Pible: Battery-Free Mote for Perpetual Indoor BLE Applications

11/27/2018 ∙ by Francesco Fraternali, et al. ∙ University of Bologna Carnegie Mellon University University of California, San Diego 0

Smart building applications require a large-scale deployment of sensors distributed across the environment. Recent innovations in smart environments are driven by wireless networked sensors as they are easy to deploy. However, replacing these batteries at scale is a non-trivial, labor-intensive task. Energy harvesting has emerged as a potential solution to avoid battery replacement but requires compromises such as application specific design, simplified communication protocol or reduced quality of service. We explore the design space of battery-free sensor nodes using commercial off the shelf components, and present Pible: a Perpetual Indoor BLE sensor node that leverages ambient light and can support numerous smart building applications. We analyze node-lifetime, quality of service and light availability trade-offs and present a predictive algorithm that adapts to changing lighting conditions to maximize node lifetime and application quality of service. Using a 20 node, 15-day deployment in a real building under varying lighting conditions, we show feasible applications that can be implemented using Pible and the boundary conditions under which they can fail.

READ FULL TEXT VIEW PDF
POST COMMENT

Comments

There are no comments yet.

Authors

This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

1. Introduction

Buildings are integrated with thousands of sensors: temperature sensors provide feedback to HVAC systems, smoke sensors provide fire safety, etc. These sensing systems are designed for wired communication and power during the design of the building itself. Even a minor change requires a domain expertise in the building’s wiring infrastructure and can be prohibitively expensive, e.g., $2500 (for Demand Response and Efficiency, 2015). Wireless sensors have emerged as the answer to this problem. With low power and communication protocols (i.e ZigBee, 6LowPAN), wireless sensors can be deployed with a multi-year battery lifetime and be used for applications such as occupancy based control (Agarwal et al., 2010). But these nodes are powered by batteries that require periodic manual replacement. As we scale to large deployments, the manual replacement of batteries becomes a bottleneck. Battery replacements can be mitigated using energy harvesting, e.g., DoubleDip measures water flow by powering itself using temperature difference (et al., 2012b). However, there are limited commercial devices that use indoor energy harvesting solutions. We highlight 3 limitations that inhibit adoption: : (i) they are designed for specific applications, (ii) they do not support standard protocols, (iii) the application quality of service (QoS) is inadequate.

In this paper, we analyze the feasibility of overcoming these limitations using commercial off the shelf components. We explore the design space of a generic energy harvesting sensor node for indoor monitoring applications with the objective of perpetual operations (Campbell et al., 2016). We designed and built Pible, a Perpetual Indoor BLE sensor node. We show trade-offs between QoS, lifetime and harvested energy that enables our prototype sensor node to work in different indoor lighting conditions. We introduce hardware solutions to increase charging efficiency and overcome cold-start operations that limit usability. Finally, we propose a local sensor-node power management solution that maximizes the application-QoS and node-lifetime. We evaluate Pible by deploying twenty nodes in five different lighting conditions, for a general set of building sensing applications such as periodic sensor measurements, e.g. temperature, and event-driven sensors, e.g. PIR. We conducted a 15-day experiment in which we demonstrate continuous operation for all the different applications and on every light condition tested. Results show that Pible is able to broadcast sensor data at an average period of 94 seconds with an average luminance of 235 lux per day.

2. Background and Related Work

Existing works in energy harvesting can be categorized into:

(i) Application Specific: Application specific indoor systems have been proposed that use energy harvesting (et al., 2012b). The company EnOcean (EnOcean, 2018) produces battery-less systems but their design is application specific. We use a generic design suitable for most applications.

(ii) Energy Harvesting + Rechargeable Battery: energy harvesting can be used to extend batteries lifetime (et al., 2012b). However, the life of rechargeable batteries is limited to a few hundred cycles (batteryuniversity.com, 2017). Hence, recent works and our Pible node exploit super-capacitors as they can support up to a million charging cycles.

