Handover Analysis

Mobile devices rely on cellular networks to get network access to support data services. Since the coverage of each cell is limited, handover between cells is essential for ensuring continuous connectivity and mobility. In addition, when the device is in the coverage of multiple cells, a proper pol- icy should handover the mobile device to a cell that provides good performance.

We perform the first large-scale study of intra-LTE handovers for cellular providers using crowd-sourced measurements of over 200 users across three major carriers for the purpose of evaluating the performance implications of existing handover algorithms and policies.

Information collected
The information we collect includes:
  • Network type and signal strength
  • Cell information
  • Radio Resource Control (RRC) Messages between devices and cells
  • Active measurement results, include ping, traceroute, UDP burst, throughput measurements
Impact on user experience
The experiment is carefully designed to minimize the influence to user experience. Measurements are triggered only when the device starts to move and a handover is likely to occur. Heavy-load throughput measurements are triggered only when the screen is off and users are not interacting with the device. We also set limit on the power/data resources we use.
Power Consumption
To control the power consumption, we keep track of power consumption of the experiments and ensure it always consumes less than 10% of total battery resources after the device is unplugged. Also, we disable active measurement when the battery is below 30%.
Data Usage
We ensure the experiments consume less than 500MB of data per day. 

Code and data
To minimize measurement overhead while capturing most of handover events, we develop a measurement framework to trigger measurement based on context. We estimate the likelihood of observing relevant events based on the device context and trigger measurements only when the probability of capturing handover events is high. This helps reduce unnecessary measurements while capturing more events of interest.

We implement the context-triggered framework on top of Mobilyzer. The source code context-triggered framework can be accessed at https://github.com/mobilyzer/Mobilyzer/tree/ContextAware/Mobilyzer.

We collect data from more than 200 users for around 6 months. The data can be accessed at https://drive.google.com/drive/folders/0B1U9sfXTfAmtajVYd3JrSUxCSnM?usp=sharing.

This project is led by RobustNet Research Group from University of Michigan at Ann Arbor.
  • Prof. Z. Morley Mao
  • Shichang Shawn Xu (xsc AT umich.edu)
  • Ashkan Nikravesh