The presented above analysis showed that the target application is potentially feasible and provided valuable insights into its implementation (e.g., the band). In the current section, we will go step-by-step through the key components of the system, justifying their selection or describing their design and evaluation.
3.4.1. Selection of a Sensor Node
This is obvious that the very low energy consumption of the IoT sensor node is crucial for the success of the targeted application. Today, there are many IoT sensor solutions available on the market, including the ones designed for energy-harvesting supply. Therefore, we start our development by selecting the appropriate sensor node device, which has to address the following challenges.
The first challenge is related to the possibility of strong variation of the input radio signal’s power resulting from, e.g., the mobility of the device, blockage of the radio channel, or the temporal pattern of the base station operation. To address this challenge, the power system of the IoT device needs to combine a DC–DC converter stabilizing the supply voltage as well as an energy buffer, which would provide the energy when no radio signal is available or when it is too weak.
Notably, the peripheral components (e.g., sensors and actuators) of an IoT node may require possessing different voltage levels. In this case, the feasibility of using such components and of generating multiple supply voltage levels need to be carefully evaluated, since each new DC voltage increases the total energy consumption, especially in the case if the desired voltage is higher than the original one. At the very same time, the use of higher supply voltages may increase the power consumption, which is especially undesirable for energy-harvesting powered systems. Therefore, a single-voltage DC–DC converter generating the minimum level of the voltage needed for the operation of the whole IoT node is typically the most efficient solution.
The energy buffer, in its turn, must store sufficient amount of energy to support the operation of the node during the period when no energy income is available. Depending on the use case and the environment of operation, as an energy buffer can serve a conventional capacitor, a supercapacitor or a secondary battery. The former is characterized by small capacity and thus can only help to handle decently short (typically—minute-order) periods of energy unavailability, but are less expensive than the other options. The supercapacitors have higher capacitance and thus can enable longer autonomy, but are much more costly. Finally, the secondary batteries feature decently high capacity and reasonable costs, but often need high current to get effectively charged.
The key requirement for the processing and the wireless communication systems is the low consumption not only in sleep mode but also in active mode. The radio technology needs to feature low peak and mean power consumptions, which calls for the technologies with low transmit power, high over-the-air rate, and simple communication protocols. The BLE or IEEE 802.15.4 are typical examples of such radio technologies. Another important aspect is the reduction of consumption for communication between the processing core and the radio. In this respect, the system-on-chip based systems outperform the ones with discrete microcontroller and radio.
Finally, the proper selection of the sensors and other peripherals is of extreme importance. Based on the discussion above, for our implementation, we selected a sensor node from Solmu Technologies (Oulu, Finland) [
62] designed and optimized specifically for energy-harvesting powered operation. A node (depicted in
Figure 3, the structure is shown in
Figure 4) has in its core a multi-radio-technology-enabled system-on-chip based on 32-bit Advanced RISC Machine (ARM) controller (consuming 100 nA in shut down and below 3 mA operating with a clock of 40 MHz). To enable the energy harvesting supply, the node is equipped with a DC–DC converter, which can convert an input voltage from 0.1 V to over 5 V to a stable DC voltage in the range from 2 to 3.6 V. Importantly, having the ten
A current on its input, the converter demonstrates the efficiency of over 80%. As the energy buffers, the tantalum capacitors storing up to 10 J of energy storage in total are used. A sensor node also has the place for a single CR2032 lithium battery, which can be used as an alternative or a backup power source option. The node is equipped with an instrumental-grade external analog-to-digital converter and with low-consuming environment sensor (i.e., temperature and humidity) and three-axis acceleration sensors.
For this node, we have developed the firmware, making the sensor node measure the acceleration, process it (i.e., calculate the Fast Fourier transform (FFT) to estimate the speed of rotation) and the temperature, and send them twice a second in BLE advertisements. The practical energy consumption profile of the node for this specific application, which was measured with an Agilent N6705 power analyzer (Santa Clara, CA, USA) and then visualized in MATLAB, is depicted in
Figure 5. The data processing and communication phases, and the key numbers related to the consumption are shown in the picture. One can see that the mean power required for stable operation of the system is 1.36 mW.
