LoRaWAN Energy Study

LoRaWAN Energy Performance & Ambient Energy Harvesting



Powering the Internet of Things



Abstract

The Internet of Things (IoT) encompasses a plethora of use-cases, devices and connection technologies. One of the key IoT enabling technologies will be the low-power wireless- connected sensor. Technology already exists that will enable wireless connected telemetry to run for 10 years off a battery little bigger than an AA cell. This paper investigated the energy use of some LoRaWAN devices and energy available in the environment and found that energy harvesting for battery augmentation or battery elimination is a feasible option today.

Keywords: LoRaWAN, photo-voltaic, energy harvesting, radio-frequency, IoT


1 Introduction

In generational terms the first generation of the IoT consisted of Machine to Machine (M2M) technology including telemetry and high volume data that had high value and therefore people were willing to spend accordingly. For example, monitoring river heights for flood mitigation or flood warnings required river height monitoring stations that would cost tens of thousands of dollars to install. Relative to the overall project cost, paying thousands of dollars for the telemetry equipment and data connectivity was acceptable.

This new and coming generation of the IoT focuses on the bulk installation of low cost telemetry equipment that is inexpensive to manufacture, inexpensive to install, inexpensive to connect and inexpensive to maintain.

This white paper addresses one aspect of the last point: “inexpensive to maintain”. Some of the new generation of IoT devices will be mains powered but to get the full benefit of the coming IoT, many sensors will use low power wireless connectivity and be battery powered. This removes the installation costs associated with using cabled data connectivity or connecting mains power. Using low power wireless devices reduces some of the administration overheads and costs normally associated with mobile network connectivity. This leaves the maintenance costs associated with battery powered devices.

While the low power wireless technologies open up the opportunity to gather daily meter readings from smart meters where the telemetry will function for 10 years without a battery change, what happens if you require 10 minute sampling of the climate of office or living space over thousands of properties? In these cases the constant battery maintenance or the alternative of installing mains power to sensors would destroy the cost benefit of a project. Augmenting the sensor’s battery with power harvested from the environment has the potential to expand the viability of IoT into a range of additional use cases.

This paper investigates the energy costs per LoRaWAN transmission from several sensors and the energy available from the environment to augment the available battery power.

The following section details a number of experiments measuring energy use and energy availability. Each experiment or group of experiments has its own mini results, discussion and conclusion as appropriate. After the experiments a brief summary identifies the key results.


2. Experiments

The experiments focus primarily on light energy and the use of Photo-Voltaic (PV) Cells as the collectors of energy to be harvested. Originally ambient Radio Frequency (RF) energy was considered but after investigating the levels of ambient RF energy across a Central Business District (CBD) and the harvesting efficiency, it was found that this was not a viable alternative. (See linked study and associated map)

The intention of the experiments was not to develop a new technology but to explore the feasibility of running IoT devices either completely off harvested energy or using the harvested energy to augment batteries or reduce the size of the required batteries. Experiments were designed to run the boundary of energy harvested versus energy required using PV cells of a size comparable to the LoRaWAN sensor.

The PV cells used initially were hobbyist PV cells sold over the counter for £1 to £2. Later some state-of-the-art PV cells designed primarily for artificial light also became available. Energy harvesters were Texas Instruments evaluation modules modified for different voltages. Energy storage used Super Capacitors (SuperCaps) and Lithium Ferrous Phosphate (LiFePO4) batteries. The wireless sensors were built from MultiTech mDot LoRaWAN modules that incorporated the LoRa wireless, an ARM processor and an embedded LoRaWAN protocol stack. A variety of sensors including temperature, humidity and gas sensors were utilized. A number of high quality commercial, battery-powered LoRaWAN room monitors were obtained and these were also utilised in the experiments.

The experiments included:

  • 1. continuously running a MultiTech mDot LoRaWAN sensor in a battery-less operation,
  • 2. running a commercial sensor using:
    • (a) battery only and
    • (b) augmented battery,
  • 3. a miscellany of mix-and-match experiments using different sensors with different available energy sources.

Experiments were conducted in Glasgow, Scotland, with whatever implications that brings with it as far as available daylight is concerned. The experiments were not rigorous laboratory experiments with precisely measured light levels and micrometer accuracy but more of a ‘does this work here and now’ attitude. The PV cells were exposed to daylight from behind double glazed tinted windows that reduced PV cell short-circuit current by 75% or more. This seemed reasonable since PV cells exposed to the weather have to deal with coatings of dust and other contaminates.

Each of the following experiments appear in their own section with mini-discussions and conclusions as appropriate. The experiment list starts with a recap of why harvesting of ambient RF energy was found to not be viable for this application.

2.1 Radio Frequency Energy Harvesting.

This work was previously documented and available at this link. While there has been a substantial investment in research around the topic of RF energy harvesting, the reality is that for the general purpose, powering of devices, RF energy harvesting is useless. Sufficient ambient RF energy is only available in close proximity to large transmitters, and as new digital transmission technologies arrive in different localities, even these sources are disappearing.

