Channel-Chart-Driven Antenna Activation for Energy-Efficient Cell-Free Massive MIMO

arXiv:2611.16996 · eess.SP · Submitted 2026-11-12 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Electrical Engineering. The day in electrical engineering: signals, systems and the hardware that carries them.

Rosa: We're looking at the paper: "Channel-Chart-Driven Antenna Activation for Energy-Efficient Cell-Free Massive MIMO".

Dev: Here's the gist of the paper. Cell-free massive MIMO can provide uniformly high system performance across the coverage area through distributed cooperation However, keeping all antennas active leads to high circuit and fronthaul power consumption.

Rosa: This paper studies energy-efficient uplink reception through antenna activation, where only a subset of distributed antennas is selected to serve the users.

Dev: To enable network-wide activation with low signaling overhead, each access point (AP) maps its locally estimated channel state information (CSI) into a lowdimensional channel-chart (CC) embedding and forwards this representation to the central processing unit (CPU) instead of the full CSI.

Rosa: Based on the received embeddings, we formulate a binary antenna-activation problem for network EE maximization and propose three low-overhead strategies.

Paper summary: Rosa: So we're looking at this paper, "Channel-Chart–Driven Antenna Activation for Energy-Efficient Cell-Free Massive MIMO". The main idea here is that cell free massive MIMO can give you really good performance everywhere by having distributed cooperation across the coverage area. But activating every single antenna uses a ton of circuit power and fronthaul power.

Dev: Yeah, so the core problem they're tackling is how to get high system performance while keeping energy use down, specifically by only picking a subset of those distributed antennas to serve the users. That's the thesis here > Taro I wonder if this stuff actually works outside of a controlled lab environment, Rosa. How robust is this antenna activation strategy when you introduce real-world fading and unpredictable interference?

Rosa: Exactly, Taro. The paper says they solve this by having each access point map its local channel state information into something called a lowdimensional channel chart embedding and send that instead of the full CSI to the central processing unit. They then use those embeddings to set up a problem for maximizing energy efficiency across the entire network > Dev That sounds like they are trying to trade off some complexity for less signaling overhead, which is always a big win in these massive systems.

Taro: But what happens when you use that embedding instead of the actual instantaneous CSI? Does it lose accuracy, and if so, how much does that accuracy loss affect the final decision on which antennas to turn on > Rosa That's what they explore with their three proposed strategies. They aren't just guessing; they are systematically comparing different ways to use those embeddings to make the activation decision > Dev Which brings us into the specific methods they propose for this antenna activation problem in equation four, which is a pretty complex fractional program > Taro So we have these three strategies: direct embedding domain activation, nearest chart point CSI approximation, and decoder based CSI reconstruction. How do those map onto what's actually happening on the channel at the user end > Rosa Well, the paper says they are testing them to see which one gives them the best trade-off among activation performance, computational complexity required to compute that activation vector alpha, and the signaling overhead toward the CPU > Dev That makes sense. So if we're talking about real-world deployment where you have limited fronthaul bandwidth, which strategy would you lean toward initially > Taro I think the decoder based reconstruction sounds like it might offer better accuracy because it tries to reconstruct the instantaneous channel state information using a neural encoder decoder, even though that adds computational load > Rosa The paper shows that they test all three against each other to figure out what's best for different priorities, and their results show substantial energy efficiency gains compared to just keeping everything active > Dev So the initial results suggest this approach is viable because it cuts down on signaling while still getting good performance, even though there's a trade-off in complexity depending on which strategy you choose >

Conclusion: Rosa: So we're wrapping up this look at that paper, "Channel-Chart–Driven Antenna Activation for Energy-Efficient Cell-Free Massive MIMO." Essentially, they show how you can use these channel charts to make the whole system much more energy efficient by being smart about which antennas to turn on.

Dev: It really boils down to using those compact embeddings instead of sending all the raw channel information back and forth, which cuts down on signaling and processing power across the network.

Taro: I mean, it’s interesting because they’re replacing that huge amount of data exchange with something much smaller that still lets you do a centralized activation problem.

Rosa: Exactly. The authors propose three different ways to use those embeddings—direct embedding domain activation, nearest chart-point CSI approximation, and decoder based CSI reconstruction—and they test them all against each other to see which one works best for different goals.

Dev: And the results show that these methods give you big energy efficiency gains over just turning on every single antenna, while only causing a small drop in how well the users actually connect.

Taro: The paper confirms that these channel chart representations can indeed support centralized energy efficient antenna activation, and it shows that picking the right strategy depends entirely on whether your main worry is minimizing signaling overhead or keeping your online processing complexity low.

Rosa: That's the big implication here—it means we can scale up these massive cell free systems without immediately drowning in power consumption and fronthaul traffic.

Dev: It opens up a path for deploying more robust massive MIMO setups where energy management becomes an integrated part of the communication loop, not just an afterthought.

Taro: But you still have to worry about the accuracy of that channel chart embedding when things get really messy out in the field, which is what we need to look at next.

