Machine Learning-Driven Blind Carrier Offset Estimation for OFDM Systems

Machine Learning-Driven Blind Carrier Offset Estimation for OFDM Systems

As wireless networks continue to evolve with the emergence of sixth-generation networks and beyond, OFDM technology is expected to be an essential system component for providing reliable connectivity in an increasingly complex environment. Its high spectrum efficiency, efficiency against impulse noise, and resilience against frequency-selective multipath fading channels make it an invaluable component. Although OFDM technology has various capabilities and uses and is widely used, it has some issues and challenges, and one of these challenges is Carrier Frequency Offset (CFO). The CFO destroys orthogonality between subcarriers and causes subcarrier misalignment at the receiver, resulting in inter-carrier interference (ICI) in OFDM systems. This interference impairs system performance by introducing errors and degrading signal-to-noise ratios. In this paper, a new Deep Neural Network (DNN) algorithm has been proposed for estimating the CFO in OFDM systems. The architecture is precisely designed for regression tasks, with a focus on accurately estimating the continuous parameter based on the input features. The multiple hidden layers facilitate the extraction of intricate patterns, while careful training options ensure efficient optimization and monitoring of the training process. The performance of the proposed DNN algorithm is assessed and compared with other subspace-based methods, ESPRIT, PM(MUSIC), and QR. Remarkably, the results demonstrate that the DNN algorithm consistently outperforms all subspace methods, especially in low SNR scenarios.