This consumption is higher than the one we expected and used in our feasibility analysis. The three reasons for it are the higher consumption of the sensors, the extra consumption for preparing the packets for the BLE, and, especially notable, the additional consumption for signal processing. The calculation of the FFT required quite substantial processing, which also brought along with it the extra consumption. Nonetheless, one can see that, even with this consumption, the application stays feasible for WPT in h1.6, even though the communication range will likely decrease to below 1.5 m.
3.4.2. Antennas for RF Energy Transfer
The antennas of both the base station and the IoT node are the key elements, affecting the efficiency of the WPT. Depending on the number of the nodes served by a single WPT transmitter and whether the location of all the IoT nodes is known and static, one can use two different approaches. The former one is based on the use of omnidirectional antennas and is intended for the case when the nodes are uniformly scattered around the base station or are highly mobile. As analyzed, e.g., in Ref. [
46], this approach is characterized by the minimum energy transfer efficiency. The use of directive antennas enables increasing the WPT efficiency (refer, e.g., to Ref. [
63]) but is sensible only if the location of the nodes is known or if they are clustered most of the time. Finally, a combination of these techniques (e.g., use of directive antennas for base station and omnidirectional on the nodes, or vice versa) can be considered for particular use cases. Note that the requirements of the real-life applications often limit the antenna dimensions, thus putting a limit on antenna’s efficiency.
Given the geometry of our target application setup, the use of directive antennas is reasonable. The conducted market analyses have shown that no directive 868 MHz band commercial antennas are available. Due to this reason, we designed and prototyped our own linearly-polarized patch antenna, layout, and dimensions, which are illustrated in
Figure 6. The reason for us basing our design on the patch antenna was twofold. First, this type of antenna is known to feature higher directivity than the dipoles. At the very same time, these antennas are still sufficiently simple and technological for mass-scale production.
The designed antenna was first simulated in SW, and then two pieces have been manufactured using a low loss Rogers RT5880 substrate (
, loss
) with 3.1 mm substrate thickness and a 35
m copper cladding. The dimensions of the ground plane are 200 × 200 mm
2 and the dimensions of the radiator are 172 × 112 mm
2. A large ground plane surface increases the antenna size but also enables a more directive beam and makes the characteristics of the antenna less dependent on mounting location and environment. This makes the antenna better fit our targeted environment. The 50 Ohm coaxial line is used to feed the antenna. One of the manufactured antennas is depicted in
Figure 7.
The simulated and measured values of S11 parameter and the radiation pattern of the designed antenna, demonstrating its performance, are depicted in
Figure 8 and
Figure 9, respectively. The resulting efficiency of the designed antenna, defined as the power radiated over a sphere compared to the power delivered to the antenna port, is on the order of 87–98% for the 860–870 MHz band. The maximum antenna gain is 7.55 dB at 865 MHz, and the width of the beam (according to the 3 dB level) is 75 degrees.
3.4.3. RF–DC Converter
The third key component of the solution is the RF–DC converter. Likewise, for the antenna, there are almost no ready-made commercial RF–DC solutions available today. The only exception to this rule is the P2110B RF energy harvesting chipset [
18] from Powercast
® (Pittsburgh, PA, USA), the cost of which exceeds 30 EUR (34 USD) for one chip [
64]. In larger quantities, the cost may reduce, but still stays rather high. Therefore, we developed our own solution, which is discussed in what follows.
The first challenge with respect to the RF–DC converter design is the selection of the diodes to be used. The components need to be picked mindful of such parameters as the threshold and reverse-breakdown voltage, element parasitics, and harmonics. The former parameter defines the minimum level of the radio signal, from which the circuit could collect the energy (i.e., the “sensitivity” of the circuit), and affects the efficiency of the energy harvesting. The junction resistances and capacitances as well as the package inductances and capacitances influence the efficiency and limit the maximum operating frequency. Finally, the cost and the dimensions of the diodes also play a role, since a single rectifier circuit may employ multiple diodes.