To put this topic to bed, consider an RF energy harvester with a 100% efficiency converting a -30 dBm RF signal to stored energy. -30 dBm is 1 microWatt. Over one year this equates to 32 Joules of energy (1x10-6 W * 3600 sec per hour * 24 hours per day * 365 days per year). However, at -30 dBm the harvester efficiency would be closer to 5% or 1.6 J PER YEAR. Increase the signal to -20 dBm and 20% harvester efficiency and the energy available over a whole year rises to 63 Joules. Compare this to the 33,000 Joules available from an AA sized 3.6V Lithium battery. Also consider that your mobile phone thinks that -80 dBm (a million times smaller than -20dBm) is a good signal.

Review the map associated with the linked white paper to assess the availability of harvestable RF energy in a typical CBD environment and consider the possible available energy in comparison to the energy use and energy availability from the following experiments.

2.2 Super Capacitor Powered mDot Sensor

In this experiment a MultiTech mDot was powered by a 3V 100mA hobbyist Photo Voltaic (PV) Cell and a Super Capacitor (SuperCap). This experiment allowed the measurement of available energy, energy use and changes to energy through the year. The PV cell was connected to the input of a Texas Instruments (TI) energy harvester evaluation module (BQ25504EVM) modified to output 4.7 volts to the storage element. The TI energy harvester was a Maximum Power Point Tracking (MPPT) device that operated using the 'open voltage method' where it momentarily open-circuits the PV cell, reads the voltage and then sets the harvesting voltage point at 80% of the measured voltage. For most of the experiment the storage element was a 2.5F 5V SuperCap. Two 2.5F SuperCaps were used to test the ability of the power system to survive low light scenarios. The mDot operated in a sleep mode and on a ten minute interval it would wake up, read the voltages across the SuperCap and the PV cell and then transmit this data over the LoRaWAN using a Spreading Factor of SF7. For most of its life the test device sat behind an east-facing window with a near-hemispherical field of view.

Diagram of Basic Energy Harvester Layout

Figure 1. Basic Energy Harvester Layout. The energy harvester was configured to manage the power to the LoRaWAN sensor and the LoRaWAN sensor was configured to monitor the PV cell voltage and the storage element voltage. The Zener Diode was to protect the energy harvester from over-voltage on the input.

The four main components of Fig.1, the PV Cell, the energy harvester module, the storage element and the LoRaWAN module can be seen in Fig.2

The complete battery-less LoRaWAN sensor

Figure 2. Test Sensor comprising 3V 100mA PV Cell, 2.5F 5V SuperCap, TI Energy Harvester and MultiTech mDot, all mounted in a Jiffy Box.

2.2.1 Results

A number of static and dynamic parameters along with summary values were measured, calculated and recorded. These are presented below and discussed in the next section.

Table I: PV Cell Measured Values
Open Circuit Voltage (direct sun)VOC3.6V
Short Circuit Current (direct sun)ISC100mA
Open Circuit Voltage (behind double glazed window)VOC3.3V
Short Circuit Current (behind double glazed window)ISC24mA



Table II. MultiTech mDot and associated circuitry parameter values
Sleep CurrentISLEEP32μA
Voltage Divider Current (reading VCAP)IVD3μA
Total sleep energy per hour0.5J @ 4V
Active Energy (calculated from Fig.3)80mJ per Tx



Table III. Operational Values (2.5F SuperCap)
Fastest recorded rate of SuperCap energy increase (for 2.5F) (See Appendix for example calculations)12.5J in 10 minutes
Rate of capacitor voltage fall while no light. (rate of fall was linear with r² = 0.999)2.285 V per day
Rate of Energy use while no light. (used a linear estimate with r² = 0.998)(should really use a 2nd order polynomial but linear estimate was sufficient)23.13 J per day
Approx. Sleep Energy (24 hours) (using 3.95V, midway between 3.2 & 4.7V)11.9 J
Approx. Sample & Tx Energy (144 samples transmitted over 24 hours)11.5 J
Approx. Total 24h Energy Use (same ballpark as estimated from the regression line)23.4 J



Oscilloscope image of current draw of mDot during Tx

Figure 3. Current draw of MultiTech mDot during wake-up, sampling and transmission.

SuperCap Voltage over the seasons

Figure.4 SuperCap (2.5F) Voltage from July (Summer) to November (Late Autumn). As the nights grew longer the capacitor discharged further.

Daily fluctuations of SuperCap voltage

Figure 5. SuperCap (5 Farad ) and PV Cell Voltages, midwinter. Day 5 2017 started with sunshine but the following days were heavily overcast and insufficient energy could be extracted to fully recharge the SuperCap. The small 0.35V platform on the end of the PV Cell Voltage on days four, five and six was due to the energy harvester extracting whatever power was left to extract at this light level. Note: The PV cell voltages are most probably the voltage while under load or 80% of the open circuit voltage at the time. Refer Fig.17

Available light vs SuperCap Voltage

Figure 6. Close up of relationship between light, as measured by the PV Cell, and the SuperCap voltage. This figure represents 6th Nov. 2016, the marked sunrise, sunset and solar midday values were the appropriate values for the site of the experiment. The light before and after sunrise and sunset approximated the Civil-Twilight. Direct sunlight would have ceased prior to midday but the energy harvested from the daylight was sufficient to keep the SuperCap fully charged to just before sunset.

