Orthogonal frequency division multiplexing (OFDM) wireless systems rely heavily on channel state estimation (CSE) to mitigate the effects of multipath channel fading. Achieving a high data rate with OFDM technology requires efficient CSE and accurate signal detection. In contrast to more traditional CSE methods that depend on a model-based strategy, machine learning (ML)-based CSE techniques have attracted increased interest in recent years due to their data-driven, learning-based flexibility. In light of this, a deep learning (DL) convolutional neural network (CNN) is utilized to acquire reliable CSE over OFDM wireless system Rayleigh-fading channels. The suggested CSE utilizes offline training to gather channel information from transmit/receive pairs. In addition, it employs pilots to provide additional guidance on channels of communication. Compared to conventional estimation approaches, the proposed CNN-based CSE shows considerable improvement in experimental results. Furthermore, the trained CNN model performs better than the state-of-the-art DL channel estimators. The simulation findings also confirm that the suggested CNN-based CSE is effective when there are fewer pilots, with/without cycle prefixes (CP), and this reduces the bandwidth required to convey the same quantity of data. In addition, there is no background knowledge of the channel's statistics in the proposed estimator. Consequently, the proposed method shows potential for addressing CSE issues in OFDM systems with a significant spectrum resource reduction.