Directly Matching an MMIC Amplifier Integrated with MIMO Antenna through DNNs for Future Networks.

Area of Science:

• Electrical Engineering
• Computer Science
• Artificial Intelligence

Background:

• Exponential growth in data communications necessitates improved linearity and impedance matching in high-frequency systems.
• Efficient power transfer from power amplifiers (PAs) to antennas is critical for transceiver performance.
• Existing methods struggle to address the complex interplay between linearity and impedance transformations.

Purpose of the Study:

• To enhance transceiver linearity and optimize impedance transformations using intelligent, data-driven methods.
• To develop a deep neural network (DNN) framework for concurrent modeling and optimization of antennas and PAs.
• To demonstrate an automated methodology for improving system-level performance in high-frequency communication.

Main Methods:

• Utilized Long Short-Term Memory (LSTM)-based DNNs for antenna load impedance forecasting across a wide frequency band.
• Employed a second LSTM-based DNN with Multivariate Newton's Method to model and optimize PAs, predicting optimal drain impedances.
• Integrated antenna and PA models for automated, system-wide optimization of linearity specifications.

Main Results:

• Successfully modeled and optimized a monolithic microwave integrated circuit (MMIC) with a multiple-input multiple-output (MIMO) antenna concurrently.
• Achieved effective enhancement of transceiver linearity within the 7.49 GHz to 12.44 GHz frequency band.
• Demonstrated the capability of DNN-based models to predict optimal drain impedances for improved PA performance.

Conclusions:

• The proposed DNN-based optimization methodology effectively enhances transceiver linearity and impedance matching.
• LSTM networks provide a powerful tool for modeling complex RF components like antennas and PAs.
• This intelligent approach offers a promising solution for optimizing high-frequency communication systems.