2.2.2 Discussion:

In designing these experiments the intention was to develop borderline viable systems so as to explore the system limitations. This particular experiment was successful in this regard. The energy stored in a single SuperCap was sufficient to easily keep the system alive through the night during summer and autumn, but not through successive heavily overcast winter days (Figs. 4 & 5). Even with a double SuperCap the system could not survive for more than a few of days of heavily overcast weather.

PV Cell Voltage: The graphs relating to PV cell voltage do not show open circuit voltage. This is the voltage at the time of sampling and may be either open circuit voltage or around 80% of open circuit voltage depending on whether the energy harvester is operating or not. In Fig.6 immediately after fully charging the SuperCap the voltage varied above and below 2V. It is likely that for some of these readings the harvester was not operating (above 2V) and for others it was operating (below 2V). Once the direct sunlight had gone the PV Cell voltage would be lower, because of less light, and lower because the reading occurred during harvesting.

Available Energy: For this solar panel and harvester combination, energy seemed to be available at a PV voltage of as low as 1.3V and above. Voltages above 2V resulted in rapid charging. The peak stored energy of 12.5 J measured over a 10 minute period (Table III) may understate the actual available energy since the measuring period was limited to 10 minutes at the end of which the SuperCap was fully charged. It was unknown if the SuperCap reached full charge at 5 minutes or 7 minutes or 10 minutes into the period. For some of the LoRaWAN devices discussed later, 12.5 J would be sufficient for two days of sampling and transmission.

Spreading Factor & Tx Power: This device was programmed to only transmit using maximum power (25mW or 14dBm) and SF7. There were periods where it was averaging three gateways per transmission including occasionally connecting via a gateway 5.5km distant. Given that the transmit power was sufficient to reach gateways at this distance, the transmit power could have been reduced to save a small amount of energy. Conversely, if this sensor was in a hard to reach locality it would have needed to use a higher SF and that would have increased the overall energy use.

Sleep Mode: Based on the measurements of Table II and III, approximately half the energy use was due to the sleep mode current augmented with the voltage divider current. A modified system could double its operational life if this sleep current could be reduced. The embedded microprocessor was capable of 5μA sleep current but the embedded software libraries did not put it into this mode. Potentially a Real-Time-Clock with sub micro-amp current could be used to control the wake up timing of the microprocessor. The voltage divider, which was necessary because SuperCap voltage exceeded the microprocessor analog-to-digital input limit, could be redesigned to only be in circuit when the reading needed to be taken.

Active Energy Use: Some of the energy used during sampling and transmission could be reduced as the microprocessor seemed to be awake for an inordinate amount of time both before and after the data transmission (Fig.3). It was thought that possibly the system was in debug mode and it was staying awake to output messages on the debug port. From monitoring the debug port there were no debug messages but the bootstrap loader was outputting an error message on each wake up. Resolving this issue could have further reduced the energy consumption of the system.

Supply Side Management: The energy harvester was set to output threshold indicators. If these levels were monitored by the microprocessor it would have been possible to shut down the system when the voltage dropped below a set threshold and then restart or wake the system when the voltage rose again. It would also be possible to optimise the sampling rate to match available energy.

2.2.3 Conclusions

As it is, if this system was operating in a subtropical or tropical zone, it could operate continuously through the year. In the test location the midwinter day length (sunrise to sunset) was just seven hours. On the border between the Earth’s temperate and subtropic zones the day length at midwinter increases to around nine and three-quarter hours. This is equivalent to day 298 on Fig.4. For its present locality, combining all the above improvement factors, this system could in effect be powered indefinitely assuming that a variable sampling interval with occasional silent periods was acceptable. If this system were modified for a sleep current of 5μA and sampling once per hour, it could theoretically survive for over four days with zero light.


2.3 Elsys ESM 5K Battery & Super Capacitor Powered

An off-the-shelf Elsys ESM 5K was used in three separate energy-use experiments:

  • 1. Using the supplied battery (½ AA 3.6V 1200mAh Lithium) the sensor was run to exhaustion.

  • 2. The sensor was powered by a 3.2V 900mAh LiFePO4 battery charged to 3.3V by a TI Energy Harvester connected to a 3V 50mA hobbyist PV Cell.

  • 3. The sensor was powered by a state-of-the-art PV Cell, 100mAh Battery and associated Energy Harvester. This PV Cell was designed for operation in ambient room lighting.


