Conditional Diffusion Model-Enabled Scenario-Specific Neural Receivers for Superimposed Pilot Schemes

TL;DR

This work tackles the data scarcity challenge for training neural receivers in superimposed pilot (SIP) wireless systems by introducing conditional diffusion models that generate scenario-specific channel data conditioned on user location and velocity. A dual U-Net DM architecture learns the distribution $\hat{q}(\mathbf{h}|\mathbf{c})$, while a consistency distillation technique accelerates sampling to near one-step generation, enabling practical deployment. The generated synthetic channels are used to augment neural receiver training, yielding substantial improvements over GAN-based and other baselines and approaching performance achieved with abundant true data. Evaluations on QuaDRiGa-simulated urban microcell channels demonstrate high-fidelity channel synthesis (low Wasserstein-2 distances, matching PDP/PAS) and significant BLER/throughput gains under SIP, highlighting the method's potential for scenario-aware, data-efficient neural communication systems.

Abstract

Neural receivers have demonstrated strong performance in wireless communication systems. However, their effectiveness typically depends on access to large-scale, scenario-specific channel data for training, which is often difficult to obtain in practice. Recently, generative artificial intelligence (AI) models, particularly diffusion models (DMs), have emerged as effective tools for synthesizing high-dimensional data. This paper presents a scenario-specific channel generation method based on conditional DMs, which accurately model channel distributions conditioned on user location and velocity information. The generated synthetic channel data are then employed for data augmentation to improve the training of a neural receiver designed for superimposed pilot-based transmission. Experimental results show that the proposed method generates high-fidelity channel samples and significantly enhances neural receiver performance in the target scenarios, outperforming conventional data augmentation and generative adversarial network-based techniques.

Figures (9)

• Figure 1: A typical urban cell containing multiple scenarios, where the uplink transmission adopts the SIP scheme. The channel sample of each scenario is displayed using grayscale map.
• Figure 2: Conditional generation process using the NN $\boldsymbol{\epsilon}_{\boldsymbol{\theta}}$. The channel samples $\hat{\mathbf{h}}_T$ and $\hat{\mathbf{h}}_{\epsilon}$ are illustrated as 2D grayscale images for clarity.
• Figure 3: U-Net architecture of the subnetwork.
• Figure 4: PDP and PAS of the channel samples from subregion R3.
• Figure 5: BLER performance and throughput of various approaches for training the neural receiver when $v = 24$ km/h.