Antibodies are immune system proteins that protect the host by selectively binding to specific antigens, including viruses and bacteria. The binding between antibodies and antigens is predominantly hinges on the complementarity-determining regions (CDR) of the antibodies. While previous research has employed deep generative model that jointly models sequences and structures of CDRs based on diffusion probabilistic models and equivariant neural networks, these efforts have not explicitly incorporated protein structure. This study introduces a novel approach, leveraging the graph structure of proteins and employing an equivariant Graph Neural Network for Antibody design inside the diffusion probabilistic models. We conduct a series of extensive experiments to assess the efficacy of our approach, evaluating both the sequences and structures of designed antibodies. Our evaluation criteria include binding affinity measured by biophysical energy functions, alongside other pertinent protein design metrics.