In the design and operation of phased array antennas - an antenna type that is made up of many identical smaller "element" antennas working in unison, controlled by the phase of each element's excitation coefficient - control is currently limiting the technology in practice. The main limiting factor comes from finding the excitation coefficients. This is particularly the case for modern arrays, for applications such as earth observation with unprecedented accuracy and telecommunications for global broadband coverage, where hundreds or even thousands of elements are used. In such applications, the reconfigurability of antenna arrays are their main advantage, but the reconfigurability relies on rapid and accurate computation of the excitation coefficient phases, which is not possible today. In this project, TICRA will implement an industrial-grade implementation of an algorithm that can, with unprecedented accuracy and speed, find excitation coefficients for a regular phased antenna array, based on requirements to the array far-field, with multiple simultaneous requirements, controlling the phase of each element ("phase-only"). The implementation will be based on Machine Learning (ML) technology, namely custom-developed deep neural networks, pre-trained using TICRAs industry-standard antenna simulation tools, allowing unprecedented accuracy and speed. The solution will allow for more efficient designs of phased arrays, when used during the design of the array, as well as allowing continuous reconfigurability when used during the operation of the array.
