A Fast Fault Diagnosis Method for RF Front-end Modules Based on Adaptive Signal Decomposition and Deep Neural Network

Large-scale Multi-Input Multi-Output (MIMO) technology is currently one of the most promising wireless transmission technologies widely used in radar, communications, and navigation. As a core component of the MIMO communication system, the multi-channel RF transceiver front-end is crucial in the overall performance of the MIMO system, for its location of faults after system abnormality is urgent for system test assurance. However, with the increasing complexity and integration of the RF circuit, the testability of the transceiver front-end and the accessibility of its internal test points have dramatically decreased. Moreover, the traditional method of obtaining test parameters and locating faults by internal test points is no longer effective. Therefore, locating the internal faults directly through the external test port of the module has become a challenging problem for RF transceiver front-end tests and diagnosis. To solve the problem, this study proposes an intelligent fault diagnosis method combining variational mode decomposition (VMD) and interleaved group convolution deep neural network (IGCDNN), which can quickly and precisely locate module faults by sampling directly at the output. This method adaptively decomposes the RF channel output response into sub-signals with independent center frequencies by VMD. These sub-signals obviously show sparsity in the frequency domain. By signal mirror extension, the decomposition effectively avoids the modal confusion and endpoint effect while achieving the accurate extraction and dimensionality reduction of the fault features. According to the decomposition, we build an intelligent fault diagnosis based on IGCDNN to realize fault location. Compared to the conventional convolutional neural network, the IGCDNN has a multi-channel parallel convolutional structure, leading to fewer parameters and lower computational complexity, significantly improving fault diagnosis efficiency. We have also conducted experiments on the single-channel RF transceiver front-end simulation circuit. The results show that this method's diagnosis accuracy and channel diagnosis time are superior to the existing diagnostic methods only using the external test port of the module, and without increasing the internal test point, the faults in the RF transceiver front-end module can be identified and located accurately and quickly.