High-Performance Multiband Ambient RF Energy Harvesting Front-End System for Sustainable IoT Applications—A Review

The increasing demand for wireless Internet of Things (IoT) calls for power efficient RF energy harvesting approach. The current dominant single-band RF energy harvesting front-end system restricts itself to a single frequency, which is at the risk of ineffective operation when the harvesting frequency is unavailable. This paper reviews and explores the alternative approach of a multiband RF energy-harvesting front-end system. It covers all essential circuitry of a multiband RF energy harvesting front-end system, starting from the recent RF surveys which investigate the typical, usable, and high-strength RF input, to the overview of the state-of-the-art antenna, impedance matching network (IMN), and RF-DC rectifier. The recent advancement of the multiple RF input harvesting abilities, reflection loss minimization, and performance improvement are comprehensively reviewed for different operating conditions, and this review also presents the advantages and disadvantages of the different circuit architecture combinations for multiband RF energy harvesting front-end systems. In summary, this review aims to fill the research gap in the further advancement of multiband RF energy harvesting towards enhancing its performance through optimal circuit integration of the front-end system.


I. INTRODUCTION
The rising of the Industrial Revolution 4.0 (IR 4.0) [1] imposes a massive amount of compact wireless sensors to provide seamless data exchange between humans and machines. The energy demand thus becomes insatiable. Conventional battery-operated devices require costly bat-The associate editor coordinating the review of this manuscript and approving it for publication was Jie Tang . tery replacement which further exacerbates safety [2] and environmental concerns [3]. Harvesting ambient energies to self-power the devices prove to be crucial for long-term sustainability [4]. The common types of energy harvesting solutions are solar [5], vibration [6], sound [7], and radio frequency (RF) [8], [9].RF is the most beneficial energy source today as the wireless communication system has been advancing rapidly due to the initialization of 5 th Generation (5G) wireless communication. Many electronic VOLUME 11, 2023 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ devices follow the massive growth trend of Information and Communication Technology (ICT) and the Internet of Things (IoT) [10], [11], [12]. Electronic devices enabled with internet have contributed to the overall increase of density in RF frequency propagated in free space. This proliferates the opportunity for wireless power transfer and energy harvesting. Hence, the recent trend in research interest is to implement self and wirelessly powered low-power electronic devices, such as wireless sensor nodes (WSN), by harvesting RF energy [2]. The on-chip integration of RF energy harvesting (RFEH) system would have considerable potential to realize a miniaturized and battery-less electronic solution for healthcare wearables, biometric sensors, automotive sensors, and other industrial applications [13], [14], [15].
The RFEH system can be categorized into frontend [16], [17], [18], [19] and back-end [20], [21], [22], [23] solutions. A front-end RFEH system consists of an RF receiver antenna, an impedance matching network, and an RF-DC rectifier as shown in Fig. 1. The front-end solution is responsible to harvest and convert the RF power into DC power. For instance, for a front-end system with 40% power conversion efficiency at −10 dBm, the output DC power is only 40 µW. In this case, the front-end system is at risk in harvesting insufficient power to start-up sensor devices, in which the start-up power expands from 10 µW up to 100 µW [24]. To compensate for this bottleneck, the back-end RFEH system is integrated to boost or pump up the DC current from the front-end RFEH system and provide sufficient output current to power up the sensor devices. Since the front-end and back-end RFEH system has different functionality and adopts different design approaches, the front-end and back-end are often separated in the discussion. Therefore, the scope of this review is targeted at the front-end of the RFEH system.
Much research work on single-band RFEH systems has reported encouraging performance in the context of high power conversion efficiency (PCE), wide power dynamic range (PDR), and high sensitivity [8], [9]. However, several limitations for the single-band RFEH systems are yet to be addressed, such as capped single-band signal power and overreliance on the single-band source. In [25], any single-band signal power is limited by the surrounding usage, leading to a maximum scale in harvesting power. At the same time, ambient single-band RF energy suffers from fluctuating power density, potentially lowering the harvested signal power. Low harvested ambient signal power significantly restricts the rectifier's power transfer performance. Thus, harvesting from multiple RF bands which are known as the multiband RFEH systems has been proposed to fulfill the power demands [26] and to maximize the output power [27], [28].
This article focuses on the investigation of the RFEH front-end system, especially the multiband RFEH front-end system. Section II discusses the recent RF surveys in the ambient environment. Section III briefly overviews the RFEH front-end circuit architecture. Section IV reviews the recently reported designs for multiband RFEH front-end circuitry. Section V discusses the circuit integration of the multiband RFEH front-end system. Section VI concludes this review paper.

II. AMBIENT RF POWER
The presence of ambient RF power is mainly due to wireless power transmission which propagates electromagnetic power over a distance from the transmitter to the receiver at free space. In wireless transmission, the transmitter converts the electrical power into RF power which is transmitted to the receiver. The scavenging of the RF power residue in the free space medium is known as RF energy harvesting [29]. RF energy harvesting can be further classified into near-field and far-field which differ based on the propagation distance. The propagation distance less than Fraunhofer Distance is considered as near-field whereas the propagation distance more than Fraunhofer Distance is known as far-field [30]. The Fraunhofer Distance can be expressed as [30]: where D is the diameter of the antenna and λ is the wavelength of the RF wave. In general, both near-field and far-field energy harvesting requires a transmitter (Tx), a medium, and a receiver (Rx). Near-field energy harvesting is the transfer of power at a short distance, generally through a magnetic field such as magnetic coupling or resonance of inductors or coils [9]. A transformer is the simplest way to explain the concept of near-field energy harvesting. Assume a basic near-field application in which the transmitter consists of the primary coil while the receiver consists of the secondary coil. The AC power is transferred from the primary coil to the secondary coil through magnetic induction where the medium is the magnetic field. Battery charging through near-field wireless power transfer via magnetic resonance is reported in [31].
