DCGAN uses convolutional and convolutional-transpose layers in the generator and discriminator, respectively. It was proposed by Radford et. al. in the paper Unsupervised Representation Learning With Deep Convolutional Generative Adversarial Networks.