Deep Learning‐based Transmitter and Receiver
Design for Grant‐Free Non‐Orthogonal Multiple
Access for IoT networks
Medium Access Protocols involves a broader spectrum (up to 60 GHz) and will allow for non-orthogonal techniques and most likely be OFDMA‐based. Nonetheless, there is a consensus in the literature that the orthogonal multiple access approach will not suffice to support the massive access from a vast number of IoT systems devices. That is because grant‐based communications suffer from collision rates over the random‐access channel that is as high as 10%, for less than ten active users. Moreover, the signalling overhead involved in establishing a link is about 30‐ 50% of the payload size, for messages less than 200 bits long. In terms of latency, the grant‐based access procedure in LTE‐A, for example, takes around 5‐8 ms in the best‐case scenario. Thus, grant-based access fails at meeting many KPIs when massive connectivity is required for short packet transmissions. Because of this, the lightweight random access protocols were heavily investigated over the past years. The throughput has improved in orders of magnitude with sophisticated yet still low‐complex transceiver algorithms. In this project, we will look into whether we can adopt the advanced deep learning technology in solving this issue and design a low‐complexity transmitter and receiver to allow multiple sensors to transmit short packets simultaneously in the network.
Road Detection and Extraction in Remote Areas via Deep-Learning-based Remote Sensing
Making maps for rural areas is a tedious task as it requires inspections to be done manually by humans present on the scene. However, with the explosion of satellite image data, we now have the opportunity to automate this task by utilizing the latest Artificial Intelligence development to detect paved roads or otherwise. This new technology can accelerate the road mapping process and quickly updating the current versions of the maps. Having a remote sensing technology that can detect roads will help monitor communities’ growth in rural areas and help authorities see the illegal roads in tropical (and sub-tropical) regions that could prevent further deforestation/degradation. In the literature, the current research focuses heavily on urban areas, focused on training convolutional neural networks (CNN) with high-resolution images. The models made for one landscape aren’t transferrable to other places. This project’s scope is to develop a deep learning model to detect paved roads in rural areas using satellite images. The model should be novel and achieve high accuracy. This is a collaborative project with Distinguished Professor Bill Laurance at James Cook University.
Smart Sensing: Computer-Vision-Based Discharge and Depth Monitoring in Water Channels.
In environmental monitoring, water depth is an essential parameter for us to understand the
waterway condition and associated ecological systems. For example, the water depth affects the
underwater light intensity, temperature variation and nutrient content. In a tropical weather
location, during the wet season, the massive rainfall event can significantly affect the water depth
of the water channel. The installation cost for one water depth sensor is about 10k+, which is
expensive. The current low‐cost sensor has been installed by the council are cost about 1k per
unit. But the collection of the data from this low‐cost sensor requires a manual process, which
means that we are not able to obtain real‐time water depth data. However, during the wet
season, the water depth in water channels changes rapidly and knowing the water depth in realtime
can help us to manage the related environmental conditions better. For example, we can
issue warning to relevant parties if the water depth at one location raised very quickly so that we
can prepare for possible events. To achieve this goal, we will research, design, and implement a
low‐cost water depth sensor that capable of collect water depth data, and transmit data to the cloud
in real‐time in this project.
Early Prediction of Age-related Neurodegenerative Diseases with Deep Learning
Healthy aging is increasingly significant as larger numbers of people are living longer. Most neurodegenerative diseases (NDDs), including Alzheimer's disease and Parkinson's disease, are positively related to individual aging. Their symptoms usually begin gradually and get worse over time. Therefore, understanding NDDs progression and early detection of age-related neural diseases are paramount for normal aging. This project will develop deep neural networks to tackle the above issue using the clinical data with the imaging features, neuropsychological test outcomes, or other diagnoses.