Universal Variable TOF Formulation

• Universal Variable TOF formulation redefines delay estimation by modeling the complex response as a function of modulation frequency, unifying single-frequency and multi-frequency approaches.
• It enhances multipath separation and mitigates phase ambiguity by leveraging Fourier inversion and spectral estimation across a broad frequency sweep.
• Practical implementation balances bandwidth, frequency spacing, and windowing to optimize resolution, unambiguous range, and noise robustness.

Searching arXiv for the target paper and closely related TOF/OCT context. Universal Variable Time-of-Flight Formulation denotes a frequency-parameterized view of active range sensing in which the measured complex response is treated as a function of modulation frequency rather than solely as a phase sample at a fixed frequency. In this formulation, Frequency-Domain Time-of-Flight (FD-TOF) is expressed as a multi-frequency extension of conventional phase-based TOF, with the measurement model written as $$m(\omega)=\sum_i \alpha_i e^{j\omega \tau_i}+n(\omega)$$ over a sampled set of angular frequencies $\omega \in \Omega \subset \mathbb{R}$. The framework connects single-path estimation, multipath separation, spectral inversion, and high-frequency operating regimes within one signal model, and it is explicitly developed in the context of "Frequency Domain TOF: Encoding Object Depth in Modulation Frequency" (Kadambi et al., 2015).

1. Signal model and the shift from phase sampling to frequency sampling

A time-of-flight system strobes a source with sinusoidal modulation at radio frequency $f$, with angular frequency $\omega=2\pi f$. The emitted signal can be written as

$$s(t)=A\cos(2\pi f t)=A\,\mathrm{Re}\{e^{j2\pi f t}\}.$$

For a single reflective path with round-trip delay $\tau$, attenuation $\alpha$, and background contribution $b$, the received signal is

$r(t)=\alpha \cos(2\pi f (t-\tau))+b,$

or, in complex analytic form, approximately

$$r(t)\approx \mathrm{Re}\{\alpha e^{j2\pi f (t-\tau)}+b\}.$$

If multiple paths contribute, the received signal becomes

$\omega \in \Omega \subset \mathbb{R}$0

with one-way scene depth related to delay by

$\omega \in \Omega \subset \mathbb{R}$1

The essential distinction between standard phase-TOF and FD-TOF is the primal sampling variable. In phase-TOF cameras, the measurement is obtained by sampling correlation as a function of lag or phase step. In FD-TOF, the lag is not the primary scan variable; instead, the system sweeps the modulation frequency $\omega \in \Omega \subset \mathbb{R}$2 and records the demodulated complex response at each sampled frequency (Kadambi et al., 2015). This recasts range recovery as a spectral estimation problem in the frequency variable.

A plausible implication is that the universal formulation is not merely a restatement of phase-TOF, but a reparameterization that places single-frequency phase estimation, multi-frequency fitting, and delay-spectrum reconstruction in the same algebraic family.

2. FD-TOF measurement and delay recovery

FD-TOF sweeps the modulation frequency over a bandwidth $\omega \in \Omega \subset \mathbb{R}$3 and, for each $\omega \in \Omega \subset \mathbb{R}$4, records the demodulated complex response. Under ideal synchronous demodulation,

$\omega \in \Omega \subset \mathbb{R}$5

where $\omega \in \Omega \subset \mathbb{R}$6 is the demodulated ambient or background contribution. In practice, $\omega \in \Omega \subset \mathbb{R}$7 is described as a slowly varying real offset that can be estimated and subtracted.

For a single path, the response reduces to

$\omega \in \Omega \subset \mathbb{R}$8

with phase

$\omega \in \Omega \subset \mathbb{R}$9

The delay is then obtained from the phase slope,

$f$0

Because $f$1 is a sum of complex exponentials across frequency, delay recovery can also be posed as inverse Fourier transformation over the frequency axis:

$f$2

Under ideal infinite-bandwidth conditions, this yields

$f$3

With finite bandwidth and discrete sampling, the Dirac impulses are broadened into the point-spread function determined by the inverse transform of the selected frequency window. Peaks in $f$4 locate distinct path delays and thereby enable multipath separation (Kadambi et al., 2015).

This formulation places two reconstruction strategies in direct correspondence: slope-based estimation from linear phase progression, and transform-based estimation from the delay-domain spectrum. The former emphasizes phase continuity and unwrapping; the latter emphasizes bandwidth, sampling density, and window design.

