SDR-based Testbed for Realtime CQI Prediction for URLLC presented at INFOCOM 2021
A demo for the article “SDR-based Testbed for Realtime CQI Prediction for URLLC” by Evgeny Khorov, Alexey Kureev, and Kirill Glinsky was presented at IEEE INFOCOM 2021, the leading international conference on network technologies. The presenter was Kirill Glinsky, a 5th-year student of the DREC MIPT and a WNL member.
Ultra-reliable low latency communication (URLLC) is one of the key areas of development for 5G networks. URLLC is proposed for use in scenarios such as telemedicine, self-driving vehicles, and other promising applications that are sensitive to traffic quality of service.
One of the key mechanisms of cellular networks is the Modulation and Coding Scheme (MCS) selection mechanism, which chooses the modulation and code rate suitable for the quality of the wireless channel. The problem of MCS selection is especially acute for rapidly changing channels typical of mobility scenarios, for example, when driving in a car. To meet such stringent requirements, such as a data transfer delay of up to 10 ms and transmission reliability of more than 99.999%, it is necessary to develop new methods of MCS selection. One of the promising approaches to solving this problem is the use of machine learning methods, in particular neural networks. However, the question of the possibility of using neural networks in systems with such high latency requirements remained unresolved, due to the high computational cost of the neural networks. Moreover, neural networks require a vast amount of training data in a variety of scenarios. The unavailability of such data made it necessary for authors to collect it themselves. In this article, scientists from Wireless Networks Lab demonstrate a testbed that solves these problems.
The experimental setup is based on the LIMESDR software-defined radio and uses the modified srsLTE framework. Due to the high reconfigurability of the developed device, the stand allows both the collection of channel quality data using base stations deployed by operators and predicting the quality of the channel using an algorithm based on a convolutional-recurrent neural network trained on previously collected data. Thanks to the use of a high-performance CPU in the software-defined radio in combination with parallel computation of the neural network, the stand can operate in real-time.
The authors have published a dataset gathered using the developed prototype for further research of channel prediction algorithms. The dataset consists of the channel quality values obtained in various mobility scenarios with a granularity of 1 ms, as well as IQ samples before preprocessing.