A PDD-Inspired Channel Estimation Scheme in NOMA Network

TL;DR

The paper addresses the challenge of predicting dynamic CSI in NOMA networks to improve handover reliability and reduce pilot overhead. It introduces a transfer learning-enabled, $RNN$-$LSTM$-based predictor that uses partially decoded data ($PDD$) as supplementary CSI, along with traditional metrics like $RSRQ$, $SNR$, and $CQI$. System-level simulations show that incorporating $PDD$ improves prediction accuracy (lower $MSE$ and higher $R^{2}$) and reduces handover-related issues such as HOF and ping-pong rates, without relying on additional pilot signaling. The work highlights the practical impact of learning-based CSI estimation in 5G-era NOMA networks and demonstrates competitive performance against baselines, with a plan for broader traffic and application scenarios in future research.

Abstract

In 5G networks, non-orthogonal multiple access (NOMA) provides a number of benefits by providing uneven power distribution to multiple users at once. On the other hand, effective power allocation, successful successive interference cancellation (SIC), and user fairness all depend on precise channel state information (CSI). Because of dynamic channels, imperfect models, and feedback overhead, CSI prediction in NOMA is difficult. Our aim is to propose a CSI prediction technique based on an ML model that accounts for partially decoded data (PDD), a byproduct of the SIC process. Our proposed technique has been shown to be efficient in handover failure (HOF) prediction and reducing pilot overhead, which is particularly important in 5G. We have shown how machine learning (ML) models may be used to forecast CSI in NOMA handover.

Figures (16)

• Figure 1: System model of NOMA-HO
• Figure 2: System model of NOMA transceiver with RNN-LSTM model
• Figure 3: Flowchart of Algorithm 1
• Figure 4: Distribution of Original Values
• Figure 5: Distribution of Augmented Values