(iv) Communication protocol: Prior energy harvesting works do not support standard protocols since they require more energy(Campbell, 2014). To facilitate integration with existing technologies, we set this as a requirement. Works in (et al., 2012a; EnOcean, 2018; et al., 2017) adopt standard protocols but either lack in perpetual operations or are application specific.

(v) Power Management: Work in (etal, 07) adopt a power management to achieve energy-neutral operations. We use a simple application-specific look-up table, and we show that Pible achieves perpetual operation under varying lighting conditions.

(vi) Intermittent: Campbell et al. (Campbell et al., 2016; Campbell, 2014) designed an indoor sensor-node that stores only the amount of energy needed to read and transmit a single data packet: the system is continuously working when light is available but it stops during dark periods. Results show that door-open detection achieves only 66% accuracy.

Table 1, shows the gaps between related work and battery motes and compare them with Pible. We categorize them into 3 buckets: (i) battery powered systems, (ii) battery and energy harvesting systems and (iii) systems using only harvested energy. Battery powered systems achieve high QoS for the applications at the expense of a periodic battery replacement requirement. Adding an energy harvesting solution to a battery powered device increases the lifetime but the batteries still need to be replaced. Prior works in energy harvesting nodes promise no battery replacement but compromise the QoS, do not support standard protocols or are application specific. Flicker (et al., 2017) does provide these functionalities, but has not been evaluated indoors where we have limited light availability.

Platform Bat/ Replace Quality Standard Application
Har Battery of Commun
Power Service Protocol
Synergy(Agarwal et al., 2010) Bat No GT Zigbee Occupancy
Trinity (et al., 2013) Bat-Har No GT Zigbee Air-Flow
DoubleDip(et al., 2012b) Bat-Har No 98-65% No Water-Flow
Enocean (EnOcean, 2018) Har Yes NA BLE Specific
Buzz (Campbell, 2014) Har Yes 66% No Door, Light
Flicker(et al., 2017) Har Yes NA BLE Periodic Sense
Pible Har Yes GT BLE Sense/Event
Occupancy
GT = Ground Truth; Bat = Battery; Har = Harvesting
Table 1. Pible comparison with State of the Art Solutions

3. Design-Space and Architecture

We select a common set of applications in smart buildings and assess their power budgets. We select commercial off the shelf components to support their QoS requirements while ensuring perpetual operation on typical indoor environment without a battery.

3.1. Indoor Monitoring Applications

We categorize indoor applications as periodic or event-driven:

Sensing Environmental Conditions: (i) 1 Sensor: We test if Pible can operate perpetually for applications with minimal energy budget such as sensing of 1 sensor. Results can be extended to sensors with similar power budget. (ii) Multiple Sensors: We extend the power budget of motes that monitor different sensors at once.

Occupancy Detection: (i) PIR Motion Sensor, Door Sensor (Agarwal et al., 2010). (ii) Bluetooth Low Energy (BLE) Beacons (et al., 2015).

Table 2 reports the power consumption of Pible’s main components and operations using a super-capacitor storage element charged at 3V. For the sensing operations (i.e. Read Temperature), the power includes the reading and BLE transmission.

Operation Power[W] Operation Power[W]
MCU-Sleep 19 PIR Detection 32
Read-Hum 51 Advertising-5s 69
Read-Temp 54 Advertising-2s 86
Read-Bar 54 Advertising-1s 106
Read-Light 47 Advertising-500ms 171
MCU+PIR Sleep 22 Advertising-100ms 648
Table 2. Power consumption of Pible key operations executed with a super-capacitor charged at 3V. The values reported are averaged on a one minute execution.

3.2. Pible-Architecture

We use a general energy harvesting architecture (Tan and Panda, 2010; et al., 2018): an energy harvester transfers power to a storage element through an energy management board (EMB). Once the energy stored reaches a usable level, the EMB powers the micro-controller that starts its operations.

