Santanu Kumar Nayak

Neuro fuzzy model for adaptive filtering of oscillatory signals

In this paper we have developed a neuro fuzzy model for adaptive filtering of oscillatory signals embedded with white noise. Such type of fuzzy adaptive filters are constructed from a set of fuzzy IF-THEN rules, which change adaptively to minimise the cost function until a desired information is available. Here we have used a generalised cost function for better convergence of the error. This algorithm is simulated on a digital signal processor in order to track the signal and to filter out the disturbances present in the signal at a particular instant of time. The system presented here, can measure both types of information like numerical as well as linguistic.

Hybrid ANN reducing training time requirements and decision delay for equalization in presence of co-channel interference

Bayesian equalizer is known to be the optimum equalizer. This paper proposes a Hybrid Artificial Neural Network (Hybrid ANN) and an algorithm to modify Decision Feedback Equalizer (DFE) function of Bayesian equalizer while equalizing in presence of co-channel interference (CCI). A combination of Artificial Neural Network and Decision Feedback Equalizer (DFE) is termed as Neural-DFE (NDFE). The results show that the decision delay and training time requirement reduces significantly by use of NDFE. This creates an advantage specifically for a mobile environment where the CCI is varying in nature and the Bayesian equalizer requires a lot of training time.

Non-linear channel equalization using adaptive MPNN

In this paper the Modified Probabilistic Neural Network (MPNN) is used for dealing with the problem of channel equalization. Some improvements are suggested for the MPNN so that it is more suitable for the current problem. Firstly, the MPNN is extended to process complex signals. Secondly, a stochastic gradient adaptation technique is proposed, such that when the network is being employed to equalize a slowly varying channel, it can self-adapt to the changing environment. Simulations have shown that the MPNN is able to effectively equalize 4-QAM symbol sequences transmitted through a non-linear, slowly time-varying channel. Finally, methods that further reduce the size of the network are proposed. Simulations show that the proposed method is able to reduce the size of the network considerably.

An intelligent instrument for tracking and adaptive filtering of oscillatory signals using Hebbian learning rules

In this paper Hebbian type of learning algorithms using total least squares method is applied for adaptive filtering techniques to remove the noise and undesired oscillatory signals at different systems. Here we have used the generalised Hebbian learning rules for initializing the internal representations of a feedforward neural network, which accelerates the convergence of supervised Hebbian learning rule. In case of constrained anti-Hebbian learning rule, the weight vectors of linear neuron unit is converged to an eigenvector which has the smallest eigenvalue. In the total least squares (TLS) method the noise rejection capability is superior to the least squares method. Here we have applied the initial sets of data for the internal representation of feedforward network which consists of bottom-up unsupervised learning process followed by top-down supervised learning process using total least squares (TLS) algorithm. For faster convergence we have included the momentum term for the updating of weights. An intelligent instrumentation scheme has been developed for on-line measurement of amplitude of oscillatory signals. The undesired oscillations of the signal is also removed by implementing neural network model (using Hebbian rules and total least square algorithm) on a digital signal processor.

Equalization of Communication Channels Using GA-Trained RBF Networks

Equalization of communication channels still remains a challenge. Radial Basis Function Neural Network (RBFNN) based equalizers are also well known in the literature. However, Design of RBFNN using traditional hit and trial is time-consuming and suboptimal in nature. Hence, this work proposes the optimal design of RBFNN equalizers using Genetic Algorithms (GA). Also, methods used in literature deals equalization problem as an optimization problem. However, this work deals the same as a classification problem. Simulation results prove the better performance of proposed equalizer.

Classification of EMG Signals Using ANFIS for the Detection of Neuromuscular Disorders

Electromyography is used as a diagnostic tool for detecting different neuromuscular diseases and it is also a research tool for studying kinesiology which is the study of human- and animal-body muscular movements. Electromyography techniques can be employed with the diagnosis of muscular nerve compression and expansion abnormalities and other problems of muscles and nervous systems. An electromyogram (EMG) signal detects the electrical potential activities generated by muscle cells. These cells are activated by electrochemical signals and neurological signals. It is so difficult for the neurophysiologist to distinguish the individual waveforms generated from the muscle. Thus, the classification and feature extraction of the EMG signal becomes highly necessary. The principle of independent component analysis (ICA), fast Fourier transform (FFT) and other methods is used as dimensionality reduction methods of different critical signals extracted from human body. These different existing techniques for analysis of EMG signals have several limitations such as lower recognition rate waveforms, sensitive to continuous training and poor accuracy. In this chapter, the EMG signals are trained using soft computing techniques like adaptive neuro-fuzzy inference system (ANFIS). ANFIS is the hybrid network where fuzzy logic principle is used in neural network. This proposed technique has different advantages for better training of the EMG signals using ANFIS network with a higher reliability and better accuracy. Discrete wavelet transformation (DWT) method is used for feature extraction of the signal.

An ideal ANFIS-based approach for classification of biomedical signals employing wavelet sub-band energy-based characteristics feature

There is a critical linkage of the error detection and classification of various biomedical signals with the diagnosis of different abnormalities. In this paper, basing on the characteristic features of wavelet sub-band energy coefficient, an ideal approach has been implemented for ECG and EEG classification. The ECG and EEG signals are pre-processed using adaptive filter and further are decomposed into time-frequency representation by the use of wavelet transformation. These extracted wavelet coefficients are helpful in calculating certain statistical parameters. For the differentiation between normal and abnormal beats, the types of EEG and ECG beats are taken into consideration. For this classification, Adaptive Neuro-Fuzzy Inference System (ANFIS) is used. The real time signals are obtained from medical and diagnostic centers for the analysis purposes.

An efficient hybrid estimator for nonlinear channels in state space

This paper presents a method for estimation of nonlinear communication channels using Gaussian wavelet estimator (GWE). Results obtained using this approach is compared with image enhanced interacting multiple model (IMM-IE) extracted from the literature. Comparison reveals that the IMM-IE and the GWE require the same number of Kalman filters, but the GWE extrapolates more initial conditions. If the performance of the two algorithms is equivalent, the IMM-IE is less complex, but GWE is technically efficient. Here, we have presented the state-space approach instead of the transfer function approach, as the state-space approach can be easily extended to nonlinear mixing systems. Moreover, the state-space approach not only gives an efficient internal description of the dynamic systems, but there also exist different possible equivalent state-space realizations.