Alternately, far-field propagation is the transfer of power at a long distance, generally through the transmission and reception of antennas. In theory, long-distance transmission leads to lower input power density. This is due to the RF wave which has an inverse square law spreading characteristic, which indicates that the power transfer is inversely proportional to the square of the distance. The loss during the free space transmission is known as the free space path loss (FSPL). In effect to the FSPL, far-field energy harvesting has lower input power density than near-field energy harvesting. The input power density can be computed with the aid of the Friis transmission equation.
where P R and P T represents the input received and transmitted power, G R and G T represents the gain of the receiver antenna and transmitter antenna, c is the speed of light, D is the frequency wave transmission distance between the transmitter and receiver and f is the frequency of the RF wave.
From (2), it is apparent that harvesting RF power at lower frequency results in higher input received power. However, this statement is not entirely accurate. The signal crowding strength of different RF bands should also be a factor of consideration. This is due to the inconsistency and volatile characteristic of the RF signal's strength. In brief, the usage and environmental factors affect signal strength [22]. The radio spectrum expands from 3 kHz to 300 GHz [32] which can be further classified into several RF bands. The density of use in the RF bands relies on the RF consumption level. A highly utilized RF band increases the scale of RF power residue in the ambient environment, leading to higher RF power density. In general, the highly utilized ambient RF bands are the Very High-Frequency band (VHF), Ultra High-Frequency band (UHF), and Super High-Frequency band (SHF) ranging from 30 MHz to 30000 MHz, which capture most of the RF applications such as GSM 2G, UMTS 3G, LTE 4G, 5G NR, ISM, WIFI-Communication, FM broadcasting, and Bluetooth.
Two main constraints are limiting the signal strength and power density of RF signals. First, the non-uniform establishment of RF facilities can affect the overall power density of the RF signals in urban, suburban, or rural areas [33], [34]. Second, the presence of blocking obstacles such as tall buildings or mountains can degrade the strength of the RF signals [35].
The RF survey summarizes the signal strength for different frequency bands detected in the ambient, measured at metro stations and residential areas in London [36], indoor and outdoor of University Covilha [37], and semi-urban and urban locations in London [38]. Fig. 2 shows the power density measured at the Northfields London Underground station [38]. The frequency bands presented in Fig. 2 are highly recommended references in designing an RF energy harvesting system due to higher RF power density.

III. RFEH FRONT-ENDS ARCHITECTURE
In general RFEH front-end circuits, consist of three main components which are the RF receiver antenna, the impedance matching network, and the RF-DC rectifier. The RF receiving antenna functions to receive the AC signal at the dedicated frequency band. The received signal is subsequently passed to the impedance-matching network for maximum power transfer achieved by minimizing the reflection loss. The matching network also acts as a passive voltage-boosting block, realized by the inductor at the IMN. Thereafter, the AC signal is transferred to the RF-DC rectifier to convert the RF power into a DC output (ideally ripplefree). The entire power transfer process of the RF energy harvesting front-end circuit is conceptually shown in Fig. 3. Some predictable losses incurs during the end-to-end transfer of the received power. Considering the losses, the output DC power can be written as: where P OUT ,DC is the output DC power and P IN,AC is the input received AC power. P LOSS,IMN is the losses from the impedance matching network and P LOSS,REC is the losses from the rectifier. Parasitic resistance of the components in use such as the capacitor, inductor or transistor contributes to the losses during the power transfer process.

A. RF RECEIVER ANTENNA
The review scope in this section is limited to the RF receiver antenna, which functions in receiving and converting the ambient RF energy into an electrical signal. An optimally designed antenna can boost the system's performance by providing higher input power. The peak input voltage, V ANT of the receiver antenna can be calculated from [39]: where R ANT is the series resistance of the receiver antenna and P R is the available input received power at the receiver antenna.
For the RFEH application, the RF receiver antenna consists of three main parameters which are the operating frequency, gain, and size. The operating frequency of the RF receiver antenna specifies the frequency band to be harvested by the receiver antenna. Harvesting a non-operating frequency band will lower the input received power for the system. The gain of the RF receiving antenna specifies the performance of the antenna in scavenging the power from the operating frequency. A high-gain antenna increases the received power which enhances the overall RFEH system [40]. The gain of the antenna can be calculated from [41]: where G is the antenna gain, ε R is the radiation efficiency and D is the antenna directivity. The antenna directivity relates to the radiation pattern and polarization of the antenna. The antenna's radiation pattern and polarization define the antenna's capability to capture the ambient RF wave power. An antenna can radiate in a unidirectional pattern or omnidirectional pattern. For harvesting ambient RF energy whose incident angle is unknown, the omnidirectional pattern with a broad effective angle signal coverage can outperform the unidirectional pattern. However, the trade-off for implementing the omnidirectional pattern is the lower antenna gain [42], [43]. Besides the radiation pattern, the polarization of the antenna can affect the antenna gain. The antenna's polarization consists of linear polarization (LP), circular polarization (CP) and elliptical polarization (EP).
To obtain high gain, the mismatch in the polarization of the transmitter and receiver antenna must be avoided [44]. The size of the RF receiver antenna indicates the space required to install the antenna, and usually relates to the antenna efficiency. Referring to [45], a smaller physical size of the antenna can reduce the antenna gain. However, the design trend of antenna concentrates on achieving a lower profile, which is also known as the miniaturization of the antenna. Therefore, it is a trade-off to be considered when designing or selecting the RF receiver antenna. The relationship between the compactness, gain and size is given as [46]: where G is the antenna measured gain and A is the antenna size.
From [47], there are several distinct antenna types and shapes for RF receiver antenna. The most preferred type is the printed antennas especially printed planar patch antennas [48] due to their compact size, high versatility, and high flexibility in integrating with rectifier circuits [47]. Basic shapes such as rectangular, square, or circular are the fundamental shape for the planar patch antenna. However, the basic shape planar patch antenna suffers from limited gain and bandwidth. To enhance the performance of the planar patch antenna, modifications to the basic shape have been proposed. The popular modification is through the implementation of slotted design [49], [50], [51], employing fractal geometry [52], [53], or designing a unique-shaped antenna [24], [54], [55], [56], [57], [58], [59].
