CFP19R88-PRJ: 978-1-7281-0173-6 ICECEIC - 2019 ANALYSIS OF THERMAL NOISE IN ANALOG COMMUNICATION SYSTEM USING MACHINE LEARNING Swarupa Das Department of ECE, B. P. Poddar Inst. of Management and Technology Kolkata, India swarupadas.munia@gmail.com Souradip Mondal Department of ECE, B. P. Poddar Inst. of Management and Technology Kolkata, India souradipmondal4@gmail.com Madhumita Sarkar Department of ECE, B. P. Poddar Inst. of Management and Technology Kolkata, India s.madhumita@yahoo.co.in Shovon Nandi Department of ECE, Bengal Institute of Technol- ogy Kolkata, India nandi.shovon@yahoo.in ABSTRACT - This paper examines the thermal noise in channel for analog communication system. Here the pattern of noise pro- duction is unveiled using Exploratory Data Analysis (EDA) and the noise is represented with the co-related parameters. The change in thermal noise with per unit change in the relevant pa- rameters is also predicted in this work using Machine Learning model. Using Recursive Feature Elimination (RFE) method the target variable(s) is/are determined by more significantly predict- ing the target variable and thermal noise. Keywords: Thermal noise, Noise power, Noise voltage, Machine learning, Linear Regression Model, Recursive Feature Elimination I. INTRODUCTION Considering practical environment of signal transmission, signals face distortions and get corrupted due to interference from other signals and various other disturbances generating noise. Thus noise are basically the unwanted signals which tend to disturb signal transmission and processing in an ana- log communication channel. The source of these unpredicta- ble noise signals can be internal or external or both. Noise from external sources includes interference from other channels, man-made noise, lightning, radiation etc. Proper system designing can eliminate or reduce these external noises [1,2]. Fig. 1 shows the Block diagram of a communi- cation system for transmitting analog signal. Internal noise is generated due to spontaneous fluctuations of microscopic motion of electrons due to temperature in an electrical circuit. It is the internal noise that creates a basic limitation in trans- mission of signals or detection of signals in a communication channel. The two most important type of internal noise are shot noise and thermal noise. Internal noise can never be eliminated but can be reduced with proper care. Thermal Noise Thermal noise is introduced in a signal due to the thermal ag- itation of charge carriers in conductors, recombination or ran- dom emission of charged particles [3,4]. Random voltage across the terminals of resistor R is generated due to random across the terminals of resistor R is generated due to random thermal motion of electrons in that resistor. That random volt- age is nothing but the thermal noise [5]. Thermal noise is also called white noise and its power spec- tral density also remains constant over a wide frequency range [6]. It is pervasive in all frequencies. So mean square value of thermal noise voltage [7] which appears across a re- sistor terminal can be defined within a given bandwidth as:  2 = 4 ∫  2 1 Here, T= Temperature in Kelvin scale K=Boltzman constant R= Resistance of the component f1 and f2=upper and lower limits of bandwidth The simplified form of above equation is:  = √4 Where, B= working bandwidth In many applications it is useful to work with thermal noise power. A model, considering a noisy resistor R as an ideal one and a noise voltage source connected in series which is further connected to a load is required to access the thermal noise. Thermal power =  2 4 = (√4 ) 2 4 = KTB In many applications it is useful to work with thermal noise power. A model, considering a noisy resistor R as an ideal one and a noise voltage source connected in series which is further con- nected to a load is required to access the thermal noise. Thermal power =  2 4 = (√4 ) 2 4 = KTB Since in a resistor, number of electrons are large enough, therefore we can assume that the random motion of these electrons are statistically independent to each other [8]. Thus according to central limit theorem it can be said that thermal noise has a Gaussian distribution with mean zero. The rest of this paper is organized as follows. Section II is devoted to the dataset creation and system design. In section III we discuss the data flow diagram, the results and some