Optimum DC bias for DC biased optical OFDM based optical wireless communication using machine learning

Abstract

Light fidelity (LiFi) is a form of optical wireless communication considered supplementary to conventional radio frequency communication. LiFi uses differ- ent orthogonal frequency division multiplexing (OFDM) to encode data, includ- ing DC-biased optical OFDM (DCO-OFDM). In DCO-OFDM, using a signifi- cant DC bias causes optical power inefficiency, while a slight bias leads to higher clipping noise. Hence, finding an appropriate DC bias level for DCO-OFDM is essential. This thesis applies machine learning (ML) algorithms to find optimum DC-bias value for DCO-OFDM based LiFi systems. For this, a dataset is gener- ated for DCO-OFDM using the MATLAB tool. A dataset takes eight attributes by changing the DC-biased value, the constellation size, and the number of sub- carriers. The records in the dataset are randomly placed. Next, ML algorithms are applied using the Python programming language. ML is used to find the essential attributes of DCO-OFDM that influence the optimum DC bias. It is shown here that the optimum DC bias is a function of several factors, including the minimum, the standard deviation, the maximum value of the bipolar OFDM signal, and the constellation size. Next, linear and polynomial regression algorithms are applied to predict the optimum DC bias value. For this, the dataset is divided into additional testing and training samples. The accuracy of the regression models is evaluated by taking root mean square error (RMSE) and the coefficient of determination known as the R2 score. Polynomial regression of different orders is also taken into consideration. Results show that the highest R2 scores for the case of linear and polynomial regressions are 0.8412 and 0.96774, respectively. The result analysis indicates that both linear and polynomial regression can be considered for finding the appropriate DC bias for DCO-OFDM. The results show that at least five features need to be considered to obtain a reliable prediction of the DC bias value. Moreover, the polynomial regression model is more reliable in predicting the DC bias value for a DCO-OFDM based LiFi. Our study shows polynomial regression of order two can predict the optimum DC bias value with a coefficient of determination of 96.77% which confirms the effectiveness of the prediction.