The slotted antenna design introduces holes or gap intervals inside the basic shapes, increasing the current path and leading to the increase of gain [60]. By adopting a slotted design, the work [49], [50], [51] measures a peak gain of 8.25 dBi, 4.3 dBi and 5 dBi respectively when cutting through a cross-shaped slot, c-shaped slot and e-shaped slot from the conventional basic shape patch antenna.
The fractal antenna implements a fractal geometry pattern in the patch antenna. The popularly used fractal geometry for RF receiver antenna is the Koch snowflake geometry [52], [53]. The design in [52] and [53] proposed a 4-iteration order and 6-iteration order Koch fractal structure with a peak measured gain of 4.5 dBi and 3.1 dBi respectively. Practically, an increment in the iterations order can increase the overall effective length which leads to a higher realized gain [53]. However, the increment in the iteration order also increases the complexity of the pattern, contributing to a trade-off of difficulty in fabrication [61].
The unique-shaped antenna introduces unique-shaped patterns such as T-shaped [54], Folded-shaped [55], Yagi-Uda [56],and Loop [24], [57], [58], [59] in the design of the antenna. In [54], an 866 MHz single-band antenna has been realized by implementing a T-shaped monopole with a coplanar waveguide feeding line, which exhibits dipole-like radiation and measures a peak gain of 2 dBi. The work [55] fully utilized the folded technique by folding the two poles, contributing to a reduction in the antenna diameter. At the same time, this folded dipole design achieved a measured gain of 1.84 dBi. Alternately, the Yagi-Uda design [56] enables a high measured gain of 7 dBi at 950 MHz. However, the Yagi-Uda structure consists of several parallel branches which require more size, hence reducing the overall antenna compactness. The loop antenna is one of the most popular antenna designs for the application of RF energy harvesting [58], [59]. In [24], a compact square loop antenna has been introduced to harvest from 868 MHz. This square loop antenna design is horizontally polarized and capable of achieving a peak gain of 0.66 dBi with a tiny antenna diameter of 34 mm. Furthermore, [57] proposed a novel meander line structure loop antenna that reduces 50% of the size respective to the conventional loop antenna. The reported design can achieve a radiation efficiency of 97% at 925 MHz, leading to a peak antenna gain of 4.22 dBi.

B. IMPEDANCE MATCHING NETWORK
In the RFEH system, frequent high impedance variation occurs due to the use of non-linear devices such as diodes or transistors in the rectifier. Therefore, an impedancematching network is integrated after the antenna to minimize the reflection loss due to impedance mismatch. In general, an impedance matching network follows a common working principle, where the real component of two integrating blocks is equal while the imaginary component of two integrating blocks is in the complex conjugate. Referring to Fig. 4, the mathematical equation can be written as, Re(Z ANT ) D Re(Z REC ) and Im(Z ANT ) D Im(Z * REC ). A high reflection loss limits the input power of the rectifier, leading to a capped output power. The undesirable capped output power degrades the rectifier's output parameters such as the output voltage, sensitivity and power dynamic range. The numerical value of the signal reflected can be referred to in the reflection coefficient equation given below [62], [22].
where is the value of the reflection coefficient, it is commonly known as the S 11,mag . Z REC is the impedance of the rectifier which is modeled as R REC + X REC , Z ANT is the impedance of antenna which is modeled as R ANT + X ANT and Z * ANT is the complex conjugate of the antenna impedance. The reflection coefficient is always lesser than 1. Assuming a reflection coefficient of 0.5, it indicates that half of the input power is reflected back. In this case, the system is unmatched. To match two integrating blocks, the reflected power must be less than 10%, which is generally referred as S 11,dB less than −10 dB [63].
On-chip impedance matching network adopts the combinations of passive components such as resistors, capacitors, inductors, and transformers. The most frequently used impedance-matching network topologies are the L-matching, Pi-matching, and T-matching [64], with the L-matching network having the highest popularity in use.
An L-matching network consists of a capacitor and inductor in an L-shaped configuration as shown in Fig. 4. In general, the antenna impedance is fixed while the impedance of the rectifier and the load impedance varies respectively to the frequency. Therefore, The L-matching network is responsible to boost up the load impedance when it is lower than the antenna impedance and pulling down the load impedance when it is higher than the antenna impedance. Thus, it contributes to the upward load impedance transformation as shown in Fig. 4(a), 4(b), and downward load impedance transformation as shown in Fig. 4(c), 4(d). Capacitors pass high-frequency signals while the inductors pass the low-frequency signals. Therefore, reversing the location of the capacitor and inductor in an L-matching network realizes a low-pass L-matching and high-pass L-matching transfer function respectively. Fig. 4(a) and Fig. 4(c) show the high-pass L-matching network while Fig. 4(b) and Fig. 4(d) show the low-pass L-matching network.
To implement impedance matching, the simplest approach is to compute the optimized parameters for the components in the matching network [65], [66], [67], [68]. Reference [65] proposed a new analytic model for the Dickson charge pump rectifier which includes the optimum design parameters for the matching network components. Subsequently, [66] stated the design consideration for the impedance matching network and proposed the general type of impedance matching networks. Nevertheless, [67] presents a detailed analysis and design considerations for lumped (inductor and capacitor) matching networks. Lastly, [68] analyzed the multistage matching networks and provides the optimization parameters.

C. RF-DC RECTIFIER
In RF energy harvesting, the power received from the antenna is low in density, requiring boosting or multiplying of the received power in order to produce sufficient voltage [69]. Also, the power received is in AC voltage, thus, rectification is necessary to convert the AC voltage to a usable DC voltage. Therefore, the integration of an RF-DC rectifier is essential for the purpose of voltage multiplication and voltage rectification of the received RF power [70].
