Enhancing Symbol Detection Accuracy in Pilot-Aided Multiple Access Networks Using the Novel Deep Learning Approach Adaptonet

• Manchuri Tulasi,
• D. Sreenivasa Rao,

Vol. 20, No. 3, March 31, 2026
10.3837/tiis.2026.03.021, Download Paper (Free):

• Deep Learning
• LSTM
• symbol detection
• Adaptive Architecture
• Communication Systems

Abstract

This paper presents AdaptoNet, a novel deep learning–based framework for enhancing symbol detection accuracy in pilot-aided multiple access OFDM systems employing Offset Quadrature Phase Shift Keying (QPSK) modulation. The proposed approach is built on a Long Short-Term Memory (LSTM)–based deep neural network with customizable hidden units, enabling dynamic adaptation to varying channel conditions. The methodology comprises three stages: training data generation using a pilot-aided OFDM model over a narrowband Rayleigh fading channel, training of the AdaptoNet model, and extensive performance evaluation. The proposed model achieves rapid convergence, attaining 100% training accuracy within 100 epochs, with training loss converging to zero around 250 iterations. Simulation results demonstrate that AdaptoNet significantly outperforms conventional Least Squares (LS) and Minimum Mean Square Error (MMSE) detection methods, achieving a Symbol Error Rate (SER) of zero at an SNR of 18 dB when a cyclic prefix is employed, and an SER of 0.0002 at 20 dB. Comparative analysis with existing methods further confirms the superiority of the proposed framework, highlighting its effectiveness, robustness, and suitability for reliable symbol detection in pilot-aided OFDM multiple access systems.

Statistics

Cite this article