Ku-Band Low-Noise Amplifier Configuration and Optimization Through BiLSTM-Based Deep Neural Networks

This work proposes an automated configuration and sizing framework for a low-noise amplifier (LNA) operating in the Ku-band. The methodology is based on the implementation of deep neural networks (DNNs), specifically a bidirectional long short-term memory (BiLSTM) classifier and a BiLSTM regression model, which are employed sequentially to first generate an appropriate LNA topology and subsequently predict the optimal design parameters of the selected configuration. To further refine the design toward multiple conflicting objectives, a multi-objective multiverse optimization (MOMVO) algorithm is integrated into the workflow to balance key performance metrics such as noise figure, and gain. The overall process is fully automated through the coordinated use of an electronic design automation (EDA) tool and a numerical analysis environment, wherein the circuit topology is synthesized and simulated in the EDA platform, while parameter optimization and decision-making are carried out in the numerical analyzer. This closed-loop framework enables rapid convergence to high-performance solutions with minimal human intervention. The practical effectiveness of the proposed methodology is validated by designing and optimizing an LNA suitable for a Ku-band small satellite receiver, achieving a noise figure below 1.5 dB while satisfying gain and stability constraints, thereby demonstrating the viability of deep learning–assisted automated radio frequency circuit design.