Govind Sharma

A model-based approach for estimation of two-dimensional maximum entropy power spectra

A stochastic model-based approach is presented for estimation of the two-dimensional maximum entropy power spectrum (MEPS) from given finite uniform array data. The method consists of fitting an appropriate two-dimensional noncausal Gaussian-Markov random field (GMRF) model to the given data using the maximum likelihood (ML) technique for parameter estimation. The nonlinear criterion function used for ML estimation is similar in structure to the function arising in the deterministic approach of Lang and McClellan. The model-based approach provides new insights into the two-dimensional MEPS estimation problem. For example, using the asymptotic normality of ML estimates, we derive simultaneous confidence bands for the estimated MEPS. It turns out that when the true correlations are generated by a noncausal GMRF model, the two-dimensional MEPS can be obtained by solving linear equations. This approach also suggests techniques for realizing two-dimensional GMRF models from the given correlation data. Several numerical examples are given to illustrate the usefulness of the approach.

Two-dimensional spectrum estimation using noncausal autoregressive models

Two-dimensional (2-D) spectrum estimation from raw data is of interest in signal and image processing. A parametric technique for spectrum estimation using 2-D noncausal autoregressive (NCAR) models is given. The NCAR models characterize the statistical dependency of the observation at location s on its neighbors in all directions. This modeling assumption reduces the spectrum estimation problem to two subproblems: the choice of appropriate structure of the NCAR model and the estimation of parameters in NCAR models. By assuming that the true structure of the NCAR model is known, we first analyze the existence and uniqueness of Gaussian maximum likelihood (GML) estimates of NCAR model parameters. Due to the noncausal nature of the models, the computation of GML estimates is burdensome. By assuming specific boundary conditions, computationally tractable expressions are obtained for the likelihood function. Expressions for the asymptotic covariance matrix of the GML estimates as well as the simultaneous confidence bands for the estimated spectrum using GML estimates are derived. Finally, the usefulness of the method is illustrated by computer simulation results.

Acoustic echo cancellation using multiple sub-filters

We present the modeling of the acoustic echo path based on the process segmentation approach. A new algorithm is proposed using the concept of decomposing the long adaptive filter into low order multiple sub-filters. Simulation results show that the decomposed LMS adaptive algorithm significantly improves the convergence rate while keeping the steady state error almost the same as that of the original long adaptive filter.

A new family of concurrent algorithms for adaptive Volterra and linear filters

A novel idea for introducing concurrency in least squares (LS) adaptive algorithms by sacrificing optimality has been proposed. The resultant class of algorithms provides schemes to fill the wide gap in the convergence rates of LS and stochastic gradient (SG) algorithms. It will be particularly useful in the real time implementations of large-order linear and Volterra filters for which both the LS and SG algorithms are unsuited.

Multipath delay estimation for acoustic echo channel

A new algorithm for multipath delay estimation based upon autocorrelation estimator (AE) has been proposed. The existing generalized autocorrelation estimator (GAE) has been modified by passing it through a sliding window of suitable shape and size resulting in improvement in performance. The performance improvement obtained because of the window has been determined for the case when the input is a filtered version of a white Gaussian signal. Accuracy percentage (AP) has been plotted against SNR. Simulation results show that the proposed scheme performs better delay estimation even at relatively low SNR

Precoder quantization for interference alignment with limited feedback

Interference Alignment is a promising technique for achieving higher rates by aligning interference at the receiver. To design such a system, the global channel state information at the transmitter (CSIT) as well as at the receiver is necessary. But in practice, it is hard to obtain this information, therefore, limited feedback is used to provide CSIT or the precoder design information to the transmitter. Conventionally, precoders are quantized at receiver by finding its best match in the codebook using chordal distance and its index is fedback to the transmitter. In this paper, instead of minimizing chordal distance, we propose algorithms with objectives that are derived from subspace alignment method, SINR maximization, or minimization of leakage interference power to measure the “goodness” of quantized vector. These algorithms achieve higher rates for small size codebooks. The rate loss has been analyzed for precoder quantization. We also find less computational intensive solution to find the desired vectors in the codebook. The simulation results show that for small codebooks, significant sumrate gains can be achieved for (2 × 2,1)3 for 2-6 bits of feedback per user, compared to quantization based on chordal distance, while for large codebooks, the chordal distance based quantization performs better.

Two-dimensional spectral estimation using spatial autoregressive models

Two-dimensional spectral estimation from raw data is of interest in signal and image processing. In this paper, a parametric technique using non causal spatial autoregressive models for spectral estimation is given. The spatial autoregressive models characterize the statistical dependency of the observation at location s on its neighbors in all directions. Once an appropriate model is fitted, the spectrum is a function of the model parameters. By assuming specific boundary conditions maximum likelihood estimates of model parameters are obtained. The usefulness of the method developed here is illustrated by resolving two closely spaced sinusoids on the plane.

Speech recognition using neural networks

The paper presents a continuous speech recognition system based on a neural network concept. An articulatory-phonetic feature extraction network (APFEN) is used for extracting articulatory-phonetic features. This is followed by a coarticulation network for reducing the effect of coarticulation present in the continuous speech. Algorithms for segmentation and then identification of phonemes are given.

Genetic max-SINR algorithm for interference alignment

In this paper, we propose interference alignment (IA) algorithms inspired by Genetic Algorithm (GA). By simulations for (2 × 2, 1)3 system, we observe that the existing max-SINR (MS) algorithm converges to different sumrates for different initializations of precoders. And the initializations for which sumrate is good, cannot be found trivially using channel state information. Also, in the case of limited feedback (LFB) of precoders, the sumrates can be achieved greater than that can be achieved using conventional chordal distance, if the precoder is selected properly along with receiver combining matrix. Therefore, in this paper, two algorithms are proposed inspired by GA: first, to make the max-SINR robust to initializations: MS-GA, and second, to achieve better sumrates in case of limited feedback: MS-GA-LFB. These optimal sumrates are obtained at the cost of increased computation complexity which is proportional to the population size chosen in the Genetic Algorithm. The simulation results show that the sum rates of the proposed algorithms match with that obtained using brute force approach to find the good initialization.

A quantization method for precoder feedback in interference channel

In this paper, we modeled the precoder feedback for interference channel as a combinatorial optimization problem. Conventionally, in limited feedback, chordal distance has been used to find the best precoder index from the codebook. But, the brute force approach on the problem shows that higher sumrates can be achieved with the same sized codebook, i.e, the same feedback bits. Thus, in this paper, we propose a method to get better sum rates in limited feedback. The simulation results show that the proposed approach perform close to brute force approach and gives better sum rates than the approaches like chordal distance, maximizing SINR, etc.