How will AI help models and simulate future wireless systems, enabling easier and faster designs?
Florent Busnoult, Senior Application Engineer (SPC, signal processing and communications), MathWorks
Future wireless systems have various layers of complexity such that design optimization takes months for wireless engineering experts using traditional modeling and design approaches. 5G-NR systems, for example, rely on beamforming techniques, massive MIMO, digital predistortion algorithms (DPD) and other specialized techniques at the baseband and RF front-end levels to ensure reliable communications. It can be difficult to optimize each of those elements individually as well as part of a whole system to reach an optimal design solution. Data driven approaches like machine learning and deep learning are well suited to solve such multi-dimensional optimization problems in a computationally efficient way.
Let’s look closer at one element of future wireless systems – the design of accurate antenna elements and phased array systems. As this task requires many time-consuming simulations to optimize the antenna parameters, it is another area where machine learning techniques show promising results. By predicting the antenna/array behavior based on its characteristics, we can increase the computational efficiency and reduce the number of necessary simulations. One such method is the surrogate model assisted differential evolution for antenna synthesis (SADEA) which carries out global optimization and employs a surrogate model built by statistical learning techniques. Similarly, deep learning can be used by engineers to select the most effective DPD models depending on the characteristics of the nonlinearities due to the power amplifier design and the operating conditions.
The complexity of accurately modeling the RF environment is another key limitation for the design and simulation of future wireless systems. Engineers typically rely on simplified channel models, but this limits the achievable performance of the designed system. Deep learning can improve on this by integrating the RF channel environment into a unified end-to-end system called joint encoder decoder or autoencoder. The idea is to build an AI algorithm combining the transmitter encoder, the RF channel and the receiver decoder as one unique piece of the system, freeing engineers from the need to design encoder optimized for different type of data and different channel conditions.
What is the future of AI and Industry 4.0 in relation to wireless systems?
Wireless systems, AI, and Industry 4.0 have a future that is intertwined. Smart factories will rely on wireless systems to gather the large amounts of data required to make decisions and diagnose issues proactively using AI. Using remote AI processing (located in edge computing devices) for real-time applications is a big challenge as it requires extreme robustness and low latency in the communication protocol. Additionally, smart factories will rely on wireless systems with AI algorithms of their own to provide improved localization within indoor environments.