⚡ electrical engineering

Over-the-Air Beamforming Design for Full-Duplex Cell-Free Massive MIMO Systems

This paper proposes a fully distributed over-the-air beamforming design for full-duplex cell-free massive MIMO systems that mitigates pilot contamination and interference through a pilot-domain projection technique and best-response updates, achieving faster convergence and higher sum rates compared to existing methods.

Original authors: Bikshapathi Gouda, Antti Tölli

Published 2026-05-11

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Bikshapathi Gouda, Antti Tölli

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a massive wireless network as a giant, busy town square where many small radio towers (Access Points) are trying to talk to and listen from many people (User Devices) all at the same time.

In older systems, these towers had to take turns: one group would talk while another listened, like a walkie-talkie where you have to say "over" before you can speak again. This paper proposes a smarter, faster way: Full-Duplex. This means the towers can talk and listen simultaneously, like a person having a face-to-face conversation where both parties speak and listen at the exact same time. This doubles the potential speed of the network.

However, doing this in a "Cell-Free" system (where towers are scattered everywhere rather than in fixed cells) creates a chaotic noise problem. Here is the breakdown of the challenges and the solution, using simple analogies:

The Problem: The "Echo Chamber" and the "Cross-Talk"

When everyone talks and listens at once, three types of noise get in the way:

Self-Interference: A tower hears its own voice so loudly it drowns out the person it's trying to listen to.
Tower-to-Tower Noise: One tower's loud voice leaks into a neighbor's microphone.
User-to-User Noise (The Big One): This is the paper's main focus. Because everyone is transmitting at once, the person you are trying to talk to is also hearing the person standing next to them. It's like trying to have a private conversation in a crowded room where everyone is shouting at once; your friend's voice is mixed with the person next to them.

In previous systems, the towers would take turns sending "test signals" (pilots) to figure out how to tune their microphones and speakers to cancel out the noise. But in this new "talk-and-listen-at-once" system, those test signals get scrambled. The "User-to-User" noise contaminates the test signals, making it impossible for the towers to figure out exactly how to tune themselves. It's like trying to tune a radio while someone is playing a drum solo right next to you; you can't hear the station clearly.

The Solution: The "Noise-Canceling Projection"

The authors developed a clever new method to clean up these test signals without needing a central boss to tell everyone what to do. They call it Over-the-Air (OTA) Beamforming.

Here is how their solution works, step-by-step:

The "Projection" Trick:
Imagine the test signals are a messy pile of colored sand. The "User-to-User" noise is a specific color of sand that ruins the mix. The authors invented a mathematical "sieve" (a projection) that the users apply to their signals before sending them back to the towers.
- The Metaphor: Think of the users as chefs. They are sending a soup back to the kitchen (the towers) to taste. But the soup has a weird spice (noise) in it. Before sending it back, the chefs use a special strainer that only lets the "good" ingredients through and blocks the "weird spice." This allows the towers to taste the pure flavor of the connection without the noise.
The "Best-Response" Dance:
Because the system is so complex, if everyone tries to fix their settings all at once, they might get stuck in a loop, constantly changing their minds.
- The Metaphor: Imagine a group of dancers trying to synchronize. Instead of everyone guessing the next move at the same time, they take turns. One dancer adjusts their step based on where everyone else is standing, then the next one does the same. This "Best-Response" method ensures they all eventually find a perfect, stable rhythm without tripping over each other.

The Results: Faster and Clearer

The paper tested this new method against older ways of doing things (where towers take turns or only look at their immediate neighbors).

Faster Convergence: The new method figured out the perfect settings in fewer steps (iterations). It's like finding the right combination to a lock in fewer tries.
Higher Speed: The total amount of data the network could handle (sum-rate) was significantly higher.
Handling the Loud Neighbors: The system was especially good at helping users who were surrounded by loud interference. In the old systems, these users would get stuck with terrible speeds; in this new system, they got much closer to the speed of everyone else.

Summary

In short, this paper teaches a network of scattered towers how to talk and listen at the exact same time without getting confused by the noise of their own voices or their neighbors. They did this by inventing a way to "filter out" the confusing noise from their test signals and by having the devices take turns adjusting their settings to find the perfect balance. The result is a faster, more efficient wireless network that works even when the environment is very noisy.

Technical Summary: Over-the-Air Beamforming Design for Full-Duplex Cell-Free Massive MIMO Systems

Problem Statement
This paper addresses the challenge of designing distributed beamforming for Full-Duplex (FD) Cell-Free Massive Multiple-Input Multiple-Output (MIMO) systems. While FD operation at Access Points (APs) allows simultaneous Uplink (UL) and Downlink (DL) transmission on the same time-frequency resource, it introduces complex interference components absent in Half-Duplex (HD) systems. Specifically, FD operation generates residual self-interference (SI) at APs, AP-to-AP coupling, and, critically, UE-to-UE cross-link interference.

