잡음 주입 기반 심층신경망을 이용한 시변 채널 예측 기법 = (Noise Injection-Based Deep Neural Network Channel Prediction for Time-Varying Channels)

다국어 초록 (Multilingual Abstract)

We propose a noise injection-based deep neural network channel prediction method for single-input single-output time-varying channels, motivated by service environments that require reliable wireless transmission, such as mobile generative-AI-based media services. The proposed method injects noise corresponding to the signal-to-noise ratio of the test environment into the training input, enabling robust channel prediction even in the presence of pilot-based channel estimation errors. Simulation results demonstrate that the proposed method achieves lower normalized mean squared error and bit error rate compared with a baseline deep neural network-based channel predictor. The results also indicate that the proposed method provides more pronounced performance gains in high-Doppler environments.