Chapter Summary

Chapter Summary

Key Points

  • 1.

    One pilot identifies all PP paths. The embedded pilot scheme places a single known impulse at (p,kp)(\ell_p, k_p) on the DD grid, surrounded by a guard region of size (lmax+1)×(2kmax+1)(l_{\max} + 1) \times (2k_{\max} + 1). Through the DD convolution, the guard region receives exactly PP scaled copies of the pilot at offsets (i,ki)(\ell_i, k_i). Reading these off yields the full channel state.

  • 2.

    Path detection is a thresholded likelihood ratio. Per-cell test: YDD[]/α2γth|Y_{DD}[\cdot]/\alpha|^2 \geq \gamma_{\text{th}}. Under i.i.d. CN\mathcal{CN} noise (Chapter 4), the threshold for target false-alarm rate PFAP_{\text{FA}} is γth=σ2ln(PFA)/α2\gamma_{\text{th}} = -\sigma^2\ln(P_{\text{FA}})/|\alpha|^2. Typical operation: PFA=103P_{\text{FA}} = 10^{-3}, pilot boost 25 dB, gives reliable detection of paths down to −15 dB relative to pilot.

  • 3.

    Estimation MSE scales with PP, not MNMN. The variance of each path-gain estimate is σ2/α2\sigma^2/|\alpha|^2, so the total MSE is Pσ2/α2P \cdot \sigma^2/|\alpha|^2 — independent of grid size. This is the estimation-side benefit of DD sparsity: an MNMN-dimensional channel is fit with only PP real parameters.

  • 4.

    Pilot overhead is 113%3\% for typical channels. For a 5G-NR-aligned OTFS system with (M,N)=(512,16)(M, N) = (512, 16) and vehicular mobility (τmax=4μs\tau_{\max} = 4\mu s, fD=500f_D = 500 Hz), the embedded-pilot overhead is roughly 2%2\% — compare with 5G NR DMRS at 7714%14\%. A 5 percentage-point net gain in spectral efficiency.

  • 5.

    Superimposed pilots buy zero overhead. With a pilot pattern P[,k]P[\ell, k] overlaid on the data (fraction ρp\rho_p of power), no DD-grid cells are reserved. Channel estimates come from correlation over the full MNMN-cell grid. MSE scales as 1/(MNρp)1/(MN\rho_p) — at typical parameters, matching or beating embedded pilots at high SNR. The trade-off is iterative joint estimation/detection at the receiver.

  • 6.

    Cell-free OTFS is where the CommIT contribution shines. In LL-AP, KK-UE distributed systems, embedded-pilot overhead scales as KηembK \eta_{\text{emb}} — prohibitive at K8K \geq 8. The superimposed-pilot design (Mohammadi–Ngo–Matthaiou–Caire 2023) achieves zero overhead via per-UE Zadoff-Chu sequences, delivering a 252535%35\% net throughput gain over OFDM DMRS at vehicular mobility. This is the engineering lever behind cell-free OTFS deployment.

Looking Ahead

With the channel known from the estimator of this chapter, Chapter 8 takes up the problem of detection: recovering the data symbols X^data\hat{X}_{\text{data}} from the DD-domain observation YDDY_{DD}. We develop message-passing detection (MP-OTFS), low-complexity LMMSE, the LCD detector, and cross-domain iterative schemes. All exploit the sparse DD channel matrix structure of Chapter 4; the path estimates from this chapter feed the detectors of Chapter 8. Together, Chapters 4-8 complete the single-link OTFS receiver.