The fundamentals of energy harvesting design
The proliferation of power-miser sensors, microcontrollers and RF transceivers is ushering in wireless sensor networks powered exclusively or supplemented by energy harvesting techniques. Ultra-low-power wireless protocols are ever more widely used and standards are in active development. Sensor networks unshackled from the ac mains and batteries open the possibility for greater reliability, less maintenance, and better safety. Applications unthinkable only a few years ago are now possible with energy harvesting techniques.
New power management products can squeeze usable power out of intermittent and often miniscule outputs from such energy harvesting transducers as TEGs (thermo-electric generators), photovoltaics, piezoelectric materials, and electromagnetics. But designing for these minute amounts of power requires a new way of specifying and analyzing power management devices.
The typical wireless sensor node includes an energy source, processor, and an RF link. The missing link in this system to date has been a means of power management. This function is necessary because transducers available to harvest power are often inconvenient to work with. They generally provide either an extremely low-voltage, low-impedance output, or a high-voltage, high-impedance output. The various elements in the typical system can be further divided into power producers/regulators (transducer and power management) and power users (everything else). Simply put, if the average energy the system can harvest exceeds the average power the remote sensor electronics requires, you have the possibility for an autonomous system.
Before initiating any design, it is worthwhile to run a quick feasibility analysis. This will determine whether energy harvesting techniques are even practical. The first step is to decide how often measurements must take place and be transmitted. We will call this the measurement frequency, F. Next, we can determine the amount of processing power necessary for handling the sensor, signal conditioning, data conversion and processing, plus the amount of power needed for transmitting data via the RF transceiver.
The accompanying table shows the typical power requirements for a widely used microcontroller and RF link system. These power requirements can vary from one manufacturer to another for a particular application. There are numerous choices, and they can be optimized for specific end applications. The table provides enough information to calculate the system duty cycle and average power. The duty cycle, D, of the system is defined as:
D = Tm + Tp + Tt × F
where Tm = measurement time, Tp = processing time, Tt = transmit time, F = measurement frequency. The average power, Pa, is simply
Pa = P × D + Ps
where P = total power and Ps = standby power, which is generally small enough to be ignored.
For example, assume the task is to design an autonomous indoor temperature sensor. This sensor will be deployed throughout a large office building and coupled with proximity sensors that can detect when a room is occupied and adjust the temperature accordingly. Deploying this type of sensor within a large building can reduce heating and cooling costs significantly.
The sensors require 500 µA at 3.3 V for 2 msec to measure temperature and detect an occupant. A low-power microcontroller must operate on this data for another 5 msec. The processor consumes 3 mA at 3.3 V when processing the data. Finally, the RF link requires 30 mA at 3.3 V for 30 msec to transmit the data. The desired measurement frequency is 0.2 Hz (one measurement every five seconds). Then
D = Tm + Tp + Tt × F
= (2 msec + 5 msec + 30 msec) × 0.2 Hz = 0.0074
P = (3.3 V × .500 µA) + (3.3 V × 0.003) + (3.3 V × 0.03)
= 110.6 mW
Pa= P × D + Ps
= 0.0074 × 0.1106 = 818 µW
Page 2 of 3
vPa is the key term that will reveal the types of energy harvesting transducers, if any, will suit this system. The accompanying table lists typical energy transducers and the typical average power they can deliver. K is a power conversion constant that takes into account the type of power management block needed to convert the transducer energy to a usable voltage, 3.3 V in this case. A perfect power converter has a K=1. K will vary with the type of transducer employed.
Generally speaking, K is proportional to the output voltage of the transducer. Transducers like TEGs put out a low output voltage and thus require an extremely high boost ratio and correspondingly high input currents. Thus K tends to be lower for these devices than for high-output-voltage transducers like Piezo elements. In our example, we can see that Pa approaches the upper range of that available from piezo transducers of a reasonable size, but is within the capabilities of TEGs and photovoltaic transducers or solar cells.
