The Cognitive Complete Knowledge Network (CogNet) system combines advances in storage, network connectivity, and device capability, with advances in analytical approaches to make next generation wireless systems refine their performance throughout the lifetime of their deployment. The CogNet architecture exploits the availability of low cost "some time some where" communications to asynchronously and automatically gather large amounts of user data by enlisting users, devices, and even the whole network as probes. The information gathered is derived from the radio state, network state, environment state, and user profiles, and is spatio-temporally tagged and archived in a distributed repository and made available to the community of participants. A user in a particular spatio-temporal region could learn about typical usage, others' experiences, network conditions, and protocol parameters by querying this repository and tune or adapt itself to achieve the desired quality of service.
We use stochastic analysis, algorithm development, performance analysis and optimization, and prototype deployment to identify globally relevant internal state data from the vast amounts generated within any mobile device; manage data transfer and archiving; design learning models and algorithms to fit the large and diverse collected dataset; develop cross-layer, cross-system adaptations to use information obtained from the collective experience; and analyze and learn the network traffic and network parameters to identify appropriate service compositioning parameters to guarantee user experience.