Existing Over-the-Air (OTA) distributed beamforming frameworks, developed for HD systems, rely on iterative bi-directional training (IBT) to reconstruct channel cross-terms without explicit Channel State Information (CSI) exchange. However, in FD systems, the simultaneous transmission of UL and DL pilots causes "UE-to-UE pilot leakage." This leakage contaminates the signals received at User Equipments (UEs), corrupting the reconstruction of the cross-term channel information necessary for AP-specific beamforming updates. Consequently, standard OTA methods fail to account for the coupled interference between UL and DL users, leading to suboptimal performance.

Methodology
The authors propose a fully distributed beamforming framework based on iterative UL and DL pilot signaling under a joint UL and DL sum Mean-Squared Error (MSE) criterion. The methodology consists of two primary components:

Perfect CSI Benchmark: The authors first derive the optimality conditions for distributed beamformers (DL precoders, DL combiners, UL precoders, and UL combiners) assuming perfect global CSI. They formulate a non-convex joint optimization problem and solve it using an alternating optimization framework. To ensure stability under strong UE-to-UE interference, they employ "best-response" updates, where each variable is optimized while others are fixed.
OTA Implementation with Pilot Projection: To implement this design without explicit CSI exchange, the authors adapt the IBT framework for FD operation. The protocol involves three time slots:
- Slot 1 & 2: Standard pilot transmission where APs and UEs exchange precoded pilots to estimate direct channels and update local combiners/precoders.
- Slot 3 (The Novel Contribution): To address pilot leakage, the authors introduce a pilot-domain projection mechanism. Before retransmitting received pilot signals to the APs for cross-term reconstruction, UEs project the received signals onto the subspace of their respective pilot sequences (DL UEs project onto the DL pilot subspace $P$ , and UL UEs onto the UL pilot subspace $Q$ ). This projection mathematically suppresses the interfering UE-to-UE components while preserving the desired signal subspace. This enables APs to accurately reconstruct the cross-terms ( $\xi$ ) required for updating DL precoders and UL combiners, despite the simultaneous transmission environment.

The final beamformers are updated using the best-response method within the alternating optimization loop to ensure convergence.

Key Contributions

FD-Specific OTA Framework: The paper extends OTA distributed beamforming to FD cell-free massive MIMO, explicitly modeling and addressing the unique interference landscape (SI, AP coupling, and UE-to-UE cross-link interference).
Pilot-Domain Projection: A novel signal processing technique is introduced to mitigate UE-to-UE pilot leakage. By projecting received signals onto pilot subspaces prior to retransmission, the method enables accurate cross-term reconstruction at APs, which is otherwise impossible in simultaneous FD pilot transmission.
Joint Optimization: The design optimizes UL and DL beamformers jointly under a sum-MSE criterion, rather than treating them separately, thereby capturing the coupling effects inherent in FD operation.
Convergence Stability: The integration of best-response updates is shown to improve convergence behavior in the presence of strong interference.

Numerical Results
Numerical simulations were conducted on a network with 16 APs (4 antennas each) and 32 UEs (16 DL, 16 UL, 4 antennas each). The proposed scheme was compared against:

Separate OTA Training: Where UL and DL are trained independently (ignoring cross-link interference).
Local MMSE: A baseline using only local CSI without cross-term reconstruction.
HD Scenario: A baseline where UL and DL operate on orthogonal resources.

The results demonstrate that the proposed FD-OTA scheme achieves:

Faster Convergence: The algorithm converges more rapidly than baseline schemes.
Higher Effective Sum Rate: The proposed method yields a higher sum of UL and DL rates compared to Separate OTA and Local MMSE designs.
Robustness to Interference: The performance gains are particularly significant for UEs experiencing strong interference. While the interference-aware UL precoder design may slightly reduce UL rates compared to Separate OTA (due to interference mitigation), the coupled updates substantially improve DL rates, leading to a net gain in total system throughput.
Resource Efficiency: Even with limited total resources, the proposed method outperforms baselines, and the performance gap widens as available resources increase because the training overhead becomes negligible relative to the gains in spectral efficiency.

Significance
The paper claims that its primary significance lies in enabling the practical deployment of FD cell-free massive MIMO systems. By solving the specific problem of pilot leakage in distributed OTA training, the framework allows APs to fully exploit the spectral efficiency gains of FD operation without requiring centralized CSI exchange or extensive fronthaul signaling. The work establishes that explicit accounting for UE-to-UE cross-link interference and the use of pilot-domain projection are essential for realizing the theoretical benefits of FD in distributed wireless networks.

The Problem: The "Echo Chamber" and the "Cross-Talk"

The Solution: The "Noise-Canceling Projection"

The Results: Faster and Clearer

Summary

More like this