In short, RF-DC rectification includes two types which are half-wave rectification and full-wave rectification. A halfwave rectifier can be made by placing a single diode in series or in parallel with the output load, which is also known as the single-series rectifier [71] and single-shunt rectifier [72] respectively. However, research from [73] shows that the half-wave rectifier produces lower efficiency as compared to the full-wave rectifier. On the other hand, there are three types of full-wave rectifiers which are the Cockcroft-Walton rectifier, Dickson-type rectifier, and cross-coupled differential-drive (CCDD) rectifier. The Cockcroft-Walton rectifier is also known as the Greinacher rectifier or Villard rectifier [74]. In this topology, the coupling capacitors for multiple stages are connected in series as shown in Fig. 5(a), leading to a significant voltage drop per stage [74]. To reduce the voltage drop, the coupling capacitors are placed in parallel for multiple stages as shown in Fig. 5(b), which is known as the Dickson-type rectifier [75]. The Dickson-type rectifier is the enhanced version of the Cockcroft-Walton topology. Unlike the Cockcroft-Walton topology which is typically used for the high-voltage application, the Dickson-type topology is suitable for low-voltage applications [9].
In that case, the two main rectifier topologies in the RFEH system are the Dickson-type rectifier and the CCDDtype rectifier. Both topologies differ in the rectification approach. In short, there are two main differences. First, a Dickson-type rectifier requires single-ended input while a CCDD-type rectifier requires differential input. Second, a Dickson-type rectifier uses the transistors as a diode while a CCDD-type rectifier uses the transistors as the switch. There are three common indicators to evaluate the performance of the rectifiers which are sensitivity, PCE, and PDR.
The sensitivity in dBm is defined as the minimum input power required for the circuit to generate a certain amount of DC output. The common DC value in evaluating the sensitivity is 1V DC output [9]. The sensitivity of the rectifier determines the lowest RF power density applied to the input of the RFEH system for proper operation in a given environment. The sensitivity is defined as: P dBm = 10 log 10 P mW (8) where P dBm is the rectifier's sensitivity expressed in dBm and P mW is the rectifier's received power expressed in milliwatts. The PCE measures the efficiency of the RFEH system through the ratio of the DC output power to the rectifier's input received power [76].
where η REC is the rectifier's PCE, P DC is the DC power at output and P REC is the rectifier's received power. The PDR of the rectifier is determined by the continual range of the input received power at which the PCE of the rectifier maintains its value above a certain percentage. The common percentage of determining the PDR is above 20% [39], [19]. The PDR defines the output stability of the system in deciding whether the RFEH system is capable of continuously maintaining the efficiency in generating the output voltage.
In reference to the indicator parameters as above, it is desirable to strive for high sensitivity, high PCE, and wide PDR. The output parameter for the RFEH system is the DC output voltage. By stacking or cascading the stage of rectifiers, the low RF input voltage can be multiplied to produce the desired DC output voltage.
A common single-stage Dickson-type rectifier consists of two diode-connected transistors, M 1 and M 2 and two capacitors, C 1 and C 2 , as shown in Fig. 6(a).
During the negative cycle, the transistor M 1 is on while the transistor M 2 is off. Thus, the capacitor C 1 charges up to a voltage value of: where V P is the peak amplitude of RF input and V TH1 is the threshold voltage of the transistor M 1 . During the positive cycle, the transistor M 1 is off while the transistor M 2 is on. At this point, the capacitor C 2 charges up to a voltage value of: or where V TH2 is the threshold voltage of the transistor M 2 . The single-stage output voltage (V OUT ,1 ) is the voltage of the capacitor C 2 , thus, the expression of a single-stage Dickson-type rectifier can be written as: The multi-stage output voltage (V OUT ,n ) of a Dicksontype rectifier can be expressed as: where n is the number of cascaded stages. Dickson-type rectifier produces a high voltage at the cost of poor sensitivity and lowers PCE [8], [9], [77]. As observed from (10), the output voltage of the Dickson-type rectifier suffers from the inevitable drop of V TH . When it is required to produce higher DC voltage, more stages are essential, resulting in more voltage drop as described in (14) which eventually degrades the sensitivity and PCE [78]. To increase the performance of Dickson-type rectifiers, research on compensation or cancellation of V TH has been conducted. The first technique is through self V TH compensation (SVC), which can be achieved by interconnecting the terminals of transistors of the same stage. References [79] and [80] proposed the SVC method where the NMOS gate terminal connects to the output terminal while the PMOS gate terminal connects to the ground. This method achieves a peak PCE of 32% at a low input power of −10 dBm. The second technique is the inter-stage V TH compensation (ISVC) by connecting the gate terminal of the transistors to the subsequent stage [81], [82], achieving a high sensitivity of −19 dBm and −20 dBm. Subsequently, the third technique is the internal V TH compensation (IVC) by adding capacitors or circuits at the output to bias the gate terminal of transistors. A passive current-mirror bias circuit to cancel out the V TH drop was proposed by [83] and [84], where a peak PCE of 36.6% have been achieved. Alternatively, [85] proposed an IVC alike method -the floating-gate technique, which reports an operation condition as low as −22.6 dBm. However, the floating-gate technique requires a pre-charge phase which is inconvenient and unattractive [86]. Lastly, there is also work reported on external V TH compensation (EVC) achieved by integrating external power supply or external control such as external clocking switched-capacitor circuit, where a sensitivity of −14 dBm has been recorded [87]. However, compensation of V TH usually comes with the trade-off in reverse leakage current. To offset the effect of reverse leakage current, [88], [89] proposed an adaptive threshold voltage compensation scheme while [90] investigated an optimization for bias voltage (V C ).
The CCDD rectifier is a topology derived from the fullwave diode-bridge rectifier. Commonly CCDD-type rectifier consists of two NMOS transistors (M N1 , M N2 ), two PMOS transistors (M P1 , M P2 ) and two coupling capacitors (C C1 , C C2 ), as shown in Fig. 6(b). The integration of coupling capacitors in the circuit allows the AC RF input signal to pass through while blocking the DC charges.
During the negative cycle, M P2 and M N1 turns on while M P1 and M N2 turns off. At this point, it creates a complete loop for the rectification. During the positive cycle, M P2 and M N1 turns off while M P1 and M N2 turns on. In this case, the complete circuit loop with the same current direction respective to the negative cycle occurs.
