SAMBA: Scenario-Adaptive Meta-Learning for mmWave Beam Alignment

In millimeter wave (mmWave) communication systems, achieving high-quality data transmission demands efficient and rapid beam alignment. Traditional deep learning-based methods, although promising, often rely on the assumption that training and testing channels share identical distribution. This assumption may not hold true in practical settings, potentially leading to significant performance degradation when the deployment environment changes. To address this issue, we introduce SAMBA, a novel meta-learning-based approach for beam alignment. SAMBA enables swift adaptation to unknown scenarios using a minimal set of newly labeled data. Specifically, we ensure our model's adaptability by leveraging Model-Agnostic Meta-Learning (MAML) for initial parameter vector determination. To handle a large number of classes present in the narrow beam selection problem, we develop a novel training task splitting
^1
1
A task refers to a specific problem within a defined scope. In our context, each task signifies a probing beam design and beam selection process for user grids in a designated range. strategy. Our deep neural network employs a complex layer that mimics a probing codebook and a narrow beam selector consisting of three dense layers. Experimental assessments involving ray-tracing scenarios
^2
2
A scenario refers to a particular set of circumstances that define the communication environment, including factors such as indoor or outdoor settings, frequency, and bandwidth of the communication channel. demonstrate SAMBA's capacity to adapt to unknown scenarios, including transitions from outdoor to indoor environments. Our results show that SAMBA outperforms learning-based baselines and exhibits superior domain adaption capabilities.