Intelligent AutoEncoder-Based Modulation: Optimizing Transmission Performance over Non-Ideal Wireless Channels

End-to-End AI-Native Physical Layer for Robust 6G MIMO-OFDM: Adaptive Constellation Shaping and Attention-Based Detection

Sixth-generation (6G) physical-layer designs require robustness against non-analytical channel distortions and hardware impairments that violate classical linear assumptions. This paper presents an AI-native framework for MIMO-OFDM systems that jointly optimizes adaptive constellation shaping and neural detection through end-to-end learning. The proposed model employs a differentiable channel layer incorporating Rayleigh fading, power amplifier nonlinearities, and phase noise, enabling gradient-based optimization of complex constellation coordinates under strict average power constraints. The receiver utilizes a Real-Valued Feedforward Neural Network with Spatial Attention (RFNN-SA) to dynamically weight fading streams and mitigate channel estimation errors. Extensive simulations demonstrate that the proposed model achieves a 3.2 dB SNR gain at BER=10⁻³ for 16-QAM, 3.8 dB for 64-QAM, and 4.1 dB at BER=10⁻² for 256-QAM over MMSE detection at equivalent operating points. Under realistic CSI uncertainty, performance degrades by only 22%, compared to 52% for classical baselines. With a 0.95 ms physical‑layer detection inference latency, the proposed architecture provides a computationally efficient and impairment-resilient foundation for practical 6G physical-layer deployments.

Deep Learning Based Autoencoder Communication System

Modern wireless communication systems rely on independently designed transmitter and receiver components a modular approach that, while well- stablished, introduces significant complexity and limits adaptability to dynamic channel conditions. This project proposes and implements an intelligent end-to-end communication system using a deep learning-based Autoencoder architecture as an alternative to conventional modulation and coding designs. The proposed system models the transmitter as an encoder neural network and the receiver as a decoder neural network, with an Additive White Gaussian Noise (AWGN) layer representing the communication channel. By training the entire communication chain jointly through back propagation, the system automatically learns optimal signal encoding and decoding strategies without manual feature engineering or predefined modulation schemes. The implementation is built using Python, TensorFlow, NumPy, Matplotlib, and a Streamlit-based interactive web interface. Three optimization algorithms — Adam, RMSProp, and Gradient Descent — are evaluated across 10,000 training iterations with varying batch sizes (256, 512, 1024) and a training SNR of 7 dB. Adam achieved the fastest convergence with a final cross-entropy loss of 0.054, outperforming RMSProp (0.113) and Gradient Descent (0.421). Block Error Rate (BLER) analysis across SNR values from 0 to 14 dB confirms substantial improvement in communication reliability at higher SNR levels., versus 0.016 and 0.009 respectively.

Deep Learning–Aided Adaptive MIMO-OFDM Receiver with Real-Time Channel Estimation and Hardware-in-the-Loop Validation on Embedded Raspberry Pi Platforms

Accurate channel estimation remains a fundamental bottleneck in the performance of any coherent Multiple-Input Multiple-Output Orthogonal Frequency Division Multiplexing (MIMO-OFDM) receiver, particularly when the system is required to operate over a wide range of signal-to-noise ratios (SNRs) and under multipath fading. In this paper, we present the design, implementation, and experimental validation of a complete MIMO-OFDM transceiver running on two Raspberry Pi 4 single-board computers connected over a Wi-Fi link, in which the conventional Least Squares (LS) channel estimator is enhanced with a four-layer feedforward Deep Neural Network (DNN). The transmitter supports adaptive Quadrature Amplitude Modulation (QAM) schemes ranging from 16-QAM to 256-QAM, which can be selected by the user through a browser-based Flask dashboard. At the receiver, the bit error rate (BER) is computed in real time, while the active processing stage is displayed on an onboard 16×2 LCD. The DNN was trained offline using 100,000 synthetic complex channel samples and reduces the channel estimation mean squared error (MSE) from 0.1810 (LS) to 0.1676, corresponding to an improvement of approximately 0.33 dB in MSE. This improvement translates into an equivalent signal-to-noise ratio (SNR) gain of approximately 1.0–1.5 dB over the LS baseline in the 22–30 dB region of the BER-versus-SNR curve for 256-QAM. The end-to-end system reliably transmits text, grayscale images, and parallel text-and-image streams across the configured channel models. To the best of our knowledge, this work represents one of the first hardware-validated demonstrations of DNN-assisted OFDM channel estimation on a low-cost embedded platform.

Turbo receiver design for differentially encoded PSK in bursty impulsive noise channels

Learning Multitone Modulation for Nonlinear Dual-Unified SWIPT Receivers

Unified receivers (URs) enable simultaneous wireless information and power transfer (SWIPT) by reusing rectified signals for both information decoding and energy harvesting (EH). Dual UR-SWIPT architectures extend this concept by producing two rectified outputs of opposite polarity. However, existing dual UR modulation schemes rely on heuristic scalar decision rules, which can be suboptimal. This letter studies the design of multitone-based modulation schemes for nonlinear dual UR-SWIPT, with the objective of minimizing the symbol error rate (SER) under average transmit power and EH constraints. Since the resulting problem is analytically intractable, we propose an autoencoder-based design that embeds the nonlinear dual UR-SWIPT channel as a fixed differentiable layer. Our framework enables the joint learning of multitone parameters and a decoding rule that operates directly on the noisy two-dimensional output, while also supporting EH-aware training to satisfy an EH requirement. Numerical results show that the learned modulations achieve 3.5 dB gain at a given SER over the baseline schemes, while the EH-aware design ensures an improved rate-energy tradeoff. The learned waveforms reveal a clear structure and exploit the output geometry for improved symbol separation.

DirectDemodNet: An End-to-End Neural Demodulator for Polarization-Diverse Underwater Visible Light Communication

We demonstrate an underwater visible light communication system using a circularly polarized 520 nm laser transmitter, 32APSK modulation, and a polarization-diverse dual-aperture receiver. An end-to-end post-equalization network, DirectDemodNet, directly maps dual-polarization received waveforms to 32APSK symbol logits, replacing conventional Least Mean Square (LMS) + Volterra equalization. By combining waveform-difference features, dual-scale dilated temporal convolutions, and multi-period positional encoding, DirectDemodNet improves nonlinear compensation and branch fusion. Extensive evaluations are conducted across data rates from 7.5 to 13.75 Gbps over a 1.2 m static underwater channel. Experiments show that DirectDemodNet broadens the forward error correction compliant operating range and provides a maximum net transmission rate gain of 4.095 Gbps over LMS + Volterra at the 7% Hard-decision Forward Error Correction (HD-FEC) threshold.