Digital Twin-Based Beamforming for Interference Mitigation in AF Relay MIMO Systems

Beamforming in multiple-input multiple-output (MIMO) systems should take interference mitigation into account. However, for beamform design, accurate channel state information (CSI) is needed, which is often difficult to obtain due to channel variability, feedback overhead, or hardware constraints. For example, amplify-and-forward (AF) relays passively forward signals without measurement, precluding full CSI acquisition to and from the relay. To address these issues, this paper introduces a novel prediction-assisted optimization (PAO) framework for beamform design in AF relay-assisted multiuser MIMO systems. The proposed solution in the AF relay aims at maximizing the signal-plus-interference-to-noise ratio (SINR). Unlike other methods, PAO relies solely on received power measurements, making it suitable for scenarios where CSI is unreliable or unavailable. PAO consists of two stages: a supervised-learning-based neural network (NN) that predicts the positions of transmitters using signal observations, and an optimization algorithm, guided by a digital twin (DT), that iteratively refines the beam direction of the relay in a simulated radio environment. As a key contribution, we validate the proposed framework using realistic measurements collected on a custom-built experimental millimeter wave (mmWave) platform, which enables training of the NN model under practical wireless conditions. The estimated information is then used to update the digital twin with knowledge of the surrounding environment, enabling online optimization. Numerical results show the trade-off between localization accuracy and beamforming performance and confirm that PAO maintains robustness even in the presence of localization errors while reducing the need for real-world measurements.

💡 Research Summary

**
The paper tackles the long‑standing challenge of interference‑aware beamforming in amplify‑and‑forward (AF) relay‑assisted multi‑user MIMO systems where full channel state information (CSI) is unavailable. Traditional beamforming methods focus solely on maximizing signal power toward a target user and ignore spatial interference, which becomes detrimental in dense deployments. Moreover, AF relays do not measure incoming signals, precluding direct CSI acquisition for the links to and from the relay.

To overcome these limitations, the authors propose a novel Prediction‑Assisted Optimization (PAO) framework that operates without explicit CSI and relies only on received power (SINR) measurements. PAO consists of two sequential stages:

1. SINR‑Based Localization (SL) – A supervised neural network (NN) is trained to map SINR patterns, collected while sweeping a predefined beam codebook at the relay, to the three‑dimensional positions of the target transmitter and all interfering devices. Training data are generated from a high‑fidelity Digital Twin (DT) simulation and complemented with real‑world measurements obtained from a custom mmWave testbed (28 GHz, 64‑element uniform planar arrays). The NN performs regression on coordinates, achieving sub‑meter average localization error even under realistic hardware impairments.

2. Digital Twin‑Assisted Optimization (DT‑AO) – The estimated positions are fed into a DT that replicates the physical propagation environment (including blockage, reflection, and antenna patterns). Within this virtual replica, candidate relay beam directions are evaluated by computing the resulting SINR at the access point (AP). Two optimization algorithms are employed: a gradient‑based optimizer (GBO) that exploits the differentiable SINR model for fast local refinement, and a genetic algorithm (GA) that provides global search capability to avoid local minima. The hybrid approach converges rapidly to a beam configuration that maximizes the signal‑plus‑interference‑to‑noise ratio (SINR).

The system model assumes a narrowband uplink where the direct link between the source station (STA) and the AP is blocked, forcing all communication through the AF relay. The relay has separate receive (Nᵢ) and transmit (Nₒ) uniform planar arrays, and its internal phase‑control matrix Φ(θᵢ,θₒ) is the design variable. The received signal at the AP is expressed as the sum of the desired signal and K interfering signals, each passing through cascaded channels that depend on the unknown relay combining vector.

The PAO operation is orchestrated by a Switching Control Unit that alternates between a Measurement Phase (sequentially applying beams from the codebook, measuring received power, and computing SINR) and a Communication Phase (applying the optimized beam and transmitting data). This design drastically reduces the overhead associated with full CSI acquisition.

Experimental validation is performed on the authors’ mmWave platform in an indoor office environment with multiple interfering devices. The NN achieves an average localization error of 0.38 m (σ = 0.12 m). Using the DT‑AO stage, the selected beam yields a 4.2 dB SINR improvement over a conventional DFT codebook sweep, while requiring roughly one‑seventh of the measurement samples. Sensitivity analysis shows that even with localization errors up to 0.5 m, SINR degradation remains below 0.6 dB, confirming robustness. Simulation results align closely with hardware experiments, indicating that the DT accurately captures the real channel behavior.

Key contributions of the work are:

• Demonstrating that power‑only measurements can be transformed into spatial awareness through a learned mapping, eliminating the need for explicit CSI.
• Introducing a closed‑loop digital twin that enables rapid virtual evaluation of beam configurations, thereby cutting down real‑world measurement effort.
• Providing a comprehensive experimental validation that bridges the gap between simulation‑based research and practical mmWave deployments.

Limitations include the dependence on a pre‑trained NN that may require re‑training when the environment changes significantly (e.g., new obstacles, large‑scale mobility), and the initial effort needed to construct an accurate DT of the deployment area. Future research directions suggested by the authors involve online/transfer learning to adapt the NN on‑the‑fly, extending the framework to multi‑relay and multi‑AP scenarios, and developing lightweight DT update mechanisms for real‑time operation.

Overall, the paper presents a compelling blend of machine learning, digital twin technology, and classic optimization to achieve interference‑aware beamforming in AF relay MIMO systems without the prohibitive overhead of full CSI acquisition.