TL;DR
This paper tackles the growing concern of mobile device energy consumption in WiFi environments by developing practical energy models. They demonstrate that both WiFi signal strength and the type/amount of network packets initiated by access points significantly impact a phone's battery drain, even when the phone is in a basic, idle state. The proposed modeling methods offer developers an accessible way to understand and predict this energy consumption.
What Problem Does This Paper Address?
As the Internet of Things (IoT) expands, mobile devices like smartphones are increasingly central to our connected lives. However, their limited battery capacity presents a persistent challenge, especially when running power-hungry applications or maintaining constant network connectivity. While much research has focused on energy consumption related to specific applications or poor signal conditions, this paper addresses a more nuanced user experience: why a phone's battery drains differently even when idle, depending on the WiFi environment. Specifically, it investigates two key factors: the impact of good WiFi signal strength (not just weak signals) and the difference in energy consumption between public and private WiFi networks. The goal is to provide understandable models for developers and users to better manage phone energy.
Methods
The researchers employed a two-pronged experimental approach to model energy consumption:
- Signal Strength-Aware Model: They conducted experiments using a Samsung Galaxy GT-S7898 phone in a basic state (WiFi on, other radios off, no apps running). By varying the distance between the phone and the WiFi Access Point (AP), they manipulated the Received Signal Strength Indicator (RSSI). Energy consumption was measured using a power meter over time. They then developed a time-based energy model that correlates energy consumption with both time and RSSI, using linear regression and curve fitting techniques.
- Packet Type/Amount-Aware Model: To differentiate between public and private WiFi networks, they focused on the types of protocol packets initiated by the AP. They identified that private networks primarily use UDP and TCP packets, while public networks also send ICMP and IGMP packets. Using a software tool to send controlled groups of these packets (UDP, TCP, ICMP, IGMP) in varying amounts (5, 10, 15, 20 packets), they measured the phone's energy consumption. This led to packet-driven energy models that linearly relate energy consumption to the number of packets for each type.
Datasets
The paper does not introduce new benchmark datasets. Instead, it relies on data collected through controlled experiments using a Samsung Galaxy GT-S7898 smartphone. The experimental data consists of:
- Measurements of phone energy consumption (in Joules) at various distances from a WiFi AP, correlating to different RSSI levels.
- Measurements of phone energy consumption when handling different types and quantities of protocol packets (UDP, TCP, ICMP, IGMP) sent by a WiFi AP.
Key Insights and Ideas
- Good Signal Strength Still Matters: Counterintuitively, even under "good" WiFi signal conditions (RSSI > -50 dBm), a stronger signal (higher RSSI, closer distance to AP) leads to lower energy consumption. This suggests that the phone's radio might be more efficient when it doesn't have to work as hard to maintain a stable connection.
- Public vs. Private Networks: It's the Packets: The type of WiFi network (public vs. private) significantly impacts energy drain, even when the phone is idle. This difference is attributed to the specific protocol packets initiated by the AP. Public networks, which often send ICMP and IGMP packets in addition to UDP/TCP, cause higher energy draw than private networks that primarily use UDP/TCP.
- Linear Relationship with Packet Count: For TCP, ICMP, and IGMP packets, the energy consumption scales linearly with the number of packets sent by the AP. This provides a predictable way to model energy usage based on network traffic.
- Accessible Modeling: The proposed modeling methods are designed to be accessible to ordinary developers. They rely on inexpensive power meters for data collection and use straightforward regression techniques, making it easier to analyze and predict phone energy draw in WiFi environments.
Key Results
- Signal Strength Model: The developed signal strength-aware model, $E(RSSI,t) = (0.009|RSSI|^2 - 0.7|RSSI| + 14.87)t + 1.76$, accurately predicts energy consumption based on RSSI and time, with high R-square values (close to 1) and low RMSE. It shows that energy consumption increases rapidly as RSSI drops below -78 dBm.
- Packet-Driven Models:
- UDP packets from APs have a negligible impact on phone energy consumption.
- TCP, ICMP, and IGMP packets show a linear relationship with energy consumption.
- ICMP and IGMP packets have a higher impact on energy consumption than TCP packets. For example, an IGMP packet contributes approximately 0.723 J to energy consumption, while a TCP packet contributes about 0.612 J.
- The models confirm that phones consume energy faster in public WiFi networks than in private ones, aligning with user experience.
Limitations and Future Work
- Phone-Type Dependency: The authors acknowledge that the created models are phone-type dependent. However, they emphasize that the methodology for creating these models is general and can be applied to any phone.
- Limited Factors: The study focused on signal strength and packet types initiated by the AP. Other factors influencing energy consumption, such as retransmission rates, error rates, transmission power, phone mobility, and the number of users in the network, were not deeply explored.
- Future Work: The authors plan to:
- Collect more experimental data on diverse phone types to improve model generalizability.
- Investigate the impact of other WiFi environmental factors.
- Develop models that consider direct factors like transmission parameters and packet exchange initiated by the phone itself.
- Incorporate factors like phone mobility and user density to better explain differences between public and private network energy consumption.
Summary updated