Yara Kafa, Luc Le Magoarou, Joumana Farah, Matthieu Crussiere

University of Rennes, INSA Rennes, CNRS, IETR-UMR 6164

eess.SP

Submitted: 2026-11-12

Updated: 2026-11-12

Journal ref: WiMob 2026 - The 22nd International Conference on Wireless and Mobile Computing, Networking and Communications., Oct 2026, Avignon (FR), France

Project page: https://nvlabs.github.io/sionna

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 77/100

The gist: Cell-free massive MIMO can provide uniformly high system performance across the coverage area through distributed cooperation However…

Key concepts

Cell-Free Massive MIMO (CF-mMIMO)
This system uses many distributed antennas across a coverage area to serve users. The goal is uniform high performance by cooperating among the distributed nodes, but this requires managing the activation of all antennas.
Channel-Chart Embedding (CC)
Instead of sending the entire instantaneous channel state information (CSI), each access point compresses its local channel knowledge into a low-dimensional representation called a channel chart embedding. This compact data is sent to the central unit instead of the full CSI.
Antenna Activation Strategies
These are three methods—Direct Embedding-Domain, Nearest Chart-Point, and Decoder-Based—that decide which antennas to activate based only on the compact CC embeddings. These strategies aim to maximize energy efficiency by making intelligent antenna selection decisions with minimal overhead.

Terminology

Summary

Cell-free massive MIMO can provide uniformly high system performance across the coverage area through distributed cooperation However, keeping all antennas active leads to high circuit and fronthaul power consumption. p. 0 This paper studies energy-efficient uplink reception through antenna activation, where only a subset of distributed antennas is selected to serve the users. p. 0 To enable network-wide activation with low signaling overhead, each access point (AP) maps its locally estimated channel state information (CSI) into a lowdimensional channel-chart (CC) embedding and forwards this representation to the central processing unit (CPU) instead of the full CSI. p. 0 Based on the received embeddings, we formulate a binary antenna-activation problem for network EE maximization and propose three low-overhead strategies. p. 0

System Model Description

The system considers the uplink of a user-centric CF-mMIMO system with L distributed APs serving K single-antenna users (UEs). Each AP l ∈ [1, L] is equipped with N antennas, yielding a total of M = LN antennas, and is connected to a CPU via fronthaul links. p. 1 The effective uplink channel between user k and AP l, incorporating the spatial selectivity induced by the directive three-dimensional antenna radiation patterns, is denoted by hkl ∈ C N. p. 1 User-centric dynamic cooperation clustering (DCC) at the AP level is adopted to limit signaling and processing complexity. p. 1

Antenna Activation Strategies

The paper proposes three low-overhead AAc strategies to solve the EE-maximizing AAc problem in (4). These strategies exploit the channel-chart embeddings forwarded by the APs as input to the AAc procedure. p. 2

The three proposed strategies are:

  1. Direct Embedding-Domain Antenna Activation (DD): This strategy performs AAc in the CC domain without reconstructing instantaneous CSI at the CPU. It uses embeddings to define a proxy EE metric that guides antenna selection with minimal computational and signaling overhead. p. 2

  2. Nearest Chart-Point CSI Approximation (NN): This strategy uses CC embeddings to obtain approximate instantaneous CSI at the CPU and performs AAc by evaluating the EE objective in (4) using the resulting channel approximation.

  3. Decoder-Based CSI Reconstruction (RC): This strategy employs a neural encoder-decoder to reconstruct the instantaneous CSI at the CPU from lowdimensional CC embeddings, which is then used for AAc.

Performance and Trade-offs

Numerical results show substantial EE gains over the all-active baseline, with only a minor degradation in sum spectral efficiency and a significant reduction in signaling overhead. The proposed greedy framework achieves EE performance close to that of the simulated annealing baseline, confirming its effectiveness for low-complexity antenna activation. p. 0 Table II summarizes the complexity comparison, showing that DD has the lowest total online computational complexity among the three strategies. p. 4 Fig. 2 summarizes the performance-overhead-complexity trade-offs, indicating that NN achieves the best EE performance and the strongest performance-overhead compromise. p. 5

Conclusion

This work demonstrated that channel charting enables scalable centralized antenna activation in CF-mMIMO systems by replacing full instantaneous CSI exchange with compact CC embeddings. The results confirm that CC-based representations can successfully support centralized energy-efficient antenna activation, with the preferred strategy depending on whether the main priority is signaling overhead, online complexity, or EE performance. p. 5 The findings confirm that CC-based representations can successfully support centralized energy-efficient antenna activation. p. 5

REFERENCES

arXiv:2611.16996v1 3 Nov 2025

--- Page 5 ---

TABLE I: Uplink fronthaul overhead per UE-AP link

TABLE II: Complexity summary of the proposed AAc strategies

--- Page 3 ---

III. ENERGY EFFICIENCY PROBLEM FORMULATION

IV. ANTENNA-ACTIVATION STRATEGIES UNDER LOW-OVERHEAD CSI EXCHANGE

V. PERFORMANCE EVALUATION

VI.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Direct Embedding-Domain Antenna Activation (DD) allows an AI system to perform antenna selection across all distributed access points directly in a low-dimensional channel chart space without needing to reconstruct instantaneous CSI at the central processing unit. This enables network-wide activation decisions with minimal signaling overhead by using a proxy EE metric derived from embedding quantities.

  2. Nearest Chart-Point CSI Approximation (NN) allows an AI system to obtain approximate instantaneous CSI at the CPU by retrieving the nearest calibration user embedding for a given user's embedding and reusing that channel realization. This approximation supports antenna activation decisions while maintaining performance closer to full-CSI benchmarks than pure proxy methods.

  3. Decoder-Based CSI Reconstruction (RC) allows an AI system to reconstruct high-fidelity instantaneous CSI at the CPU from low-dimensional embeddings using a pre-trained autoencoder, followed by a greedy antenna activation search. This strategy aims for superior EE performance by recovering channel magnitude and phase information, albeit with higher computational complexity due to the decoder.

Related papers