Hidden Node-Aware Dynamic Spectrum Access using Deep Learning for Coexisting Aeronautical Communication Systems


We propose a novel approach based on deep learning to address the hidden node problem which occurs in the coexistence of aeronautical communication standards. The modern aeronautical communication standard L-band Digital Aeronautical Communications System (LDACS) in Air-Air (A/A) mode needs to share spectrum with the Distance Measuring Equipment (DME), which is a legacy system. As DME is safety-critical, causing interference on it must be avoided for all newly proposed aeronautical systems spectrally coexisting with it. Recently, cognitive radio techniques have been proposed for LDACS A/A to access spectrum dynamically and to overcome the limitations of static approaches. For this, a Recurrent Neural Network (RNN) was trained to predict idle time slots on those frequency bands, where both systems operate. By exploiting patterns in the spectrum access of DME, a promising amount of idle resources could be predicted. However, previous approaches would perform poorly in a real-world deployment, as they did not take the hidden node problem into account.

This paper formulates the hidden node problem for the case that an LDACS A/A user is within communication range of a DME ground station, but not within range of all airborne DME users connected to it. Through statistical analysis, we underline the problem’s significance in practical cases. We simulate the coexistence of the two systems from a channel access perspective, taking signal propagation and the behavior of the ground station into account. Further, we present an RNN that is able to predict the channel access of hidden nodes. The key idea of our algorithm is to exploit the fact that while DME request pulses from airborne users may appear as hidden, response pulses from the ground station will be visible. Our results show that by inferring DME request channel activity from the response channel, the hidden node problem can be overcome effectively. By using our approach, nearly the same performance can be achieved as in the idealized case where all nodes are visible.

David Kopyto

Maschinelles Lernen für die Nachrichtentechnik, End-to-End Learning, Autoencoder, Constellation Shaping, Compressed Sensing

Daniel Stolpmann
Doktorand & MLE-Webmaster

Machine Learning for Communication Networks, Low Latency & High Throughput Communication, Mobile Edge Computing, Satellite Communication, Network Coding

Sebastian Lindner

Maschinelles Lernen für Kommunikationsnetze, Verteilte Algorithmen, Kanalzugriff bei ad-hoc Netzen, Reinforcement Learning, Zuverlässigkeitsanalyse, Modellierung

Gerhard Bauch
Andreas Timm-Giel
Professor & Präsident

Communication Networks, Performance Evaluation, Modelling, Machine Learning for Communication Networks