HybNet: A Hybrid Deep Learning-Matched Filter Approach for IoT Signal Detection

Abstract

Random access schemes are widely used in IoT wireless access networks. They enable a reduced complexity and overcome power consumption constraints. Nevertheless, random access results in high packet losses which are caused by overlapping transmissions. Signal detection methods for digital modulation techniques are typically based on the well-established <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">matched filter</i> , which is proven as the optimal filter under additive white Gaussian noise for minimizing error probability. However, with the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">colored</i> interference arising from the overlapping IoT transmissions, deep learning approaches are being considered a suitable alternative. In this paper, we present a hybrid framework, dubbed as <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">HybNet</i> , that alternates between deep learning and match filter pathways based on the perceived interference level. This helps the detector to work in a broader range of conditions, optimally leveraging both the matched filter and deep learning advantages. We compare the performance of several possible data modalities and detection architectures concerning the interference-to-noise ratio, demonstrating that leveraging domain knowledge by pre-processing the input data in conjunction with the proposed HybNet surpasses the complex conjugate matched filter performance under interference-limited scenarios.