Fault Diagnosis of Communication Systems using Deep Neural Networks

Research output: Contribution to journalJournal articleResearchpeer-review


In 5G new radio and 6G, the primary communication technology of the front end is the active phased array (APA).
To enable stable operation, fault diagnosis of front ends is crucial.
The fault diagnosis methods of front ends, to date, rely on frequency domain radiation patterns and synthetic equations to identify the possible faulty antenna elements in the APA.
This requires a stable measurement setup with multiple strictly controlled measurement probes in a specific plane, which is time consuming, complex, and costly.
This paper proposes a novel method that only needs a single probe in one measurement point for a fast and accurate diagnosis of the faulty elements and components in APAs. This is achieved by using Deep Neural Networks (DNN) to classify the baseband in-phase and quadrature signals with more signal features that cannot be easily captured by conventional methods.
High-accuracies of up to 99\% for correct fault detection and recognition have been demonstrated in situ with a commercial 28 GHz APA under three groups of test scenarios including, on-off antenna elements, phase variations, and magnitude attenuation variations.
Further, an investigation in the robustness of the model with varying signal to noise ratio (SNR) shows a stable fault detection accuracy above 90\% with a low SNR of down to 4 dB.
Original languageEnglish
JournalIEEE Transactions on Antennas and Propagation
Publication statusIn preparation - 5 Jul 2021


Dive into the research topics of 'Fault Diagnosis of Communication Systems using Deep Neural Networks'. Together they form a unique fingerprint.

Cite this