In conventional wireless communications, the environment is a chaotic adversary. Signals bounce off buildings, trees, and moving vehicles, arriving at a receiver via multiple paths. These paths can interfere constructively or destructively, causing the signal strength to fluctuate wildly over time and space—a phenomenon known as small-scale fading. To combat this unpredictability, traditional systems rely on complex channel estimation, rapid power control, and conservative scheduling.
However, when we drastically scale up the number of antennas or participating nodes in a wireless network, this randomness undergoes a profound statistical transformation. Instead of becoming more chaotic, the channel stabilizes. The random fluctuations average out, and the wireless link begins to behave like a static, predictable, and nearly deterministic wire. This phenomenon is known as channel hardening.
The Intuition: The Law of Large Numbers Over the Air
To understand why channel hardening happens, consider a single-input multiple-output (SIMO) system where a transmitter sends a signal to a receiver equipped with $N$ antennas. The channel can be represented as a vector $\mathbf{h} = [h_1, h_2, \dots, h_N]^T$, where each $h_i$ is a complex random variable representing the fading coefficient of the path to the $i$-th antenna.
The total received signal power is proportional to the squared Euclidean norm of the channel vector, $\|\mathbf{h}\|^2 = \sum_{i=1}^N |h_i|^2$. If the propagation paths to the different antennas are independent and identically distributed (i.i.d.) with a mean power of $\mathbb{E}[|h_i|^2] = 1$, the Law of Large Numbers dictates that as the number of antennas $N$ grows large:
$$\frac{\|\mathbf{h}\|^2}{N} \to 1 \quad \text{as } N \to \infty$$
More formally, a channel exhibits hardening when the variance of its normalized channel gain shrinks to zero relative to its mean:
$$\lim_{N \to \infty} \frac{\text{Var}(\|\mathbf{h}\|^2)}{\left(\mathbb{E}[\|\mathbf{h}\|^2]\right)^2} = 0$$
As Andrews et al. outlined in their foundational perspective What Will 5G Be?, this asymptotic stabilization means that small-scale fading effectively disappears.
In a comprehensive retrospective, Björnson et al. explain in Twenty-Five Years of Signal Processing Advances for Multiantenna Communications that this deterministic behavior fundamentally simplifies radio resource management. Because the channel gain becomes predictable, base stations no longer need to track instantaneous channel state information (CSI) on the downlink with high frequency. This drastically reduces signaling overhead and improves spectral efficiency, a property that Senel et al. leveraged in NOMA Versus Massive MIMO in Rayleigh Fading to demonstrate that standard Massive MIMO often outperforms more complex non-orthogonal multiple access (NOMA) schemes.
From Active Antennas to Intelligent Reflecting Surfaces
While channel hardening was originally discovered in the context of active Massive MIMO base stations, recent research has extended this principle to passive electromagnetic structures, such as Large Intelligent Surfaces (LIS) and Intelligent Reflecting Surfaces (IRS). These surfaces consist of arrays of passive, software-controlled elements that alter the phase shifts of reflecting waves to dynamically reshape the wireless environment.
The hardening effect remains robust in these architectures. Jung et al. analyzed this in Performance Analysis of Large Intelligent Surfaces (LISs): Asymptotic Data Rate and Channel Hardening Effects, proving that the fluctuations in a system's mutual information decay rapidly as the number of reflecting elements increases. This ensures highly reliable data rates even in the presence of pilot contamination, as the authors further demonstrated in Performance Analysis of Active Large Intelligent Surfaces (LISs): Uplink Spectral Efficiency and Pilot Training.
[User] ---- (Random Fading Paths) ----> [ IRS / LIS Array ] ---- (Sum of Reflections) ----> [Base Station]
* Fluctuations cancel out
* Aggregate channel hardens
However, physical deployment introduces constraints. How large must an IRS be to achieve this deterministic behavior? Bereyhi et al. addressed this in How Should IRSs Scale to Harden Multi-Antenna Channels? and Channel Hardening of IRS-Aided Multi-Antenna Systems: How Should IRSs Scale?. The authors proved that channel hardening can be successfully induced even when the physical dimensions of the surface scale sub-linearly relative to the number of elements, and even when those elements exhibit spatial correlation.
