Chapter Summary

Chapter 8 Summary

Key Points

• 1.

  The FDD bottleneck is fundamental. In FDD massive MIMO, the DL pilot overhead scales as $\tau_d \geq N_t$ and the feedback overhead scales as $B_{\text{fb}} \propto N_t$. For $N_t \gg T_c$, these overheads can consume the entire coherence block, making FDD massive MIMO impractical without compression.

• 2.

  Angular-domain sparsity enables compressed feedback. Real-world channels occupy a low-dimensional subspace of the angular domain, with effective rank $r_k \ll N_t$ determined by the angular spread. Compressed sensing with random projections reduces the feedback dimension to $O(r_k \log(N_t/r_k))$, exploiting this sparsity.

• 3.

  Codebook-based feedback (5G NR Type I/II) is the deployed solution. Type I selects a single DFT beam (low overhead, moderate accuracy). Type II combines $L$ beams with per-subband amplitude/phase coefficients (higher overhead, 3–5 dB better MU-MIMO performance). The quantization error decays as $2^{-B/(N_t-1)}$, requiring $B \propto N_t$ to maintain a target quality.

• 4.

  Deep learning CSI compression (CsiNet) outperforms classical methods at the same compression ratio by learning encoder and decoder functions matched to the channel distribution. The encoder is lightweight (single FC layer, suitable for UE deployment), while the decoder is heavier (convolutional RefineNet at BS). The key limitation is environment specificity — retraining is needed for new deployments.

• 5.

  JSDM is the most principled FDD solution. By pre-beamforming to the group covariance eigenspace, JSDM reduces both DL pilot overhead and feedback dimension from $N_t$ to $r_g$, while preserving the full multiplexing gain. It requires only partial reciprocity (UL covariance estimation), which is available in FDD. The overhead reduction factor $N_t/r_g$ is large in macro-cell deployments with narrow angular spread.

• 6.

  The rate loss from CSI quantization creates an interference floor. Imperfect CSI causes residual multi-user interference that limits the achievable rate, analogous to pilot contamination in TDD. The overhead–accuracy tradeoff (more feedback bits improve rate but consume more resources) is the central design problem in FDD massive MIMO.