.png){kind=icon}
.png){kind=icon}
Return
Physics-Guided 1D-CNN Inverse Mapping From S-Parameters to THz SWS Geometry
.png){kind=icon}
PRE
AI
J
Y
Y
J
Z
S
Y
DOI：10.1109/led.2026.3707551.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This letter presents a novel physics-guided one-dimensional convolutional neural network (1D-CNN) inversion method to directly predict the six geometric parameters of slow-wave structures (SWSs) from their S-parameters. This method employs a frequency screening strategy to eliminate redundant noise and extract effective electromagnetic features. A weighted Huber loss function is introduced to increase the weight of the key parameter (waveguide width) and improve the prediction accuracy. Validated on a 1 THz folded waveguide SWS, the method achieves predictions within machining error tolerances using cold test data from the optimized 890–1100 GHz band. The reconstructed S-parameters agree well with cold test results, and beam-wave interaction simulations show that the predicted structure, operating at its optimal beam voltage, yields 39.2% higher output power than the original design. This method supports processing error analysis of SWSs and provides a reference for subsequent hot test.
Keywords:
Terahertz (THz) technology
traveling wave tube (TWT)
slow wave structure (SWS)
convolutional neural network (CNN)
S-parameters inverse mapping
Journal
IF:
4.5
Papers:
614
Citations:
2.3W
Organization
Cited Papers