2.3.1 Data Collected for Run to Exhaustion.

For the run to exhaustion the data transmitted by the Elsys ESM 5k was received by the LoRaWAN Network Server and later downloaded for analysis. It was found that the sensor had transmitted using different SF so the transmissions at different SF were counted since the energy use changes from SF to SF. Where there was missing data the message SF was estimated based on surrounding messages. A digital storage oscilloscope was used to inspect the current draw during sensing and transmission.

Elsys EMS 5k Current Draw during Tx

Figure 7. Current drawn by Elsys ESM 5K transmitting using SF7. Estimated energy @3.6V was 18mJ per transmission. Estimated energy due to sleep current: 91mJ per hour or 2.2J per day.



Table IV. Battery (3.6V ½ AA 1200mAh) Run to Exhaustion § ‡
TransmissionmJ/TxEst. Count Total Energy (Joules)
SF718 1520 27.36
SF822 102422.53
SF931 1024 31.74
SF1051 1008 51.41
SF1189 970 86.33
SF12166 28508 4732.33
Total 34054 4952 J †
Sleep Energy 90 days ~ 196 J

§ According to the self reports the battery voltage started at 3.5V so the initial state was unknown.
† About 270J of the total energy was due to listening for acknowledgement messages (ACKs).
‡ Originally on a 1 minute interval but this self adjusted due to duty cycle requirements as the sensor changed from SF7 to SF12. The sensor was reprogrammed to a 5 minute interval after the first fourteen thousand transmissions.




Elsys EMS-5k Graph of Battery Voltage

Figure 8. Elsys ESM 5K Voltage Reading. Left side of graph, running the battery to exhaustion. Right side of graph, powered by PV Cell & 900mAh LiFePO4 battery.


2.3.2 Discussion (Run to Exhaustion)

In analysing this data the first thing that was noted was that the sensor, although configured not to ask for Class-A LoRaWAN acknowledgements, appeared to be listening for Rx-1 and Rx-2 ACKs (The current pulses at 1 second and 2 seconds after the Tx in Fig.7). In analysing the logs it also appeared that the sensor was self adjusting the transmission SF as if it was looking for responses to know that it was in contact with the network. The sensor would start transmitting at SF7 for around 95 transmissions then switch to SF8 for 64 transmissions, then to SF9 etc., finally ending up at SF12 where it stayed. It was assumed that this was a programming issue in earlier release sensors.

The sensor was started using SF7 on one-minute intervals but due to the auto stepping this ended up as SF12 with a duty-cycle imposed interval of two minutes twenty-eight seconds.

After approximately 14000 samples the sensor was reprogrammed to sample on five-minute intervals. This is visible in the Fig.8 graph at around day 272 of 2016 where the line density changes. As the battery voltage started dropping the sensor started restarting by itself. It would cycle through SF7 to SF12 and after arriving at SF12 it might only manage a few SF12 transmission before restarting. Eventually it was restarting during SF11 transmissions or earlier in the sequence.

Overall the sensor transmitted approximately 34,000 readings. This was an estimate based on packets received and either on the the length of gaps in the record where the sensor was taken away out of range or the difference in up-counts if there hadn’t been any sensor restarts.

From the results in Table IV, the transmissions accounted for nearly 5000 Joules of energy drawn from the battery. If all these transmissions happened at SF7 they would have accounted for 613 J. Alternately, 4952 J could have powered 275,000 SF7 transmissions which, if transmitted on a 5 minute interval, would take 955 days. In that case the device would have also used around 2000 J in sleep mode energy, which it didn’t have in this situation.

The sensor self-reported the starting voltage as 3.5V so it may be that the battery was not at full capacity. A new 3.6V 1200mAh battery should store approximately 15,500 J and this device stopped after consuming just over 5000 J.

Check the http://elsys.se website for specifications for the current sensor family.

2.3.3 Hobbyist PV cell Powered Activities

This experiment used a lower rated PV Cell compared to SuperCap experiment covered in section 2.2. Although the PV Cell of Table I was a 100mA 3V Cell, the 50mA Cell results recorded in Table V were only 8% to 13% lower.

Table V: 50mA Hobbyist PV Cell Measured Values
Open Circuit Voltage (direct sun)VOC3.6V
Short Circuit Current (direct sun)ISC92mA
Open Circuit Voltage (behind double glazed window)VOC3.2V
Short Circuit Current (behind double glazed window)ISC21mA



Discussion: Hobbyist PV Cell & 3.2V 900mAh LiFePO4 Battery.

The original intention for this experiment was to use a 250mAh or 400mAh LiFePO4 so as to be better able to view the charge and discharge cycle. Unfortunately those cells were not available at the time so the 900mAh batteries were used instead.

In the first of these experiments the hobbyist PV cell was laying flat on the sill of an East facing window. The sensor voltage reading appears on the right side of Fig.8, starting around day 337 of 2016 and heading past mid February of 2017. During December the sun was never more that 12.3 degrees above the horizon therefore although there was direct sunlight (through the window) on the PV Cell it was at a very large angle of incidence (low glancing angle). Day length through the second half of December hovered around 7 hours. By the end of January the length of day had increased to more than eight and a half hours and the sun was reaching nearly 17 degrees at solar noon. A detailed inspection of the battery voltage compared to the time of day indicated that there were little charging spikes during the sun-lit periods.

