Puncturing-Based Resource Allocation for URLLC and eMBB services via Unsupervised Deep Learning

In this work, we establish a puncturing-based resource allocation framework to enable the coexistence of the Ultra-Reliable and Low-Latency Communications (URLLC) and the enhanced Mobile Broad Band (eMBB) services for the next generation of mobile communications networks. Since the optimal resource allocation depends on channel gains, we aim to find an optimal mapping from the channel gains to the bandwidth and transmit power allocated to both types of services. To solve the problem, we use a deep neural network to represent the mapping and train the parameters using an unsupervised learning method, where the Lagrangian function serves as the loss function. Simulation results show that, compared with the optimal solution obtained from exhaustive search, our proposed method can achieve a learning precision of more than 98.71%. When compared with the existing random puncturing scheme, our scheme can realize an average long-term data rate gain for eMBB of up to 351.1% under URLLC packet size of 512 bits.