⚡ electrical engineering

Performance Analysis of Cell-Free Massive MIMO under Imperfect LoS Phase Tracking

This paper proposes a Rician fading model with a phase-error penalty factor and a corresponding linear MMSE estimation framework to analyze and derive tractable spectral efficiency bounds for cell-free massive MIMO uplink networks under realistic imperfect LoS phase tracking conditions.

Original authors: Noor Ul Ain, Lorenzo Miretti, Renato L. G. Cavalcante, Slawomir Stanczak

Published 2026-02-27

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Noor Ul Ain, Lorenzo Miretti, Renato L. G. Cavalcante, Slawomir Stanczak

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

The Big Picture: The "Cell-Free" Orchestra

Imagine a future where your phone doesn't connect to just one cell tower. Instead, it connects to hundreds of tiny, distributed speakers (Access Points or APs) scattered all over a city, like streetlights or lampposts. This is called Cell-Free Massive MIMO.

In this system, all these speakers work together like a massive orchestra to play a single, perfect song (your data) to you. Because they are everywhere, there are no "dead zones" or "cell boundaries." Everyone gets a great signal.

The Problem: The "Tuning" Issue

For this orchestra to sound perfect, every speaker needs to know exactly when to play their note and how loud to play it. In technical terms, they need to know the phase of the signal.

The Ideal Scenario: Imagine a conductor who can hear every single instrument perfectly and tell them exactly when to play. This is "Perfect Phase Knowledge."
The Worst Scenario: Imagine the conductor is blindfolded and has no idea what the instruments are doing. They just guess. This is "Completely Unknown Phase."

The Real World Problem:
In reality, the conductor isn't blindfolded, but they aren't perfect either. Their hearing is slightly off due to:

Hardware glitches (the instruments are slightly out of tune).
Movement (the musicians are walking around).
Sync errors (their watches are slightly different).

This creates a "fuzzy" estimate. The paper asks: What happens to the music if the conductor is only mostly right, but not 100%?

The Paper's Solution: A New Way to Listen

Previous research mostly looked at the two extremes: "Perfect Conductor" or "Blindfolded Conductor." This paper fills the gap by studying the "Fuzzy Conductor."

Here is how they did it, broken down into three simple steps:

1. The "Rotated" Signal (The Model)

The authors created a new mathematical model. Imagine the signal as a spinning arrow.

In a perfect world, the arrow points exactly where it should.
In this paper, they admit the arrow is rotated slightly because the "conductor" (the system) made a small guess.
They also added a "penalty factor." Think of this as the signal getting a little bit quieter or "muffled" because the system isn't 100% sure of the direction.

2. The "Smart Guess" (Channel Estimation)

To fix the muffled signal, the system needs to estimate what the real signal looks like.

Old Way: If the system didn't know the phase, it would just ignore the direction and guess the average (which is often wrong).
New Way: The authors invented a Linear MMSE Estimator.
- Analogy: Imagine you are trying to hear a friend in a noisy room. If you know their voice is slightly distorted by a specific echo, you can "undo" that echo in your brain to hear them clearly.
- This new estimator uses the "fuzzy" information the system does have to make a much better guess than just ignoring the phase entirely. It bridges the gap between "Perfect" and "Blind."

3. The "Virtual Uplink" (Beamforming)

Once the system guesses the signal, it needs to combine the audio from all 100 speakers to send it to you. This is called Beamforming.

Calculating the perfect combination is mathematically impossible when the signal is "fuzzy" and "non-Gaussian" (a fancy way of saying the math gets messy and breaks standard rules).
The Trick: The authors created a "Virtual Uplink."
- Analogy: Imagine you are trying to solve a puzzle with missing pieces. Instead of giving up, you pretend the missing pieces are just "static noise." You solve the puzzle assuming the noise is there, and then you use that solution on the real puzzle.
- By treating the uncertainty as extra noise, they can use standard, easy math to design the beamformers (the strategy for combining the speakers).

The Results: Why It Matters

The authors ran simulations (computer tests) to see how well this new method works.

The Finding: Even if the system only has a rough idea of the phase (like a 15-degree error), the performance is much better than having no idea at all.
The Analogy: It's like driving in fog. If you have no idea where the road is, you crash. If you have a GPS that is slightly off (but still points generally the right way), you can still drive safely and fast. You don't need perfect vision to get to your destination; you just need some vision.
Centralized vs. Distributed:
- Centralized: One big brain controls all speakers. It handles the "fog" very well.
- Distributed: Each speaker has its own small brain. It struggles a bit more with the fog, but the new method still helps it perform surprisingly well.

The Takeaway

This paper is a roadmap for 6G networks. It tells engineers:

"Don't wait for perfect hardware that never makes mistakes. That doesn't exist. Instead, build systems that are smart enough to work even when the 'conductor' is slightly out of tune. Even a little bit of knowledge about the signal direction is worth a lot."

By using their new math, future cell-free networks will be more robust, efficient, and capable of delivering high-speed internet even when things aren't perfect.

1. Problem Statement

Cell-free massive MIMO (mMIMO) is a leading candidate for 6G networks, utilizing distributed access points (APs) to serve users without cell boundaries. In dense deployments, Line-of-Sight (LoS) propagation is highly probable, significantly impacting system performance.

However, a critical challenge arises in acquiring reliable Channel State Information (CSI) when LoS phases fluctuate due to oscillator drift, user mobility, and synchronization errors. Existing literature typically analyzes these systems under two extreme, unrealistic assumptions:

Perfect Phase Knowledge: Assumes LoS phases are known exactly, ignoring estimation errors.
Completely Unknown Phases: Assumes no phase information is available, discarding valuable deterministic LoS information.