CCDD-type rectifier achieves high peak PCE but narrow PDR [8], [9], [77]. This transpires as the CCDD-type rectifier suffers from unwanted reverse leakage current at high RF input [17]. When the number of stages increases, the output voltage may be higher than the input voltage. In this case, the reverse leakage current becomes severe, degrading the overall performance. To increase the performance of CCDDtype rectifier, research in eliminating or lowering the effect of reverse leakage current have been initiated. There are two main approaches: to detect and provide the counter solution to the high reverse leakage current or to build a high resilience circuit that is capable to withstand the effect of reverse leakage current. For the detection and counter solution, [91] and [92] proposed a feedback network to detect the strength of reverse current leakage and control the gate biasing of the transistors, where a peak PCE of 65% is achieved. In the second approach, [24] and [93] proposed complementary MOSFET with the aid of thick-oxide MOSFET to reduce the reverse leakage current, where a peak PCE of 75.2% has been reported. Subsequently, [94] and [95] proposed the integration of bootstrapping capacitors to minimize the leakage current effect. An 80% and 87% peak PCE have been recorded from the two projects, respectively.
In comparison to the two topologies, additional consideration should be accounted for in making the selection. The number of transistors required for the two topologies is different. A single-stage Dickson-type rectifier requires two transistors, while a single-stage CCDD-type rectifier requires four transistors. With the same dimension of the transistor, the transistor count of the CCDD-type rectifier doubles in comparison to the Dickson-type rectifier, with the penalty of increased area consumption. However, a single CCDD-type rectifier may achieve a higher PCE than the Dickson-type rectifier. Thus, the CCDD-type rectifier could potentially achieve the same PCE as the Dickson-type rectifier by comparatively adopting fewer stages, thus requiring a smaller area.

IV. MULTIBAND RFEH FRONT-END CIRCUITRY
Similar to the typical RFEH front-end system, the multiband RFEH front-end system consists of three main components, which are the multiband antenna, multiband impedance matching network and multiband rectifier. There are two main challenges in designing the multiband RFEH front-end circuit. The first challenge is that the front-end system must be capable in harvesting multiple frequency from the ambient environment. To cope with this, multiband antenna designs are needed. VOLUME 11, 2023 The design challenge is also foreseen due to the non-linear characteristics of the rectifier [96], [97]. The rectifier is a non-linear device where the impedance varies respectively to the change in the operational frequency. In this case, a multiband or wide band impedance matching network is necessary to reduce the reflected power loss. Various impedance-matching networks have been proposed to harvest single-band frequency [98], [99], [100]. However, in the context of multiband RFEH systems, the single-band impedance matching network, which is initially targeted to match a single frequency, will produce significant reflection loss due to the impedance mismatch at a non-matching frequency. The working principle of the RF-DC rectification process is the same as with the single-band or multiband RFEH systems.

A. MULTIBAND RADIO FREQUENCIES
To design a multiband RF energy harvester, the first step is the selection of the targeted harvesting frequency. Referring to Fig. 2, 433 MHz, 900 MHz, 1.8 GHz, and 2.4 GHz are the four most commonly harvested center frequencies due to their density of use, leading to high signal strength. However, depending on the location and condition, the preferred frequency may have a relatively higher signal strength. In addition, there is also another potential harvesting frequency, such as 875 MHz, 1.9 GHz, and 2.5 GHz. In summary, the multiband RF is usually selected from the five listed center frequencies as highlighted above with a tolerable range.
However, a hidden trend can be observed by observing the improvement of GSM 2G, UMTS 3G, LTE 4G to 5G NR, where the preferred frequency increases from around 900 MHz to the current widely used 5G frequency of around 4 GHz. The current highest 5G frequency can be up to around 50 GHz. Therefore, this outlays a considerable gap of novelty to be discovered.

B. MULTIBAND RF RECEIVER ANTENNA
Antennas with multiband features can be designed easily by implementing multiple single-band antenna designs (MA) to harvest multiple frequencies simultaneously, as shown in Fig. 7(a). The example of the single-band antenna can be referred from Section III. However, such designs usually require a larger area. Therefore, a single antenna design with a multiband feature (SA) has been proposed.
A single antenna with the multiband feature can be classified into two categories: multi-narrowband or broadband design. The multi-narrowband antenna is designed to precisely create resonance at the desired frequency to harvest multiple frequencies with minimal losses, as shown in Fig. 7(b). On the contrary, the broadband antenna is designed to create a wide resonance covering all the desired frequencies to achieve multiband harvesting, as shown in Fig. 7(c).
Designing a wide lossless region for broadband antenna has its pros and cons. The advantage is that the wide broadband approach has higher adaptability compared to the multi-narrowband approach due to broad exposure to ambient frequency. However, the trade-off of using a broadband antenna is that the reflection loss is higher than the multi-narrowband antenna. This is because narrowband antennas can be optimized to harvest a specific frequency band. Moreover, the broadband approach has difficulty in producing an adequate wide region for two frequencies that are far apart, such as 900 MHz and 2.4 GHz. Therefore, the selection of an antenna may depend on different application scenarios.
For multi-narrowband harvesting of 900 MHz and 1.8 GHz, [101] proposed a dual-band second monopole antenna with the respective measured gain of 1.8 dBi and 2.06 dBi. The size of the proposed antenna is 9600 mm 2 . To miniaturize the antenna size, [102] proposed a dualband inverted-L (IL) shaped monopole antenna which is eight times smaller than the dual-band second monopole antenna with a size of 1200mm 2 . As a trade-off, the proposed design comes with a lower gain of 1.46 dBi and 1.91 dBi at 900 MHz and 1.8 GHz respectively. Subsequently, [103] proposed a 2500mm 2 dual-band monopole antenna design. This design has recorded the highest gain of 2.36 dBi among the three monopole designs. Other than dual-band harvesting from 900 MHz and 1.8 GHz, [104]  There are also several proposed design to harvest from tri-band and quad-band. Reference [106] proposed a triband slotted/slitted antenna to harvest energy from 2 GHz, 2.5 GHz and 3.5 GHz. The measured gain for the tri-band are 7 dBi, 5.5 dBi and 9.2 dBi respectively. To realise tri-band harvesting, the planar implementation requires a large size of 25600 mm 2 . Comparatively reporting with the same size of 25600 mm 2 , [107] proposed a quad-band bow-tie antenna to harvest from 850 MHz, 1.8 GHz, 2.18 GHz, and 2.4 GHz. The highest measured gain for this quad-band design is 4.82 dBi at 2.4 GHz. Improvising from [80] and [108] proposed a quad-band fork-shaped antenna to harvest from 950 MHz, 1.81 GHz, 2.42 GHz, and 2.9 GHz. The design recorded the highest measured gain of 7.6 dBi for 2.42 GHz. However, the implementation requires a larger size of 48400 mm 2 .
