Core Concepts
The author proposes a CNN auto-encoder for T-NOMA, achieving superior performance over the SVD method by approximately 2 dB in ideal conditions.
Abstract
The content discusses the design of an auto-encoder for downlink time-offset faster-than Nyquist signaling NOMA channels. It compares the proposed CNN AE with the SVD method, highlighting performance gains and complexity differences. The study explores various training modes, loss functions, linear connections, and impacts of imperfect CSI on BER performance.
The authors introduce novel CNN AE architectures for T-NOMA systems with linear complexity in sequence length. They propose a modified loss function combining cross-entropy and Q-function to improve BER. Additionally, experiments show that linear connections between encoder and decoder components enhance performance significantly.
Furthermore, the study evaluates the impact of timing errors and imperfect CSI on system performance. The results indicate error floors in BER due to data-dependent noise from timing offset errors and CSI estimation errors.
Stats
The proposed CNN AE surpasses the SVD method by approximately 2 dB in a T-NOMA system.
In the presence of channel state information (CSI) error variance of 1% and uniform timing error at ±4% of the symbol interval, the proposed CNN AE provides up to 10 dB SNR gain over the SVD method.
Simulations show that the modified loss function achieves SNR gains of up to 1 dB over the CE loss function alone.
Experiments show that architectural innovations achieve additional SNR gains of up to 2.2 dB over standard serial CNN AE architecture.