A deep learning-based digital key distribution method for digital mobile communication networks has been designed. In the key distribution stage, adjacent contacts are connected through broadcast information, and each broadcast message requests the use of identifiers for identity verification. Based on deep learning, a digital key distribution protocol for digital mobile communication networks is established to determine the effectiveness of key distribution based on the quantum bit error rate. Comparative experiments have shown that this method has more uniform key distribution and higher communication security and can be applied in practical life.