3. Relation to conventional phase-based TOF

At a single modulation frequency $f$5, phase-based TOF represents distance through the wrapped phase $f$6 as

$f$7

with $f$8 phase wrapping every

$f$9

Within the universal variable-TOF formulation, this is the special case $\omega=2\pi f$0, meaning that the measurement is available at only one angular frequency. FD-TOF generalizes this by sampling many frequencies and estimating $\omega=2\pi f$1 either from the linear relation $\omega=2\pi f$2 or from the delay profile recovered by inverse transformation (Kadambi et al., 2015).

The data identify several consequences of this generalization. First, using multiple frequencies allows unambiguous $\omega=2\pi f$3 estimation within a range set by $\omega=2\pi f$4, even when single-frequency phase would wrap. Second, coherent use of many frequencies improves SNR and mitigates phase sampling errors at high $\omega=2\pi f$5. Third, the multi-frequency setting admits least-squares and parametric spectral estimators that have no analogue in the strictly single-frequency case.

A standard multi-frequency least-squares estimator fits the unwrapped phases $\omega=2\pi f$6 measured at frequencies $\omega=2\pi f$7 according to

$\omega=2\pi f$8

When multiple paths are present, parametric spectral estimators such as Prony and MUSIC/ESPRIT can be applied directly to $\omega=2\pi f$9 to estimate $$s(t)=A\cos(2\pi f t)=A\,\mathrm{Re}\{e^{j2\pi f t}\}.$$0.

This suggests that the universal formulation is best understood as a hierarchy: phase-TOF is the degenerate single-frequency limit, FD-TOF is the finite multi-frequency case, and chirp-coded or wideband modulation occupies the continuous-spectrum extension of the same model class.

4. Resolution, unambiguous range, and spectral trade-offs

Let the swept bandwidth be $$s(t)=A\cos(2\pi f t)=A\,\mathrm{Re}\{e^{j2\pi f t}\}.$$1, sampled with spacing $$s(t)=A\cos(2\pi f t)=A\,\mathrm{Re}\{e^{j2\pi f t}\}.$$2 at $$s(t)=A\cos(2\pi f t)=A\,\mathrm{Re}\{e^{j2\pi f t}\}.$$3 points. The fundamental scaling laws in the variable-TOF formulation are bandwidth laws.

The delay resolution is set by the inverse swept bandwidth,

$$s(t)=A\cos(2\pi f t)=A\,\mathrm{Re}\{e^{j2\pi f t}\}.$$4

and the corresponding distance resolution is

$$s(t)=A\cos(2\pi f t)=A\,\mathrm{Re}\{e^{j2\pi f t}\}.$$5

For a rectangular window over $$s(t)=A\cos(2\pi f t)=A\,\mathrm{Re}\{e^{j2\pi f t}\}.$$6, the delay-domain point-spread function is proportional to $$s(t)=A\cos(2\pi f t)=A\,\mathrm{Re}\{e^{j2\pi f t}\}.$$7. The exact constant depends on the chosen window, and the ICCP paper reports $$s(t)=A\cos(2\pi f t)=A\,\mathrm{Re}\{e^{j2\pi f t}\}.$$8 for its own definition and window; the common rule-of-thumb $$s(t)=A\cos(2\pi f t)=A\,\mathrm{Re}\{e^{j2\pi f t}\}.$$9 captures the scaling with bandwidth (Kadambi et al., 2015).

Discrete frequency sampling produces periodic replicas in delay, separated by

$\tau$0

which translates to

$\tau$1

Accordingly, tighter frequency spacing increases unambiguous range, whereas larger total bandwidth improves resolution.

Windowing controls sidelobe behavior. A rectangular window yields high sidelobes and a sinc-shaped point-spread function. Tapered windows such as Hann, Hamming, and Blackman suppress sidelobes at the cost of wider main lobes and hence poorer resolution. The number of samples $\tau$2 controls the sidelobe floor and numerical stability of estimators, while the span $\tau$3 controls main-lobe width.

The separability condition for multipath follows directly from these relations: multiple paths appear as distinct peaks in $\tau$4 provided their delays satisfy

$\tau$5

or, equivalently,

$\tau$6

This is the operational criterion under which FD-TOF converts multipath from an ill-conditioned phase mixture into a resolvable delay spectrum.