Yet, scaling hardware indefinitely yields diminishing returns. Zhang et al. showed in Outage Probability and Finite-SNR DMT Analysis for IRS-Aided MIMO Systems: How Large IRSs Need to be? that the performance benefits and outage probability reductions of increasing IRS size saturate rapidly. When properly optimized, however, the resulting stable power gains can be combined with other access technologies; for instance, Xie et al. demonstrated in STAR-RIS Aided NOMA in Multicell Networks: A General Analytical Framework With Gamma Distributed Channel Modeling that simultaneously transmitting and reflecting surfaces can provide full 360-degree predictable coverage.
The Nuances of Selection and Routing
Does channel hardening always occur at the same rate? Not necessarily. The speed and nature of hardening depend heavily on how the antennas are utilized.
Consider Transmit Antenna Selection (TAS), where a system dynamically chooses a subset of the best available antennas to reduce hardware costs. In Rate of channel hardening of antenna selection diversity schemes and its implication on scheduling, Bai et al. discovered that selection schemes exhibit a slower rate of channel hardening compared to traditional MIMO systems. Because the system deliberately picks the best channels, it artificially preserves channel variability over a longer duration as antennas are added. This slower hardening is actually a feature, not a bug: it allows the system to exploit multi-user scheduling gains in low-SINR environments.
Despite this selection bias, the underlying statistical convergence remains intact. Asaad et al. proved in Asymptotics of Transmit Antenna Selection: Impact of Multiple Receive Antennas that as the total pool of transmit antennas grows, the variance of the mutual information still shrinks relative to the mean, confirming that selection-based systems ultimately achieve a highly predictable, Gaussian-distributed capacity.
This predictability ripples up the network stack, altering higher-level routing paradigms. In An Energy-Based Comparison of Long-Hop and Short-Hop Routing in MIMO Networks, Lo et al. demonstrated that channel hardening fundamentally shifts the traditional energy-reliability trade-off. In fading environments, short-hop routing is typically preferred to avoid deep fades over long distances. But because channel hardening eliminates these deep fades, long-hop routing becomes significantly more energy-efficient when networks are bound by strict delay constraints.
The Blessing of Scale in Over-the-Air Computing
A compelling modern application of channel hardening is found in Over-the-Air Computation (AirComp), particularly for Federated Learning (FL). In AirComp-FL, multiple edge devices transmit their local machine learning model updates simultaneously over the same frequency channel. The wireless network acts as a natural analog computer, summing the signals in the air so that the base station receives the aggregate gradient directly.
Traditionally, this approach is highly sensitive to channel distortions, requiring meticulous per-device power control to ensure that fading doesn't bias the sum. However, Das et al. challenge this limitation in Rethinking Federated Learning Over the Air: The Blessing of Scaling Up. They show that as the number of participating clients scales up, a multi-transmitter channel hardening effect occurs. The individual fading fluctuations of the clients average out across the network, stabilizing the aggregate signal and allowing accurate convergence without complex, expensive power adjustments or instantaneous channel estimation.
Client 1 (Fading h₁) ---\
Client 2 (Fading h₂) ----\ ---> Air Superposition ---> Base Station receives stable,
Client 3 (Fading h₃) ----/ (h₁ + h₂ + h₃ + ...) hardened aggregate gradient
Client N (Fading h_N) --/
To shield this distributed computing process from external disruptions, Shi et al. demonstrated in Combating Interference for Over-the-Air Federated Learning: A Statistical Approach via RIS that an intelligent surface can be used to artificially enforce channel hardening and favorable propagation. By optimizing passive phase shifts based on statistical properties rather than instantaneous tracking, the surface suppresses both accidental and malicious interference, ensuring robust and unbiased gradient aggregation.
Ultimately, channel hardening demonstrates that scale transforms quality. By replacing the unpredictable physics of a single propagation path with the predictable mathematics of large numbers, it turns wireless channels into reliable, deterministic resources.
Key Foundational Work
Rate of channel hardening of antenna selection diversity schemes and its implication on scheduling (2009) — contributed 2 concept elements to this direction.
An Energy-Based Comparison of Long-Hop and Short-Hop Routing in MIMO Networks (2010) — contributed 1 concept elements to this direction.