Efficient Signal and Modulation Classification for High Mobility Wireless Communication System

Abstract

The escalating demand for high-mobility applications, such as vehicle-to-everything (V2X) networks, high-speed rail, and internet of things (IoT), drives the evolution of sixth-generation (6G) wireless communication systems. These systems must perform reliably in doubly selective channels, where severe Doppler shifts and multipath fading cripple traditional orthogonal frequency-division multiplexing (OFDM) due to high inter-carrier interference. Discrete Zak transform based orthogonal time frequency space (DZT-OTFS) and affine frequency division multiplexing (AFDM) offer robust alternatives. DZT-OTFS maps symbols in the DD domain for enhanced resilience, while AFDM’s chirp-based subcarrier design optimizes spectral efficiency in time-varying channels. Signal classification, modulation classification, and channel equalization are critical for adaptive 6G transceivers but face challenges, including high computational complexity, reliance on accurate or complete channel state information (CSI), and performance degradation in low SNR or imperfect CSI scenarios. This thesis proposes four novel algorithms for DZT-OTFS and AFDM systems, delivering lowcomplexity solutions to enhance 6G communication efficiency and adaptability. We develop a scalable blind signal classification technique to identify single-carrier, OFDM, and DZT-OTFS waveforms in doubly selective channels without requiring pilots. This method exploits higherorder cumulants to distinguish the underlying distributions of random processes and leverages the discrete Zak transform’s Gaussianity-inducing properties, using a threshold-based framework for effective classification. Additionally, we propose a modulation classification algorithm for DZTOTFS signals, which relies on imperfect CSI and higher-order cumulants to distinguish the underlying distribution, also employing a threshold-based framework for accurate classification. Monte Carlo simulations in additive white Gaussian noise (AWGN) and impulsive noise environments demonstrate superior accuracy and lower complexity compared to deep learning-based approaches, enabling waveform coexistence in unified transceivers. Next, we introduce a low-complexity linear equalization algorithm for DZT-OTFS, leveraging the banded structure of the time-domain channel matrix and a zero prefix to minimize inter-block dependencies. Incorporating Bunch-Huffman decomposition, the algorithm reduces computational complexity from cubic to linear with respect to number of subcarriers, enabling parallelization and modulation-independent operation. Simulation results in MATLAB validate its performance, showing improved BER over minimum mean square error (MMSE) and zero forcing (ZF) equalization techniques, supporting low-latency transceivers in high-mobility scenarios. We also propose a deep neural network-based surrogate model for modulation classification in reconfigurable intelligent surface (RIS) aided OTFS systems, approximating ML decision rules with reduced complexity. The architecture combines multikernel convolutional neural networks (CNNs) for local DD domain feature extraction, VOLO attention mechanisms, vision transformers (ViTs) for hierarchical spatial dependencies, and a Kolmogorov-Arnold Network (KAN) classification head for precise decision boundaries. Evaluated on a MATLAB-generated dataset over an EVA channel, the model outperforms CNN-ViT baseline, demonstrating robustness against imperfect CSI for real-time 6G applications. Finally, we develop a blind modulation classification method for AFDM, designed for doubly selective environments without prior CSI. The technique uses the DAFT to process received symbols, followed by 12th-order cumulant estimation to differentiate modulations like binary phase shift keying (BPSK), quadrature phase shift keying (QPSK), amplitude shift keying (2ASK), and 8PSK, and a hybrid 8th/14th-order cumulant to distinguish quadrature amplitude modulation (16QAM) from QPSK. Monte Carlo simulations benchmark its computational efficiency and accuracy, highlighting its potential for single-symbol classification and addressing AFDM’s chirp-based structure. These algorithms enhance existing approaches by delivering low-complexity solutions that improve spectral efficiency, reduce latency, and enable adaptive transceivers for 6G applications, including V2X, cognitive radio, and spectrum surveillance. To the best of our knowledge, this is the first work to propose modulation classification for AFDM and to integrate such a diverse set of algorithms for DZT-OTFS and AFDM, advancing waveform-agnostic 6G systems.