A New Receiver Design: Simultaneous Wireless Power Transfer With Modulation Classification

This work proposes a new simultaneous wireless power transfer and modulation classification (SWPTMC) scheme, appropriate for internet of things (IoT) and military applications. The problem of SWPTMC is investigated for various modulation formats, i.e, quadrature phase-shift-keying (QPSK), 16-pulse amplitude modulation (16-PAM), π/4-QPSK, minimum shift keying (MSK), offset QPSK (OQPSK), and 16-quadrature amplitude modulation (16-QAM). We propose a new receiver architecture that incorporates conventional power splitting under a linear model with a certain level of sensitivity. The blind modulation classification algorithm is based on the higher-order cumulants and cyclic cumulants of the received signal. The cyclic cumulants use the non-zero cycle frequency position, while the higher-order cumulants use threshold values for classifying modulation formats. Monte Carlo simulations are carried out to validate the accuracy of the proposed SWPTMC scheme.


I. INTRODUCTION
E NERGY-efficient transmission is one of the primary concerns in modern wireless networks, such as wireless sensor networks (WSN) and Internet of Things (IoT) due to the limited lifespan of fixed energy supplies. Wireless power transfer (WPT) through radio-frequency (RF) signals has recently emerged as a candidate technology for providing power to remotely located sensors and IoT devices [1]. The WPT concept is extended to the simultaneous wireless information and power transfer (SWIPT), which allows data and power to be transmitted through the same electromagnetic waveform. SWIPT has recently gained significant attention as an integrated approach for information decoding (ID) and energy harvesting (EH) [2], [3]. Practical SWIPT receivers have been proposed by using time-switching (TS) and power splitting (PS) schemes [4].
Classification of modulation formats is a key process of smart receivers to ensure correct demodulation. It plays a significant role in the military, intelligence, and civilian applications [5]- [9]. The blind modulation classification (MC) algorithms can be divided into two general categories, i.e., likelihood-based (LB) and feature-based (FB) algorithms. The LB algorithms require signal preprocessing tasks and suffer from higher computational complexity [5]. On the other hand, FB algorithms are easier to implement and have lower computational complexity [6]- [9]. The higher-order cumulants and moments-based MC algorithms employ threshold values to classify the modulation formats [6]- [8]. The cyclic cumulantsbased methods discussed in [8], [9] are more robust and use non-zero cycle frequencies of the received signals to identify the modulation schemes.
In this paper, we propose a new receiver architecture which simultaneously harvests power and performs blind MC; this joint design introduces a new concept which is called simultaneous wireless power transfer and modulation classification (SWPTMC). We consider the linear harvesting model with a certain level of sensitivity. The proposed MC algorithm can be applied to a wide range of modulations, i.e., quadrature phase-shift-keying (QPSK), 16-pulse amplitude modulation (16-PAM), π/4-QPSK, minimum shift keying (MSK), offset QPSK (OQPSK), and 16-quadrature amplitude modulation (16-QAM) over Rayleigh fading channels. It is based on the combination of higher-order cumulants and second-order cyclic cumulants. The second-order cyclic cumulants use the non-zero cycle frequency as a feature to classify MSK and OQPSK modulation formats. The remaining modulation formats are identified by fourth and eight-order cumulants. To the best of the authors' knowledge, this is the first work in the literature focused on SWPTMC scheme.

II. SIGNAL MODEL
We consider a SWPTMC system consisting of a single transmit and receive antenna as shown in Fig. 1. The transmitter sends the modulated signal with an average transmission power of P tx in each time slot. The receiver divides the received signal into two components, i.e., one is used for the blind MC with PS ratio 1 − ρ and the other one is for EH with power ratio ρ, where ρ ∈ [0, 1] is the PS factor. The transmitted continuous-time domain passband signal is given as where φ is the carrier phase and f c is the carrier frequency. The in-phase (I) and quadrature (Q) components of the transmitted signal s(t) are given as: is the root raised cosine (RRC) pulse shape filter, and T is the symbol period. For QPSK, OQPSK, MSK: The discretetime received signal over additive white Gaussian noise (AWGN) and Rayleigh fading channel can be written as where . At the receiver, we consider a linear EH model with a certain level of sensitivity and the power received at the EH circuit is P r during a symbol period T . The amount of harvested power can be expressed as [4] P Eh = η (P r − P th ) , P r ≥ P th ; 0, where η ∈ [0, 1] is the RF-EH conversion efficiency and P th is the RF-EH sensitivity level.
III. PROPOSED RECEIVER ARCHITECTURE The proposed separate receiver splits the received signal into two streams as shown in Fig. 1. One stream for the blind MC with PS ratio 1 − ρ and the other stream is for EH with PS ratio ρ, described in detail below.