3.2.1. Platform System on Chip, Antenna and Sensors

We select the TI CC2650 chip that supports multiple communication protocols (e.g. 6LoWPAN, BLE) and consumes 1 A in standby mode. We equip our board with temperature, light, humidity, pressure, reed switch, accelerometer, gyroscope, and a PIR motion sensor.

3.2.2. Energy Storage

batteries have a short cycle-life of 1000 recharges. To increase lifetime, we adopt super-capacitors as they support up to 1 million recharges (batteryuniversity.com, 2017). However, super-capacitors’ choice is not trivial since: (i) Energy stored drops linearly on discharge (batteryuniversity.com, 2017). (ii) super-capacitor size is proportional to charging time using a constant charging current. (iii) Large super-capacitors increase node dimensions leading to packaging and aesthetic issues. (iv) The leakage current increases with size. We tested 3 super-capacitors with capacitance of 0.22F, 0.44F and 1F at 3.6V.

3.2.3. Energy Harvester

We use solar light since it has high power density w.r.t other energy sources (Tan and Panda, 2010) and it is available in buildings. To match our worst case scenario (Table 2), we use the indoor solar panel AM-1454 that harvests 71 W with 300lux.

3.2.4. Energy Management Board (EMB)

For our design, we select the BQ25570 from TI since it includes two programmable DC/DC converter: (i) an ultra-low-power boost converter (V) and (ii) a nano-power buck converter (V) that can support up to 110mA output current. The BQ25570 has a programmable power good output signal (V) that indicates when the super-capacitor reaches a user-set voltage level. We set this signal to 2.1V. The V converter is highly efficient when the storage element voltage level is above 1.8V but it is inefficient under this threshold (i.e. ‘cold-start’). The cold-start can be frequent if energy availability is low and can occur when adding or moving sensor nodes. In Figure 1-left, we show the charging time of the 3 super-capacitors sizes using V under 750lux. It takes 1 day for the 220mF SC to charge from 0 to 3.6V and almost the entire charging-time is spent to exit cold-start operations (0 to 1.6V). The second DC-DC converter (V) charges the super-capacitor at a much faster rate and the 1F element exits cold-start operations after 2.2 hours at 750 lux (Figure 1-right). However, V is more energy consuming after cold-start operations are over. Hence, we switch between the two charging modes using the circuit in Figure 2-right.

Figure 1. Charging different Super-Capacitors Size: Left- 750 lux with VBAT, Right: 750 lux with VBuck
Figure 2. Our circuitry to switch between V and V.

When the super-capacitor voltage level is low, the EMB raises V and V. Since the V signal is low, the inverter (P-2 and N-3) blocks P-1 while N-2 is conducing and charges the super-capacitor through V. Using a gate-source voltage threshold of 1.7V, N-2 charges the super-capacitor till 2.1V by setting V to 3.8V. At this point, the V signal turns on and connects V to the storage element through P-1. Finally, N-1 powers the MCU once the super-capacitor reaches 2.1V (Figure 2-left).

3.2.5. Wireless Communication Protocol

We use Bluetooth Low Energy (BLE) as it offers advantages for indoor environments (et al., 2015): (i) BLE is more energy efficient w.r.t other technologies (e.g. ZigBee) in transmitting small data packets called advertisements. (ii) advertisements can be used to perform indoor occupancy detection (et al., 2015).

Figure 3. Pible: a Perpetual Indoor BLE Mote

3.3. Wireless Sensor Network Architecture

Pible-nodes send a data packet to the closest Base Station (BS) that stores and sends the data to the cloud for post-processing. We used a Raspberry PI equipped with a BLE USB dongle as a Base Station. To monitor the nodes’ status, each Pible node sends to the BS information such as sensor data, QoS state, and voltage level.

