiver is assumed to be able to simulta- DC level. Choosing a = E {E {vd (t)}} = −vout , we can write
P∞ P∞ i/2
neously decode information and harvest energy. We consider id (t) = i=0 ki vin (t)i = i=0 ki Rs y(t)i . Truncating the
a multi-carrier/band transmission (with single-carrier being a expansion to order 4, the DC component of id (t) is obtained
special case) consisting of N orthogonal subbands where the as iout ≈ k0 + zDC where
nth subband has carrier frequency fn and equal bandwidth Bs , zDC = k2 Rs E E y(t)2 + k4 Rs2 E E y(t)4
   
(3)
n = 0, . . . , N − 1. The carrier frequencies are evenly spaced
such that fn = f0 + n∆f with ∆f denoting the inter-carrier with k2 = 0.0034 and k4 = 0.3829 for a typical Schottky
frequency spacing (with Bs ≤ ∆f ). The channel, over which diode with is = 5µA and n = 1.05, and vt = 25.86mV .
the signal is transmitted, is assumed to be flat fading and is The reader is referred to [10],[14] for more details on this EH
subject to AWGN. Throughout the paper, we assume that the model.
effect of noise is negligible for energy harvesting purposes. III. M ODULATION D ESIGN FOR SWIPT
A. Transceiver Design Considering a flat fading channel with gains given as hn = 1
Considering the multi-carrier transmission, the channel in- for n = 1, ..., N , we first study the scaling of the harvested
put at time t is given as power at the receiver with respect to the number of carriers,
(N −1 ) as well as the variation in the phase range of the transmitted
X
j2πfn t
signal. Next, we consider the M PSK modulation design for
x(t) = ℜ Xn e (1) SWIPT purposes and study the achievable information rate
n=0
for the considered M PSK modulation. Finally, we consider
jΦn optimization of the harvested power under an average power
with Xn = Sn e , where Sn and Φn refer to the amplitude
and phase of the nth carrier at frequency fn . The transmitter constraint and a received information rate constraint.
is subject to a transmit power constraint E |Xn |2 ≤ P .
A. Scaling Law for Uniformly Distributed Random Phase
The transmitted signal x(t) propagates through a multipath
channel. The received signal at the receiver is modelled as An analytical expression for zDC in (3) is now derived for
(N −1 ) the multi-carrier signal with uniform
p power allocation across
X
j2πfn t the carriers with Sn = s = 2P/N , n = 0, . . . , N − 1
y(t) = ℜ hn X n e , (2) and the phases of different carrier frequencies chosen to be
n=0
independent identically distributed (i.i.d.) as Φn ∼ U [−δ, δ],
where hn is the channel frequency response at frequency fn . n = 0, . . . , N −1. For a flat fading channel, the received signal
y(t) is given by
B. Antenna Equivalent Circuit Model
r N −1
The antenna model reflects the power transfer from the 2P X
y(t) = cos(2πfn t + Φn ). (4)
antenna to the rectifier through the matching network. As N n=0
The second and fourth order terms in zDC in (3) are sim-
plified by taking the time average and expectation over the
randomness of the signal, resulting in

3k4 Rs2 P 2 X
zDC = k2 Rs P + E {cos (Θ)} , (5)
2N 2 n0 ,n1 ,n2 ,n3
n0 +n1 =n2 +n3

where Θ denotes the random term in (5) given by

Θ = Φn0 + Φn1 − Φn2 − Φn3 . (6)
The resultant probability density function (p.d.f.) of Θ is
obtained by convolving four uniform distributions Φni ∼ Fig. 3: Asymmetric 4PSK constellation with δ = π/3. MAP
U [−δ, δ], i = 0, . . . , 3. The exact distibution of Θ can also decision boundaries are shown by red lines.
be approximated with a normal distribution. In [17], it is
shown that the sum of I i.i.d. and uniformly distributed random support of the random variable Θ is the set of values {θk =
variables converges to the normal distribution extremely fast. 2δk
−4δ + M−1 , k = 0, 1, . . . , 4(M − 1) : δ < π} for the
Accordingly, we approximateP3 the distribution
P3 of Θ in (6) by a case of asymmetric M PSK and {θk = (−4(M−1)+k)π , k =
M
normal distribution N ( i=0 E{Φni }, i=0 V ar{Φni }) with 0, 1, . . . , 4(M − 1)} for the case of symmetric M PSK.
2
E{Φni } = 0 and V ar{Φni } = δ3 . To account for the discrete distribution of the phases Φni ,
Next, the expected mean of  the random variable cos(Θ) is the scaling law in (7) is modified as follows:
2
obtained as E{cos(Θ)} = ℜ E{ejΘ } = e−2δ /3 .

