Ferkans — Interactive Telecom Tutor

Scaling Up to Hundreds of Antennas

Conventional MIMO systems (Chapters 15--17) operate with a modest number of antennas, typically $N_t = 2$ -- $8$ . A radical departure emerges when the base station is equipped with $M = 64$ -- $256$ or even thousands of antennas, while serving $K \ll M$ single-antenna users simultaneously. This regime, known as massive MIMO, is not merely a quantitative scaling of existing techniques: it produces qualitative changes in system behaviour. Channels harden, user channels become nearly orthogonal, simple linear processing becomes near-optimal, and the effects of uncorrelated noise and fast fading vanish. These phenomena are the foundation of 5G NR and beyond.

Definition:
Massive MIMO System Model

A massive MIMO system consists of a base station (BS) with $M$ antennas that simultaneously serves $K$ single-antenna user terminals on the same time-frequency resource, where $M \gg K$ . The uplink received signal at the BS is:

$\mathbf{y} = \sum_{k=1}^{K} \sqrt{p_k}\,\mathbf{h}_k x_k + \mathbf{n} = \mathbf{H}\mathbf{D}_p^{1/2}\mathbf{x} + \mathbf{n}$

where $\mathbf{y} \in \mathbb{C}^{M}$ , $\mathbf{H} = [\mathbf{h}_1, \ldots, \mathbf{h}_K] \in \mathbb{C}^{M \times K}$ is the channel matrix, $\mathbf{D}_p = \text{diag}(p_1, \ldots, p_K)$ collects the user transmit powers, $\mathbf{x} \in \mathbb{C}^K$ contains the unit-power data symbols, and $\mathbf{n} \sim \mathcal{CN}(\mathbf{0}, \sigma^2 \mathbf{I}_M)$ .

Each channel vector decomposes as:

$\mathbf{h}_k = \sqrt{\beta_k}\,\mathbf{g}_k$

where $\beta_k$ is the large-scale fading coefficient and $\mathbf{g}_k \sim \mathcal{CN}(\mathbf{0}, \mathbf{I}_M)$ is the small-scale fading vector under i.i.d. Rayleigh fading.

The regime $M \gg K$ is the defining feature of massive MIMO. In practice, $M/K \geq 5$ -- $10$ is sufficient to observe the key asymptotic behaviours. A typical 5G NR deployment uses $M = 64$ antennas serving $K = 8$ -- $16$ users.

,

Definition:
Channel Hardening

A MIMO channel exhibits channel hardening if the normalised channel gain concentrates around its mean as $M$ grows:

$\frac{\|\mathbf{h}_k\|^2}{M} \xrightarrow{M \to \infty} \beta_k \quad \text{almost surely}$

Equivalently, the variance of the normalised gain vanishes:

$\text{Var}\!\left[\frac{\|\mathbf{h}_k\|^2}{M}\right] = \frac{\beta_k^2}{M} \to 0$

When channel hardening holds, the effective channel seen by each user after linear combining behaves as a deterministic scalar, eliminating the need for downlink pilots and fast power control.

The term "hardening" comes from the observation that the random fading channel starts to behave like a deterministic (hardened) channel. The effective channel fluctuations decrease as $1/\sqrt{M}$ , so with $M = 100$ antennas, fading variations are reduced by a factor of 10 compared to single-antenna reception.

,

Theorem: Channel Hardening in i.i.d. Rayleigh Fading

Let $\mathbf{h}_k = \sqrt{\beta_k}\,\mathbf{g}_k$ where $\mathbf{g}_k \sim \mathcal{CN}(\mathbf{0}, \mathbf{I}_M)$ . Then:

$\frac{\|\mathbf{h}_k\|^2}{M} \xrightarrow{\text{a.s.}} \beta_k \quad \text{as } M \to \infty$

More precisely, for any $\varepsilon > 0$ :

$\Pr\!\left[\left|\frac{\|\mathbf{h}_k\|^2}{M} - \beta_k\right| > \varepsilon\right] \leq \frac{\beta_k^2}{M\varepsilon^2}$

The channel gain $\|\mathbf{h}_k\|^2 = \beta_k \sum_{m=1}^{M}|g_{mk}|^2$ is a sum of $M$ i.i.d. unit-exponential random variables scaled by $\beta_k$ . By the law of large numbers, this sum divided by $M$ converges to its expectation $\beta_k$ . With many antennas, the randomness of small-scale fading is averaged out, leaving only the deterministic large-scale fading $\beta_k$ .

