Ferkans — Interactive Telecom Tutor

The Antenna Economy

The classical view of wireless networks is pessimistic: spectrum is scarce, users compete for it, and interference limits capacity. Massive MIMO turns this picture on its head. By deploying many more antennas at the base station than there are active users, we gain a resource that was always free but went unexploited: spatial degrees of freedom.

This section develops the capacity scaling argument from first principles. The message is simple: with $N_t$ antennas serving $K$ users, the sum capacity grows linearly in $\min(N_t, K)$ — and in the massive MIMO regime $N_t \gg K$ , it grows linearly in $K$ with a pre-log factor of order $N_t$ .

Definition:
Multi-User Massive MIMO Uplink Model

Consider a single-cell system with a base station (BS) equipped with $N_t$ antennas serving $K$ single-antenna users, $N_t \gg K \geq 1$ . In the uplink, the $N_t \times 1$ received signal vector is

$\mathbf{y} = \mathbf{H}\mathbf{s} + \mathbf{w},$

where:

$\mathbf{H} \in \mathbb{C}^{N_t \times K}$ is the uplink channel matrix, with $k$ -th column $\mathbf{h}_k \in \mathbb{C}^{N_t}$ being the channel from user $k$ to all BS antennas.
$\mathbf{s} = (s_1, \ldots, s_{K})^T$ is the transmitted symbol vector, with $\mathbb{E}[|s_k|^2] = P_k$ .
$\mathbf{w} \sim \mathcal{CN}(\mathbf{0}, \sigma^2\mathbf{I}_{N_t})$ is i.i.d. receiver noise.

The per-user SNR is $\text{SNR}_{k} = P_k / \sigma^2$ .

The downlink model is the Hermitian transpose: $\mathbf{y}_k = \mathbf{h}_k^H \mathbf{x} + n_k$ , where $\mathbf{x} \in \mathbb{C}^{N_t}$ is the transmitted signal vector. We focus on the uplink first; the downlink follows by duality in Section 1.4.

Massive MIMO

A base station architecture with $N_t$ antennas where $N_t \gg K$ , typically $N_t \geq 64$ with $K \leq 20$ . The large antenna surplus enables channel hardening, favorable propagation, and near-optimal linear processing.

Definition:
Ergodic Sum Rate and Degrees of Freedom

With Gaussian inputs and a known channel $\mathbf{H}$ , the ergodic uplink sum rate (in bits per channel use) is

$C_{\text{sum}}(\mathbf{H}) = \log_2 \det\!\left(\mathbf{I}_{N_t} + \frac{1}{\sigma^2} \mathbf{H} \mathbf{P} \mathbf{H}^{H}\right),$

where $\mathbf{P} = \text{diag}(P_1, \ldots, P_{K})$ is the diagonal power matrix. The ergodic sum rate is $\bar{C}_{\text{sum}} = \mathbb{E}_{\mathbf{H}}[C_{\text{sum}}(\mathbf{H})]$ .

The multiplexing gain (degrees of freedom) is

$d = \lim_{\text{SNR} \to \infty} \frac{\bar{C}_{\text{sum}}}{\log_2 \text{SNR}} = \min(N_t, K).$

For $N_t \geq K$ (the massive regime), $d = K$ : every user gets exactly one independent spatial stream regardless of how many BS antennas exist. More antennas improve the pre-log SNR multiplier, not the DoF.

Theorem: Sum Rate Linear Scaling with $tn{ntx}$

Assume i.i.d. Rayleigh fading: the entries of $\mathbf{H}$ are i.i.d. $\mathcal{CN}(0, \beta_{k})$ where $\beta_{k}$ is the path-loss of user $k$ . Fix $K$ and let $N_t \to \infty$ with equal power $P_k = P$ for all $k$ . Then the ergodic sum rate satisfies

$\bar{C}_{\text{sum}} = \sum_{k=1}^{K} \log_2\!\left(1 + \frac{P\,N_t\,\beta_{k}}{\sigma^2}\right) + o(N_t) \;\longrightarrow\; K\log_2(N_t) + \mathcal{O}(1).$

In particular, the sum rate grows as $K \log_2 N_t$ bits per channel use.

