The principles of automatic modulation classification (AMC) are used in the different applications such as cognitive radio (CR), adaptive communication, electronic reconnaissance, and noncooperative communication. Of all these applications, it is a great challenge to determine the type of modulation of a received unknown signal for which no information is available regarding the signals’ parameters. This research mainly focuses on the AMC problem, particularly on enhancing the efficiency of radio frequency spectrum sensing using a deep convolutional neural network (DCNN). The proposed DCNN model is structured to efficiently classify different modulation types. Initially, the signals are separated into 16 sub-band signals, and then these sub-band signals are fed to the DCNN model. The DCNN architecture consists of multiple layers with the skip connections that helps in mitigating the vanishing gradient problem, and asymmetric kernels reduce computational complexity. The model is trained using the DeepSig:RadioML 2018. 01A dataset and optimized using stochastic gradient descent (SGD) with momentum. The results demonstrate the model's robustness and efficiency in classifying modulation types under signal-to-noise ratio (SNR) of 10 dB. The simulation results show that the proposed DCNN-AMC achieved better classification accuracy with 98.97% at 10 dB of SNR.