4. Power Management Algorithm

A mote with a 1F super-capacitor that advertises every 100ms lasts 1.9 hours without energy harvesting. Our algorithm dynamically changes the rate of advertising depending on energy availability to increase lifetime. It uses a simple sensor specific lookup table, and a lighting availability prediction to set the sensing rate. All the algorithmic decisions are made inside Pible and are the following:

Setting the Sensing Quality: We use the super-capacitor voltage level to make a coarse adaptation of the QoS to use. We divide the usable voltage level (from 3.6V to 2.1V) into 7 states to maintain MCU memory requirements low. Table 3 shows the relation between the voltage level and QoS selected for the sensors in Pible. The different sensing periods are manually assigned to different QoS levels based on the power measurements in Table 2 and empirical observations and requirements given in indoor sensing literature (Campbell, 2014). Event-driven sensors need to be continually operating to send a packet as an event occurs. To save power, we turn off the sensors for a fixed period as soon as an event occurs. The longer the switched off period, the more likely an event will be missed, but it will also save more power. For advertising applications, BLE indoor localization systems exploit advertisement rates between 0.1s to 0.9s (aislelabs.com, 2015). We assign QoS levels to best meet this requirement.

Light Intensity Prediction: Prediction of light intensity is used to refine the next QoS. The system stores and compares the last 5 light intensity levels: if the light read is close to 0 (i.e. light off) or decreasing, the algorithm decreases the QoS state while if the light intensity is increasing the QoS state is increased.

Super-Cap Voltage Level Prediction: we store the last 5 voltage values of the super-cap and if the voltage level decreases or remains stable over time, the algorithm lowers the QoS by a level while if it is increasing the system increases the QoS. 5 levels of voltage and light intensity help us capture the short term trends while keeping the MCU compute and memory requirements low. Between two sensing intervals, the system goes to sleep.

QoS Voltage QoS QoS-PIR QoS
State [V] Sensings [s] Detection [s] Advertising [s]
7 3.6 - 3.4 20 10 0.1
6 3.4 - 3.2 40 20 0.2
5 3.2 - 3.0 60 30 0.4
4 3.0 - 2.8 120 60 0.64
3 2.8 - 2.6 240 120 0.9
2 2.6 - 2.4 300 300 2
1 2.4 - 2.1 600 600 5
Table 3. Relation between Voltage Level and QoS to enable applications requiring sensing and user-position.

5. Experimental Results

We report experimental results of Pible motes deployed in a real building. By avoiding cold start-operation (e.g. Figure 2), we use a 1F super-capacitor for all our experiments.

5.1. Pible in the Wild

We first describe the specifics of the sensors used.

1 Sensor: We configured Pible for sensing pressure. 5 Sensors: We configured Pible for sensing 5 sensors: light, ambient-temperature, object-temperature, pressure and humidity. PIR Detection: To compare its performance, we placed a battery powered PIR node near Pible. Broadcasting BLE Advertisements: To measure the performance, an external Base Station queries the nodes for light, QoS State, and super-cap voltage levels every 10 minutes.

Table 4 quantifies the average-QoS per day achieved by placing 20 Pible-nodes on 5 different indoor locations. For each node, we collect data for ¿ 15 days.

1 Sensor 5 Sensors Advertising PIR
Light QoS Light QoS Light QoS Light Event[%]
[lux] [s] [lux] [s] [lux] [s] [lux] Detect
Door 121 337 112 564 175 1.9 116 71
Center 246 128 227 251 312 0.9 395 94
Office
window 7k 79 9k 95 6k 1.3 8k 87
Stairs 235 94 238 160 241 0.6 32
Access
Confer 1k 75 427 286 1k 0.8 1k 97
Room
Table 4. Position-Application-QoS Trade-Off for 15 Pible-nodes. Data are averaged per day.

The application executed and the placement of the node play a fundamental role on the QoS achieved. While nearby a window, Pible sends 5 sensor data every 95 seconds on average and 1 sensor data every 79 seconds. The Door case has an average light of 112 lux and sends 5 sensors data every 564s.

