Classification of Radio Signals on a neuromorphic processor in Space

In this project, we will build a neural network on a spiking neuromorphic device ROLLS [1] with the aim to learn to identify and categorise different power levels of the radio signal corresponding to specific signal modulation types. We will use the open source RadioML [2] dataset to generate spike trains by amplitude thresholding as input to the neuromorphic device. The ROLLS chip [1] is a mixed signal analog/digital device that features on-chip plasticity circuits, which were used previously to learn simple frequency patterns [5].

This project will be part of our collaboration with NASA Glenn Research Centre. The real-time power predictor based on different channel models [4], developed in the project, will augment NASA’s multi-objective Reinforcement Learning Neural Network (RLNN) for the use as a radio-resource-allocation-management controller [3]. Classifying radio signals enables software defined radios (SDRs) to dynamically allocate resources, i.e. power and spectrum, and to adaptively code and modulate them. When impairment events are detected, mitigation techniques can be deployed to minimise the impact on the quality of service [2].

[1] Qiao, Ning, et al. "A reconfigurable on-line learning spiking neuromorphic processor comprising 256 neurons and 128K synapses." Frontiers in neuroscience 9 (2015).

[2] https://github.com/radioML/dataset

[3] Wong, Lauren J., William C. Headley, and Alan J. Michaels. "Estimation of transmitter I/Q imbalance using convolutional neural networks." Computing and Communication Workshop and Conference (CCWC), 2018 IEEE 8th Annual. IEEE, 2018.

[3] Hackett, Timothy M., et al. "Implementation of a space communications cognitive engine." Cognitive Communications for Aerospace Applications Workshop (CCAA), 2017. IEEE, 2017.

[4] Ferreira, Paulo Victor Rodrigues, Randy Paffenroth, and Alexander M. Wyglinski. "Interactive multiple model filter for land-mobile satellite communications at Ka-band." IEEE Access 5 (2017): 15414-15427.

[5] Kreiser, R., Moraitis, T., Sandamirskaya, Y., & Indiveri, G. (2017, October). On-chip unsupervised learning in winner-take-all networks of spiking neurons. In Biomedical Circuits and Systems Conference (BioCAS), 2017 IEEE (pp. 1-4). IEEE.

Requirements

Programming skills in C++ and Python,
Interest in AI, electrical circuits, neuronal systems, and machine learning

Contact

rakrei(at)ini.uzh.ch

© 2018 Institut für Neuroinformatik