Automatic modulation classification of MIMO signals
 

Modulation classification of MIMO signals is a challenging problem due to interference between source signals. An intuitive solution to modulation
classification in MIMO is to first separate source signals and perform modulation classification on separated sources. Separation of sources has been

an active research topic through years and some widely used approaches are Independent Component Analysis (ICA) and clustering. In classifying modulation types, likelihood- and feature-based methods are the two main approaches. In this project, we work on feature-based modulation classification approaches through the use of Support vector machine (SVM). SVM's are statistical learning machines which learn through training data, generalize a pattern and classify the feature based on this generated pattern. Compared to other fixed threshold feature-based approaches, SVM employs adaptive thresholds which improve the correct classification rates.
 

More details

Geolocation in Multipath Environment

(Possible applications: geolocation in sensor networks, geolocation in wireless networks)

In this work, we seek to develop techniques for accurate geolocation of RF sources through the utilization of multiple sensors spatially distributed. Two main approaches are studied. In non-coherent geolocation, the source localization is obtained from estimates of the signal time of arrival (or time difference of arrival if the time reference of the source is not available). Coherent source localization is achieved by exploiting the phase information contained in signals at various points in space. It is shown that coherent localization has the ability to provide highly accurate localization, but has to contend with ambiguity sidelobes. Our investigations are focused on ways to alleviate the sidelobe problem in coherent localization and to improve the accuracy in non-coherent localization.

The work is funded by U.S. Army Ft. Monmouth and it is carried out in collaboration with Princeton University.

 A  metric derived based on maximum likelihood estimation utilizing phase information is capable of localizing sources with an accuracy of the order of the source carrier wavelength. However, the maximum likelihood metric is subject to high ambiguity sidelobes. Several methods to reduce the sidelobes are investigated. The most basic way of reducing sidelobes is increasing the number of sensors. However, that might be difficult in many practical situations. In this project, we study various other ways to control sidelobes in localization systems utilizing phase information. An alternative to increasing the number of sensors is equipping each sensor with an antenna array. Another method is to place the sensors (or some of them) in a moving platform and utilize multiple locations to observe the signals transmitted by the emitting source. The latter method has the disadvantage of processing non-coherently the signals obtained from different sensor locations.  The capability of multiple sensors to act to reduce sidelobes is degraded when the sensors are widely varying distances from the source. This motivated the development of a localization technique that is robust to distance variations (as long as a required SNR is maintained). The technique  is based on the variance of the phase measured at the different sensors.

Publications:

1.      V. M. Chiriac, A. M. Haimovich, S. C. Schwartz, and J. A. Dabin, "Performance bound for localization of a near field source," in the Conference Record of the Forty-Third Asilomar Conference on Signals, Systems and Computers, 2009, pp. 130-135.

2.      V. M. Chiriac and A. M. Haimovich, "Ziv-Zakai lower bound on target localization estimation in MIMO radar systems," in the IEEE Radar Conference, 2010, pp. 678-683

The coherent source localization approach was shown to offer high accuracy but at the same time it has to contend with high ambiguity sidelobes. Since the non-coherent localization approach doesn't have the sidelobe issue, but its accuracy is significantly lower, ways to improve accuracy offered by the non-coherent approach are still of interest, especially in multipath propagation environments. That is because the non-coherent approach involves time delay estimation, which is particularly challenging in multipath environments, where the line of sight signal component becomes obscured by the multipath reflections. Promising results in improving the localization accuracy were obtained by exploiting the sparsity of the multipath channels. Consequently, part of our work is focused on exploiting the sparse representation framework to improve the accuracy of the non-coherent localization approach.
 

Publications:

1.      C. R. Comsa, J. Luo, A. M. Haimovich, and S. Schwartz, "Wireless localization using time difference of arrival in narrow-band multipath systems," in Proceedings of the IEEE International Symposium on Signals, Circuits and Systems, ISSCS 2007, Iasi, Romania, 13-14 July 2007, vol. 2, pp. 469 - 472.

2.      C. R. Comsa, A. M. Haimovich, S. Schwartz, Y. Dobyns, and J. A. Dabin, "Source localization using time difference of arrival within a sparse representation framework," submitted to the International Conference on Acoustics, Speech and Signal Processing, ICASSP 2011, Prague, Czech Republic, 22-27 May 2011.

Blind Equalization & Blind Source Separation (BSS) of MIMO Channels

The objective of blind equalization (or BSS) is to recover the unknown input sequence to the unknown channel based solely on the probabilistic and statistical properties (higher order statistics) of the input sequence.

There are a number of blind equalization methods been presented. Among others, constant modulus algorithm (CMA) is a particular Godard algorithm minimizes certain cost function, which was also developed independently. With a number of demonstrated successes in real applications and its relatively good convergence property, the CMA equalizer quickly gained popular acceptance due in part to its simplicity and its effectiveness. Independent Component Analysis (ICA) is a relatively new approach for implementing blind source separation. In essence, it is a statistical and computational technique for revealing hidden factors (independent components) underlie signals. For a noise free MIMO system, using ICA we can recover the channel matrix up to an unknown scaling factor and permutation of the input source signals.

In this project, we will carry out an in-depth investigation into these techniques, working on the performance and computational complexity issues related to them.
 

Nonconvex Sparse Localization: Beyond Local Methods

We address the problem of source localization within a sparse representation framework. Among other properties, the sparse approach has been shown to achieve high resolution localization. We propose a branch-and-bound algorithm to tackle the resulting nonconvex problem. Thanks to a non-trivial problem reformulation, the proposed algorithm is efficiently performed on a dual domain in order to achieve performances beyond that possible with state-of-the-art local methods. Rather than aiming to reach a local optimal solution, the proposed algorithm characterizes the topology of multiple local solutions, in order to increase localization accuracy in low signal-to-noise scenarios and/or in presence of multiple or extended targets.