Chapter Summary

Chapter Summary

Key Points

  • 1.

    The cache-aided cloud-RAN architecture places caches of size MM at each of NENN_\text{EN} edge nodes, connected to a central cloud via per-EN fronthaul capacity CFC_F. The aggregate caching gain is t=NENM/Nt = N_\text{EN} M/N.

  • 2.

    Normalized delivery time (NDT) Δ\Delta is a dimensionless latency metric normalized to the infinite-fronthaul MU-MIMO baseline. NDT = 1 is the ideal; NDT > 1 reflects the architecture's bottleneck.

  • 3.

    NDT formula. Achievable upper bound: Δ=max(1,K(1μ)/(NENLEN(1+t))+K(1μ)/(NENC))\Delta = \max(1, K(1-\mu)/(N_\text{EN} L_\text{EN}(1+t)) + K(1-\mu)/(N_\text{EN} C)). Combines the downlink cooperative Lampiris-Caire DoF term with the fronthaul transfer term. Cut-set lower bound matches when t=0t = 0.

  • 4.

    Cache-fronthaul substitutability. Iso-NDT contours are hyperbolae in the (μ,C)(\mu, C) plane. Doubling either resource reduces its contribution to NDT by half. Operators choose along the Pareto frontier based on cost structure.

  • 5.

    The CommIT NDT framework (Sengupta-Tandon-Simeone, 2017, with Caire-adjacent follow-ups) is the unified language for cache- aided C-RAN analysis. It cleanly isolates the cache-fronthaul tradeoff from the underlying SNR.

  • 6.

    Saturation behavior. μ1\mu \to 1: NDT = 1 (full cache, no fronthaul needed). CC \to \infty: NDT reduces to the Lampiris- Caire DoF formula for cooperative NENLENN_\text{EN} L_\text{EN}-antenna BC.

  • 7.

    Deployment implications. 5G NR C-RAN with NEN=2-8N_\text{EN} = 2\text{-}8 and μ0.1-0.3\mu \approx 0.1\text{-}0.3 achieves NDT 3\leq 3 at typical fronthaul C=5-20C = 5\text{-}20 files/use. Design tools benefit from the NDT framework when sizing fronthaul and cache budgets.

Looking Ahead

Chapter 9 treats multi-server coded caching: multiple independent transmitters (e.g., cooperating base stations) each holding the full library, coordinated at the placement/delivery level. The multi-server MAN scheme extends the single-transmitter rate formula and sheds light on the differences between shared- cache and dedicated-cache models. Chapters 10-11 then move to D2D networks where users themselves are transmitters.