Chapter Summary

Chapter Summary

Key Points

• 1.

  Sensing is (short-horizon) channel estimation. In the DD-sparse representation, the channel is determined by the scatterer geometry. A good sensing algorithm produces a channel estimate of CRB-level accuracy — the pilot overhead of classical systems becomes largely redundant. This is the SAC paradigm.

• 2.

  Scene-to-channel map is deterministic. Given $\hat\Theta^{(t)} = \{(\hat\tau_i, \hat\nu_i, \hat\theta_i, \hat\phi_i, \hat a_i)\}$, the predicted channel $\hat{\mathbf{h}}^{(t+1)} = \mathbf{H}_{DD}(\hat\Theta^{(t+1|t)})$ has MSE equal to the sensing CRB plus process-noise contribution. No additional estimation noise is introduced by the map — the gain over classical pilot-based estimation is pure.

• 3.

  Prediction horizon $T_{\text{pred}}$ is set by maneuver dynamics. $T_{\text{pred}} \approx \nu_{\text{tol}}\lambda/(2\sigma_a)$ for kinematic noise $\sigma_a$. Automotive: 10-50 ms; LEO: 0.1-1 ms; pedestrian: 100 ms-1 s. SAC gain is largest where prediction horizon most exceeds coherence time.

• 4.

  IMM handles multi-model dynamics. Interacting multiple model filter runs CV + CA + CT in parallel, weighted by maneuver likelihood. Achieves near-optimal prediction across diverse target motion patterns. Cost: $k\times$ compute for $k$ models.

• 5.

  SAC spectral efficiency gain scales with mobility. $G \approx (1 - T_{\text{fr}}/T_{\text{pred}})/(1 - T_{\text{fr}}/T_c)$. Pedestrian: $\sim 1\%$. Vehicular mmWave: $\sim 25\%$. LEO: $\sim 50\%$. Above break-even velocity ($\sim 10$ m/s at 28 GHz), SAC is decisively better than classical.

• 6.

  Beam prediction enables beam-sweep-free operation. $P_{\text{correct}} = \Phi((B/2 - \dot\phi T_{\text{pred}})/\sigma_\phi)$. For automotive mmWave with moderate mobility, beam prediction succeeds at $\geq 95\%$. Classical SSB sweeps are replaced by sensing-driven beam selection — 5× less overhead for beam management.

• 7.

  PRA enables $5\times$ URLLC efficiency. Predictive resource allocation uses the SAC horizon to pre-schedule. URLLC reservation drops from worst-case (20-30%) to horizon-specific ($\sim 2.5\%$), freeing 20% of BS capacity for eMBB. Joint SAC + PRA is the framework that makes URLLC-scale reservation efficient.

• 8.

  CommIT contribution: Liu-Caire 2022 + Cui-Yuan-Caire 2023 + Zhao-Liu-Caire 2023. Together these three papers establish the mathematical foundation for sensing-assisted communication in OTFS. The DD-domain structure is essential throughout — without it, the scene-to-channel map is not deterministic, and the predictive gains collapse.