For broadband harvesting, [109] introduced a 64-element dual-circularly-polarized spiral-shaped antenna array with a broad harvesting frequency from 2 GHz to 18 GHz. Due to the presence of 64 spiral elements array, the realization of this design requires a comparatively large size of 33300 mm 2 . On the other hand, [110] proposed a 1 × 4 Quasi-Yagi antenna array that produces a lossless broadband region that includes the desired frequency of 1.8 GHz and 2.1 GHz. The work measures a gain of 13.3 dBi, which is the highest among the reviewed design. However, the bottleneck of this solution is in the large size of 38000 mm 2 . To minimize the size, [111] proposed a planar cross dipole dual-polarized antenna while [112] proposed a bent triangular omnidirectional antenna to produce a lossless broadband region to harvest from their desired dual-band frequency respectively. Both works report a smaller size of 4900 mm 2 and 7708 mm 2 correspondingly. However, their respective measured gains are lesser than [110], which are 2.5 dBi for 1.8 GHz and 3.95 dBi for 2.5 GHz as in [111] and 2 dBi for both 950 MHz and 1.85 GHz in [112].

C. MULTIBAND IMPEDANCE MATCHING NETWORK
The impedance matching network is essential in the RFEH front-end system to eliminate the unwanted impedance mismatch reflection loss. In the implementation of an IMN-less rectenna design as shown in Fig. 8, the need for the IMN network is eliminated. IMN-less rectenna is a device with a customized antenna and rectifier to convert RF energy into DC voltage output. References [113] and [114] proposed a tunable rectenna capable of tuning the antenna's impedance and matching it conjugately with the impedance of the rectifier. In this design, the impedance-matching network block can be excluded from the multiband circuit. However, the integration of a rectenna in the multiband RFEH system has its downsides. The antenna has to be specially customized, contributing to higher cost, complexity, and low adaptability.
In brief, this section lists the state-of-the-art multiband IMN. As the harvesting frequency and the impedance of the rectifier are completely different, the size of the multiband IMN designs cannot be used to indicate the performance of the IMN designs. Therefore, to ensure fairness in comparison, the reflection loss S 11,dB will be the only indicator in determining the performance of the design. As highlighted, the multiband IMN design with S 11,dB less than −10 dB for both bands will be considered as matched.
Abstaining the need in customizing the antenna, RFEH system researchers proposed the use of a transmission line (TL) or microstrip line [115], [116], [117], [118], [119], [120] as the impedance matching network and integrates the diode as the rectifier to build a simple multiband RFEH system, as shown in Fig. 9. Schottky diodes are the preferred selection for the rectifier in this method [121], [122]. In general, the transmission line impedance matching network can be classified into single branch matching networks (SBMN) or multiple branch matching networks (MBMN). Both methodologies differ in the number of branches required for multiband matching.
For the SBMN implementation, several designs have been proposed resonating at 915 MHz and 2.45 GHz. Reference [115] proposed a series combination of Pi-matching TL network and L-matching TL network. This design requires a total of five TL segments with an approximate average segment dimension of 1.3mm/29.3mm (width/length) and an approximate matching network circumference of 295mm. To minimize the size, [116] proposed a novel stepped impedance stub matching network built from three TL segments with an approximate average segment dimension of 1.3mm/12.2mm and an approximate matching network circumference of 76mm. Smaller TL segment dimension leads to an improvement in size optimization. In addition, [117] reported a three-TL segment single-stage T-matching TL network design with an approximate average segment dimension of 0.7mm/26.5mm and an approximate matching network circumference of 147mm. The design outstands for its smallest width of the TL segment. All three SBMN designs recorded S 11,dB of −40 dB, −22.5 dB and −25 dB respectively, complying to the matching condition of less than −10 dB.
In addition, several MBMN designs have been reported. Reference [118] proposed a dual-branch T-junction coupled line matching network to harvest energy from a dual band of 915 MHz and 2.45 GHz. Both branches consist of three TL segments, leading to a total of six TL segments. The approximate average segment dimension is 1.9mm/20.9mm. Respective of a total of six TL segments, the design requires a bigger approximate matching network circumference of 299mm as a trade-off. Subsequently, [119] proposed a triple branch impedance matching network built from a single radial stub, a single short stub, and a TL segment at each branch, contributing to a total of nine TL segments. This design comes with a small approximate average segment dimension of 3.4mm/12.5mm. However, the use of nine TL segments contributes to a larger value of 219mm for the approximate matching network circumference. Both the multiple branch design measures a S 11,dB of −20 dB and −21 dB respectively.
In brief, the TL matching network can achieve a matched condition between the antenna and the rectifier. In addition, the TL matching network is known to have greater efficiency in impedance matching of frequencies higher than 2.6 GHz compared to the LC-based matching network [123]. From the research work [123], the insertion loss for the TL matching network and LC-based matching network are benchmarked, and the result shows that the insertion loss for the TL matching network and LC-based matching network is similar in the frequency of 2.6 GHz and less. Beyond the frequency of 2.6 GHz, the TL matching network has a significant advantage over the LC-based matching network and can achieve a lesser insertion loss. However, most of the congested harvesting frequencies, are lower than 2.6 GHz, concluding that the use of a TL matching network is less significant. Furthermore, the downside of using the TL matching network is the larger required size due to the off-chip PCB integration.
Apart from the off-chip PCB approach of rectenna, stripline, and microstrip line implementation, the CMOS approach adopts passive components such as resistor, inductor, capacitor, or transformer in designing the multiband impedance matching network.
