Chapter Summary

Chapter 3 Summary: Channel Estimation and Pilot Design

Key Points

• 1.

  TDD Training Protocol. The coherence interval $\tau_c = T_c \cdot B_c$ is divided into pilot ($\tau_p$), uplink data ($\tau_u$), and downlink data ($\tau_d$) phases. Orthogonal pilot assignment within a cell requires $\tau_p \geq K$. Pilot overhead $\tau_p/\tau_c$ represents a non-recoverable spectral efficiency loss that grows with the number of users and decreases with the coherence interval.

• 2.

  LS vs. MMSE Estimation. The LS estimator $\hat{\mathbf{h}}_k^{\text{LS}} = \mathbf{y}_k/\sqrt{p_u\tau_p}$ requires no statistical knowledge but treats all $N_t$ dimensions equally. The MMSE estimator exploits the covariance matrix $\mathbf{R}_k$ to achieve MSE $= \sum_i \lambda_i\sigma^2/(p_u\tau_p\lambda_i + \sigma^2) \leq N_t\sigma^2/(p_u\tau_p)$. At high SNR, MMSE outperforms LS by the factor $N_t/r_k$ where $r_k$ is the effective channel rank — up to 20+ dB for spatially correlated channels.

• 3.

  Pilot Contamination. In multi-cell systems, finite pilot pools force pilot reuse across cells. Co-pilot users in other cells corrupt the channel estimate coherently: $\mathbf{y}_{1,k} = \sqrt{p_u\tau_p}\sum_\ell\mathbf{h}_{\ell,k} + \mathbf{n}_{1,k}$. This causes an SINR floor that does not vanish as $N_t \to \infty$ — both desired signal and contamination grow as $N_t$, keeping their ratio constant.

• 4.

  Pilot Assignment Algorithms. Greedy and graph-coloring algorithms assign pilot sequences to minimize contamination by separating users with high covariance overlap. The contamination metric $\rho_{k,k'} = \|\mathbf{R}_{1,k}^{1/2}\mathbf{R}_{\ell,k'}^{1/2}\|_F^2 / (\|\mathbf{R}_{1,k}\|_F\|\mathbf{R}_{\ell,k'}\|_F)$ quantifies the severity of interference. Good assignment raises the SINR floor but cannot eliminate it when the pilot pool is smaller than the total user count.

• 5.

  Pilot Decontamination via Subspace Projection. When co-pilot users have non-overlapping angular windows, their spatial covariance matrices are orthogonal: $\mathbf{R}_{1,k}\mathbf{R}_{\ell,k} = \mathbf{0}$. Projecting the pilot observation onto range$(\mathbf{R}_{1,k})$ eliminates contamination exactly: $\mathbf{P}_{1,k}\mathbf{h}_{\ell,k} = \mathbf{0}$ a.s. for all $\ell \neq 1$.

• 6.

  Massive MIMO Has Unlimited Capacity (Caire 2018). When covariance subspaces are asymptotically orthogonal — a generic condition for large arrays — the pilot contamination SINR floor vanishes and per-user rate grows without bound as $N_t \to \infty$. Pilot contamination is not a fundamental capacity limit but an artifact of ignoring spatial structure. Spatial correlation is a resource, not a nuisance.