Work

Correlated Noise-Aware Iterative MPA-CNN Detection for SCMA Systems

Year: 2026
Type: article
Abstract: Correlated noise severely degrades the detection performance of conventional message passing algorithm (MPA)-based sparse code multiple access (SCMA) detection due to the mismatch between the practical noise statistics and the white Gaussian assumption adopted in conventional MPA detection. To address this issue, a correlated noise-aware iterative MPA-convolutional neural network (MPA-CNN) detection framework is proposed for SCMA systems. In the proposed framework, the residual noise reconstructed from preliminary MPA detection is exploited as the input of a lightweight convolutional neural network, which is trained to estimate and suppress the correlated noise components. The refined received signal is subsequently fed back to the MPA detector for iterative detection enhancement. By jointly optimizing the CNN architecture and loss function, the proposed scheme not only suppresses the residual noise power but also reshapes the residual noise distribution toward a more Gaussian-like form, thereby improving the compatibility with subsequent MPA detection. Furthermore, the proposed framework is extended to generalized frequency division multiplexing (GFDM)-SCMA systems, where the colored noise introduced by zero-forcing equalization can be naturally incorporated into the proposed correlated-noise-aware detection framework. Simulation results demonstrate that the proposed detectors achieve significant bit error ratio (BER) performance gains over conventional MPA-based schemes under correlated noise environments in both SCMA and GFDM-SCMA systems.
Language: English
Open Access status: hybrid
Estimated APC paid: $2,645
OpenAlex ID: W7224001030
FWCI:
Cites: 51
Cited by:
Related to: 10