Second, one has to decide the architecture and the number of stages in the rectifying circuitry. Some of the possibilities are reported and discussed, e.g., in Ref. [
11]. Importantly, the number of stages needs to be specified based on the expected range of the input radio signal’s power. Specifically, more stages enable collecting the energy from weaker signals, but at the same time introduce some losses which affect the efficiency of the system when harvesting the power from stronger signals. In addition, the more stages increase the dimensions and the price of the system. The former challenge can be addressed by employing a multi-path or switching rectifiers (e.g., the ones proposed in Ref. [
65]); nonetheless, this makes the design of a rectifier even more complex and costly.
Finally, the rectifier has to be matched with the antenna to prevent the energy reflections to the environment. Such matching is a non-trivial task due to the nonlinearity of the diode’s characteristics and namely the variation of the diode’s impedance depending on the power and frequency of the input radio signal and needs to account for the specifics of the Printed Circuit Board (PCB) and the manufacturing processes.
For this study, we have designed and produced own energy harvesting circuitry shown in
Figure 10. The form factor of the PCB is such that it can be connected to our in-house developed modular IoT HW platform [
2,
66]. For the board, we used FR4 with 0.8 mm substrate height and 0.18
m trace thickness. The substrate dielectric is
. With these parameters, 1.4 mm wide line provides 50 Ohms impedance. We applied a Dickson charge pump built using the Avago HSMS 285C pair serial connected diodes (San Jose, CA, USA), which are optimized for low voltage operation. The schematic of the designed RF–DC converter board is illustrated in
Figure 11.
After assembling the PCB, its matching to the antenna was manually tuned. Namely, we used the Anritsu MS4623B vector network analyzer (Atsugi, Kanagawa, Japan) to estimate the input impedance and used it to calculate the matching components. Furthermore, the matching components were fine-tuned to fight the imperfections and tolerances of the components themselves, as well as to address the effects of manual component assembly. The measured reflection coefficients for the designed board are illustrated in
Figure 12. We also measured the efficiencies of the designed RF–DC converter board for the different power levels—the respective results are presented in
Table 4. As one can see, the performance is in the same range as that of the state-of-the-art works, discussed in
Section 2. In addition, one can see that the designed board demonstrates the peak efficiency for the input signals with units-of-milliWatt power levels, which is well in line with what is needed for our application. Note that the reported measurement results were obtained having the resistive load of 9 kOhm connected to the circuit. The level of the load was selected based on the mean power consumption of the sensor node running our target application. A few iterations of the RF–DC converter design were executed to find the appropriate balance between the efficiency and the level of the output voltage, which needs to exceed the minimum threshold voltage for sensor node’s DC–DC converter.
3.4.4. RF Power Source and Gateway
The two final components of our envisaged system are the wireless power transmitter and a gateway for receiving the reports. We start with the latter. Since we have selected the BLE to be the radio access technology in our test system, this provides us with a sheer diversity of options for the gateway. Specifically, to receive the advertising packets sent by our device, the dedicated BLE transceivers or the mobile devices (i.e., tablets, smartphones or laptops) can be used. For our test case, we have developed two gateway options. The former is a dedicated HW BLE gateway, which monitors and logs the data sent by the WPT sensor for further processing. Another option bases on an Android-based tablet computer, for which we developed a simple application to read the data received from BLE and display them on the screen in real time.
Finally, the strong-enough source of the radio signal to power the designed sensor solution was needed. At first, we attempted to find a commercial solution. The only one available in the market is the TX91501 device [
67] from Powercast
® (Pittsburgh, PA, USA). Unfortunately, this device did not fit us since it:
Operates in 915 MHz US ISM band,
Has a fixed power level of either 1 or 3 Watt,
Does not enable duty-cycling,
Does not permit the connection of external antennas.
Therefore, we decided to design the solution matching better our needs. This was done based on our IoT modular HW platform [
2,
66]. Specifically, we used a node composing a microcontroller and an 868 MHz ISM-band LoRaWAN radio transceiver (the RN2483 from the Microchip (Chandler, AZ, USA) in our case). The transceiver was forced into the test mode, transmitting the radio carrier. The microcontroller can switch the transmission on and off, thus enabling to implement duty-cycling. Since the radio transceiver can transmit with the maximum power of only 14 dBm, we connected to it an external amplifier—the Kalmus 710FC (Souderton, PA, USA). This device amplifies the signal to the level of 27 dBm (500 mW) and applies it to an external antenna—the one we described earlier.