5. OCT analogy and universal reconstruction strategies

The spectral-domain interpretation of FD-TOF is explicitly compared to spectral-domain interferometric ranging in optical coherence tomography. In OCT, the detected interferometric signal versus wavenumber $\tau$7 is modeled, omitting DC terms, as

$\tau$8

where $\tau$9 are reflector depths. Identifying $\alpha$0 shows the same dependence on a spectral variable as in FD-TOF (Kadambi et al., 2015).

Fourier transformation with respect to $\alpha$1 recovers reflectivity versus depth in OCT. Its axial resolution depends on optical source bandwidth $\alpha$2 around center wavelength $\alpha$3, and for a Gaussian spectrum is given by

$\alpha$4

The analogous FD-TOF relation is

$\alpha$5

Both are therefore bandwidth-limited ranging systems in which depth resolution is set by the inverse spectral span.

Within the universal variable-TOF model,

$\alpha$6

three reconstruction strategies are identified:

Strategy	Form	Condition emphasized
Phase-slope estimation	$\alpha$7	Reliable phase unwrapping, linear phase behavior
Fourier inversion	$\alpha$8	Sufficient bandwidth and sampling density
Parametric spectral estimation	Prony/MUSIC/ESPRIT	Small number of discrete paths, high SNR, correct model order

Fourier inversion requires sufficient bandwidth $\alpha$9 and sampling density $b$0 to achieve target $b$1 and sidelobe performance. Phase-slope estimation requires coherent reference stability and negligible dispersion so that the phase remains linear in $b$2. Parametric methods can deliver super-resolution beyond $b$3 if SNR is high and model order is correct, but their stability depends on conditioning, path spacing $b$4, and SNR (Kadambi et al., 2015).

6. Noise, high-frequency operation, and implementation constraints

The measurement model with noise is

$b$5

where $b$6 is complex, zero-mean, i.i.d. Gaussian with variance $b$7 per real and imaginary component. For a single path of amplitude $b$8, the Fisher information for $b$9 accumulates across frequencies proportionally to the mean-square bandwidth. Under uniform sampling over a span $r(t)=\alpha \cos(2\pi f (t-\tau))+b,$0 with $r(t)=\alpha \cos(2\pi f (t-\tau))+b,$1 points and fixed $r(t)=\alpha \cos(2\pi f (t-\tau))+b,$2,

$r(t)=\alpha \cos(2\pi f (t-\tau))+b,$3

where $r(t)=\alpha \cos(2\pi f (t-\tau))+b,$4 and $r(t)=\alpha \cos(2\pi f (t-\tau))+b,$5 depends on sampling distribution, window, and nuisance parameters. The canonical scaling is quadratic in $r(t)=\alpha \cos(2\pi f (t-\tau))+b,$6 and linear in $r(t)=\alpha \cos(2\pi f (t-\tau))+b,$7: doubling bandwidth quarters the variance. Coherent integration across frequency increases SNR and reduces estimator variance (Kadambi et al., 2015).

The data further identify practical noise sources: phase noise in the local reference, sampling jitter in $r(t)=\alpha \cos(2\pi f (t-\tau))+b,$8, and ambient light contribute to $r(t)=\alpha \cos(2\pi f (t-\tau))+b,$9 and to a slowly varying bias $$r(t)\approx \mathrm{Re}\{\alpha e^{j2\pi f (t-\tau)}+b\}.$$0. FD-TOF is described as robust to ambient because demodulation and subsequent coherent processing suppress DC and low-frequency background. Phase-slope estimators require reliable phase unwrapping, whereas sweep-based spectral inversion sidesteps explicit unwrapping.

The required hardware architecture comprises a frequency-agile illumination source and reference clock, a sensor able to lock in synchronously at each $$r(t)\approx \mathrm{Re}\{\alpha e^{j2\pi f (t-\tau)}+b\}.$$1 and output complex $$r(t)\approx \mathrm{Re}\{\alpha e^{j2\pi f (t-\tau)}+b\}.$$2, and coherent control that maintains a stable phase reference across the full sweep. The principal trade-offs involve dwell time per frequency, sample count $$r(t)\approx \mathrm{Re}\{\alpha e^{j2\pi f (t-\tau)}+b\}.$$3, total sweep duration, detector bandwidth, front-end linearity, calibration of frequency-dependent amplitude and phase distortions, and exposure consistency when using an integrating “slow” camera.