Although this experiment only ran for 80 days, experience with the SuperCap powered sensor suggested that there was no point continuing this experiment. The morning dose of direct sunlight was sufficient to keep this sensor running indefinitely.

2.3.4 Specialist PV Cell with Energy Harvester & 100mAh battery.

Because the supplied specialised PV Cell was integrated with the energy harvester the figures in Table VI were taken from the associated data sheet.

Table VI Specialist PV Cell Data Sheet Values (@200 Lux of Fluorescent Light)
Open Circuit Voltage VOC4.61V
Short Circuit Current ISC43.8μA
Energy Harvester charging current available.IOUT39μA



Elsys EMS 5k Battery Voltage - Specialist PV Cell

Figure 9. Elsys ESM 5K Voltage Reading. Powered by special PV cell and 100mAh battery. For the first 5 days the PV faced due north. On day 61 it was turned to face North-East and captured some slant sun in the early morning. On day 73 it was turned to face east and returned to North-East on day 86 (27 March 2016)

This experiment used a device supplied where the PV cell and incorporated energy harvester and battery was designed for maximised response under artificial lighting. The battery voltage was 4V but this was dropped down to around 3.4V for the Elsys using a forward biased silicon diode. The PV Cell’s dimensions were smaller than the top of the Elsys sensor (Fig.11).

This experiment started on 24/25 February 2017, the self-monitored supply voltage was graphed in Fig.9. Initially the solar panel was placed on a window sill and faced directly north (no direct sunlight). The sensor was transmitting temperature, humidity, light and battery readings every five minutes using SF12. The voltage was dropping so on the 2nd March (day 61 on the graph) the PV cell was turned to face North-East (45 degrees towards the East). This would have equated to some sunlight starting at an angle of incidence of 56 degrees and increasing to 90 degrees after two and a half hours. The battery voltage was still declining so on day 73 the PV Cell was set to face due east. This arrested the battery discharging and led to accelerated charging over the following days. The PV Cell was returned to a north-east orientation on day 86 (27 March). At this point the sun was rising at 84 degrees giving an incidence angle of 39 degrees and over 3 hours of slanting sun on the panel. Note that this was inside double glazed windows and as the sun angle approached orthogonal to the panel it was increasing its angle of incidence to the window therefore the actual angles of incidence given are only approximate.

Daily charge and discharge

Figure 10. Closeup View of Daily Charge and Discharge Cycle: Elsys EMS 5K powered by specialised PV Cell & 100mAh battery. March 13 2017, PV Cell facing north-east.

From a look at the daily cycle of Fig.10, the sun rose at 94 degrees from north giving an angle of incidence to the panel of 49 degrees. Sunrise at this point would have been delayed as the taller buildings to the east blocked the initial view. Some sun would have remained on the panel until just before 9AM but the battery voltage continued to increase until 9:20AM. Possibly the bright sun on the white window sill immediately in front of the panel was supplying sufficient reflected light. Through the day the ambient light was sufficient to maintain the battery voltage until an hour before sunset when the battery started discharging.

Comparing the size of the PV cell and the sensor

Figure 11. Size comparison of the Elsys EMS 5K and the Specialist PV Cell.

In theory this PV Cell and associated energy harvester and battery should be able to sustain this Elsys EMS 5K LoRaWAN sensor indefinitely under lower light conditions. During the above test the sensor was transmitting at SF12 and would have required 50J of energy per day. Changing the SF to SF7 would reduce the energy requirement to 7.25J per day. If the ambient daylight was sufficient to hold the battery voltage constant at a load of 50J/day then it should have sufficient surplus to charge the battery against a load of 7.5J/day. Increasing the sample interval from 5 minutes to 10 minute would reduce the daily energy requirement to less than 5J/day. According to the specification sheet, this is within the supply capacity of this PV Cell and associated Energy Harvester operating under artificial lights for 10 hours per day (@200Lux: 39μA x 4V x 3600s/h x 10h = 5.62J)

At the time of writing (July - mid summer) this sensor / harvester combination is running with SF7 transmissions at five minute intervals with the PV cell facing due north. The overnight rate of battery voltage fall has changed from 0.107V per day for SF12 transmission to just 0.027V per day for SF7 transmissions. Energy harvesting has continued through the day even on overcast and heavily overcast days.

2.3.5 Conclusions

This sensor could operate continuously using either of the PV cell options tested here. From the SuperCap experiments we know that in a single 10 minute period of direct sunshine as much as 12.5 J were available for storage. Using the 50mA Hobbyist PV Cell and 900mAh battery there didn’t appear to be any noticeable discharge through the darkest period of winter and small charging spikes were noticeable on most days. This supply could maintain this sensor indefinitely under the substantial sensing and transmission load of 5 minute sampling with SF12 transmission.

