Existing E2E systems would suffer performance degradation when applied in the environments where the channel conditions deviate from the trained ones. To address this problem, we propose a comprehensive method to develop an E2E communication system that possesses strong adaptability under dynamically varying unknown environments. From the system structure prospective, our system consists of the impulse generative adversarial network (IGAN) and transceiver. Within the E2E system, a novel transceiver is proposed, which maps information bit sequences to symbol vectors to introduce the inter-symbols correlation such that the signals can be detected without pilots. Hence, the transceiver is more flexible in a changing fading environment while achieving higher spectrum efficiency. On top of that, the model-agnostic meta-learning (MAML) is utilized to meta-train the IGAN to enable fast adaptation to practical noise samples. The transceiver is then general-trained based on the designed fading channel set that includes various time and frequency selectivity and the simulated noise set. Simulation results show that our proposed E2E system can outperform conventional approaches both in terms of BER, overheads and adaptability.
An End-to-End Communication System with Environmental Adaptability
2023-10-10
1398070 byte
Conference paper
Electronic Resource
English
Environmental adaptability evaluation model for stratospheric airships
Online Contents | 2013
|Research on Environmental Adaptability of Stratospheric Airship Materials
Springer Verlag | 2024
|Environmental Damage and Environmental Adaptability of the Aircraft in Marine Atmosphere
DOAJ | 2016
|Device for testing environmental adaptability of unmanned aerial vehicle
European Patent Office | 2021
|