Figure 4. Pible in the Wild: (left) Advertising in a Stair Access Building Hall, (right) 5 Sensors in a Conference Room

Figure 4-left shows Pible used for BLE advertising and placed in a stairway where internal lights are always on for security reasons. Even if the average-light per day is similar to other locations (i.e. center of office), the QoS is better due to consistent light availability. The power management algorithm adapts the QoS by fully-charging the super-capacitor and then by increasing the QoS up to the maximum allowed. This validates the usefulness of the lookup table. Figure 4-right shows results of Pible placed in a conference room while sensing 5 sensors. There are no windows in the room, and due to the intermittent light availability due to the presence of people, we place the node close to an internal light (i.e. 2500 lux) otherwise, the node would not have sustained continual operations. Applications such as advertising and one sensing perform well even in the presence of low light (Table 4). For PIR event detection, the placement of Pible in the center of an office or in a conference room detects occupancy events with an accuracy respectively of 94% and 97%. In these conditions, the light available is adequate to recharge the super-capacitors and most events are detected. Stair access case achieves only 32% since too many people pass through the area and time between events is not enough to recharge the super-capacitor.

5.2. Comparison with State of the Art Solutions

The QoS, depends on the application - for periodical sensing sensors (light) it is to sensing rate, for event-driven sensors (PIR) it is the success rate of event capture. We compare the QoS achieved by Pible w.r.t. battery powered systems (Agarwal et al., 2010) and a pure solar energy harvesting powered architecture (PEH) for buildings (Campbell, 2014), since they represent the extremes of our design space. We consider the location Center of Room and emulate the performance of the baselines.

System QoS [s] QoS [s] PIR Events QoS [s] Working
1 Sensor 5 Sensor Detect [%] Advertise Operations
Battery 60 60 100 0.1 to 0.9 few months
Pure EH (Campbell, 2014) 139 309 96 NA when light-on
Pible 128 251 94 0.6 to 0.8 Perpetual
Table 5. Pible QoS Comparison with State of Art Solutions: Battery System end Pure Energy Harvesting Architecture

5.2.1. Quality of Service

Sensing 1 Sensor: PEH (Campbell, 2014) uses a 1.56mJ storage capacitor to sense and send a sensor data without any standard communication protocol. Pible needs 3.20 mJ to sense and send a data using BLE. Since Pible uses the light sensor to monitor the node, we simulate a PEH system that sense and send 2 sensor data (i.e. 6.40mJ). Furthermore, we compare Pible to a battery system from MicroDAQ that sends sensor data every minute using BLE. Results are reported on Table 5. The PEH sends data with an average of 139s and Pible 128s. The results are better for Pible since the PEH stops sending data when the light is off. Sensing 5 Sensors: In this case, the PEH needs to accumulate 8mJ for the 5 sensors. Table 5 shows that a PEH can send data with an average per day of 309s while Pible of 251s. PIR Detection: A PIR detection together with a BLE transmission of light and QoS requires 5.12mJ. By looking at the time-stamp and data given by the battery system, 96% of the events were captured. PEH provides a good accuracy in detection since most of the events happen when light is on. Occupancy Detection: BLE state of the art localization systems exploit advertisement rates between 0.1s and 0.9s (aislelabs.com, 2015). Table 4 shows that the average QoS per day spans between 0.6 and 0.9 when Pible is placed in a Center of Office, Windows or Conference Room. We posit that occupancy based triggers can tolerate latencies of up to 1 second and the Pible QoS is acceptable for these applications.

5.3. Limitations

5.3.1. Manual creation of lookup table and thresholds:

we manually write the QoS for each sensing application given the super-capacitor voltage level (i.e. Table 3

). As a future work, we will explore machine learning methods to automatically configure the sensors to different lighting and application demands.