Based on the results above the much smaller specialist PV cell and associated energy harvester and 100mAh battery should also be able to sustain this sensor indefinitely but at a more limited sensing and transmitting regime.

2.4 Miscellaneous PV Cell Energy Harvesting

A number of experiments were initially set up in parallel and then later the combinations of energy-system and LoRaWAN were re-arranged and new systems introduced. Some of these experiments were useful in identifying the limits of the combinations.

The combinations recorded here are:

  • 1. 50mA 3V Hobbyist PV Cell with 900mA LiFePO4 battery powering a MultiTech mDot with temperature sensor. No direct sunlight.
  • 2. Specialist PV Cell and 4V 100mAh battery powering MultiTech mDot. No direct sunlight and direct sunlight.
  • 3. 50mA 3V Hobbyist PV Cell with 900mA LiFePO4 battery powering an Elsys ERS. With direct sunlight.

2.4.1 MultiTech mDot Powered by LiFePO4, 50mA 3V Hobbyist PV Cell and TI Harvester

This experiment used a MultiTech mDot programmed to turn on and read a DHT11 temperature and humidity sensor every ten minutes. The mDot then transmitted those readings along with the battery voltage and PV Cell voltage via the LoRaWAN using SF7 transmissions. It did not request acknowledgements. Despite the additional hardware the energy use per transmission approximated that of the mDot in the SuperCap experiment. The sleep current was also similar.

The TI harvester was modified to partially charge the battery to 3.35V (LiFePO4 battery chargers can go up to 3.6V). Initially the sensor was placed on a window sill facing east until the battery was charging at 3.35V. The sensor was then moved to the opposite side of the room and placed on the top of a cupboard so that it faced the eastern floor to ceiling windows. Due to the distance from the windows and the position at the back edge of the cupboard the PV Cell view of the windows only accounted for four degrees of view in the vertical plane. There was a nearby light fitting but it didn’t directly light this location.

battery voltage for mDot powered by PV Cell and LiFePO4 battery

Figure 12. Hobbyist PV Cell and LiFePO4 powered mDot. On the left of the figure, the sensor charged while sitting on a window sill. The following 90 days the sensor sat on top of a cupboard.

Clearly, at less than 1.5V the PV cell voltage was insufficient to charge the battery or sustain the load and the battery slowly discharged. After the sensor ceased communicating it was placed in the east facing window to recharge and the sensor restarted. From the close up view of the PV Cell voltage and the battery voltage in Fig.13 (below) it could be seen that PV Cell voltages above 1.5V were putting some charge into the battery, probably sufficient to sustain the base battery voltage through the next 24 hours. PV Cell voltages above 2.0 Volts were sufficient to substantially raise the base battery voltage and amount of charge (see day 306).

This sensor continued in this mode through to around the 24th February when battery voltage gains began to accumulate. After another 30 days the battery voltage finally reached the charging voltage.

Figure 13. PV Cell in east facing window attempting to recharge the LiFePO4 battery while sustaining the sensor.

Conclusions

The indirect light of the test location was insufficient to power this device. At first appearance the size of the windows made them appear as a good source of light but ultimately it wasn’t enough to maintain the battery in equilibrium. The approximately 90 day discharge was longer than running the sensor directly off a similar battery charged at 3.35V so some energy harvesting was occurring. Reducing the energy used as discussed previously would prolong the battery discharge and probably allow some charging to occur at lower light levels. For example, in the above graph, day 301 has a PV voltage above 1.5V but below 2V. It can be seen to be increasing the charge on the battery and then the voltage falls again during the night. A smaller energy use would increase the energy available to charge the battery.

2.4.2 Specialised PV Cell and 100mAh Battery powering mDot

In this experiment the specialised PV Cell and associated Energy Harvester was used to power a MultiTech mDot sending messages on a 10 minute cycle. The PV cell was located at a similar location to the sensor in the above experiment (2.4.1) but on the front edge of the cupboard giving the PV cell a 12 degree vertical view of the floor-to-ceiling eastern windows.

Battery voltage specialised PV Cell powering mDot

Figure 14. Specialised PV Cell and 100mAh Battery powering MultiTech mDot. The left hand side of the graph was the sensor sitting on top of a cupboard with no direct sunlight. The right hand side of the graph the sensor was sitting in an east facing window with direct sunlight in the morning. The period covers all of August and nearly 12 days of September.

The specialised PV Cell and associated circuitry was unable to support the energy requirements of the mDot at the initial location. The energy usage of the mDot would have been similar to the previous mDot experiments. The left hand side of Fig.14 represents approximately 18 days of use from a partially charged 100mAh battery. The sensor was moved to a window position and restarted. Each of the humps on the right side of the Fig.14 graph represent one day. It took a fortnight of late August morning sunshine to get the battery back to a fully charged condition.

