Chapter Summary

Chapter 1 Summary: Why Massive MIMO

Key Points

• 1.

  Capacity scales as $K\log_2(N_t)$. With $N_t$ BS antennas serving $K$ users, the uplink sum rate grows as $K\log_2(N_t)$ bits/s/Hz for i.i.d. Rayleigh fading. The multiplexing gain is $K$ (not $N_t$); extra antennas increase each user's effective SNR linearly, adding $\log_2(N_t)$ bits per user. This is not a bound — it is achievable with simple linear processing.

• 2.

  Channel hardening: $\|\mathbf{h}_k\|^2/N_t \to \beta_k$ almost surely. With $N_t$ antennas, the normalized channel gain concentrates around its mean (path-loss) with variance $\beta_k^2/N_t$ for i.i.d. channels. The channel behaves like a deterministic scalar, enabling open-loop scheduling and reliable QoS without instantaneous CSI feedback. Hardening fails for pure LoS (rank-1 covariance); correlated channels harden more slowly.

• 3.

  Favorable propagation: $\mathbf{h}_k^H\mathbf{h}_j/N_t \to 0$ for $k \neq j$. Channel vectors of different users become asymptotically orthogonal. The residual interference decays as $\beta_k\beta_j/N_t$. For correlated channels, orthogonality requires that users occupy different angular sectors (non-overlapping eigenspaces of $\mathbf{R}_k$ and $\mathbf{R}_j$).

• 4.

  Linear processing is asymptotically optimal. MRC achieves near-optimal SINR because signal power grows as $N_t^2$ while interference grows as $N_t$ (favorable propagation). ZF eliminates interference exactly at all $N_t$ but amplifies noise. Both MRC and ZF achieve the same $K\log_2(N_t)$ asymptotic scaling as the optimal SIC detector — the main operational advantage of massive MIMO.

• 5.

  TDD reciprocity is the enabling technology. TDD pilot overhead is $\tau_p \geq K$ (independent of $N_t$). FDD requires $\tau_p \geq N_t$ downlink pilots plus $N_t \times K$ feedback scalars — infeasible for large $N_t$. TDD reciprocity allows the BS to use uplink-estimated channels directly for downlink precoding, without any feedback from the users.

• 6.

  Pilot contamination is not fundamental. Marzetta's original formulation identified pilot contamination as a hard barrier. Caire (2018) showed that with spatially correlated channels, MMSE channel estimation exploiting covariance structure eliminates pilot contamination, enabling unlimited capacity scaling.