2026· IEEE Journal of Selected Topics in Electromagnetics, Antennas and Propagation· Vol 2, pp. 477-488· 0 citations· 23 references
TL;DR
This work presents a direct-learning architecture of a neural network (NN) digital predistortion (DPD) linearizer for a multiple-input multiple-output (MIMO) system while maintaining low complexity compared to a single-input single-output (SISO) system.
Abstract
This work presents a direct-learning architecture of a neural network (NN) digital predistortion (DPD) linearizer for a multiple-input multiple-output (MIMO) system. The array-propagated neural network (APNN) technique accounts for nonlinear cross-modulation distortion due to antenna coupling in MIMO systems while maintaining low complexity compared to a single-input single-output (SISO) system. The behavior of the MIMO antenna array is modeled using a NN to train the DPD NNs through memory backpropagation. The linearization methods were evaluated across two test cases using 65 nm complementary metal-oxide-semiconductor (CMOS) power amplifiers (PAs). The first case utilized two PAs under a worst case 4 dB coupling scenario, transmitting 20 MHz 802.11ac Wi-Fi signals with a 10 dB peak-to-average power ratio (PAPR) at a 1 GHz center frequency and an average power of 8.5 dBm. Results demonstrate an error vector magnitude (EVM) improvement from −15.1 to −40.6 dB and from −16.2 to −39.4 dB for the two chains, respectively, with computational complexity increases of 35% and 32% compared to the SISO model. The second test case employed a three-antenna array, in which mutual coupling was introduced according to element proximity. In this scenario, 80 MHz 802.11ax Wi-Fi signals with an 11 dB PAPR were transmitted at an average power of 9 dBm. The proposed approach yielded EVM improvements from −22.1 to −39.7 dB, from −19.1 to −40.4 dB, and from −20.5 to −40.2 dB. In this configuration, only cross samples from adjacent channels were included, demonstrating the scalability of the APNN method.
Due to the continuous increase in communication bandwidth and the use of highly efficient yet nonlinear power amplifiers, Digital Predistortion (DPD) algorithms are becoming increasingly complex. In particular, neural network (NN) based DPD approaches using Phase-Normalized NN architectures often incur substantially higher computational costs than widely deployed polynomial-based methods, such as the Memory Polynomial (MP) and Generalized Memory Polynomial (GMP) models. To bridge this gap between research performance and practical implementation, we propose a low-complexity Feature Selection NN DPD architecture. The proposed method employs an offline feature-engineering pipeline based on the Least Absolute Shrinkage and Selection Operator (LASSO) and the Minimum Redundancy Maximum Relevance (MRMR) algorithm to construct a compact and informative input representation. Using measured wideband FR3 power amplifier datasets that are publicly released with this work, we demonstrate up to 30% reduction in computational complexity while maintaining comparable linearization performance.
Cel Thys, Rodney Martinez Alonso, A. Alsarraf et al.· 0 citations
Wireless physical neural networks (WPNNs) embed neural computation directly into analog hardware, offering lower energy consumption and latency than conventional digital implementations. In this paper, we propose a deep WPNN in which nonlinear activations are realized by a multi-hop multiple-input multiple-output (MIMO) relay network, in which each relay implements a trainable complex linear gain and bias, followed by the power amplifier's intrinsic nonlinearity acting as an activation function. The cascade of multiple relays therefore realizes an over-the-air fully connected network whose parameters can be trained end-to-end. We develop two transceiver designs for different channel state information (CSI) availability scenarios: a least squares (LS)-based scheme requiring only receiver-side CSI, and a singular-value-decomposition (SVD)-based scheme requiring both transmitter-side and receiver-side CSI. Simulation results show that the proposed architecture enables accurate over-the-air inference for image classification. In particular, the results highlight the advantage of exploiting hardware nonlinearity for enhanced inference capability.
In this study, we examine and contrast the effectiveness of different artificial neural network (ANN) topologies for power amplifier (PA) digital pre-distortion (DPD). In particular, we investigate long short-term memory (LSTM) networks, gated recurrent units (GRU), recurrent neural networks (RNN), con-volutional neural networks (CNN), and fully connected neural networks (DNN). For training and assessment, a dataset comprising measured input and output signals from a commercial NXP Doherty PA working in the 3.6–3.8 GHz region with a 16-QAM OFDM signal is utilised. Normalised mean squared error (NMSE), adjacent channel power ratio (ACPR), and model complexity are used to evaluate the models. Simulation results show that while CNNs offer a favorable trade-off between linearization performance and model complexity, GRU and LSTM architectures achieve the best overall NMSE and ACPR improvements, albeit with a higher number of parameters. Power spectral density and AM/AM characteristic analyses further confirm the superior linearization performance achieved using recurrent gated structures.
Hafsa Laakouri, M. Ouadefli, A. Tribak et al.· EPJ Web of Conferences· 0 citations
This article proposes a novel, low-rate digital predistortion (DPD) for sub-6-GHz 400-MHz modulated signal transmission. In this model, a low-rate input signal is concurrently fed into multiple parallel cascaded model branches to generate the low-rate predistorted signal. Each branch contains: a dual-function finite impulse response (FIR) filter for interpolation and memory effect compensation, multiple parallel Volterra series-based submodels that are selectively activated based on the current input state and are optimized in complexity by a pruning algorithm, and low-rate filters. Additionally, a model parameter extraction method is introduced for this model. Based on this model, a high-precision DPD system can be obtained with low system processing rates. Experimental validation was performed using 5G new radio (NR) signals with 400-MHz modulation bandwidth on a 3.4–3.8-GHz gallium nitride (GaN) Doherty power amplifier (PA) at low system processing rates of 983 and 491 MSPS, respectively. Measurement results demonstrated that the proposed technique achieves good performance with lower complexity than existing models, with a normalized mean-square error (NMSE) and an adjacent channel power ratio (ACPR) of approximately −41 dB and −46 dBc at 983 MSPS, respectively, and an NMSE of approximately −40 dB at 491 MSPS.
Xiaoyu Lu, Tong Tong, Yucheng Yu et al.· IEEE transactions on microwa...· 0 citations
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.
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.
Twinkle Srusti J K, Padmajadevi G, D. K C et al.· 2026 6th International Confe...· 0 citations