Each user's SINR grows linearly with $N_t$ because the BS can coherently combine $N_t$ received copies of the user's signal, yielding an $N_t$ -fold beamforming gain. The interference from other users is suppressed by the spatial orthogonality of their channels (favorable propagation — Section 1.3).

Show Hint

Apply the identity $\det(\mathbf{I} + \mathbf{AB}) = \det(\mathbf{I} + \mathbf{BA})$ to move from $N_t \times N_t$ to $K \times K$ determinant.

For large $N_t$ , use the law of large numbers: $\frac{1}{N_t}\mathbf{H}^{H}\mathbf{H} \to \text{diag}(\beta_{1}, \ldots, \beta_{\ntn{nusers}})$ .

With a diagonal Gram matrix, the determinant factors into a product and the sum rate becomes a sum of per-user log terms.

Proof

Determinant identity

Apply the matrix determinant lemma: for any $m \times n$ matrix $\mathbf{A}$ , $\det(\mathbf{I}_m + \mathbf{A}\mathbf{A}^H) = \det(\mathbf{I}_n + \mathbf{A}^H\mathbf{A}).$ Thus $C_{\text{sum}} = \log_2 \det\!\left(\mathbf{I}_{K} + \frac{1}{\sigma^2}\mathbf{H}^{H}\mathbf{H}\,\mathbf{P}\right).$ This reduces the computation from an $N_t \times N_t$ determinant to a $K \times K$ one.

Gram matrix concentration

For i.i.d. $\mathcal{CN}(0, \beta_{k})$ entries in column $k$ of $\mathbf{H}$ , the $(j, k)$ entry of $\frac{1}{N_t}\mathbf{H}^{H}\mathbf{H}$ is $\left[\frac{\mathbf{H}^{H}\mathbf{H}}{N_t}\right]_{jk} = \frac{1}{N_t}\mathbf{h}_j^H\mathbf{h}_k \xrightarrow{a.s.} \begin{cases}\beta_{k} & j = k \\ 0 & j \neq k \end{cases}$ by the strong law of large numbers (this is precisely favorable propagation, proved in Section 1.3). So $\frac{1}{N_t}\mathbf{H}^{H}\mathbf{H} \to \boldsymbol{\Lambda} = \text{diag}(\beta_{1}, \ldots, \beta_{\ntn{nusers}})$ .

Asymptotic sum rate

Substituting the asymptotic Gram matrix: $C_{\text{sum}} \approx \log_2 \det\!\left(\mathbf{I} + \frac{N_t}{\sigma^2}\boldsymbol{\Lambda}\mathbf{P}\right) = \sum_{k=1}^{K} \log_2\!\left(1 + \frac{N_t\,P\,\beta_{k}}{\sigma^2}\right).$ For large $N_t$ , $\log_2(1 + N_t\,\rho\,\beta_{k}) \approx \log_2(N_t) + \log_2(\rho\,\beta_{k})$ , so the sum rate grows as $K\log_2(N_t) + \mathcal{O}(1)$ . $\blacksquare$

,

Key Takeaway

Degrees of freedom are bounded; SNR is not. With $N_t$ BS antennas and $K$ users, the spatial DoF is $\min(N_t, K) = K$ (in the massive regime). But each user's effective SNR grows linearly as $N_t \to \infty$ , giving $K\log_2(N_t)$ scaling in sum rate. This is the fundamental reason massive MIMO is transformative.

Sum Rate vs. Number of BS Antennas

Explore how the ergodic sum rate scales with $N_t$ for MRC, ZF, and MMSE combining at different SNR and user counts. The $K\log_2(N_t)$ asymptote is shown as a dashed reference.