The Gap: Real-world systems operate in an intermediate regime where LoS phases are tracked but with residual uncertainty. Current frameworks fail to model this "imperfect tracking" scenario, particularly regarding how partial phase knowledge affects channel estimation and beamforming in non-Gaussian settings.

2. Methodology

The authors propose a comprehensive framework to bridge the gap between idealized assumptions and practical limitations.

A. System and Channel Model

Architecture: A user-centric cell-free mMIMO network with $L$ APs (each with $N$ antennas) serving $K$ single-antenna users.
Rician Fading with Imperfect Tracking: The channel $h_{k,l}$ is modeled as:
$h_{k,l} = \bar{h}_{k,l}e^{j\theta_{k,l}} + \tilde{h}_{k,l}$
Where $\bar{h}_{k,l}$ is the deterministic LoS signature, $\tilde{h}_{k,l}$ is the NLoS component, and $\theta_{k,l}$ is the LoS phase.
Imperfect Estimation: The system observes an estimate $\hat{\theta}_{k,l} = \theta_{k,l} - \varepsilon_{k,l}$ , where the error $\varepsilon_{k,l}$ is modeled as a uniform random variable $\varepsilon_{k,l} \sim U[-\delta, \delta]$ .
Phase Penalty Factor: The authors derive a deterministic factor $\rho_{error} = \text{sinc}(\delta)$ (for $\delta > 0$ ) that attenuates the magnitude of the estimated LoS component based on tracking accuracy.

B. Channel Estimation

Conditionally Linear MMSE Estimator: Since the channel and pilot signals are not jointly Gaussian under imperfect phase knowledge, the standard Bayesian MMSE estimator is intractable. The authors derive a Conditionally Linear MMSE (CL-MMSE) estimator.
Unification: This estimator acts as a unified framework:
- As $\delta \to 0$ (perfect tracking), it reduces to the standard Bayesian MMSE estimator.
- As $\delta \to \pi$ (unknown phases), it reduces to the zero-mean model estimator.
Error Characterization: The estimator explicitly accounts for the covariance of the estimation error, which depends on the phase uncertainty $\delta$ .

C. Beamforming Design via Virtual Uplink

To handle the non-Gaussian nature of the channel for beamforming optimization, the authors introduce a Virtual Uplink Model:

Concept: The true channel is replaced by its estimate $\hat{h}_{k,l}$ , and the estimation error covariance is embedded into the noise term.
Benefit: This transforms the problem into a tractable Gaussian setting, allowing the derivation of closed-form solutions for:
1. Centralized MMSE Beamforming: The CPU has access to global estimates.
2. Distributed MMSE Beamforming: Each AP uses local estimates and global phase statistics.
Evaluation: These virtual beamformers are applied to the actual physical channel model to compute rigorous Spectral Efficiency (SE) bounds.

D. Performance Metrics

The paper evaluates performance using two standard bounds on achievable SE:

Use-and-Then-Forget (UatF): A lower bound assuming the decoder uses average channel statistics.
Optimistic Ergodic Rate (OER): An upper bound (genie-aided) assuming the decoder knows the instantaneous effective channel.

3. Key Contributions

Realistic Channel Modeling: Introduction of a Rician fading model that explicitly incorporates imperfect LoS phase tracking via a bounded uniform error distribution and a deterministic phase-noise penalty factor.
Unified Estimator: Derivation of a linear MMSE channel estimator that continuously interpolates between the "perfect knowledge" and "zero-mean" extremes, providing a mathematically rigorous solution for intermediate tracking scenarios.
Virtual Uplink Framework: Development of a virtual uplink model that enables the derivation of tractable centralized and distributed MMSE beamformers in non-Gaussian settings, embedding estimation errors into the noise covariance.
Comprehensive Analysis: Providing the first rigorous performance benchmarks for cell-free mMIMO under partial LoS phase knowledge, bridging the gap between theoretical ideals and hardware limitations.

4. Numerical Results

Simulations were conducted with 100 APs, 40 users, and a Rician factor $\kappa = 5$ .

Impact of Phase Uncertainty ( $\delta$ ):
- Even coarse phase tracking (e.g., $\delta = 15^\circ$ ) yields substantial performance gains compared to the "no phase knowledge" scenario.
- Performance degrades gracefully as $\delta$ increases. Moderate errors result in distributions close to the ideal case.
- Significant degradation only occurs at large phase uncertainties (e.g., $\delta = 90^\circ$ or $180^\circ$ ).
Centralized vs. Distributed:
- Centralized beamforming is more robust to severe phase errors.
- However, distributed beamforming still benefits significantly from partial phase knowledge.
- At high Rician factors ( $\kappa = 100$ ) with accurate tracking, distributed beamforming approaches the performance of centralized beamforming.
Bound Tightness: The gap between the UatF (lower) and OER (upper) bounds is small, especially when LoS phases are well-tracked, indicating that the lack of instantaneous CSI at the decoder causes negligible loss in high-LoS regimes.

5. Significance

This work is significant for the design of 6G cell-free networks for several reasons:

Practicality: It moves beyond idealized assumptions, offering design insights for systems where hardware impairments and mobility prevent perfect phase synchronization.
Design Guidelines: It demonstrates that investing in phase tracking mechanisms (even if imperfect) is highly beneficial. Systems should not discard LoS phase information even if it is noisy.
Framework Utility: The proposed "virtual uplink" approach provides a generalizable tool for designing beamformers in other non-Gaussian channel estimation scenarios.
Benchmarking: It establishes rigorous performance baselines for future research, clarifying the trade-offs between tracking accuracy and system complexity.

In conclusion, the paper proves that partial knowledge of LoS phases is a valuable asset in cell-free massive MIMO, and the proposed framework allows network designers to optimize beamforming strategies under realistic tracking constraints.