Introduction
We aim to work towards the dream of perpetual flight, UAVs that can stay aloft for weeks or even months without fuel as part of the fight against climate change.
UAVs capable of harvesting atmospheric energy could transform environmental monitoring, disaster response, and global connectivity, while leaving a zero-carbon footprint. Why is it important? We can capture high resolution data that satellites can't. This data is essential for many applications, from searching for environmental data related to global warming to search and rescue during disasters.
This PhD will focus on developing predictive soaring technologies that allow UAVs to anticipate and exploit soaring conditions in a timely and efficient manner. A key part of the project will be to explore the most effective low-power sensing strategies for detecting these conditions, ranging from acoustic and pressure sensors to alternative novel methods. These sensors will then be combined with spatio-temporal machine learning and environmental propagation models to forecast temperature fields and turbulence at a distance.
The challenge is to design algorithms that are efficient enough to run onboard UAVs yet powerful enough to predict soaring opportunities in real time. Beyond UAVs, the AI techniques you will develop, such as spatio-temporal modelling, uncertainty-aware prediction, and sensor–model fusion, have broader applications in climate modelling, renewable energy forecasting, environmental monitoring, and autonomous robotics operating in complex dynamic environments. If you are excited by pushing the boundaries of AI for sustainability, this PhD offers the chance to apply your skills to one of aviation’s most ambitious frontiers.
We hypothesise that an actively emitted, low-power acoustic pulse (when processed with advanced signal conditioning and ML-based classification) can infer vertical wind patterns and energy gradients in the lower atmosphere. Such echolocation-inspired sensing, tuned for specific atmospheric frequency bands, can be embedded into small UAVs without the energy and weight penalties of conventional systems.
Similar to established Sonic Detection and Ranging (SODAR) systems [1], transmitted acoustic pulses are scattered by temperature and density fluctuations associated with air and turbulent eddies, with a fraction of this energy backscattered and received as echoes. The time delay of these echoes yields scatter height, while the Doppler shift provides radial wind velocity information [1]. Understanding the physics of acoustic scattering in high-altitude turbulence is central to the success of this project. If these processes are mischaracterised, the system could be optimised for the wrong frequency ranges, transmit powers, or signal structures, severely limiting its effectiveness. To mitigate this risk, we include a dedicated physics work package to quantify scattering coefficients, capture the effects of intermittency, and model stratospheric absorption under realistic conditions. This approach not only safeguards the performance of the proposed miniaturised SODAR on UAVs but also establishes a validated knowledge base that can benefit the broader atmospheric sensing community.
Recent studies on wave propagation in random, time-varying media show that nonlinear multiple-scattering dynamics can emerge, reinforcing the need for optimised excitation strategies that account for temporal disorder [2]. By carefully selecting frequency bands with minimal absorption and employing low-duty-cycle modulation to capture nonlinear multi-scattered responses, a low-power SODAR can maintain sensitivity comparable to conventional systems while using significantly less transmitted energy, making it lightweight, UAV compatible, and suitable for high-altitude operation. Operating from an airborne platform introduces an additional challenge, as propeller wash, structural vibrations, and unsteady aerodynamic loading raise the noise floor compared to stationary ground-based systems. To mitigate these effects, the system must be strategically positioned on the UAV to minimise exposure to self-generated noise, while the signal chain will incorporate advanced processing techniques such as adaptive noise suppression, pulse compression with modulated waveforms, and coherent averaging across multiple pulses. This represents one of the open research questions to be addressed in the study.
Methodology
The supervisory team have used ML techniques [3], learning from highly nonlinear systems, we can design a system that works effectively with a minimal set of actuators and sensors. However their suitability for a highly nonlinear and unpredictable condition like Turbulence need to be checked and new models need to be built. To minimise energy consumption, the acoustic excitation mode and pulse modulation will be carefully selected and emitted at low duty cycles, using burst-modulation schemes that balance signal strength with power constraints. Pulse compression and digital signal processing techniques such as matched filtering and time-frequency transforms will be employed to enhance the signal-to-noise ratio of received echoes. The system may require the use of adaptive scheduling and beamforming methods to improve directional sensitivity and spatial resolution.
Objectives and Outcomes
Design a reliable, compact, energy-efficient acoustic actuator–sensor pair capable of atmospheric echolocation.
Develop ML-based algorithms to classify return signals and infer vertical flow conditions.
Validate echolocation sensing against reference probes in a controlled lab and outdoor environments.
The proposed system will consist of compact, multi-directional acoustic emitters paired with sensitive receivers capable of capturing atmospheric reflections from actively generated pulses. These units will be mounted on UAV platforms or test rigs and designed to operate in frequency bands optimal for atmospheric return detection, typically in the infrasonic or low-frequency audible range (20 Hz–2 kHz). The achievable sensing range will depend on both transmit power and excitation frequency, with the system designed to reach distances of over 500 m.
References:
[1] M. A. Kallistratova, “Acoustic waves in the turbulent atmosphere: A review,” J. Atmos. Oceanic Technol., vol. 19, no. 8, pp. 1139–1150, 2002.
[2] R. Pierrat, J. Rocha, and R. Carminati, “Causality and instability in wave propagation in random time-varying media,” Phys. Rev. Lett., vol. 134, no. 23, p. 233801, 2025.
[3] L. Yule, N. Harris, M. Hill, and B. Zaghari, “An experimental study of machine-learning-driven temperature monitoring for printed circuit boards (PCBs) using ultrasonic guided waves,” NDT, vol. 3, no. 1, p. 1, 2025, doi: 10.3390/ndt3010001.