Reduced-complexity transform-domain adaptive algorithm with selective coefficient update

This paper proposes a new low-complexity transform-domain (TD) adaptive algorithm for acoustic echo cancellation. The algorithm is based on decomposing the long adaptive filter into smaller subfilters and employing the selective coefficient update (SCU) approach in each subfilter to reduce computational complexity. The resulting algorithm combines the fast converging characteristic of the TD decomposition technique and the benefits of the SCU of low complexity with minimal performance losses. The improvement in convergence speed comes at the expense of a corresponding increase in misadjustment. To overcome this problem, a hybrid of the proposed algorithm and the standard TD LMS algorithm (TDLMS) is presented. The hybrid algorithm retains the fast convergence speed capabilities of the original algorithm while allowing for low final MSE. Simulations show that the hybrid algorithm offers a superior performance when compared to the standard TDLMS algorithm with less computational overhead.     

Leaky LMS algorithm: MSE analysis for Gaussian data

Despite the widespread usage of the leaky LMS algorithm, there has been no detailed study of its performance. This paper presents an analytical treatment of the mean-square error (MSE) performance for the leaky LMS adaptive algorithm for Gaussian input data. The common independence assumption regarding W(n) and X(n) is also used. Exact expressions that completely characterize the second moment of the coefficient vector and algorithm steady-state excess MSE are developed. Rigorous conditions for MSE convergence are also established. Analytical results are compared with simulation and are shown to agree well     

On transient error surfaces of output error IIR adaptive filtering

Addresses the convergence of IIR output error adaptive algorithms. We argue that convergence to global or local optimum is simultaneously determined by the shape of the transient error surface and the convergence speed of the algorithm. This argument is confirmed by simulation examples comparing the convergence of LMS-based and LS-based algorithms     

Combined regressor methods and adaptive IIR filtering

An open issue in adaptive infinite-impulse response (IIR) filtering is that of convergence to a global minimum in the presence of observation noise when the system is insufficiently modeled . It is well known , that algorithms based on equation error (EE) formulation contain a single minimum that may be biased whereas, algorithms based on output error (OE) ensure the existence of an unbiased global minimum in presence of local minima. Recently, there have been several attempts to combine these formulations in order to ensure the existence and uniqueness of an unbiased minimum. Works presented here, EEOE and modified EEOE (MEEOE), are such attempts in the context of system identification. We will show, analytically and through simulations, the convergence properties of the MEEOE approach, in the context of system identification.     

A separable low complexity 2D HMM with application to face recognition

In this paper, we propose a novel low-complexity separable but true 2D Hidden Markov Model (HMM) and its application to the problem of Face Recognition (FR). The proposed model builds on an assumption of conditional independence in the relationship between adjacent blocks. This allows the state transition to be separated into vertical and horizontal state transitions. This separation of state transitions brings the complexity of the hidden layer of the proposed model from the order of (N/sup 3/T) to the order of (2N/sup 2/T), where N is the number of the states in the model and T is the total number of observation blocks in the image. The system performance is studied and the impact of key model parameters, i.e., the number of states and of kernels of the state probability density function, is highlighted. The system is tested on the facial database of AT&T Laboratories Cambridge and the more complex facial database of the Georgia Institute of Technology where recognition rates up to 100 percent and 92.8 percent have been achieved, respectively, with relatively low complexity.     

A robust variable step-size LMS-type algorithm: analysis andsimulations

A number of time-varying step-size algorithms have been proposed to enhance the performance of the conventional LMS algorithm. Experimentation with these algorithms indicates that their performance is highly sensitive to the noise disturbance. This paper presents a robust variable step-size LMS-type algorithm providing fast convergence at early stages of adaptation while ensuring small final misadjustment. The performance of the algorithm is not affected by existing uncorrelated noise disturbances. An approximate analysis of convergence and steady-state performance for zero-mean stationary Gaussian inputs and for nonstationary optimal weight vector is provided. Simulation results comparing the proposed algorithm to current variable step-size algorithms clearly indicate its superior performance for cases of stationary environments. For nonstationary environments, our algorithm performs as well as other variable step-size algorithms in providing performance equivalent to that of the regular LMS algorithm