Assuming the battery was nearly discharged the harvester restored 1440J in 14 days at an average of 102 J per day. The mDot also required 23 J per day so the total generated was 125 J per day. Assuming the battery only accepted 50% of full charge the harvester restored 720 J at 51 J per day plus the 23 J per day required for the mDot for a total energy harvest of 74 Joules per day.

As indicated earlier, this suggests that there are light conditions where this supply could source adequate power to power LoRaWAN devices.

2.4.3 ELSYS ERS powered by 50mA 3V PV Cell, TI Harvester and 3.2V 900mAh LiFePO4 Battery.

As in some previous experiments the TI Harvester was configured to charge at approximately 3.35V. The Elsys ERS was programmed to read and transmit the temperature, humidity, light, CO2 and battery voltage. The sensor transmitted packets to the LoRaWAN every five minutes. 99% of data was transmitted at SF10.

Like the Elsys EMS 5k this sensor was programmed not to request ACKs but regardless it appeared to be listening for ACKs. Using the data supplied on the Elsys website and calculating at a voltage of 3.35V, to transmit the above data would cost 38 mJ per transmission at SF7 and 83 mJ per transmission at SF10. These equate to 11 J per day for SF7 and 24 J per day at SF10. Refer to Appendix Table A-I for sample calculations.

The measured sleep current was 35μA or 10J per day. The specification on the Elsys website indicated that the sleep current for the sensor should be 4μA. The sensor used in this experiment was an early device and the design may have changed.

Elsys ERS current draw during Tx

Figure 15. Elsys ERS current during sample and transmission. This sample included Light, Temperature, Humidity and Battery Voltage, transmission occurred at SF7. The transmission current was too large for this screen and had to be read at a different scale. Like the Elsys EMS 5k, the Elsys ERS appeared to listen for Rx-1 and Rx-2 acknowledgements at one and two seconds after the transmission (not shown in this graph).

Battery Voltage for Elsys ERS powered by Hobbyist PV Cell and 3.2V LiFePO4

Figure 16. Battery voltage of 3.2V LiFePO4 powering Elsys ERS. Gaps in the record at the right before day 140 were due to removing the device for measuring.

Although the test period did not cover the depths of winter, this combination of PV Cell, harvester and battery powered the Elsys EMS 5k for 3 months immediately prior to this test. The Elsys EMS 5k was using more energy per transmission (166 mJ vs 83 mJ ) with less sleep energy (2.2 J per day vs 10 J per day) therefore the Elsys ERS, despite the use of the CO2 sensor, presented a lighter load and this combination of power system would appear to have the capacity to easily maintain this sensor.


3. Discussion

One of the limitations of the documented experiments was that they all appeared to require some daily exposure to ‘direct’ (via the window) sunlight. This limitation could be offset by observing that in the SuperCap experiment the energy harvester was able to equal or exceed the energy requirements on daylight alone, without direct sunlight. Detailed inspections of daily PV cell voltages identified many examples of the energy harvester switching on and off within 30 minutes of sunset (Fig.17). This indicated that even in those light conditions the energy harvester did not need to run continuously to match the energy requirements of the mDot. Days 6 and 8 of Fig.5 reinforce this, these were heavily overcast days but the energy harvester was still able to exceed demand for some periods on these days.

Difference between PV Cell Open Circuit Voltage and Voltage during energy harvesting

Figure 17. Energy Harvester switching on and off is visible in the switching between two voltage curves. Midday occurred at 186.5 on the x-axis.

Undoubtedly, just as exposing the PV cells to sunlight is a ‘brute force’ approach, the addition of extra PV Cells or SuperCaps could enhance the energy availability but at an additional cost. The use of the LiFePO4 cells was an attempt to expand the energy storage past the daily cycle limit of the SuperCap and still give visibility to the energy cycle. By only partially charging the LiFePO4 cell the energy harvesting and storage cycle could be expanded to tens of days rather than single days. In the example of Fig.12 this extended to approximately 90 days but the available light was insufficient to significantly offset the energy use. For a properly designed augmented battery solution, the design should incorporate a specific LiFePO4 charging circuit that could cycle the battery intelligently.

As an alternative to using a rechargeable battery the energy harvesting could augment the energy supplied by a non-rechargeable cell. If both an energy harvester and battery were powering the circuit via appropriate diodes, then neither source would lose energy to the other. By setting the energy harvester voltage a few hundred millivolts above the battery voltage, the battery would remain out of circuit whenever the harvester had energy available. Using the example of the Elsys ESM 5K and the specialist PV cell; under artificial light @ 200 Lux the energy harvester can put out 39μA (Table VI). If 7μA was consumed in sleep mode (Fig.7) then in five minutes the energy harvester could store away 38mJ ( (39μA-7μA) x 4V x 300 s) . This is sufficient to power a transmission at SF7 (18mJ), SF8 (22mJ) or SF9 (31mJ) (Table IV). A small SuperCap (100mF) would be sufficient in this application.

