Abstract

With the speeding up of the fifth generation (5G) new radio (NR) worldwide commercialization, one of the paramount questions for operators and vendors is how to optimize the radio links, considering the widely diverse scenarios envisioned. One of the key pillars of 5G has been an unprecedented flexibility on the configuration of the radio access network (RAN) on scenarios that include cellular, vehicular, and industrial networks among others. This flexibility has its main exponent on link adaptation (LA), which has evolved into a multi-domain technique where a plethora of parameters, like numerology, bandwidth part, radio frequency beam, power, modulation and coding scheme (MCS) or multiple antenna precoding can be adapted to the instantaneous link conditions. Although such enhancements open the door to a significant performance improvement, they also pose many challenges to LA optimization. In this article, we first present the signaling aspects of NR technology for multi-domain LA and the challenges that need to be faced. Then, we explore the latest advances on LA for wireless networks. We envision a combination of machine learning (ML) tools with multi-domain LA as a key enabler for 5G and beyond networks. Finally, we investigate emerging ML approaches for LA and present a promising application of ML for LA that is assessed with simulations. With this scheme, the training is performed at the network side to relieve the user equipment (UE) to do such a complex task. It is shown with simulations that our ML approach outperforms the well known outer loop link adaptation (OLLA) algorithm in terms of instantaneous block error rate (BLER), while reaching the same average spectral efficiency (SE). Interestingly, it is shown that the proposed scheme only requires 4 bits to represent the features used to train the model, which makes it suitable for implementation in real systems with limited feedback.

Highlights

  • The main driving factor behind the fifth generation (5G) new radio (NR) is to provide a single wireless technology for a fully connected society

  • This flexibility has its main exponent on link adaptation (LA), which has evolved into a multi-domain technique where a plethora of parameters, like numerology, bandwidth part, radio frequency beam, power, modulation and coding scheme (MCS) or multiple antenna precoding can be adapted to the instantaneous link conditions

  • We have focused on AMC to show the potential of the application of machine learning (ML) techniques to LA, since the MCS is one of the domains with higher impact on the block error rate (BLER) and throughput, which are key metric in wireless communications

Read more

Summary

INTRODUCTION

The main driving factor behind the fifth generation (5G) new radio (NR) is to provide a single wireless technology for a fully connected society. A flexible numerology design was needed to combat the impact of impairments like phase noise or IQ imbalance and the non-linearity of high power amplifiers (HPAs), since their impact is more noticeable at higher frequency bands [1] It is especially relevant the case of uplink (UL) where two types of waveforms, cyclic prefix orthogonal frequency division multiplexing (CP-OFDM) or transform precoding, can be selected depending on the link conditions. Besides the traditional triplet of MCS, power control, and digital MIMO precoding of previous standards, 5G NR allows selecting the numerology, BWP, waveform, and RF beam to maximize the efficiency of radio links. These new degrees of freedom increase the complexity on LA since more parameter sets must be jointly optimized.

SIGNALING ASPECTS OF LA IN 5G NR
RF IMPAIRMENTS
CLASSIFICATION OF ML TECHNIQUES
EMERGING APPLICATIONS OF ML TO LA AND MAIN CHALLENGES
CONCLUSIONS
Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call