۱- Introduction

۲- Related works

۳- Contribution and structure of this work

۴- Theoretical modeling

۵- Deep learning based resource allocation method

۶- Simulation and performance analysis

۷- Conclusion

References

Abstract

Wireless personal communication has become popular with the rapid development of 5G communication systems. Critical demands on transmission speed and QoS make it difficult to upgrade current wireless personal communication systems. In this paper, we develop a novel resource allocation method using deep learning to squeeze the benefits of resource utilization. By generating the convolutional neural network using channel information, resource allocation is to be optimized. The deep learning method could help make full use of the small scale channel information instead of traditional resource optimization, especially when the channel environment is changing fast. Simulation results indicate the fact that the performance of our proposed method is close to MMSE method and better than ZF method, and the time consumption of computation is smaller than traditional method.

Introduction

In 5G wireless communication systems, how to make full use of precious bandwidth, power and antenna resource has become a critical topic in recent studies. The official standardization organization 3GPP has recently published the official released standard [1] and add some key features to improve system throughput and reduce the latency. The target of 5G is to spread the bandwidth and make flexible use of system resources to achieve better performance, resources in time, spectrum and spatial domain are jointly combined and optimized. Traditional studies on 5G are mainly about the proof of mathematical bound [2, 3], and then provide heuristic methods to approximate the proved bound. However, there is hardly effective operations to reach the bound considering only existing coding, modulation, antenna selection, etc. To solve this problem, researchers tried to introduce the famous Artificial Intelligence (AI) technology. The growing discussions about deep learning in AI has brought opportunities to improve system performance in 5G related works. Relying on the study process of deep learning, benefits of resource allocation could be obtained using the dedicated neural network. So there are still quite space to approximate the theoretical bound using the resource allocation.