A layered infrastructure for mobility-aware best connectivity in the heterogeneous wireless internet

Paolo Bellavista, Antonio Corradi, Carlo GiannelliMOBILWARE 2008 · OpenReview

Introduction

Modern mobile devices are equipped with multiple wireless interfaces, including IEEE 802.11 (WiFi), Bluetooth, and cellular networks (3G/UMTS). Despite this hardware diversity, most devices follow rigid, hard-coded priorities—such as defaulting to WiFi whenever available—without considering the specific context of the user or the requirements of active applications. The "Always Best Connected" (ABC) concept posits that a device should intelligently switch between interfaces to optimize for factors like battery life, cost, and connection stability. However, defining what is "best" is inherently subjective: a streaming video application prioritizes high bandwidth, while a background synchronization task may prioritize energy efficiency.

To address these challenges, this paper introduces MAC (Mobility-Aware Connectivity), a two-layered middleware architecture that incorporates mobility-awareness and per-application requirements into a unified, low-overhead framework.

The MAC Middleware Architecture

The MAC architecture is designed to separate node-wide concerns from application-specific needs through two distinct layers:

  1. Connector Manager (CoM) - Bottom Layer: This layer manages node-wide requirements. It monitors available network interfaces and filters out unsuitable connections based on device-level constraints, such as battery life, security trust levels, and the user's current mobility state.
  2. Channel Selector (ChaS) - Top Layer: This layer addresses per-application requirements. It receives the list of "suitable" channels from the CoM and selects the optimal interface for a specific application's needs, such as prioritizing bandwidth for FTP transfers or durability for VoIP calls.
The MAC middleware architecture showing the interaction between the Channel Selector, Connector Manager, and Mobility & Peer Estimator.

By treating other mobile devices as potential "peer connectors," the system significantly expands the connectivity map beyond traditional fixed infrastructure. This decentralized approach requires no specialized network support, making it immediately deployable on existing hardware.

Mobility Estimation and Signal Processing

A critical component of the MAC system is the Mobility & Peer Estimator (MPE). Mobility serves as a proxy for connection durability; for instance, if a user is moving rapidly, short-range Bluetooth connections are discarded to avoid the overhead of frequent handovers.

To estimate mobility without the high power consumption of GPS, the MPE monitors Received Signal Strength Indication (RSSI) fluctuations. The process involves:

  • Noise Filtering: The system applies a Discrete Fourier Transform (DFT) to convert RSSI data into the frequency domain, filters out high-frequency noise, and uses an Inverse DFT (IDFT) to reconstruct a clean signal.
  • State Classification: A state machine classifies the user as either "Still" or "In Motion." To prevent "bouncing" (rapidly toggling between states due to minor signal fluctuations), the system employs a hysteresis model with distinct thresholds for entering and exiting the motion state.
The Mobility & Peer Estimator (MPE) signal processing pipeline and state machine.

Evaluation and Results

The researchers evaluated the MAC middleware using both a custom simulation environment and a real-world prototype. The simulation featured 17 Access Points in a hexagonal grid with varying levels of signal noise, while the prototype was tested on Windows and Linux platforms supporting WiFi and Bluetooth.

Key findings include:

  • High Detection Accuracy: In low-to-moderate noise scenarios, the system achieved a "Long Term Hit Rate" for mobility detection between 94% and 99% after a brief stabilization period.
  • Robustness: Even under high signal noise (5dB standard deviation), the system maintained a 65% accuracy rate, validating the effectiveness of the DFT-based filtering.
  • Operational Efficiency: The system is computationally light, ensuring minimal impact on battery life. While there is a detection lag of 4 to 13 seconds, this is considered sufficient for the coarse-grained decision-making required by the middleware.

Conclusion and Future Work

The MAC middleware demonstrates that mobility-aware, per-application network selection is feasible with low overhead. By separating node-level and application-level logic, the system prevents individual apps from negatively impacting device-wide resources.

Future research will focus on reducing detection lag for high-speed mobility scenarios and integrating a wider range of technologies, such as WiMAX and 4G. Additionally, the authors aim to develop "Continuity Management" features to enable seamless data packet handovers, ensuring uninterrupted service during interface transitions.

Summary updated

BibTeX
@inproceedings{bellavista2008layered,
  author = {Paolo Bellavista and Antonio Corradi and Carlo Giannelli},
  title = {{A layered infrastructure for mobility-aware best connectivity in the heterogeneous wireless internet}},
  booktitle = {MOBILWARE},
  year = {2008},
  url = {https://openreview.net/forum?id=tDthDEPBwo},
}
Figure