An Introduction to Deep Learning for the Physical Layer

By Tim O'Shea, Jakob Hoydis. 2017

2,620 citations recorded by the source corpus.

We present and discuss several novel applications of deep learning for the physical layer. By interpreting a communications system as an autoencoder, we develop a fundamental new way to think about communications system design as an end-to-end reconstruction task that seeks to jointly optimize transmitter and receiver components in a single process. We show how this idea can be extended to networks of multiple transmitters and receivers and present the concept of radio transformer networks as a means to incorporate expert domain knowledge in the machine learning model. Lastly, we demonstrate the application of convolutional neural networks on raw IQ samples for modulation classification which achieves competitive accuracy with respect to traditional schemes relying on expert features. This paper is concluded with a discussion of open challenges and areas for future investigation.

Subfields: Aerospace Engineering, Artificial Intelligence, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Aerospace Engineering

In Artificial Intelligence

In Signal Processing

Citations in the corpus

Deep Learning for Distributed Optimization: Applications to Wireless Resource Management (2019, neutral): “To overcome the drawbacks of traditional optimization techniques, deep learning (DL) frameworks have been recently investigated in wireless resource management [7]–[12] and end-to-end communication system design [13]–[17].”
Meta-Learning in Neural Networks: A Survey (2020, background): “For example by learning coding systems that exceed the best hand designed codes for realistic channels [280].”
White Paper on Broadband Connectivity in 6G (2020, methodology): “End-to-end optimization of a communication system using ML is introduced in [115], where the system is interpreted as an autoencoder that jointly optimizes the transmitter and receiver components in a single process to minimize the end-to-end reconstruction error.”
Automatic Modulation Classification: A Deep Learning Enabled Approach (2018, background): “recognition is proposed in [28], [29], and the simulations with a dataset called RML2016.”
Deep Learning in Mobile and Wireless Networking: A Survey (2018, methodology): “[349] Modulation recognition Radio transformer network O’Shea and Hoydis [350] Modulation classification CNN Gonzalez and Vallina [351] Network data monetisation Unknown Kaminski et al.”
Machine Learning in the Air (2019, methodology): “An autoencoder is trained for coding and modulation over an additive white Gaussian noise channel in [40], and it is shown to have a performance very close to conventional coding and modulation scheme in short blocklengths.”
Learning to Rest: A Q-Learning Approach to Flying Base Station Trajectory Design with Landing Spots (2018, methodology): “Note that the DNNs we envision for message sharing maybe compared with DNNs used in autoencoder fashion in [8].”
Modulation Classification Based on Signal Constellation Diagrams and Deep Learning (2019, neutral): “In [7], [9], and [12], adaptation of DL to the complex temporal signal domain is studied.”
Team Deep Neural Networks for Interference Channels (2018, positive): “In [21], [22], the use of deep learning to design the physical layer is discussed and it is in particular shown how the TX, the channel, and the RX can been seen as a single DNN and trained as an autoencoder, which is a model combining encoding and decoding that are learned jointly.”
Learning to Cooperate in Decentralized Wireless Networks (2018, methodology): “Note that the DNNs we envision for message sharing maybe compared with DNNs used in autoencoder fashion in [8].”
Machine Learning for 5G/B5G Mobile and Wireless Communications: Potential, Limitations, and Future Directions (2019, background): “The optimization of a large non-linear NN with many degrees of freedom for a simple high-level objective has been attempted in [61].”
Model-Driven Deep Learning for MIMO Detection (2020, background): “Recently, it has been applied in physical layer communications [11]–[13], such as channel estimation [14]–[16], CSI feedback [17], signal detection [18]–[26], channel coding [27], [28], and end-to-end transceiver design [29], [30].”
End-to-End Deep Learning of Optical Fiber Communications (2018, methodology): “We implement the complete fiber-optic communication system and transmission chain including transmitter, receiver and channel as a complete end-to-end ANN, as suggested in [18], [19].”
Deep Learning for Intelligent Wireless Networks: A Comprehensive Survey (2018, background): “O’Shea and Hoydis [51] proposed a novel idea which treats the physical layer as an end-to-end autoencoder.”
Deep Learning-Based Channel Estimation (2018, background): “…systems, some approaches have been proposed to enhance the performance of different conventional algorithms including modulation recognition [7], channel encoding and decoding [8], signal detection [9], channel equalization [10], channel state information (CSI) feedback [11] and channel…”
Application of Machine Learning in Wireless Networks: Key Techniques and Open Issues (2018, background): “In [29] and [30], the applications of deep learning in the physical layer are summarized.”
Deep Learning Framework for Wireless Systems: Applications to Optical Wireless Communications (2018, methodology): “, M} is first mapped to a one-hot vector [6], which is a zero vector except for the bth element equal to 1, and then is processed by several fully connected layers.”