The CMOS-based SBMN focuses on a series combination of several matching networks, such as the implementation of two L-matching networks in series to match with dual-band [124]. Besides the combination of series networks, SBMN may also employ a tunable approach, as shown in Fig. 11; however, this requires an extra loop or circuit to release the control signal, increasing the circuit complexity and uncertainty. Alternately, the CMOS-based MBMN focuses on a parallel combination of several single-band matching networks. For instance, a dual-band matching network can be built from two single-band matching networks, where each single-band matching network is responsible to match with a single frequency.
In reference to SBMN, [124] proposed a fully on-chip series double L-matching network produce two resonant frequencies at 915 MHz and 2.4 GHz. The reflection loss, S 11,dB are −37 dB and −33 dB for the dual frequency respectively. The work [125] proposed an on-chip tunable single L-matching network. Adopting a tunable LC matching network, the S 11,dB reflection loss can be as low as −36 dB and −40 dB at 900 MHz and 1.9 GHz respectively. Alternatively, [126] proposed a tunable 4-bit capacitor bank based broadband impedance matching network to match with a frequency range between 700 MHz to 900 MHz. The work reported a reflection loss S 11,dB of less than −10 dB in the defined frequency range. In brief, onchip design are capable to maintain the perfect matched condition. However, on-chip impedance matching network suffers from low Q-factor for the inductor and capacitor due to high losses in the substrate [135]. To address this issue, several work in employing off-chip inductor and capacitor are proposed [70], [127], [128], [129], [130], [131], [132], [133], [134]. Reference [127] proposed a novel impedance matching network with a 1:100 transformer and a single LC matching network. Reference [128] proposed two external SMD inductors as the matching network and fine-tuned it to match at dual frequencies of 900 MHz and 1.8 GHz. Reference [129] proposed a 900 MHz and 2.4 GHz dual-band impedance matching network built from two L-matching networks. Due to the differential RF input, a total of four offchip inductors and capacitors are used. The reflection loss, S 11,dB is measured as −15.8 dB and −17.8 dB at 900MHz and 2.4GHz, respectively. Subsequently, [130] recorded a reflection loss of −24.8 dB and −23.5 dB for 902 MHz and 2.45 GHz respectively using an off-chip matching.
In reference to on-chip MBMN design, [131] proposed a dual-parallel L-matching network to match with 915 MHz and 1.85 GHz respectively. The work [134] proposed a Pi-matching network to match at 915 MHz and an L-matching to match at 2.4 GHz. This work reported a reflection loss S 11,dB of −16 dB and −20 dB for both of the bands. Additionally, there are also some reported design on off-chip MBMN, such as dual-parallel branch design [133] and triparallel branch design [70], [132].

D. MULTIBAND RF-DC RECTIFIER
Unlike the single-band RFEH system, the multiband RFEH system aims to harvest from multiple frequencies, resulting in multi-branch input power at the rectifier block. Concurrently, a multi-branch rectifier (MBR) can be proposed where each branch is individually responsible for processing a single branch of input power. The output of each branch combines to form a larger output [70], [134]. However, MBR encounters current leakage between the branch due to the inherent potential difference. In addressing this setback, [131] proposed a diode summation network and switch summation network to control the leakage current in between the branch, hence, improving the output summation voltage. Moreover, [136] introduced an intelligent shortcut switching summation network that can improve the power summation performance at higher input power. Furthermore, [137] analyzed and introduced a new connection for the diode summation network by connecting a single output node to all other output nodes with the usage of a properly oriented and low leakage diode. As simplified in Fig. 14(a) and (c), the diode or switch summation network is implemented as a power combiner to allow the summation of power for a multiband RFEH front-end system [131], [136], [137].
Alternately, the multiband RFEH system could combine the input power at an early stage, leading to a single branch input power to the rectifier block. The implementation promotes a single-branch rectifier (SBR) realization. An SBR design benefits the Dickson-type which suffers from poor sensitivity due to the presence of threshold voltage. With the aid of high input power, the threshold voltage can be easily overcome [138].
However, the CCDD rectifier suffers from reverse leakage current, especially at high input power. Thus, the SBR design will worsen the reverse leakage current of the CCDD-type rectifier, leading to a drop in the output voltage [139]. In addressing this bottleneck, methods to eliminate the reverse leakage current have been presented in Section III. These methods can be adopted for multiband energy harvesting schemes. For instance, [128], [131], and [134] proposed a multiband RF-DC rectifier with threshold voltage selfcompensation to reduce the reverse leakage current. The proposed technique would improve the performance of the CCDD-type rectifier for multiband energy harvesting.

V. MULTIBAND RFEH FRONT-END SYSTEM CIRCUIT INTEGRATION
The rectenna design, as in [113], [114], and [140] will not be considered in this part. The reason is that the rectenna design requires a specially customized antenna that will not fit with the standard sensor nodes and IoT solution available in the market designed to connect with a standard 50 antenna.
A multiband RFEH front-end system will use the combination of the multiband circuitry listed in Section IV. There are two initial types of multiband antenna design (SA and MA), two types of multiband IMN design (SBMN  and MBMN), and two types of multiband rectifier design (SBR and MBR). Therefore, eight different multiband front-end system combinations are possibly realized as shown in Fig. 12.    From the eight combinations, two considerations have to be followed to avoid unwanted output voltage drop. The first consideration is in between the impedance matching network and the rectifier. Based on the rectifier cascading scheme analysis reported in [141], the output DC voltage is dependent on two main parameters which are the number of cascaded rectifier stages, and the rectifier input voltage: where V OUT is the output voltage, V IN,REC is the input voltage to the rectifier, N is the number of cascaded rectifier stages, and the Losses which includes the threshold voltage losses and matching losses. Therefore, the SSM-type and MSM-type where SBMN connects with MBR divide the input current to each branch of the rectifiers which directly results in the output current drop as given in (15). Assuming the SSM-type, the output voltage can be derived from [141] where V OUT ,SSM is the output voltage for the SSM-type and N B is the number of rectifier branches. Comparing (15) and (17), the output DC voltage drops with a ratio of (1/4N B ).
