Deep learning-based decoding in non-orthogonal multiple access (NOMA) for high altitude platform systems (HAPS)

High altitude platform systems (HAPS) offer flexible communication solutions compared to ground-based systems. The non-orthogonal multiple access (NOMA) in HAPS facilitates effective multi-user communication. However, using successive interference cancellation (SIC) for decoding presents significant drawbacks, such as the domino effect if the initial user fails to decode correctly. This study proposes convolutional neural network (CNN)-based methods for HAPS to enhance multiuser detection and overcome drawbacks from SIC. Our findings indicate that user positions and movement patterns impact system performance, with closer nodes requiring lower signal-to-noise ratios (SNRs) for a bit error rate (BER) of 10–3. Increased user numbers raise SNR requirements for adjacent cluster head (CH) nodes, emphasizing placement importance. An inverse relationship between BER and SNR, influenced by power allocation and channel conditions, is crucial for optimization. CNN-based systems show superior scalability and robustness, supporting up to seven users, compared to four for conventional systems. Despite slightly higher SNR needs, CNN-based systems achieve more users, making them ideal for smart cities and IoT applications. Additionally, CNN-based systems outperform traditional methods for next-generation networks.