Modular System Design for Uncertainty-Aware Engineering Architectures

Abstract

Modular system design is becoming increasingly important in modern engineering systems, particularly in wireless communication and Multiple Input Multiple Output (MIMO) receiver architectures, where complex interactions, high-dimensional data, and uncertainty in operating environments significantly affect the performances. However, challenges such as data heterogeneity, feature redundancy, high dimensionality, and unstable channel conditions limit reliable prediction and efficient signal processing. To address these issues, this research proposes a scalable and uncertainty-aware modular model for engineering architectures. The MIMO Receiver Reliability Dataset used in the research comprised 7,980 samples, consisting of wireless communication records from the MIMO receiver environment. Z-score normalization is applied to standardize input features and improve convergence stability, while Principal Component Analysis (PCA) is employed to reduce dimensionality and retain the most informative features. A Capuchin Search Algorithm-tuned Dynamic Backpropagation Neural Network (CSA-DBPNN) is developed and implemented using Python to enhance adaptive learning and uncertainty-aware prediction. The CSA optimization improves parameter tuning and convergence efficiency, while the DBPNN captures nonlinear relationships in complex wireless environments. Experimental results demonstrate superior performance with 9.2 ms latency, 135.8 Mbps throughput, 1.4% packet loss rate, and 0.91 feature entropy, outperforming the previous results, thereby confirming its effectiveness for CSA-DBPNN models. The proposed model effectively enhances robustness, reduces uncertainty, and improves predictive reliability in MIMO wireless communication systems.