Adaptive Broadcasting for Similarity Queries in Wireless Content Delivery Systems

Abstract

We present a new adaptive and energy-efficient broadcast model to support flexible responses to client queries. Clients do not have to request documents by name, since they may know the characteristics of the documents but not the document names or IDs. In our model, clients specify requirements through attributes, and servers broadcast documents that match client requests at a prespecified level of similarity. A given document may satisfy several clients, so the server broadcasts a minimal set of documents that achieves a desired level of satisfaction in the client population. The server obtains randomized feedback from clients and adapts its broadcast program accordingly. Clients use a selective tune-in scheme based on approximate indexing to conserve energy. Our model captures client interest patterns efficiently and accurately and scales very well with the number of clients while reducing the overall client average waiting times. The selective tune-in scheme reduces client energy consumption greatly, with a modest wait time increase.