Truncation-Based Updating Approach Enhances Convergence of LTAT in Non-Orthogonal Multiple Access

In this article, we explore the design of power allocation schemes for short packet communications under stringent block error rate (BLER) constraints. Short packet communications are crucial in 6G wireless communication systems as they enable real-time mobile communication services/applications. However, classical Shannon information theory no longer applies due to finite blocklengths. To address this challenge, we propose a deep reinforcement learning (DRL)-based method that significantly outperforms existing power allocation schemes.
A. Background
The timeliness of information transmission is becoming more important in today’s society. Short packet communications are emerging as a key technology in 6G wireless communication systems to enable real-time mobile services/applications, such as sensory interconnection and autonomous driving. However, classical Shannon information theory does not apply to short packet communications due to finite blocklengths.
B. Maximization of LTAT
We analyze the impact of the length of sub-codeword on the average BLER in Fig. 6. The simulation and analytical results match perfectly, revealing that the gap between the asymptotic and simulation results decreases as either γ increases or L increases.
C. Convergence of DDPG-Based Algorithm
We investigate the convergence of the DRL-based algorithm in Fig. 8, which shows the convergence curve of BLER versus the number of training epochs. The truncation-based updating approach outperforms the non-truncation approach in terms of convergence and stability. Moreover, the average BLER value is lower than the bound ¯P = 0.01.
D. Comparison with Existing Methods
We compare the performance of the DRL-based method with existing power allocation schemes in Fig. 9, which plots the convergence curve of BLER versus the number of training epochs. The DRL-based method significantly outperforms the GP-based method, especially under low SNR conditions. However, both methods exhibit comparable performance under high SNR conditions.
E. Conclusion
In conclusion, this article proposes a DRL-based power allocation scheme for short packet communications under stringent BLER constraints. The proposed method significantly outperforms existing methods, especially under low SNR conditions. The truncation-based updating approach improves the convergence and stability of the algorithm. This work paves the way for further research in developing practical and efficient power allocation schemes for short packet communications in 6G wireless communication systems.