The system environment will usually dictate the type of transducer. In our example, PV transducers are impractical because there is no always-available light source. The application is at the upper end of feasibility for piezo transducers, so the best choice seems to be a TEG. TEGs utilize the Seebeck effect to generate a voltage across their output terminals when exposed to a temperature differential.
To continue our example, assume the use of a 50-mm2 TEG. One side of the TEG will mount to the HVAC duct in the ceiling while the other side will be exposed to room temperature air. Because TEGs have a low thermal resistance, it's often challenging to develop a suitable ΔT across them, so the room temperature side will employ a heat sink. Measurements have shown that an HVAC duct surface will average 38°C in the winter (heating) and 12°C in the summer (cooling) with an average room temperature ambient of 25°C.
Careful measurements have shown the ΔT across the TEG is ~±10°C when mounted to the duct with a heat sink. The manufacturer's data sheet reveals that the TEG Vout with a 10°C dT is 180 mV. The TEG output resistance (Rout) is 2.5 Ω. The power available to the load is maximum when the TEG Rout = power converter (or load) Rin.
If we assume our power management circuit has an Rin near 2.5Ω, then the maximum power available to the power converter input is (180 mV)2/(2.5 Ω × 4) = 3.24 mW. Our power converter constant K is 0.4, so the total power available to the remote sensor 3.3 V output is 3.24 mW × 0.4 = 1.3 mW. Because 1.3 mW is comfortably above the previously calculated Pa of 818 µW, we can generate enough power to operate.
The next challenge is to find a power management circuit able to convert the extremely low output voltage of the TEG to the required 3.3 V. A further complication is that the input voltage (TEG output) can be either positive or negative 180 mV, depending on whether the duct surface is hot or cold. While it may be possible to develop a discrete circuit to meet this challenge, it is often difficult to devise a solution that meets requirements for manufacturability, small size and reliability.
Further, circuit design is extremely sensitive to stray capacitance, and the entire circuit must operate at micropower levels to hit the rated K factor. Fortunately there are integrated solutions that handle the job. One is the LTC3109 which can operate from inputs as low as ± 30 mV and will produce any of four (2.35, 3.3, 4,1 or 5 V) pre-programmed output voltages, Vout. A switchable Vout is provided to power sensors only when necessary. The LTC3109, also includes a power manager that is useful for storing and using excess harvested energy. Because our typical load power is less than the available energy, any excess energy can be stored for later use on a large capacitor, Cstore.
The nearby figure shows the 3.3-V output of the LTC3109 before, during and after a measurement/transmit cycle. The capacitor on Vout is sized based on the acceptable voltage droop for one measure/transmit cycle. In our example, we've determined that a voltage droop of 300 mV is acceptable on the 3.3V output. Using the values obtained previously, we can calculate the required Cout:
Cout= (Iload-Iavg) × TL/VD
= (30 mA × 30 msec + 500 µA × 2 msec + 3 mA × 5 msec)-(1.3 mW/3.3 V) / 0.3 V
= 1.74 mF,
This value is in the range of a nominal 2,200 µF capacitor. Here Iload = sum of all the load currents on the 3.3 V output, A; Iavg = average output current of the LTC3109, A; TL = duration of the load pulse, sec; VD = acceptable voltage droop, V. The actual droop as illustrated in a nearby figure turns out to be much less than 300 mV. This is because the measurement took place on a simple system that had a higher output capacitance and a short current transmit pulse.
In the case of a short interruption of the energy coming in from the harvesting transducer, the LTC3109 operates from the storage capacitor, Cstore. There is no limitation on the value of Cstore, so it can be sized for whatever system holdup time is desired.
The basic design procedure here applies to other types of energy harvesting transducers. Power management circuits that interface with piezo elements (high voltage ac), electromagnetic (coil/magnet) and solar cells are all readily available today. In all cases, designers must first determine the average load power to see if autonomous operation is feasible.
Operating environment will always dictate what types of energy harvesting transducers are suitable, and average load power will further narrow the choices.