The experimental insight highlighted in the ICCP paper is that a 10–30 MHz sweep observes only a fraction of a cycle in the primal frequency domain, making frequency estimation harder. Nonetheless, measurable differentiation of depths was demonstrated and the concept was verified. Simulations showed FD-TOF maintains lower percent error than phase-TOF at very low SNR, while phase-TOF can be superior at high SNR (Kadambi et al., 2015).

7. Comparative significance and scope of the universal formulation

The universal variable-TOF formulation consolidates several TOF modalities into a single representation. Phase-based TOF corresponds to the single-$$r(t)\approx \mathrm{Re}\{\alpha e^{j2\pi f (t-\tau)}+b\}.$$4 specialization. FD-TOF corresponds to the multi-frequency case. Chirp-coded and wideband modulation are incorporated by treating the signal spectrum over $$r(t)\approx \mathrm{Re}\{\alpha e^{j2\pi f (t-\tau)}+b\}.$$5 and applying the same inversion principles after coherent demodulation.

Several comparative points follow directly from the source formulation. Reduced phase wrapping arises because FD-TOF estimates $$r(t)\approx \mathrm{Re}\{\alpha e^{j2\pi f (t-\tau)}+b\}.$$6 from phase slope or spectral inversion rather than from a single wrapped phase. The unambiguous range scales as $$r(t)\approx \mathrm{Re}\{\alpha e^{j2\pi f (t-\tau)}+b\}.$$7 rather than $$r(t)\approx \mathrm{Re}\{\alpha e^{j2\pi f (t-\tau)}+b\}.$$8. Multipath resolution arises because paths separated by at least $$r(t)\approx \mathrm{Re}\{\alpha e^{j2\pi f (t-\tau)}+b\}.$$9 become resolvable as distinct peaks. High-frequency compatibility follows because, as $\omega \in \Omega \subset \mathbb{R}$00 increases into hundreds of MHz or GHz, precise phase stepping becomes electronically challenging, whereas FD-TOF leverages frequency sweep and coherent processing. Broadband flexibility follows because arbitrary frequency grids, windows, and parametric estimators can be used (Kadambi et al., 2015).

The quantitative illustration given in the source contrasts single-frequency phase-TOF at $\omega \in \Omega \subset \mathbb{R}$01 MHz, where

$\omega \in \Omega \subset \mathbb{R}$02

with FD-TOF using $\omega \in \Omega \subset \mathbb{R}$03 MHz, for example a 10–110 MHz sweep, where

$\omega \in \Omega \subset \mathbb{R}$04

Under different window conventions the ICCP paper reports approximately $\omega \in \Omega \subset \mathbb{R}$05 m. Increasing $\omega \in \Omega \subset \mathbb{R}$06 by a factor of $\omega \in \Omega \subset \mathbb{R}$07, to $\omega \in \Omega \subset \mathbb{R}$08 GHz, improves axial resolution by the same factor and allows separation of paths differing by a few centimeters. The unambiguous range remains

$\omega \in \Omega \subset \mathbb{R}$09

so that $\omega \in \Omega \subset \mathbb{R}$10 MHz yields $\omega \in \Omega \subset \mathbb{R}$11 m, independent of the center frequency.

A common misconception is that FD-TOF merely replicates phase-TOF with more measurements. The universal formulation suggests a narrower and more technical statement: FD-TOF changes the estimation problem from single-phase decoding to frequency-domain inference over $\omega \in \Omega \subset \mathbb{R}$12, with different ambiguity structure, different noise accumulation, and explicit delay-domain resolvability. Another plausible implication is that the main conceptual contribution is not only a new architecture, but a unifying spectral viewpoint in which delay estimation, multipath separation, and bandwidth-limited resolution become manifestations of the same model.

In summary, the universal variable-TOF formulation is the representation

$\omega \in \Omega \subset \mathbb{R}$13

under which phase-based methods appear as the single-frequency case and FD-TOF exploits frequency diversity to estimate $\omega \in \Omega \subset \mathbb{R}$14 through phase slope, Fourier inversion, or parametric spectral recovery. Its performance limits are governed by bandwidth $\omega \in \Omega \subset \mathbb{R}$15, frequency spacing $\omega \in \Omega \subset \mathbb{R}$16, windowing, and SNR; its principal advantages are reduced phase wrapping, multipath separability, and improved robustness at high modulation frequencies (Kadambi et al., 2015).