Meta-learning applications for machine-type wireless communications
Issa, Hebatalla (2023-07-04)
Issa, Hebatalla
H. Issa
04.07.2023
© 2023 Hebatalla Issa. Tämä Kohde on tekijänoikeuden ja/tai lähioikeuksien suojaama. Voit käyttää Kohdetta käyttöösi sovellettavan tekijänoikeutta ja lähioikeuksia koskevan lainsäädännön sallimilla tavoilla. Muunlaista käyttöä varten tarvitset oikeudenhaltijoiden luvan.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:oulu-202307042834
https://urn.fi/URN:NBN:fi:oulu-202307042834
Tiivistelmä
Machine Type Communication (MTC) emerged as a key enabling technology for 5G wireless networks and beyond towards the 6G networks. MTC provides two service modes. Massive MTC (mMTC) provides connectivity to a huge number of users. Ultra-Reliable Low Latency Communication (URLLC) achieves stringent reliability and latency requirements to enable industrial and interactive applications. Recently, data-driven learning-based approaches have been proposed to optimize the operation of various MTC applications and allow for obtaining the desired strict performance metrics. In our work, we propose implementing meta-learning alongside other deep-learning models in MTC applications. First, we analyze the model-agnostic meta-learning algorithm (MAML) and its convergence for regression and reinforcement learning (RL) problems. Then, we discuss uncrewed aerial vehicles (UAVs) trajectory planning as a case study in mMTC and RL, illustrating the system model and the main challenges. Hence, we propose the MAML-RL formulation to solve the UAV path learning problem. Moreover, we address the MAML-based few-pilot demodulation problem in massive IoT deployments. Finally, we extend the problem to include the interference cancellation with Non-Orthogonal Multiple Access (NOMA) as a paradigm shift towards non-orthogonal communication thanks to its potential to scale well in massive deployments. We propose a novel, data-driven, meta-learning-aided NOMA uplink model that minimizes the channel estimation overhead and does not require perfect channel knowledge. Unlike conventional deep learning successive interference cancellation (SICNet), Meta-Learning aided SIC (meta-SICNet) can share experiences across different devices, facilitating learning for new incoming devices while reducing training over- head. Our results show the superiority of MAML performance in addressing many problems compared to other deep learning schemes. The simulations also prove that MAML can successfully solve the few-pilot demodulation problem and achieve better performance in terms of symbol error rates (SERs) and convergence latency. Moreover, the analysis confirms that the proposed meta-SICNet outperforms classical SIC and conventional SICNet as it can achieve a lower SER with fewer pilots.
Kokoelmat
- Avoin saatavuus [34589]