Modulación adaptativa para un sistema OFDM, utilizando técnicas de Machine Learning

Abstract

Adaptive Orthogonal frequency division multiplexing (OFDM) is a modulation scheme that dy- namically adjusts the system’s modulation based on real-time channel conditions, optimizing bandwidth usage and maintaining an optimal Bit Error Rate (BER). This integrative curricular work analyzes the BER of a conventional adaptive OFDM system and one enhanced with Machine Learning (ML) techniques. To implement adaptive OFDM with ML, a training database is generated by analyzing performance in terms of BER and Signal to Noise Ratio (SNR) for five modulation schemes [QPS, 8-PSK, 16-QAM, 32-QAM, 64-QAM], establishing the boundaries and parameters. Three models are trained: decision tree, k-nearest neighbors (classification technique), and robust linear regression (regression technique). The performance of each mo-del is compared based on BER and throughput. The results show that ML techniques improve the performance of the adaptive OFDM system, enabling precise modulation scheme selection and efficient adaptation to channel conditions. The adaptive OFDM system with ML can switch modulation, achieving high throughput and a BER lower than 0.001 with an SNR of 17 dB or higher. Among supervised ML techniques, classification models like decision tree and k-nearest neighbors outperform linear regression in prediction accuracy, achieving a lower BER with a lower SNR and high throughput.