1) Modulation Classification:
The received signal at MC circuit is y m [n] = √ 1 − ρȳ [n]. After down conversion, we can obtained the lowpass discrete signal as where w[n] is the lowpass AWGN with zero mean and variance of σ 2 w . The classification performs in three stages: (a) at the first-stage, we use the second-order cyclic cumulants to identify OQPSK and MSK modulation formats; (b) at the second-stage, π/4-QPSK modulation is identified using the fourth-order cumulant; (c) at the last-stage, we employ eightorder cumulant to identify the remaining modulation formats. The second-order cyclic cumulant of the baseband signal r m [n] for different modulation formats are discussed below.
The Fourier series coefficient of the second-order timevarying correlation function, c [rm,2,1] [n; τ ], is known as cyclic cumulant, and at τ = 0 lag can be expressed [9] C [rm,2,1] [α; 0] where α is the cycle frequency and N is the received signal length. Once C [rm,2,1] [α; 0] is determined, frequency estimation is given by [9] From (6), we obtain the second-order cycle frequency for MSK atf b = 2f s . For QPSK, π/4-QPSK, 16-QAM, and 16-PAM modulation formats, we get the same feature value, i.e., f b = f s and there is no peak for OQPSK,f b = 0 [9], where f s is the symbol rate. To classify the remaining modulation formats, we further examine their distinctive features in the second-stage. The fourth-order cumulant with zero conjugations can be expressed as [6] c [rm,4,0] From (7), we can easily differentiate π/4-QPSK and the rest of modulations as shown in Table I. At the final-stage, to identify the remaining modulation formats, we use eight-order cumulant with four conjugations, which can be expressed as [7] c [rm,8,4] Table II. , where w c [n] is the rectenna circuit AWGN noise with zero mean and variance of σ 2 c . From the received signal y e [n], we can find the power for the i-th constellation point as P ri ≈ ρ aP xi , assuming that the power harvested from the passband and circuit noise is negligible, where a is the fading power gain and P xi is the transmit power for the i-th constellation point. By using (3) and averaging at the fading, we get the average harvested power as where M is the constellation size of the modulation formats. IV. SIMULATION RESULTS AND DISCUSSION In this section, we evaluate the average harvested power and success rate of the modulation classifier for the proposed SWPTMC scheme using Monte Carlo simulations. The intermediate carrier frequency, sampling rate, symbol rate, and oversampling factor are set to 5 MHz, 50 MHz, 1 MHz, and 50, respectively. The roll-off factor of the RRC pulse shape filter is 0.5. The Rayleigh fading channel with L = 4 channel taps is considered and the receiver RF-EH sensitivity threshold and conversion efficiency are set to 0 dBm and 0.8, respectively. The variances of AWGN σ 2 v , σ 2 w , and σ 2 c are set to 0.06, 0.1, and 0.1, respectively. The performance is evaluated for 1000 iterations and in each iteration the number of symbols is set to 2000.
Figs. 2 and 3 show the average harvested power (P Eh ) and the percentage of correct classification (P cc ) for the modulation formats considered, as a function of ρ over a Rayleigh fading channel. It is known that a modulation scheme with higher peak-to-average power ratio (PAPR), increases the average harvested power [4]. The PAPR of PSK based modulations, 16-QAM, and 16-PAM modulation formats are given as ψ PSKs = 1, ψ 16-QAM = 1.8, and ψ 16-PAM = 2.64, respectively. Hence, the average harvested power follows: Eh , as shown in Fig. 2. The classification performance trade-off has been observed in Fig.  3 for 0.5 < ρ ≤ 1. Thus, for ρ ∈ [0, 0.5] we can simultaneously harvest power with slightly affecting the classifier performance.
V. CONCLUSION A SWPTMC receiver architecture has been proposed and implemented for different modulation formats over Rayleigh fading channels. The EH circuit uses a linear model with certain level sensitivity and the MC circuit uses cumulants and cyclic cumulants at the baseband level to classify QPSK, 16-PAM, MSK, OQPSK, 16-QAM, and π/4-QPSK modulation formats. The results highlight that for 0 < ρ ≤ 0.5 we can harvest power without affecting the classifier performance.