International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 08 Issue: 07 | July 2021 www.irjet.net p-ISSN: 2395-0072 © 2021, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 2057 Fitting Sum Of Exponentials Model Using Differential Linear Regression Molise Mokhomo 1 , Naleli Jubert Matjelo 2 1 Mukuru, Cape Town, Western Cape, South Africa. 2 National University of Lesotho, Department of Physics and Electronics, P. O. Roma 180, Lesotho ---------------------------------------------------------------------***--------------------------------------------------------------------- Abstract In this paper, we present the problem of fitting a sum of exponentials (growth or decay) to some data using linear regression. We start by transforming the problem from the commonly adopted nonlinear regression formulation and present it as a linear regression problem using differential formulation. This allows the problem to be solved in closed form. We demonstrate here by fitting a dataset to a two-component sum of exponentials and show its performance in both noisy and noise-free datasets. The method is shown to be sensitive to noise and this due to the noise amplification inherent in any differential formulation. Key Words: Exponential Growth, Exponential Decay, Regression, Noise Amplification, Model Fitting, Parameter Estimation. 1. INTRODUCTION Fitting models to data comes frequently in data-based research. Model fitting makes extensive use of parameter estimation methods to obtain the optimal model parameters associated with the given data set. Some of the disciplines making use of parameter estimation include system identification, characterization, behavioral analysis, model-based control, state estimation, forecasting, smoothing, and filtering [1], [2], [3]. One of the models used to estimate almost many smooth continuous functions is the sum of the exponentials model (SEM), composed of the weighted average of exponentials [4]. Fitting SEM to data is often considered as a nonlinear regression problem from which there exists no closed- form solution to the problem hence the need for iterative optimization algorithms for solving this problem [4], [5]. In this paper, we adopt a differential formulation of the SEM which allows the parameter estimation problem to posed as a linear regression problem, solvable in closed- form. The approach is based on exploiting the differential equation satisfied by the SEM when formulating the regression problem. In this work we consider the two- component SEM to demonstrate the differential-based linear regression formulation of the SEM fitting problem. This method is related to the method based on integral equations in [6]. The rest of this paper is organized as follows. Section 2 presents a two-component SEM and the resulting differential linear regression problem formulation. Section 3 presents model-fitting simulation results and discussion. Section 4 concludes this work with a summary of major findings and some remarks. 2. LINEAR REGRESSION MODEL FOR SEM 2.1 Sum of Exponentials Model A general one-dimensional -component SEM  can be represented as follows,   ∑  (1) with as the  component weight, and being related to the  component decay or growth rate. We restrict ourselves to  however, the same concept applies to higher values of . With  we have the following model with four unknown parameters to be determined from the data,     (2) 2.2 Sum of Exponentials Model In Differential Form Differentiating equation (2) twice and relating it with its first two derivatives we obtain the following homogeneous ordinary differential equation with constant coefficients,     (3) with parameters and given by,    (4)  (5) 2.3 Differential Linear Regression Model Given a data  of size (i.e.        ) to fit an SEM, we can formulate the linear regression problem using equations (3-5) as shown in the cost function below,  ∑       (6) The resulting solution to this linear regression problem is given by the following matrix equation,