Proof

Mean and variance of the channel gain

Since $|g_{mk}|^2 \sim \text{Exp}(1)$ , we have $\mathbb{E}[|g_{mk}|^2] = 1$ and $\text{Var}[|g_{mk}|^2] = 1$ . Therefore:

$\mathbb{E}\!\left[\|\mathbf{h}_k\|^2\right] = \beta_k \sum_{m=1}^{M}\mathbb{E}[|g_{mk}|^2] = M\beta_k$

$\text{Var}\!\left[\|\mathbf{h}_k\|^2\right] = \beta_k^2 \sum_{m=1}^{M}\text{Var}[|g_{mk}|^2] = M\beta_k^2$

Normalised gain statistics

For the normalised gain $Z_M = \|\mathbf{h}_k\|^2/M$ :

$\mathbb{E}[Z_M] = \beta_k, \qquad \text{Var}[Z_M] = \frac{\beta_k^2}{M}$

The coefficient of variation is:

$\frac{\sqrt{\text{Var}[Z_M]}}{\mathbb{E}[Z_M]} = \frac{1}{\sqrt{M}}$

which vanishes as $M \to \infty$ .

Convergence by Chebyshev and the SLLN

By Chebyshev's inequality:

$\Pr\!\left[|Z_M - \beta_k| > \varepsilon\right] \leq \frac{\text{Var}[Z_M]}{\varepsilon^2} = \frac{\beta_k^2}{M\varepsilon^2} \to 0$

For almost sure convergence, note that $Z_M = \beta_k \cdot \frac{1}{M}\sum_{m=1}^{M}|g_{mk}|^2$ . By the strong law of large numbers applied to the i.i.d. sequence $\{|g_{mk}|^2\}_{m=1}^{\infty}$ :

$\frac{1}{M}\sum_{m=1}^{M}|g_{mk}|^2 \xrightarrow{\text{a.s.}} 1$

and therefore $Z_M \xrightarrow{\text{a.s.}} \beta_k$ . $\blacksquare$

,

Channel Hardening as M Grows

Watch the empirical distribution of

\|\mathbf{h}_k\|^2/M

concentrate around

\beta_k = 1

as the number of BS antennas increases from

M = 4

to

M = 256

. The histogram narrows dramatically, demonstrating that the random fading channel "hardens" into a deterministic scalar.

Channel hardening: the normalised gain distribution concentrates around

\beta_k

as

M

grows.

Channel Hardening Effect

Observe how the empirical distribution of $\|\mathbf{h}_k\|^2/M$ concentrates around $\beta_k = 1$ as the number of BS antennas $M$ increases. For $M = 4$ the gain fluctuates widely; by $M = 256$ the distribution is tightly concentrated, demonstrating channel hardening.

Parameters

K

(number of users)1

Historical Note: Marzetta's Unlimited Antennas Vision

2010

The concept of massive MIMO was introduced by Thomas L. Marzetta of Bell Labs in his landmark 2010 paper "Noncooperative Cellular Wireless with Unlimited Numbers of Base Station Antennas." Marzetta showed that letting $M \to \infty$ while keeping $K$ fixed eliminates all effects of uncorrelated noise and fast fading, leaving only inter-cell interference from pilot contamination as the fundamental performance bottleneck. This paper, which initially encountered scepticism due to the seemingly impractical antenna counts, sparked a decade of research and ultimately became the theoretical foundation of 5G NR massive MIMO deployments. Marzetta's insight was that the asymptotic regime is not merely a mathematical convenience but is practically approachable: many of the predicted benefits manifest with as few as $M = 64$ antennas.

Common Mistake: The i.i.d. Rayleigh Model is an Idealisation

Mistake:

Deriving all massive MIMO results under i.i.d. Rayleigh fading ( $\mathbf{g}_k \sim \mathcal{CN}(\mathbf{0}, \mathbf{I}_M)$ ) and treating them as exact predictions for real deployments.

Correction:

The i.i.d. Rayleigh model is analytically convenient and captures the essential scaling laws, but real propagation channels are spatially correlated: the channel covariance $\mathbf{R}_{k} = \mathbb{E}[\mathbf{h}_k\mathbf{h}_k^H]$ is not proportional to the identity. Spatial correlation has several effects:

Channel hardening weakens: The variance of $\|\mathbf{h}_k\|^2/M$ depends on $\text{tr}(\mathbf{R}_{k}^{2})/(\text{tr}(\mathbf{R}_{k}))^2$ , which exceeds $1/M$ for rank-deficient $\mathbf{R}_{k}$ .
Favourable propagation may fail: If $\mathbf{R}_{i} \approx \mathbf{R}_{j}$ , the channels do not become orthogonal.
MMSE estimation improves: Correlated channels enable subspace-based estimation that can overcome pilot contamination (Caire, 2018).

Always validate massive MIMO results against spatially correlated channel models (e.g., one-ring, 3GPP SCM) before drawing deployment conclusions.

Massive MIMO

A multi-user MIMO system in which the base station is equipped with $M \gg K$ antennas, where $K$ is the number of simultaneously served users. The large antenna excess enables channel hardening, favourable propagation, and near-optimal performance with simple linear processing.

Channel Hardening

The phenomenon whereby the normalised channel gain $\|\mathbf{h}_k\|^2/M$ concentrates around its deterministic mean $\beta_k$ as $M \to \infty$ , causing the random fading channel to behave as a deterministic scalar channel.

Massive MIMO Fundamentals