(2N 2 + 1) 2
Finally, noting that there are N (2N 2 + 1)/3 terms in the zDC (ξ) = k2 Rs P + k4 Rs2 P ξ, (8)
sum of (5), the scaling law for zDC for the signal y(t) in (4) 2N
P4M−4
is obtained as where ξ = k=0 cos(θk )Pr(Θ = θk ). Note that the
(2N 2 + 1) 2 −2δ2 harvested energy is maximized when ξ = 1.
zDC (δ) ≃ k2 Rs P + k4 Rs2 P e 3 . (7) For convenience, in (8) we can further write ξ = cos(θ)p̄T ,
2N
where cos(θ) = [cos(θ0 ) . . . cos(θ4M−4 )] and p̄ = [Pr(Θ =
It is observed that the harvested power given by the zDC
θ0 ) . . . Pr(Θ = θ4M−4 )] are the support of the random
metric increases with N but decreases with δ. The rate of
variable Θ and the corresponding probabilities obtained as
scaling of zDC with N is diminished with enlarging the
p̄ = p∗p∗ ṕ∗ ṕ, respectively. In the latter, p = [p0 . . . pM−1 ]
phase range, until it becomes nearly flat for Φn ∼ U [−π, π].
denotes the p.m.f. of the symbols in M PSK constellation using
Moreover, the case of δ = 0 is equivalent to the scaling
the notation pm = Pr(X = xm ) and ṕ = [pM−1 . . . p0 ] is the
exhibited by the deterministic multisine signal in a frequency-
reversed version of the vector p. The reversion is necessary to
flat channel in [10], whereas the case of δ = π leads to a
account for the subtractions of the two random variables Φni
scaling behaviour similar to the multi-carrier signal modulated
in the expression for Θ in (6).
with CSCG inputs in [14]. Thus, departing from the in-phase
condition for maximum power transfer makes it possible to C. Achievable Information Rate
simultaneously transfer information with a phase-modulated Consider a multi-carrier transmission with N carriers over
multi-carrier signal. a flat fading deterministic channel. We aim at maximizing the
B. Asymmetric M PSK Modulation Design for SWIPT mutual information I(X; Y) between the input vector X =
[X0 . . . XN −1 ] and the output vector Y = [Y0 . . . YN −1 ],
The scaling law for zDC (7) motivates the use of modulation when the transmitter utilizes asymmetric M PSK modulation
schemes, in which the symbols are distributed in the limited √
with s = γ, where γ denotes the signal-to-noise ratio (SNR).
phase range. Accordingly, this allows to obtain higher gains We assume that different carriers (subchannels) are statistically
in harvested power. QN −1
independent of each other, i.e. Pr(Y|X) = n=0 Pr(Yn |Xn ).
Motivated by the WPT observations in Section III.A, for Under this assumption and due to the fact that Xn , n =
asymmetric M PSK modulation, the symbols are given as 0, . . . , N − 1 are i.i.d., the achievable rate at the receiver is
Xn ∈ {xm = se±j(2m+1)δ/(M−1) , m = 0, 1, . . . , M 2 − 1 : given by
δ < π}, n = 0, . . . , N − 1. Note that choosing δ = π yields ∆
IN = I(X; Y) = N I(X0 ; Y0 ). (9)
the standard symmetric M PSK modulation Xn ∈ {xm =
se±j(m+1)π/M , m = 0, 1, . . . , M
2 − 1}, n = 0, . . . , N − 1. As Accordingly, in the following, we equivalently focus on
an example, an asymmetric 4PSK constellation with δ = π/3 maximizing I(X0 ; Y0 ). For clarity, we omit the subchannel
is illustrated in Fig. 3, along with the decision regions for index. The channel input and output are related through Y =
maximum a posteriori (MAP) decoding at the receiver side. X +Z, where Z is an AWGN with standard CSCG distribution
The resultant probability mass function (p.m.f.) of Θ in (6) CN (0, 1). The capacity of this channel is obtained by maxi-
is found by discrete convolution of the p.m.f.s of Φni . The mizing the mutual information I(X; Y ) = H(Y ) − log2 (πe).
 5

