Model-aided distributed shallow learning for OFDM receiver in IEEE 802.11 channel model
- PDF / 1,102,158 Bytes
- 10 Pages / 595.276 x 790.866 pts Page_size
- 20 Downloads / 163 Views
(0123456789().,-volV)(0123456789(). ,- volV)
Model-aided distributed shallow learning for OFDM receiver in IEEE 802.11 channel model Messaoud Ahmed Ouameur1
· Anh Duong Tuấn Lê1 · Daniel Massicotte1
© Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Deep learning (DL) has been recognized as an instrumental tool for the design of future communication systems. Since it is still not clear whether a fully data-driven end-to-end communication learning approach would eventually outperform the traditional ones in terms of performance and complexity, it is argued that the optimal design needs to be tackled by taking the benefits of both model-based and data-driven approaches and by leveraging the concept of transfer learning. However, the grand question lies in how this can be implemented efficiently. As such, this paper proposes an efficient end-to-end OFDM based receiver learning approach based on distributed data-driven and model-based approaches. The approach relies mainly on augmenting a typical OFDM receiver’s processing blocks with a shallow neural network as a data-driven stub to improve its performance. Relying on a two-phases training approach, the last receiver’s processing stage benefits from the transfer learning approach to improve its performance. Limiting the scope to a typical OFDM transmission where the DL-based methods fail, the proposed model-aided shallow learning receiver shows performance improvements compared to the baseline structure. Keywords Shallow learning · Deep learning · OFDM receiver · Training · Model-based · Data-driven · Transfer learning
1 Introduction Machine learning (ML) in general and deep learning (DL) in particular are identified as indispensable tools for the design of future communication systems [1, 2]. The authors in [1] have provided a thorough discussion on ML-based approaches for wireless communication networks’ design and operation, whereas, the authors in [2] have envisioned to use ML as an instrumental tool to enable a truly intelligent massive MIMO. Both works agreed on the fact that D. Massicotte holds the Chaire de recherche sur les signaux et l’intelligence des syste`mes haute performance. & Daniel Massicotte [email protected] Messaoud Ahmed Ouameur [email protected] Anh Duong Tua´ˆ n Leˆ [email protected] 1
Laboratoire des Signaux et Syste`mes Inte´gre´s, Department of Electrical and Computer Engineering, Universite´ du Que´bec a` Trois-Rivie`res, 3351, Boul. des Forges, Trois-Rivie`res, QC, Canada
the grand question is not whether ML will be integrated but rather how and when this integration will be implemented. Recent contributions advocate for the potential of using DL to break the bottleneck of communication systems [3–6]. Even if most signal processing algorithms have solid and well-established roots in statistics and information theory for mathematically tractable models, it remains that a practical system has many impairments and non-linearities, which can be roughly captured by such models [7].
Data Loading...
