Prediction of UHF-RFID Tag Performance Utilizing Deep Learning Regression

RFID is a mature and widespread technology, posing the backbone of today's supply chain. Especially UHF-RFID is very common in logistics and production environments, due to the high read rates and range. However, this comes at the cost of higher susceptibility to detuning effects and performance degradation caused by materials in close proximity. Therefore, the application surface and material have a great impact on the performance of RFID tags. To increase the reliability of RFID systems and enable more accurate and efficient planning, this paper proposes a novel approach to predict the complex mutual effects utilizing deep artificial neural networks to solve a multivariable regression problem. First, training data is generated using full-wave simulation techniques and considered datasets and input features are introduced. Further, the neural network architecture and optimized hyper-parameters of the model are presented. Finally, the simulation results are evaluated in comparison to the predictions of the deep learning model. Superior computational performance providing fair accuracy compared to conventional simulation techniques can be attested.