Existing models of wireless transmission errors are generally simple two-state Markov models, e.g. the Gilbert-Elliott model. Developed in the 1960s, these models have questionable relevance to modern networks. In particular, the parameters for these models have largely persisted through a tradition of use in academic papers, without being substantiated by actual measurements from real modern networks. These models also often fail to accurately characterize the distribution tail, which is critical when attempting to establish network coverage across whole environments.

This project will first collect a comprehensive set of data recording the behavior of operational wireless networks. This will consist of information available at the driver level and above, including: 1. packet-level analysis of errors (e.g. monitoring which packets broadcast using UDP are successfully received), 2. bit-level analysis of errors (e.g. modifying NIC drivers to retain errored frames for comparison at the bit level to what is known to have been transmitted), and 3. signal-strength analysis. We intend to start in the 2.4GHz band (using PRISM-based 802.11b NICs that allow capture of errored frames), and then cover the 5GHz band (using packet-level analysis alone if NICs don't allow deeper access). We will also make environmental measurements (of floor plan, spectrum use, etc) to allow attribution of transmission characteristics to environmental causes.

Wireless LAN error characteristics can vary widely, depending on device characteristics (e.g. antenna design and orientation) and environmental factors (e.g. floor plan) and variables (e.g. humidity, noise sources, and movement of people). To confidently characterize wireless LAN behavior, we will need to conduct numerous experiments in different locations and at different times. We intend to develop an experimental plan that allows our large numbers of undergraduate students (130 p.a. in the EE&T school, 200+ p.a. in the CSE school) to make measurements as part of the practical component of their courses. The equipment proposed for this project will allow these 'labs' to be conducted by groups of students in varied locations (e.g. home, work, etc).

This project will then map the observed errors to a specific mathematical model of the communications channel. This will be taken to be an n-state Hidden Markov Model (HMM). The thrust of this exercise will be the development of mathematical tools to estimate the parameters (number of states, transition matrix etc) of the HMM. The research will involve optimization strategies using the Maximum Likelihood criteria, and use of the Expectation Maximization algorithm.

Facilities

• Platforms:
  • Laptop computers
  • HP iPAQ Pocket PCs
• Network Interface Cards:
  • 802.11b PCMCIA cards from Avaya/Lucent, D-link and Cisco
  • 802.11a CardBus NICs from Intel and Netgear
  • CompactFlash 802.11b NICs are also available
• Access Points:
  • 3 Avaya AP-3 access points with 802.11a/b interfaces, and range extender antennas.
  • Cisco 350 series AP
  • D-link AP

Student activities

Results