Learning Wireless Networks
A variety of systems have been deployed in recent years to provide
users access over a wired or wireless network to a wide variety of communication, information and entertainment services. These systems
have largely been individually designed and deployed and optimized around information that is largely gathered within the elements of the
particular system. While these systems demonstrate the feasibility of extending new innovative services they fail to systematically gather and retain information useful to the effective use of the larger set of networked systems that the end-users seek access to.
There is a need to identify gaps in the awareness of attributes such as spectrum availability, usage patterns, interference conditions that limit user satisfaction and seek ways to plug these gaps by (a) gathering and preserving large amounts of user observations in a persistent manner and (b) developing new inference algorithms to extract and furnish the needed information from this data set.
Two contemporaneous trends are very relevant to this: (1) rapid drop in the cost of storage; gigabytes of memory will become commonplace in cellular devices in less than a year making it possible to use large amounts of storage to assist with the data gathering and archiving and (2) advances in algorithm design for data analysis.
It is now possible to create a cognitive network that combines advances in storage, connectivity and processing to construct a wireless network in which users share information to enhance end-user experience. We envisage a large number of mobile end-user devices serviced by a combination of licensed and unlicensed band operators. End-users will actively gather, store and ultimately furnish relevant information to a community of their peers. The information gathered is derived from the state of their radio, the state of their protocol
stacks and the nature of their traffic and usage patterns and other such sources. Raw information gathered, will be spatio-temporally
tagged, compacted and relayed over a possibly out of band communication channel to a repository where the data will reside
persistently for others to utilize. This persistent repository, which could be distributed physically, is organized as a single coherent
entity that can be queried over the cyberinfrastructure!. A user passing through a particular spatio-temporal cell could
learn about the attributes of that environment such as typical usage, user experiences, and network conditions and adjust their use.
The need for this architecture is largely driven by user challenges in evaluating and optimizing wireless services. Typically, users make service decisions based on testimonials from others based on their experience. Experiences and beliefs frequently shared range from binary judgments on the utility of a service ("Black Berry on the Verizon network really works") to fine grained details about scheduling of transmissions in certain localized spatio-temporal domains (I will email you the attachment when I get off the flight). Existing web repositories of this type of information build on manual data entry of anecdotal user experience not objectively verified. However, this anecdotal approach is the current "best practice" with even the FCC advising users when choosing a service provider to "Ask neighbors, work colleagues and friends who have similar calling patterns about their experiences with different service providers and plans. "
The essence of this approach is to exploit inexpensive storage and the availability of low cost "some time some where" communications to asynchronously and automatically gather large amounts of user data by enlisting users as probes. Once a good amount of data has been gathered algorithms that mine this data set can be developed to generate meaningful ways to improve user experience.
In contrast to most deployed systems, this system is not designed to tightly optimize a particular objective. The gathering of the data, the algorithms to generate useful hints and the use of these hints by mobile end users and others are loosely coupled in this open architecture. No one user is critical to the functioning of the system but the quality of improvements generated depends on the number of users that participate in this system and the quality of the information they furnish.