To compensate for this, more stages of rectifier at each branch are required, inevitably increasing the total size. The second consideration assumes MBMN connects to SBR, as shown in Fig. 13. This causes undesired impedance matching problems as the matching branch in MBMN are now parallel to each other. The inductance and capacitance for each branch are combined to form single inductance and single capacitance, leading to a match at a single frequency. Thus, the SA-MBMN-SBR (SMS-type) and MA-MBMN-SBR (MMS-type) are less favorable. From the two considerations, it concludes that the IMN and rectifier must have the same number of branches to avoid the division of rectifier input power and prevent the undesired matching problem. Alternately, the remaining four combinations are used separately in two different cases. The first case utilizes the advantage of a multiband RFEH system with the ability to scavenge multiple input power from multiple frequencies simultaneously, which leads to higher output power. To harvest multiple frequencies concurrently, the system must consist of an MA design. With this regard, the potential options are the MA-MBMN-MBR (MMM-type) and MA-SBMN-SBR (MSS-type) as shown in Fig. 14(a) and Fig. 14(b) respectively. The former design sums up the output power from each rectifier branch, resulting in a larger single output power. The latter design sums up the received power from the antenna, leading to a larger single input power for the following IMN and rectifier. About these two circuit combinations, harvesting multiple frequencies simultaneously improves the output PCE [142], circuit sensitivity [70], and dynamic range [127], [132], [142]. Reference [142] reports an improvement of 8%, 18%, and 25% in PCE when combining the tri-band compared to the respective single band. Reference [70] reports a triple parallel branch RFEH front-end system using a 3-stage gate crosscoupled (FGCC) that produces a 4.6 dB enhancement in the sensitivity when combining three bands compared to a single band. On the other hand, [127], [132] reports a 30 dB wide operation range by summing the output from dual band from the RFEH front-end system.
Besides, the second case has the advantage of a multiband RFEH system with the ability to operate in multiple harvesting frequencies, leading to a wider harvesting system. Unlike the first case, the second case does not require harvesting multiple frequencies at the same time. In the absence of these constraints, the system can use the SA design. The potential options are the SA-MBMN-MBR (SMM-type) and SBMN-SBR (SSS-type) as shown in Fig. 14(c) and Fig. 14(d) respectively. By adopting the SSS-type [124], [130] and SMM-type [134], the circuit benefits in terms of sensitivity and dynamic range as shown in Table 4.
In brief, the four different combinations of topology in harvesting multiband frequency results in a dedicated output voltage formulation in reference to (15). The derivation is based on the assumption that the system is matched, with the matching loss neglected. In reference to Table 5, with the exact amount of rectifier stage, MSS-type obtains the highest output voltage due to multiple antennae that harvest multiple sources simultaneously, leading to the increase of input received power to the rectifier block. Subsequently, the SSS-type and MMM-type lead in the output voltage. The SSS-type with a single antenna limits the received power from a single source, leading to a slightly lower performance than the MSS-type. The MMM-type topology integrates multiple branches with the distributed amount of VOLUME 11, 2023 cascaded rectifier, thus degrading the boosting. However, the SSS-type and MMM-type topology achieves higher output voltage than the SMM-type. The SMM topology limits the input received power per time, further degrading the output voltage. However, it is worth noting that although the SSS-type and MMM-type topology exhibits the same output voltage expression, the design consideration is entirely different. The MMM-type multiband RFEH system has a similar architecture as in a combination of multiple singleband RFEH systems. On contrary, the SSS-type adopts an entirely new circuit architecture that only consists of singlebranch multiband components. At the same time, the design approach for an SSS-type topology is completely different from the existing single-band system, thus wider research opportunity is established.
Based on Table 6, the four types of multiband RFEH frontend circuit enables good performance. The ideal performance matrix for the multiband RFEH front-end circuit tends to be a high PCE, high sensitivity, and wide power dynamic range. However, based on Fig. 15(a), most of the research work exhibits a trade-off between PCE, power dynamic range, and sensitivity. The work [117] shows a high PCE of 81.7% and a wide power dynamic range of 24 dB, but with a comparatively lower sensitivity at 0 dBm. Alternately, [131] has achieved good sensitivity of −33 dBm with the trade-off of lower PCE around 43.2%, simultaneously sacrificing the power dynamic range to 9 dB. In summary, good conduction leads to high sensitivity, however, at higher input power, PCE drops along with the power dynamic range due to the reverse current flow from the output to the input. Thus, the design trend emphasizes PCE enhancement. Nevertheless, sensitivity is a critical performance in the RFEH and is foreseen as a future research opportunity. Fig. 15(b) shows the PCE performance achieved by each harvesting frequency in the state-of-theart multiband RFEH front-end system. Fig. 15(b) shows the research direction of prior works in the range of 90 MHz, 900 MHz, 1900 MHz, and 2400 MHz bands.

VI. CONCLUSION
This article reviews the recent state-of-the-art multiband RF energy harvesting front-end system, presented from the investigation of the RF survey in RF transmission and RF bands to the overview of the front-end architecture with the stateof-the-art multiband antenna design, multiband impedance matching network design, and multiband RF-DC rectifier design. The performance of prior state-of-the-art architecture is summarized and compared to investigate the advantage and disadvantages of different circuit architectures for multiband RFEH front-end systems. Subsequently, a section on the integration between the three core blocks of the multiband RFEH front-end has been discussed. In addition, critical considerations in designing a multiband RFEH front-end system are also provided to realize optimum performance targeting ultra-low power biomedical chips [13], [14], [15]. In summary, this article fills the research gap in the further advancement of multiband RF energy harvesting towards enhancing its performance through optimal circuit integration of the front-end system. He is currently an Associate Professor with the School of Engineering, Asia Pacific University of Technology and Innovation, working in the area of micro-and nano-electronics. Since 2009, he has been publishing most of his research work in the IEEE TRANSACTIONS, the American Institute of Physics, the American Physical Society, and Nature Group journals. His research interests include semiconductor micro-and nano-fabrication, quantum dot devices, cryogenic temperature measurements, single electron transistors, quantum computation, and silicon microdosimeters.