DISTRIBUTED FEATURE-BASED MODULATION CLASSIFICATION USING WIRELESS SENSOR NETWORKS Pedro A. Forero, Alfonso Cano and Georgios B. Giannakis Dept. of ECE, University of Minnesota, Minneapolis, MN 55455 ABSTRACT Automatic modulation classification (AMC) is a critical prerequisite for demodulation of communication signals in tactical scenarios. Depending on the number of un- known parameters involved, the complexity of AMC can be prohibitive. Existing maximum-likelihood and feature-based approaches rely on centralized processing. The present paper develops AMC algorithms using spatially distributed sensors, each acquiring relevant features of the received signal. Individual sensors may be unable to extract all relevant features to reach a reliable classification decision. However, the cooperative in-network approach developed enables high classification rates at reduced-overhead, even when features are noisy and/or missing. Simulated tests illustrate the performance of the novel distributed AMC scheme. I. I NTRODUCTION Automatic modulation classification (AMC) of digital modulations amounts to identifying the constellation used by a digital communication system. It plays a key role in military applications ranging from eavesdropping and jamming to cognitive and software defined radios. Once a signal has been detected, the AMC algorithm chooses the modulation format from a pool of possible candidates. AMC has been widely studied and many approaches have been proposed; however, formidable challenges remain unsolved especially when the number of modulation formats is large and many unknown parameters are involved [3], [6], [1], [11], [12]. Among the notable AMC approaches, likelihood-based (LB) algorithms characterize the likelihood function of the received waveform conditioned on a particular constellation format. Selecting a modulation then reduces to testing multiple hypotheses. Viewing unknown parameters as ran- dom, leads to an optimal decision in the Bayesian sense. Work in this paper was supported by the USDoD ARO Grant No. W911NF-05-1-0283; and also through collaborative participation in the C&N Consortium sponsored by the U. S. ARL under the CTA Program, Cooperative Agreement DAAD19-01-2-0011. The U. S. Government is authorized to reproduce and distribute reprints for Government purposes notwithstanding any copyright notation thereon. 978-1-4244-2677-5/08/$25.00 2008 IEEE However, the associated computational complexity reduces the range of applicability of these AMC schemes to cases where large processing units are available [3]. Suboptimal alternatives reduce complexity at the expense of lowering performance. Feature-based (FB) approaches on the other hand, rely on a set of features to perform the classification task [3]. Among the several FB-AMC algorithms available, those based on the methods of cumulants and moments incur manageable computational complexity relative to LB alternatives while still delivering reliable classification per- formance [3], [11], [12], [10]. Most existing LB and FB algorithms require data to be available at a central processing unit. With the surge of wireless sensor networks (WSNs), there has been a growing interest towards decentralized detection, estimation and classification schemes. Their advantages range from easier self-deployment to higher resilience and prolonged lifetime in hostile environments. This paper develops a cumulant- based distributed AMC algorithm using WSNs. Sensors are deployed in space and enabled to detect and process signals from their surrounding environment. By exchang- ing suitable information with their single-hop neighbors, individual sensors are able to classify the constellation format with accuracy as high as if they had received all other sensors’ data. Although all sensors are acquiring the same underlying signal (same transmitted symbols), the received signals at different sensors are different due to the effect of the propagation channel. This allows the novel algorithm to exploit spatial diversity which in turn boosts the classification performance. As part of the distributed AMC task, a distributed clus- tering algorithm along with a suitable model order selection criterion are also introduced. The distributed clustering algorithm is based on the distributed k-means algorithm in [4] and its objective is to enhance classification performance at high SNR. The rest of this paper is organized as follows. The problem of AMC using WSNs is formulated in Section 2. Section 3 deals with cumulant features for AMC, while Section 4 introduces the novel distributed AMC approach based on cumulant features. A distributed clustering ap- proach is developed in Section 5 to determine the order of the modulation format and improve performance at high 1 of 7