Asynchronous Network Signal Sensing and Classification Techniques: ANDRO's Automatic Modulation Classification with Multiple Sensors (AMCMS) Capabilit

This effort considers the problem of Automatic Modulation Classification (AMC) using multiple asynchronous sensors in non-cooperative environments under low signal-to-noise ratio (SNR) regimes. The goal is to improve and demonstrate the performance of AMC systems on various weak signal scenarios in a multi-cast environment that a traditional single sensor would not be able to readily classify. Candidate approaches involve both distributed decision fusion as well as centralized data fusion of asynchronous sensor data for multi-hypothesis modulation classification. This effort will build upon the results of a prior phase study, where the asymptotic behavior of distributed modulation classification systems was analyzed and conditions under which asymptotic probability of error goes to zero were derived. Upper and lower bounds for probability of error were derived based on Chernoff and Bhattacharyya error exponents and Monte Carlo sampling techniques. The optimal fusion rule for multi-hypothesis testing was developed and comparisons were carried out with the majority fusion rule. A maximum likelihood based centralized fusion problem was also formulated where each sensor experiences a different SNR and the network is asynchronous, i.e. each sensor has a non-identical phase, frequency and timing offset. In this effort, a novel centralized data fusion algorithm with multiple asynchronous sensors will be developed. The problem of asynchronous sensor data fusion for AMC using multiple sensors is considered untapped. A novel Distributed Automatic Modulation Classification (DAMC) technology and innovative approach are presented that exploits asynchronous multiple sensor data in the most effective way possible for this purpose. For distributed fusion, the proposed technology, based on a theoretical understanding of independence and dependence (Copula theory), will enable the development of novel fusion methodologies for maximized classification performance. Additionally, for time critical applications, sequential classification procedures will be explored. Time and computational complexity are also considered in proposed algorithms in view of limited computational resources. New algorithms will be developed and existing ones will be enhanced in this phase, along with trials for distributed signal sensing. Classification hardware prototypes will be tested and a robustness test of the new methods will be presented. The development of real-time software implementing this system will be produced and demonstrated. The completion of this phase will result in a mature DAMC technology, which will be inserted into selected hardware and undergo operational tests with real world signal transmission and reception in a fully functional, distributed sensor network.