Prediction Of Interference Path Loss From PEDs Inside An Aircraft Cabin To Exterior Fuselage-mounted Antennas Based On Hybrid Neural Network

With the rapid development and promotion of internet and mobile communication service,Airlines are increasingly interested in providing in-fight wireless services in the premise of safetyflight, to satisfy passengers’ demand of mobile communication and internet service onboard. However,the radiation emissions from portable electronic devices (PED), such as mobile phones, laptopscarried by passenger, expose a potential threat to aircraft airborne radio navigation system. So, how torealize the compatibility between portable electronic devices onboard and airborne equipmentsbecome the hotspot of the aircraft design for PED tolerate. The interference path loss (IPL) is the oneof the main indicators of Aircraft indoor-outdoor EMC problems. As the IPL measurement is highlycostly or even impractical, and numerical modeling approaches prove too complex with limitations, aparticular model based on hybrid neural networks is proposed to predict the IPL in this thesis.The main researches of this thesis are as follow:1. The interference mechanism between PEDs inside an aircraft cabin and exteriorfuselage-mounted antennas is investigated, and the measurement of IPL is analyzed. An efficientmodel based on hybrid neural networks is established.2. The forward neural network is applied to predict the IPL. A simple model of B757isestablished. The S21between PEDs and antennas is well predicted in the frequency band0~4.5GHzwith the trained samples from CST simulation.3. Focusing on the defects of local minimum in gradient algorithm in the optimization of neuralnetworks weights, we use two kinds of intelligent algorithms to optimize neural network weights, anddiscuss the two optimization algorithms combined with neural network, the setting of parameters andoperator selection in detail. The performance in multi-dimensional optimization comparation betweengenetic algorithm and particle swarm optimization is carried out through numerous repetitive tests.4. The existing IPL metrical data and the common law in different antenna systems of the IPL areanalyzed. Some simulation examples are given. By analyzing these simulation examples, neuralnetworks combined with particle swarm optimization is founded performing well in predicting theoverall trends of IPL in various aircraft antenna systems, which is proposed as a quick and efficientassessment approach for aircraft electromagnetic compatibility design.