Detection algorithm for waveguide connection and probe contact states based on machine learning in frequency up to 1.1 THz

The connection state of waveguides (WGs) is one of the major error factors in RF measurements. The stability of WG connections during operation strongly depends on the operator's skill. However, WG connections in the terahertz band up are particularly challenging, making it difficult to perform proper connections. This paper proposes a detection technique for the WG connection state in the frequency range up to 1.1 THz. The proposed technique enables stable measurement operation by integrating the algorithm into a terahertz measurement system. Rectangular WGs and RF probes with on-wafer measurement systems were used for the demonstrations. The proposed method utilizes the local outlier factor method and linear least-mean-square learning. This technique can detect the connection state of the WG even without requiring training data from the device under test (DUT) itself.