Yepuganti Karuna

Design of data dependent filter bank using sequential matrix diagonalization algorithm

Subband coding (SBC) is a one of the popular application to filter banks to achieve multi-channel data compression. There are data independent and data dependent filter banks in the literature. In this paper, we focus on the design of data dependent M-channel maximally decimated Paraunitary (PU) filter bank using polynomial eigen value decomposition (PEVD) technique for SBC. To design this type of filter bank, we adopted a family of iterative algorithms i) Sequential matrix diagonalization (SMD) ii) Maximum element sequential matrix diagonalization (ME-SMD). The potential benefits of designed data dependent PU filter bank are strong decorrelation and spectral majorization. The SMD algorithm maximizes the coding gain at every step to minimize the subband quantization noise and to achieve the optimality of Filter bank. The performance of the proposed algorithms compared with Sequential best rotation (SBR2) and Modified sequential best rotation (SBR2C) with respect to convergence, number of iterations and coding gain. The performance of designed data dependent filter bank based on proposed algorithms are also compared with Karhunen-Loeve (KL) transform coder as it is one of data dependent optimal data compression scheme and traditional data independent PU filter banks.

Design of matrix wiener filter for noise reduction and speech enhancement in hearing aids

This paper proposes an algorithm for matrix wiener filter for speech processing in case of hearing aids. Hearing aid (HA) devices currently used are less efficient as environmental noise already added into speech signal corrupts it and heard by listener. Introducing wiener filter in HA device allows reducing noise to particular level. By increasing number of secondary channels can enhance the performance of the device. Matrix wiener filter is a technique which considers matrix combination of speech and noise correlations and has better performance than multichannel wiener filter. A multi input multi output (MIMO) system having multiple target speech signals is used, it utilizes matrix wiener filter for nullifying noise in speech which estimates the speech coming towards listener. Wiener matrix coefficients are calculated from speech and noise correlation matrices. Simulation results prove that matrix wiener filter performs better compared to multichannel and single channel wiener filter.

Real and complex Eigen functions: An analysis of the 2-D Fourier transform

In this paper, we extend the concept of Eigen functions of the Fourier transform discussed by P. P. Vaidyanathan [9]. But for a scaling factor, if there is no change in the function after the application of the Fourier transform, we term them as Eigen functions of the Fourier transform. Gaussian is the well known Eigen function. But, there are various other functions that can be developed systematically. The theory developed by P. P. Vaidyanathan for 1-D signals is extended for 2-D case in this paper. We also presented extensions for the complex signal case.

IIR deconvolution from noisy observations using Kalman filtering

In this paper, we reconstruct the input signal of an IIR filter from the noise corrupted output signal. We perform two operations parallely. One deconvolution and the other, noise removal. We show how to use Kalman filter to perform this task. We develop theory for a very general scenario of reconstructing an ARMA process from its noise corrupted IIR filtered output. We develop augmented state space equations combining the state space equations of the ARMA process and the IIR filter, which are required to apply Kalman filter. The simulation results show clear improvement in the signal-to-noise ratio.

Broadband subspace decomposition of convoluted speech data using polynomial EVD algorithms

The Polynomial EVD (PEVD) was developed to achieve broadband subspace decomposition as a part of two-stage convolutive Blind Source Separation (BSS) algorithm. It has the ability to accomplish strong (total) decorrelation and spectral majorization on convolutive signals. We explore different algorithms for constructing FIR paraunitary (PU) matrices with the aim of performing broadband subspace decomposition. We adopt a set of new iterative PEVD algorithms for this task: a) Sequential matrix diagonalization (SMD) b) Maximum element sequential matrix diagonalization (ME-SMD). We also present a procedure to find out the total number of source signals in the convolved data, without having prior knowledge, based on the energy of individual polynomial eigen values. This helps us to find out exact signal and noise subspaces. To measure the performance of PEVD for broadband subspace decomposition, we use the diagonalization performance measure and subspace estimation measure. We present the results both for simulated data and for actual convolved speech data in the presence of noisy environment to show the effectiveness of the adopted algorithms over existing SBR2/SBR2C algorithms in the literature.

Design of multichannel wiener filter for speech enhancement in hearing aids and noise reduction technique

This paper proposes an algorithm for design of multichannel wiener filter for the application of hearing aids (HA). Present hearing aid devices amplify the speech signal which is corrupted by disturbances and noise from the same environment, resulting degraded speech quality and less efficiency of such devices. Application of wiener algorithm in HA device provides improved speech quality with less complexity. Noticeably enhanced speech quality can be obtained if multiple wiener filter channels are used. Multichannel wiener filter algorithm for speech considers scalar combination of noise inputs to filter and speech correlations. A single target speech system having multiple noise inputs to the filter is designed for estimation of degrading signal or noise. It allows extraction of pure speech by nullifying the estimated noise from corrupted speech of the pilot channel. Enhanced speech signal can be observed at the output which is available for listener who uses HA device. Filter coefficients are extracted from input noise and corrupted speech correlation matrices.

Design of maximally decimated linear phase orthogonal filter bank using iterative SVD technique and its applications

Filter Banks traditionally designed by the factorization method which is complex and time consuming. Singular valued decomposition technique is introduced to design Linear Phase Orthogonal filter bank which yields efficient and fast results in time critical applications. Signal splitting and efficient reconstruction of the original signal is studied achieving perfect reconstruction condition due to the orthogonal nature of designed filter bank. Ultimate aim of proposed methodology is to design two approximation scenarios in terms of square matrices which apparently maps into the filter bank matrix. Singular valued decomposition is iteratively employed to compute both of these square matrices until achievable convergence is achieved between to alternate iterations. Proposed algorithm starts with the design of least square filter bank and ends with the filter bank which ensures orthogonality conditions as well as all filters in the filter bank satisfying linear phase condition. Thus, faster design of the filter bank along with linear phase orthogonal condition yields multiple applications in signal processing, image processing, communication etc.

Design of multicarrier pulses maximizing SINR by ISI/ICI shaping for the Doubly Dispersive channel

Multicarrier modulation (MCM) is a popularly used transmission technique. This paper aims at designing the multicarrier modulation pulse for communication over a Doubly Dispersive (DD) channel. For a DD channel, Inter Symbol Interference (ISI)\Inter Carrier Interference (ICI) can be completely eliminated only when we design the pulses with the knowledge of the channel state. But for a quickly varying doubly dispersive channel, the channel state cannot be tracked by the transmitter. So instead of designing a pulse which suppresses all the ISI\ICI, the proposed technique designs the MCM pulses which allows a tolerable ISI\ICI within the target pattern and eliminates all the ISI\ICI outside the target pattern. By properly selecting the target pattern, it is possible to treat the allowed ISI\ICI by using simple equalization/decoding techniques.