Apart from the obvious hardware changes to enhance the energy efficiency of the LoRaWAN sensors, understanding the energy availability offers opportunities to explore energy optimisation through intelligent algorithms. The Elsys environmental monitors can be programmed for different sample rates for each sensor type and the sample rates can be different to the data transmission interval. If this kind of programmability was managed in the unit itself then the unit could be designed to monitor available ambient energy and available stored energy, and then, intelligently adjust the sampling rates and transmission rates. This functionality would expand the viability of energy harvesting into more locations. Without intelligent energy management a LoRaWAN sensor located in an energy viable zone could disappear from the network due to a small change in local conditions. With intelligent energy management the sensor adjusts its operation to match the available energy resources.

The next step would be extending the sensor intelligence to include programming the sensor by task rather than sample rate or transmission rate. A sensor tasked with monitoring office space identifies that the lights are off and therefore sampling rates can be drastically reduced. Some environmental variables may have higher inertia than others. On weekends in an empty office, the rate of change of CO2 concentration would reduce therefore the CO2 sensing rate could be reduced accordingly. A number of machine learning techniques could be applied to this scenario.


4 Future Work or Potential Future Work

Some of these experiments could be improved on for the future, for instance it would have been good to have mounted identical PV Cells on some of the sensors to capture the PV Cell open circuit voltage. This would have allowed a clearer identification of light levels and at the same time identified the periods of operation of the energy harvester.

A number of improvements were commented on in the text including:

  • Modifying the mDot embedded software to monitor voltage levels and adjust the sample period,

  • Modifying the mDot embedded software to put it to sleep when voltages drop below a set threshold and then wake on an interrupt from the energy harvester when the voltage levels have risen to a working level again.

  • Using a low power real time clock or low power programmable clock circuit to wake the mDot to allow the operation in the 5μA sleep mode.

  • Using the commercial sensors at lower spreading factors and longer sample periods to allow the testing of the energy harvesters in situations of lower light levels, particularly situations without any direct sunlight.

Some of these quicker-to-implement experiments have been set up and are currently running. Some of the others will be set up in the near future. These potential experiments will allow the development of knowledge of more marginal energy zones. Together they will form the groundwork for intelligent energy management in the sensor.


5 Summary

From the experiments conducted here there several points to take away:

  • Continuously operating, battery powered, high sampling rate, environmental sensors using LoRaWAN wireless connectivity are viable using energy harvesting for battery augmentation.

  • Provided the combination of sampling rate, transmission parameters and access to light are balanced, a PV Cell with physical dimensions compatible with the sensor can power the sensor indefinitely.

  • Depending on system requirements battery-less LoRaWAN connected sensors are completely possible. (An indoor battery-less LoRaWAN sensor is possible, even in Glasgow!)

There are definitely limitations but the existence of new PV cells designed for artificial light and the reducing current requirements of electronics indicates that even in the marginally lit areas, LoRaWAN sensors powered by energy harvesting are feasible. The possibility currently exists for a LoRaWAN environmental sensor with incorporated PV Cell to be unobtrusively “installed” by sticking it to an office light fitting. While the lights are on it measures and transmits, while the lights are off it increases the sample interval or turns off completely while waiting for the stored energy to rise sufficiently before starting up again.


Appendix (Example Energy Calculations)



Energy stored in a Capacitor: w=CV²/2 (Joules)
Where w is the energy in Joules
C is the capacitance in Farads
V is the voltage in Volts

equation to calculate energy change in capacitor

Example: A 2.5F Super Capacitor has discharged from 4.74V to 3.5V

worked equation to calculate energy change in capacitor


Energy used by a device:

Voltage is the measure of energy per unit charge (Joules per Coulomb).

Current is the rate of charge flow (Coulombs per second)

Power (Watts) is the rate of energy use (Joules per second)

P=VI = Joules per Coulomb x Coulombs per second => Joules per second.

Energy w=Pt = Power x time. Joules per second x seconds = Joules.

Therefore: In a circuit with a supply of 3.6V that draws 30mA for 20ms and 60mA for 70ms

w = 3.6V*30mA*20ms + 3.6V*60mA*70ms
w = 3.6V* (30mA*20ms + 60mA*70ms) = 17.3mJ




Worked Example for Energy use by Elsys ERS (data from Elsys.se website)

Table A-I Energy per SF10 Transmission (Elsys-ERS )
EventCurrent (mA)Time (ms)Energy @ 3.35V
Transmit @ SF1038 420 0.053466 J
Temperature Reading15 200.001005 J
Humidity Reading15 20 0.001005 J
Light Reading15 20 0.001005 J
CO2 Reading130 60 0.026130 J
Total~ 83mJ

Sleep Energy Example: Sleep current @ 3.35V = 35μA.    Calculate Energy per Day

w=3.35V * 35μA *3600 sec/hour * 24 hours = 10.13 J


 

Dr Andrew Wixted,
Glasgow Caledonian University

This work funded by Stream Technologies and Innovate UK as a Knowledge Transfer Partnership with Glasgow Caledonian University


Back