4.5
Φn =0
4 Φn ~ U[-π/12,π/12]
Φn ~ U[-π/6,π/6]

zDC [µA]
3.5
Φn ~ U[-π/4,π/4]
3 Φn ~ U[-π/3,π/3]
Φn ~ U[-π/2,π/2]
2.5
Fig. 4: Rectenna with a single series diode rectifier. Φn ~ U[-π,π]
2

R 1.5
The output entropy H(Y ) = − log2 (pY (y))pY (y)dy is 0 1 2 3 4 5
log 2(N)
computed by using Monte-Carlo or numerical integration
[18], where the output p.d.f. pY (y) is given as pY (y) = (a) zDC obtained numerically (solid lines) and with scaling law (dash
√ 2
PM−1 1 PM−1 −|y− γejφm | lines) as a function of log2 (N ).
m=0 pX (xm )pZ (y − xm ) = π m=0 pm e .
×10 -6
3.5
D. Rate-Energy Region Optimization
Φn = 0
We can now define the achievable rate-energy (or, more 3
Φn ~ U[-π/12,π/12]
∆
accurately, rate-DC current) region as CR−IDC (po X (xm )) =
Φn ~ U[-π/6,π/6]
n 2.5

P DC [W]
PM−1 Φn ~ U[-π/4,π/4]
(R, IDC ) : R ≤ IN , IDC ≤ zDC , m=0 pm = 1 . Φn ~ U[-π/3,π/3]
In order to identify the rate-energy region, we formulate 2 Φn ~ U[-π/2,π/2]

the optimization problem as an energy maximization problem Φn ~ U[-π,π]

1.5 16PSK
subject to rate constraint over the p.m.f. of input symbols: QPSK

max zDC (10) 1

{pm } 0 1 2 3 4 5
log 2(N)
subject to IN ≥ R, (11)
M−1 (b) Average DC power PDC as a function of log2 (N ).
X
pm = 1, (12) Fig. 5: Results for i.i.d. uniformly distributed carrier phases.
m=0

where zDC and IN are given in (8) and (9), respectively.

The rate constraint is convex, however, the objective func- numerically by averaging over several hundred symbol periods
tion is non-convex, as it is in essence a signomial function using (3). The power P is set as −20dBm and N carriers are
in the variables {pm } as some coefficients cos(θk ) of the centered around 5.18GHz with a frequency gap ∆f = B/N
products of powers of {pm } are negative in ξ in (8). Since the for B = 10MHz. The symbol period is set as T = 1/∆f .
rate constraint cannot be represented as a posynomial or a sig- Fig. 5a also illustrates the scaling law in (7) for zDC as a
nomial, the optimization problem (10)-(12) is not compatible function of the number of carriers N for different values of
with standard Signomial Geometric Programming (SGP) tools δ. Apart from the small N region and δ = π, there is a good
[19]. A locally optimal solution can still be obtained efficiently match between the analytical and the numerical results. The
in MATLAB with Sequantial Quadratic Programming (SQP) inaccuracy between the two is due to the fact that when the
algorithm [20] that is a quasi-Newton method for solving carriers with the same index i = j (and, consequently, phase
inequality-constrained nonlinear programming problems. Φni = Φnj ) are combined, they contribute constructively to
the channel output – this effect is not captured by the analytical
IV. S IMULATION R ESULTS approximation in (7).
We first present analytical and numerical results for har- To validate the scaling laws of Fig. 5a, the rectenna circuit
vested energy with multi-carrier transmission and uniformly of Fig. 4 is simulated by using the input data signals with
distributed random phases. We then show optimization results uniformly distributed random phase. The PSpice simulations
for the rate-energy region with M PSK modulation. All ob- results in terms of the DC output power (Fig. 5b) confirm the
servations are confirmed by PSpice simulations of a practical analytical and numerical results (Fig. 5a). The PDC behaviour
rectenna circuit. The rectenna (Fig. 4) is optimized for the naturally has a more saturated form due to the non-optimality
input signal composed of 4 in-phase carriers centered around of the circuit design for large N (because of the choice of the
5.18GHz and the input power level of −20dBm. finite output capacitor and load). In addition, it is noted that
QPSK and 16PSK modulations with equiprobable signaling
A. Results for Random Uniformly Distributed Carrier Phases perform equivalently to the case of Φn ∼ U [−π, π]. The
Fig. 5a represents zDC as a function of the number of harvested DC power is indeed very sensitive to randomness
carriers N with the phase of each carrier chosen to be i.i.d. in phase, with the losses due to phase variations increasing
distributed as Φn ∼ U [−δ, δ]. The value of zDC is obtained with N .
 V. C ONCLUSION
δ=π/12, N=8
2.4 δ=π/6, N=8 In this paper, we optimized the probabilities of symbols for
2.3
δ=π/4, N=8 the symmetric and the proposed asymmetric PSK modulations,
δ=π/3, N=8
2.2 δ=π/2, N=8 thereby enlarging the SWIPT rate-energy region considerably
QPSK, N=8, as compared to the equiprobable symmetric PSK. Such mod-
zDC [µA]