5.3.2. Operations with no light

With no light and by using the lowest QoS, Pible maintains operations for 19 hours by advertising, 27 hours for 5 sensing applications and 31 hours for 1 sensor applications. During our 15-days experiment, we were always able to achieve perpetual operations since presence of people was constant. If light is not present for more than those times, the node stops working. This can be avoided by (i) moving Pible to a closer source of light; (ii) using a bigger super-capacitor or a bigger solar panel.

6. Conclusion and Future Work

We presented Pible: a battery-free mote for Perpetual Indoor BLE applications build out of commercial off the shelf components. Pible leverages ambient light and a power management algorithm to maximize the quality of service while satisfying perpetual working operations. We tested Pible in a real-world environment, showing that Pible maintained continual operations for 15 days and for 5 different lighting conditions. As a future work, we will explore different prediction mechanisms.

Acknowledgments

This work is supported by the National Science Foundation grants CSR-1526237, TWC-1564009 and BD Spokes 1636879

References

  • (1)
  • Agarwal et al. (2010) Yuvraj Agarwal, Bharathan Balaji, Rajesh Gupta, Jacob Lyles, Michael Wei, and Thomas Weng. 2010. Occupancy-driven energy management for smart building automation. In Proceedings of the 2nd ACM workshop on embedded sensing systems for energy-efficiency in building. ACM, 1–6.
  • aislelabs.com (2015) aislelabs.com. 2015. https://www.aislelabs.com/reports/beacon-guide/. (2015).
  • batteryuniversity.com (2017) batteryuniversity.com. 2017. batteryuniversity.com/learn/article/whats_the_role_of_the_supercapacitor. (2017).
  • Campbell et al. (2016) Bradford Campbell, Joshua Adkins, and Prabal Dutta. 2016. Cinamin: A Perpetual and Nearly Invisible BLE Beacon (EWSN ’16). Junction Publishing, USA, 2.
  • Campbell (2014) B. et al. Campbell. 2014. An Energy-harvesting Sensor Architecture and Toolkit for Building Monitoring and Event Detection (BuildSys ’14). New York, NY, USA.
  • EnOcean (2018) EnOcean. 2018. https://www.enocean.com/en/products/. (2018).
  • et al. (2015) D. Schwarz et al. 2015. Cosero, Find My Keys! Object Localization and Retrieval Using Bluetooth Low Energy Tags. Springer International Publishing, Cham.
  • et al. (2018) F. Fraternali et al. 2018.

    Scaling Configuration of Energy Harvesting Sensors with Reinforcement Learning

    (ENSsys ’18).
  • et al. (2013) Feng Li et al. 2013. Powering Indoor Sensing with Airflows: A Trinity of Energy Harvesting, Synchronous Duty-cycling, and Sensing (SenSys ’13).
  • et al. (2017) J. Hester et al. 2017. Flicker: Rapid Prototyping for the Batteryless Internet-of-Things (SenSys ’17). ACM, New York, NY, USA, Article 19, 13 pages.
  • et al. (2012a) J. Song et al. 2012a. Energy consumption analysis of ZigBee-based energy harvesting wireless sensor networks. In 2012 IEEE ICCS.
  • et al. (2012b) P. Martin et al. 2012b. DoubleDip: Leveraging Thermoelectric Harvesting for Low Power Monitoring of Sporadic Water Use (SenSys ’12). New York, NY, USA, 14.
  • etal (07) Kansal etal. 07. Power Management in Energy Harvesting Sensor Networks. Trans.Embed.Comput.Syst. (Sept. 07).
  • for Demand Response and Efficiency (2015) Cypress Envirosystems. 2015. Retrofitting Existing Buildings for Demand Response and Energy Efficiency. 2015. (2015).
  • Tan and Panda (2010) Yen Kheng Tan and Sanjib Kumar Panda. 2010. Review of energy harvesting technologies for sustainable WSN. In Sustainable wireless sensor networks. Intech.