베이지안 최적화를 적용한 심층 학습기반의 빔 형성 기술 (Deep Learning-Based Beamforming Design with Bayesian Optimization)

서지정보

· 발행기관 : 한국전자파학회
· 수록지 정보 : 한국전자파학회 논문지 / 36권 / 1호 / 86 ~ 94페이지
· 저자명 : 유호근, 임재혁, 오선진, 이의혁, 정대교, 김성권, 오현수, 이재훈

초록

본논문은배열안테나의성능을향상시키기위해심층학습기반의빔형성기술을탐구하며, 계산비용을최소화하면서신호대비간섭및잡음비율(SINR, signal to interference plus noise ratio)을 개선하는 것을 목표로 한다. 본 연구에서는심층학습을활용해수신신호의자기상관행렬에서최적의빔형성가중치벡터를예측하는방법을제안한다. 심층학습중, 완전연결신경망(FNN, fully connected neural networks), 합성곱 신경망(CNN, convolutional neural networks), 순환 신경망(RNN, recurrent neural networks)을 사용해 수신 신호의 자기 상관 행렬로부터 가중치 벡터를출력하도록학습하고, 베이지안최적화를통해각모델의하이퍼파라미터를튜닝하였다. 실험결과, FNN 모델이원하는SINR을유지하면서간섭을최소화하는빔을형성하는데가장우수한성능을보였다. 특히FNN 모델은다양한신호대비잡음비율(SNR, signal to noise ratio) 조건에서 최적 가중치벡터대비SINR의열화가0.32 dB에불과하여, 심층학습기반빔형성에적합한모델로확인되었다. 이연구는배열안테나에서심층학습을활용한빔형성기술이효율성과정확성을높일수있는것을확인하였다.

영어초록

This study explores deep-learning-based beamforming techniques to enhance the performance of array antenna systems with the aim of reducing computational costs and improving the signal-to-interference-plus-noise ratio (SINR). In this study, a method is proposed to predict the optimal beamforming weight vector using deep learning. This approach involves training fully connected neural networks (FNN), convolutional neural networks (CNN), and recurrent neural networks (RNN) to learn the autocorrelation matrix of the received signals and output the weight vector. Bayesian optimization was employed to tune the hyperparameters of each model. The experimental results indicated that the FNN model exhibited the best performance in forming beams that maintained the desired signal while minimiz ing interference. Specifically, the FNN model demonstrated an SINR difference of only 0.32 dB from the theoretical optimal weight vector across various signal-to-noise ratio (SNR) conditions, confirming its suitability for deep learning-based beamforming. This study highlights the potential of deep-learning-based beamforming techniques to improve both the efficiency and accuracy of array antenna systems.

참고자료

· 없음