2.1 optimized p.m.f.

2 δ=π/12, N=1 ulation design is enabled by nonlinear modeling of the EH
δ=π/6, N=1
1.9 δ=π/4, N=1
at the receiver. The gain in harvested energy due to using
1.8
δ=π/3, N=1 optimized modulation design is additive to the increase in the
δ=π/2, N=1
1.7 QPSK, N=1,
ECE achieved by utililizing multiple carriers.
optimized p.m.f.
1.6
0 0.5 1 1.5 2
QPSK, R EFERENCES
uniform p.m.f.
Information rate per subband [bits/s/Hz] [1] H. Sakaki, S. Yoshida, K. Nishikawa, and S. Kawasaki, “Analysis of
rectifier operation with FSK modulated input signal,” in Proc. IEEE
(a) Analytical CR−IDC as a function of δ for N ={1, 8} and M = 4. Wireless Power Transfer Conf., 2013, pp. 187–190.
2.6 [2] H. Sakaki, T. Kuwahara, S. Yoshida, S. Kawasaki, and K. Nishikawa,
“Analysis of Rectifier RF-DC Power Conversion Behavior with QPSK
and 16QAM Input Signals with WiCoPt System,” in Proc. Asia-Pacific
2.4 Microw. Conf., 2014, pp. 603–605.
δ=π/12 [3] H. Sakaki, K. Nishikawa, S. Yoshida, and S. Kawasaki, “Modulated
δ=π/6 scheme and input power impact on rectifier RF-DC Efficiency for
P DC [µW]

