IIT Madras is a public technical and research university located in Chennai, Tamil Nadu. Founded in 1959, it is recognised as an Institute of National Importance.

IIT Madras has been ranked as the top engineering institute in India for four years in a row by the National Institutional Ranking Framework of the MHRD

It currently offers undergraduate, postgraduate and research degrees across 16 disciplines in Engineering, Sciences, Humanities and Management. About 596 faculty belonging to science and engineering departments and centres of the Institute are engaged in teaching, research and industrial consultancy.

IIT Madras

- In this paper, we present fast maximum likelihood (FML) estimation of parameters of multiple exponentially damped sinusoids. The FML algorithm was motivated by the desire to analyze data that have many closely spaced components, such as the NMR spectroscopy data of human blood plasma. The computational efficiency of FML lies in reducing the multidimensional search involved in ML estimation into multiple 1-D searches. This is achieved by using our knowledge of the shape of the compressed likelihood function (CLF) in the parameter space. The proposed FML algorithm is an iterative method that decomposes the original data into its constituent signal components and estimates the parameters of the individual components efficiently using our knowledge of the shape of the CLF. The other striking features of the proposed algorithm are that it provides procedures for initialization, has a fast converging iteration stage, and makes use of the information extracted in preliminary iterations to segment the data suitably to increase the effective signal-to-noise ratio (SNR). The computational complexity and the performance of the proposed algorithm are compared with other existing methods such as those based on linear prediction, KiSS/IQML, alternating projections (AP), and expectation-maximization (EM). © 1996 IEEE.

Department of Electrical Engineering

World University root node

IEEE Transactions on Signal Processing

Estimation of parameters of exponentially damped sinusoids using fast maximum likelihood estimation with application to NMR spectroscopy data

We review the standard frequency domain methods for modulation classification as a baseline for comparison with new approaches. A new procedure based on the singular value decomposition of a particular data matrix is developed. This matrix has some very interesting properties that facilitate a solution of a difficult problem of QPSK versus MSK classification. Performance of the new method is assessed via simulations and compared against those in the published literature.

Proceedings Volume 2563, Advanced Signal Processing Algorithms

Classification of digital modulation types

We argue that an important aspect of the human speech signal is scaling in the frequency domain. We discuss the two physical mechanisms responsible for the scaling. The first mechanism is that when we have a harmonic signal whose fundamental is frequency modulated then the spectrum is the sum of scaled functions. The second comes about from the consideration that while different speakers have very different size vocal tracts (for example an adult and a child), we none the less produce speech which is similar in some sense. We will argue and present evidence to show that the speaker differences result in scaling in the frequency domain. We further discuss how one can handle scale processing.

Proceedings of SPIE - The International Society for Optical Engineering

Scale-invariant speech analysis via joint time-frequency-scale processing

We show that signals which are harmonically related to a fundamental frequency which has a finite bandwidth is modeled by the sum of scaled functions. We present a method to study such signals and give an example which illustrates the method.

Scale and harmonic-type signals

When a periodic waveform is modulated the resulting signal consists of a fundamental which has broadness in frequency and the sum of functions that are scaled and/or shifted versions of the fundamental. We present a method to study such functions and present an example using a speech signal.

Scopus

Joint representation in time and frequency scale for harmonic type signals

A computationally efficient fast maximum-likelihood (FML) estimation scheme, which makes use of the shape of the surface of the compressed likelihood function (CLF), is proposed. The scheme uses only multiple one-dimensional searches oriented along appropriate ridges on the surface of the CLF. Simulations indicate that the performances of the proposed estimators match those of the corresponding maximum-likelihood estimators with very high probability. The approach is demonstrated by applying it to two different problems. The first problem involves the estimation of time of arrival and Doppler compression of a wideband hyperbolic frequency modulated (HFM) active sonar signal buried in reverberation. The second problem deals with estimating the frequencies of sinusoids. A threshold analysis of the proposed scheme is carried out to predict the signal-to-noise ratio (SNR) at which large estimation errors begin to occur, i.e., the threshold SNR, and its computational complexity is discussed.

IEEE Journal of Oceanic Engineering

Fast maximum likelihood estimation of signal parameters using the shape of the compressed likelihood function

A system of neural networks in a multistage architecture is used to resolve the components of a transient signal. The design is motivated by a desire to use smaller networks that compute a binary decision allowing the use of the Multilayer Perception Design Algorithm for faster and more effective training, and to cascade these simple networks into a pipeline architecture for efficient implementation. Simulation results are compared with the Multistage Subspace technique that utilizes all of the information in the signal model. The networks are trained with examples of one signal component in noise at a specified noise level. The resulting Multistage Neural Network is able to generalize to different noise levels and multiple signals without additional training. The neural network detector and feature extraction system localizes the arrival time and frequency for each sufficiently strong transient signal present.

Proceedings - ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing

Multistage neural network structure for transient detection and feature extraction

In this paper propose a multistage procedure to achieve good local resolution of significantly overlapping components in a transient signal. We assume the received transient signal to be a superposition of modulated and time-shifted versions of a single one-sided exponential function whose damping coefficient is known. At each stage of the proposed procedure we coarsely identify (as compared to the next stage) broad regions in the time-frequency (t-f) plane that have significant signal energy. These correspond to possible regions where signal components may be present. In each succeeding stage, they focus on these coarsely identified signal regions to further identify subregions in them that have significant energy. The detection of signal components in the sub-regions is done in the presence of possible interference from neighboring regions. Implementation of this method is based on the use of generalized eigenvectors of signal and interference correlation matrices. The detection statistics is a linear combination of the magnitude squared projections of the received data onto the individual generalized eigenvectors. © 1992 IEEE.

ICASSP, IEEE International Conference on Acoustics, Speech and Signal Processing - Proceedings

Resolving the components of transient signals by a multistage procedure

In structuring a Multilayer Perceptron (MLP) Neural Network the order of the number of hidden nodes and the starting weights can greatly affect the network training and performance. For the transient detection problem, the Subspace Inhibition Filter Algorithm concepts are used to design the neural network with a small number of nodes in a single hidden layer and compute a starting point for the weight training. An algorithm is presented to calculate the minimum number of hidden nodes required and compute a good set of starting weights for the MLP. Simulation results show the feasibility of the algorithm and the improvements in training and performance for detecting transients. © 1992 IEEE.