Modeling of In-House CENELEC A-Band PLC Channel using Fritchman Model and Baum-Welch Algorithm Ayokunle Damilola Familua and Ling Cheng School of Electrical and Information Engineering, University of the Witwatersrand, Private Bag 3, Wits. 2050, Johannesburg, South Africa. Email: ling.cheng@wits.ac.za Abstract—One of the many challenges which power line communication (PLC) has to overcome is noise and disturbances. Amidst the four frequency bands classified by the European CENELEC standard, the A-band is the most noise susceptible characterized by noise sources consisting of various electrical appliances. This paper aims at modeling the noise and distur- bances present on the in-house CENELEC A-band based on experimental measurement. The outputs are the channel models of the three major types of noise present on the PLC. These models are achieved through the use of Fritchman model, to depict the power line channel and evaluate the noise impairment caused by the different types of noise. The Baum-Welch algorithm is implemented for the Fritchman model parameter estimation through likelihood evaluation by computing the probabilities of three different noise observation sequences obtained from the experimental measurement. These channel models can then be used to evaluate and optimize coding and modulation schemes for PLC. Index Terms—Baum-Welch Algorithm, CENELEC, Fritchman model, noise measurement, power line communications I. Introduction Power line communications are classified as broadband for broadband applications with wide bandwidth ranging from 1- 30 MHz and narrowband with low frequency bandwidth rang- ing from 3-148 kHz [1], for narrowband applications in Europe under the CENELEC standardization and below 450kHz in Japan [2]. Despite the fact that broadband over power line (BPL) has attracted extensive interest and research over the last decade due to its use for high speed internet access and ‘last mile’ access services to all homes and offices, narrowband PLC is still highly relevant. Narrowband PLC utilizes the existing power line networks that were originally meant for electrical power transmission and distribution by providing a communication media and a means of interconnecting and controlling home appliances through the power outlets already and universally present in every rooms of the homes. Its relevance can also be found in various other narrowband applications namely: its historic use for control of power grid through meter reading, street lighting control, control of ground-lights of airport runways [3], home automation, and its application in communications of control data with 40 kb/s in street car/subway systems on 750V direct current networks [4]. In spite of the so many advantages and applications narrow- band PLC offers, it must overcome some technical challenges posed by the properties of the power line network itself and some other factors that makes it a harsh environment for signal transmission. One of this many challenges is noise and disturbances on the power line. CENELEC A-band with 3- 95 kHz bandwidth is plagued with a lot of noise and distur- bances emanating from various electrical appliances connected across the network or from noise sources outside the network coupled unto the network. The use of these appliances is not coordinated as users connect or switch them on and off at will and the noise introduced causes burst error and loss of data at the receiving end. The characteristics of this noise and disturbances change with time of day, place, source and frequency. Hence, the need for measurement and modeling of the channel remains highly essential. In this paper, we show noise measurement and modeling of the three major categories of noise present on the CENELEC A narrowband with this paper aimed at answering the question of how to formulize and model the noise and disturbances present on the in-house CENELEC A-band based on experimental measurement. The outputs are experimental channel models and not simulation based channel models. This aim is achieved by carrying out experimental measurement of the three major classes of noise present on the PLC namely: background noise, narrow band interference and impulse noise. The output se- quences (error sequences or observations) for these three major types of noise are generated separately from the measurement data, analyzed and presented using a three-state partitioned hidden Markov model (HMM), namely Fritchman model [5]. The motivation behind the use of a Fritchman model is because it is the most widely used model to depict the power line communication channel due to the severe impairments. An HMM training algorithm, Baum-Welch algorithm [6], is then employed for likelihood evaluation and parameter estimation of the given Fritchman model. Hence, these channel models can then be used to select and design appropriate error correction codes and/or modulation techniques to mitigate the effect of noise to increases accuracy and improves the performance of the overall power line communication. It can also help in testing, analyzing and evaluating these coding and modulation techniques. Other related works that has been 2013 IEEE 17th International Symposium on Power Line Communications and Its Applications 978-1-4673-6016-6/13/$31.00©2013 IEEE 173