2.2
δ=π/4
WiCoPT system,” in Proc. 45th Eur. Microw. Conf., 2015, pp. 60–63.
δ=π/3
δ=π/2
[4] F. Bolos, J. Blanco, A. Collado, and A. Georgiadis, “RF energy
2
QPSK, harvesting from multi-tone and digitally modulated signals,” IEEE Trans.
optimized p.m.f. Microw. Theory Techn., vol. 64, no. 6, pp. 1918–1927, Jun. 2016.
1.8 [5] I. M. Kim and D. I. Kim, “Wireless information and power transfer: rate-
energy tradeoff for equi-probable arbitrary shaped discrete inputs,” IEEE
Trans. Wireless Commun., vol. 15, no. 6, pp. 4393–4407, Jun. 2016.
1.6
0 0.5 1 1.5 2 [6] R. Zhang and C. K. Ho, “MIMO broadcasting for simultaneous wireless
Information rate per subband [bits/s/Hz] information and power transfer,” IEEE Trans. Wireless Commun., vol.
12, no. 5, pp. 1989–2001, May 2013.
(b) R − PDC region as a function of δ for N = 8 and M = 4. [7] J. Park and B. Clerckx, “Joint wireless information and energy transfer
in a two-user MIMO interference channel,” IEEE Trans. Wireless
Fig. 6: Results for rate-energy region with M PSK modulation. Commun., vol. 12, no. 8, pp. 4210–4221, Aug. 2013.
[8] K. Huang and E. G. Larsson, “Simultaneous information and power
transfer for broadband downlink systems,” IEEE Trans. Signal Process.,
vol. 61, no. 23, pp. 5972–5986, Dec. 2013.
B. Rate-Energy Region with M PSK Modulation [9] W. Liu, X. Zhou, S. Durrani, and P. Popovski, “SWIPT with practical
modulation and RF energy harvesting sensitivity,” in Proc. IEEE Int.
Fig. 6a illustrates the achievable rate-energy region for Conf. Commun., 2016, pp. 1–7.
4PSK in various phase range limits for a single (N = 1) [10] B. Clerckx and E. Bayguzina, “Waveform design for wireless power
transfer,” IEEE Trans. Signal Process., vol. 64, no. 23, pp. 6313–6328,
and multiple (N = 8) carrier transmission and SNR of Dec. 2016.
20dB. The total information rate IN is normalized w.r.t. the [11] Y. Huang and B. Clerckx, “Large-scale multiantenna multisine wireless
bandwidth N Bs . Hence, the x-axis refers to a per-subband power transfer,“ IEEE Trans. Signal Process., vol. 65, no. 21, pp.
5812–5827, Nov. 2017.
rate. Transmitting over multiple carriers results in higher ECEs [12] B. Clerckx and E. Bayguzina, “A low-complexity adaptive multisine
– this benefit is added onto the gains due to optimization of waveform design for wireless power transfer,” IEEE Antennas Wireless
the symbols’ p.m.f. relative to the non-optimized equiprobable Propag. Letters, vol. 16, pp. 2207–2210, 2017.
[13] J. Kim, B. Clerckx, and P. D. Mitcheson, “Prototyping and
QPSK, which provides a lower bound for harvested energy. It experimentation of a closed-loop wireless power transmission
is also noted that it is possible to significantly enlarge the rate- with channel acquisition and waveform optimization,” in Proc. IEEE
energy region by limiting the phase range of symbols. Smaller Wireless Power Transfer Conf., 2017, pp. 1–4.
[14] B. Clerckx, “Wireless information and power transfer: nonlinearity,
phase ranges δ = {π/12, π/6} maximize the energy metric waveform design and rate-energy tradeoff,” IEEE Trans. Signal
most but do not allow to achieve the maximum normalized rate Process., vol. 66, no. 4, pp. 847–862, Feb. 2018.
of 2 bits/s/Hz. It can also be derived analytically that for all [15] M. Varasteh, B. Rassouli, and B. Clerckx, “Wireless information and
power transfer over an AWGN channel: nonlinearity and asymmetric
values of δ, ξ = 1 when only one symbol is transmitted with Gaussian signaling,” in Proc. IEEE Inf. Theory Workshop, 2017, pp.
probability 1 giving maximum zDC of 2.47µA for N = 8. 181–185.
In order to validate those results, the rectenna circuit of Fig. [16] M. Varasteh, B. Rassouli, and B. Clerckx, “On capacity-achieving
distributions over complex AWGN channels under nonlinear power
4 is fed with the input data signals generated for N = 8, with constraints and their applications to SWIPT,” arXiv:1712.01226, 2017.
the p.m.f. of 4PSK symbols optimized for a set of δ values [17] F. Killmann and E. von Collani, “A note on the convolution of the
and at a given rate requirement. The rate-PDC region graph uniform and related distributions and their use in quality control,”
Economic Quality Control, vol. 16, no. 1, pp. 17–41, 2001.
(Fig. 6b) confirms the analytical (Fig. 6a) rate-zDC region [18] U. Madhow, Fundamentals of Digital Communication, New York:
results. This validates the EH model and modulation design Cambridge University Press, 2008.
and highlights the importance of using optimized modulation [19] M. Chiang, “Geometric programming for communication systems,”
Found. Trends Commun. Inf. Theory, vol. 2, pp. 1–154, 2005.
design in order to increase the ECE of a practical rectenna [20] P. T. Boggs and J. W. Tolle, “Sequantial Quadratic Programming,” Acta
circuit in the SWIPT configuration. Numerica, vol. 4, pp. 1–51, 1995.