Chapter Summary

Chapter Summary

Key Points

• 1.

  Fractional Doppler is the generic case, not a perturbation. Physical path Dopplers $\nu_i$ do not align to the DD grid: each path has $\nu_i T = k_i^{\text{int}} + \epsilon_i$ with $|\epsilon_i| < 1/2$. For frame durations $T$ long enough to give useful Doppler resolution, the fractional part is typically 30-50% of the grid spacing — a large effect.

• 2.

  Fractional Doppler creates inter-Doppler interference (IDI). A single path produces a Dirichlet-kernel response spread across neighboring Doppler bins, not a single cell. At $\epsilon = 0.3$: main lobe captures 68% of energy, neighbors get 32%. At $\epsilon = 0.5$: main lobe captures 60%, neighbors get 40%. The resulting channel has $P(2k_{\text{IDI}}+1)$ effective taps instead of $P$.

• 3.

  BEM converts fractional to integer Doppler. Approximating the time-varying channel by a weighted sum of $Q_{\text{BEM}}$ complex exponentials reduces each fractional path to $Q_{\text{BEM}}$ integer-Doppler paths. At $Q_{\text{BEM}} = 3$, BEM captures $\geq 95\%$ of IDI energy. After BEM, the integer- Doppler detectors of Chapter 8 apply directly.

• 4.

  Windowing suppresses sidelobes at cost of main-lobe width. Applying a Hamming or Blackman window along the Doppler axis before SFFT reduces sidelobes from $-13$ dB to $-43$ dB (Hamming) or $-58$ dB (Blackman). Main lobe widens by 36-73%. Negligible compute overhead.

• 5.

  Oversampled DD grid quadratically reduces IDI. Using grid dimensions $(\alpha M, \beta N)$ with $\beta \geq 2$ halves the effective fractional offset. IDI power drops by factor $1/\beta$, at the cost of $\beta\log\beta$-factor more compute. Practical for sensing (Chapter 11), moderate for data communications ($\beta \leq 2$).

• 6.

  Fractional Doppler does not destroy diversity. The Surabhi- Chockalingam-Caire theorem: with $Q_{\text{BEM}} \geq 2k_{\max}+1$, fractional OTFS retains diversity order $\min(P, 2k_{\max}+1)$ — typically $P$ (full diversity). Real-world penalty: $\sim 1$ dB from the ideal Chapter 9 result, using BEM $Q = 3$ + windowing.

• 7.

  Practical mitigation is $3-5\times$ compute, $1$-$2$ dB penalty. Standard deployment: Hamming window + BEM $Q = 3$ + (optional) oversampling $\beta = 2$. Compute overhead: $\sim 6\times$ baseline integer-Doppler MMSE. SNR penalty: $\sim 1$ dB compared to ideal integer Doppler. Still $5$–$8$ dB better than OFDM — the decisive mobility advantage is preserved.