Parameters

Users

K

4

SNR (dB)10

Combining scheme

Show

K\log_2(N_t)

asymptote

Example: Scaling Calculation: 64 vs 256 Antennas

A BS serves $K = 8$ users. All users have path-loss $\beta = -100$ dB and transmit at $P = 23$ dBm. The noise figure is $7$ dB, noise PSD is $-174$ dBm/Hz, and bandwidth is $20$ MHz. Compare the sum-rate lower bound with $N_t = 64$ vs $N_t = 256$ antennas using the asymptotic formula of Theorem $tn{ntx}$ $t n n t x$ " data-ref-type="theorem">TSum Rate Linear Scaling with $tn{ntx}$ .

Solution

Compute per-user SNR

Receiver noise power: $\sigma^2 = -174 + 10\log_{10}(20\times10^6) + 7 = -174 + 73 + 7 = -94$ dBm. Received signal power (at antenna): $P_r = 23 - 100 = -77$ dBm. Per-user per-antenna SNR: $\rho = P_r - \sigma^2 = -77 - (-94) = 17$ dB $= 50$ .

Sum rate with $N_t = 64$

Using the asymptotic formula: $C_{64} \approx 8 \cdot \log_2(1 + 64 \cdot 50) = 8 \cdot \log_2(3201) \approx 8 \times 11.6 = 93 \text{ bits/s/Hz}.$

Sum rate with $N_t = 256$

$C_{256} \approx 8 \cdot \log_2(1 + 256 \cdot 50) = 8 \cdot \log_2(12801) \approx 8 \times 13.6 = 109 \text{ bits/s/Hz}.$ $

Interpretation

Quadrupling the antennas ( $64 \to 256$ ) adds $2$ bits/s/Hz per user, consistent with $\Delta C = 8\log_2(4) = 16$ bits/s/Hz. The gains are real but diminishing in absolute SNR terms — the massive MIMO advantage is primarily in serving many users simultaneously, not in single-user rate maximization. $\blacksquare$

Common Mistake: More Antennas Do Not Give More Degrees of Freedom

Mistake:

A common misconception is that adding antennas to the BS continuously increases the number of simultaneous streams that can be supported.

Correction:

The spatial multiplexing gain (DoF) is $\min(N_t, K)$ . With $K$ single-antenna users, no matter how many BS antennas are deployed, only $K$ independent data streams can be transmitted simultaneously. Extra antennas improve beamforming gain (higher per-user SNR) and interference suppression, but they do not increase the number of streams. To serve more users simultaneously, you must increase $K$ , which requires more time-frequency resources for pilot sequences.

Why This Matters: Massive MIMO in 5G NR and Beyond

The 3GPP NR standard (Release 15+) specifies "Full-Dimension MIMO" with up to 32 CSI-RS ports at sub-6 GHz and up to 64 ports at mmWave. Commercial 5G base stations from Ericsson, Huawei, and Nokia deploy 64-element active antenna units (AAUs) capable of serving 8–16 simultaneous users. The theoretical analysis in this chapter predicts performance within 1–3 dB of field measurements once realistic propagation (spatial correlation, pilot contamination) is accounted for in Chapters 2–5.

Historical Note: The Birth of Massive MIMO: Marzetta's 2010 Paper

2010–present

The concept of massive MIMO was introduced by Thomas Marzetta in a landmark 2010 IEEE Transactions on Wireless Communications paper titled "Noncooperative Cellular Wireless with Unlimited Numbers of Base Station Antennas." Marzetta showed that if the BS has infinitely many antennas, intra-cell interference vanishes, the effects of uncorrelated noise and fast fading disappear, and the system performance depends only on pilot contamination.

The paper was initially met with skepticism — deploying 100+ antennas seemed impractical. Within five years, prototypes at Rice, Lund, and Bristol proved the concept experimentally, and by 2018 it became standard in 5G NR. Marzetta received the 2017 IEEE Communications Society Heinrich Hertz Award for this work.

Quick Check

A BS has $N_t = 128$ antennas serving $K = 10$ single-antenna users. What is the maximum multiplexing gain (DoF)?

128

10

64

1280