Summary

Chapter 18 Summary: Massive MIMO

Key Points

• 1.

  Channel hardening ensures that $\|\mathbf{h}_k\|^2/M \to \beta_k$ almost surely as $M \to \infty$, causing the random fading channel to behave as a deterministic scalar. This eliminates the need for downlink pilots and fast power control, as the effective channel fluctuations decrease as $1/\sqrt{M}$.

• 2.

  Favourable propagation guarantees that user channel vectors become asymptotically orthogonal under i.i.d. Rayleigh fading: $\mathbf{h}_i^H\mathbf{h}_j/M \to 0$ for $i \neq j$. The Gram matrix $\frac{1}{M}\mathbf{H}^{H}\mathbf{H} \to \mathbf{D}_\beta$ becomes diagonal, and the matched filter alone achieves SIR $\approx M/(K-1)$.

• 3.

  Linear processing becomes near-optimal: MR, ZF, and MMSE combining/precoding all converge to the same rate $R_k \approx \log_2(1 + Mp_k\beta_k/\sigma^2)$ as $M \to \infty$ with $K$ fixed. The use-and-then-forget bound provides rigorous achievable rate expressions that depend only on channel statistics.

• 4.

  Pilot contamination is the fundamental bottleneck of multi-cell massive MIMO: when cells reuse pilot sequences, the channel estimate is corrupted by co-pilot users in other cells, creating coherent interference that scales as $M^2$ — the same rate as the desired signal. This produces a finite rate ceiling $R^{\infty} = \log_2(1 + \beta_{\ell k}^2/\sum_{j \neq \ell}\beta_{jk}^2)$ that cannot be overcome by adding antennas.

• 5.

  Max-min power control with MR combining has a closed-form solution: $p_k^{\star} = \eta/\beta_k$ (channel inversion). This equalises all users' SINRs by compensating for large-scale fading differences, dramatically improving the worst-case rate at the expense of sum rate.

• 6.

  Cell-free massive MIMO distributes antennas across the coverage area as access points connected to a CPU, eliminating cell boundaries. Macro diversity ensures that every user is close to at least some APs, providing $5$--$10\times$ better 95%-likely rate than co-located deployments with the same total antenna count.

• 7.

  Energy efficiency is quasi-concave in $M$: there exists an optimal $M^{\star}$ that balances the beamforming gain (rate grows as $\log M$) against the circuit power cost (linear in $M$). The optimum shifts to larger $M^{\star}$ as hardware becomes more power-efficient. Massive MIMO can achieve 10$\times$ better bits/joule than legacy MIMO through spatial focusing of energy.