Uniform Clock Process: Theory & Applications

• Uniform Clock Process is a term that represents a family of clocks exhibiting homogeneous timing properties across various domains, including stochastic processes, relativistic kinematics, digital hardware, and distributed algorithms.
• In stochastic analysis, it refers to a logarithmically rescaled clock process governed by a functional central limit theorem, while in hardware and distributed algorithms it denotes equal phase partitioning and uniform activation probabilities.
• The concept extends to minimal-clock models and clock tomography, emphasizing distinctions between intrinsic dynamics, calibration uniformity, and the reconstructibility of timing signals.

Searching arXiv for the cited paper and closely related usages of “uniform clock” / “clock process” to ground the article. “Uniform clock process” is not a single canonical object across the literature surveyed here. In stochastic analysis, it denotes a logarithmically rescaled clock attached to a positive self-similar Markov process, with a functional central limit theorem governing long-time fluctuations. In relativistic kinematics, it denotes a family of clocks in a uniformly accelerated frame whose synchronization offsets and relative rates remain stable for as long as the acceleration tensor is constant. In digital timing hardware, it denotes a multiphase interpolation scheme that partitions one clock period into equal timing bins. In distributed algorithms, it distinguishes uniform from non-uniform Poisson activation laws for gossip updates. In minimal-clock and clock-tomography work, the term is approached more indirectly, through regular state progression or through waiting-time distributions constrained by a decay envelope rather than by exact periodicity (Caballero et al., 2020, Scarr, 2019, Qi et al., 2015, Jafarizadeh, 2015, Robu et al., 2019, Nurgalieva et al., 2024).

1. Range of meanings

Across these domains, the common element is not a single definition but a shared concern with clocks whose behavior is homogeneous in an appropriate technical sense. The relevant sense of homogeneity depends on the underlying formalism: self-similar time change, frame synchronization, equal phase partition, activation symmetry, or controlled waiting-time tails.

Domain	Clock object	Sense of uniformity
Positive self-similar Markov processes	$$t\mapsto (\log T)^{-1/2}\left(\int_1^{T^t}\frac{dr}{X_r^\alpha}-\frac{t\log T}{\alpha p}\right)$$	Functional limit uniform over compact logarithmic-time intervals
Uniform acceleration	Clocks at rest in one accelerated frame	Constant position-dependent time dilation and stable synchronization
FPGA time-to-digital conversion	Four 1 GHz clocks with phases $0^\circ,45^\circ,90^\circ,135^\circ$	One period divided into eight equal parts
Gossip algorithms	Node Poisson clocks with rates $\lambda_i$	Uniform means $P_i=1/N$
Minimal and physical clocks	Finite-state clocks or finite-dimensional waiting-time processes	Regular counters or well-behaved timing statistics

This range matters because the phrase can otherwise be misleading. In some settings, “uniform” refers to equal phase spacing or equal activation probabilities; in others, it refers to a functional scaling limit or to the existence of a stable synchronization law. A plausible implication is that the phrase is best understood relationally: it characterizes how a clock is sampled, compared, or embedded in a larger dynamical structure, rather than naming a single universal stochastic process.

2. Uniform rescaled clocks in stochastic-process theory

In the most technically developed use of the term, a clock process is attached to a positive self-similar Markov process (pssMp) of index $\alpha>0$. Such a process $(X,\mathbb Q_a)_{a>0}$ satisfies

$$\left(\{b\,X_{b^{-\alpha}t},\, t\ge 0\}, \mathbb Q_a\right)\stackrel{d}{=} \left(\{X_t,\, t\ge 0\}, \mathbb Q_{ba}\right),$$

and, when it never hits $0$, Lamperti’s representation writes it as the exponential of a Lévy process under a time change determined by the inverse of an exponential functional. The associated clock is

$T^{(X)}(t)=\int_0^t \frac{ds}{X_s^\alpha},$

with inverse $A^{(X)}$, and the Lamperti identity links it to the Lévy inverse by $0^\circ,45^\circ,90^\circ,135^\circ$0 (Caballero et al., 2020).

Under the assumptions that the Lévy process $0^\circ,45^\circ,90^\circ,135^\circ$1 has Laplace exponent $0^\circ,45^\circ,90^\circ,135^\circ$2, that $0^\circ,45^\circ,90^\circ,135^\circ$3, and that the drift $0^\circ,45^\circ,90^\circ,135^\circ$4, the first-order asymptotics are a law of large numbers: $0^\circ,45^\circ,90^\circ,135^\circ$5 for every $0^\circ,45^\circ,90^\circ,135^\circ$6, and similarly under the entrance law $0^\circ,45^\circ,90^\circ,135^\circ$7. The clock therefore grows like $0^\circ,45^\circ,90^\circ,135^\circ$8 at leading order.

The main refinement is the invariance principle for the logarithmically rescaled, centered clock. Under $0^\circ,45^\circ,90^\circ,135^\circ$9, as $\lambda_i$0,

$\lambda_i$1

where $\lambda_i$2 is standard Brownian motion and

$\lambda_i$3

The same limit holds under $\lambda_i$4 for every $\lambda_i$5, provided one of two recurrence or ergodicity conditions on $\lambda_i$6 holds: either there exists $\lambda_i$7 such that $\lambda_i$8, or there exists $\lambda_i$9 such that $P_i=1/N$0. This theorem proves the conjecture in Remark 4 of Demni–Rouault–Zani and gives an explicit asymptotic variance in terms of $P_i=1/N$1 rather than an abstract covariance (Caballero et al., 2020).

The proof identifies the centered clock as an additive functional of the generalized Ornstein–Uhlenbeck process

$P_i=1/N$2

which is stationary, Markov, and ergodic under $P_i=1/N$3, with invariant law

$P_i=1/N$4

A key identity is

$P_i=1/N$5

This converts the clock problem into a stationary additive-functional problem, to which Bhattacharya’s functional CLT is applied through the Poisson-equation pair $P_i=1/N$6 and $P_i=1/N$7, with $P_i=1/N$8. The asymptotic variance is first written as

$P_i=1/N$9

and then evaluated explicitly from the Mellin transform recursion $\alpha>0$0 for $\alpha>0$1.

A broader stochastic-process use of “clock” appears in local-time penalization for one-dimensional diffusions. There, a clock is “a family of random times parametrized by $\alpha>0$2 in a directed set such that $\alpha>0$3 a.s.” Examples include the exponential clock $\alpha>0$4, the hitting-time clock $\alpha>0$5, and the inverse local time clock $\alpha>0$6. The limiting penalized process generally depends on the chosen clock, so there is no canonical universal clock in that theory (Profeta et al., 2016). This contrast is significant: in the pssMp setting, the logarithmic rescaling isolates a specific “uniform clock process,” whereas in local-time penalization the clock is a design choice that changes the asymptotic law.

3. Uniform acceleration and stable synchronization

In relativistic kinematics, the phrase is tied to the structure of uniformly accelerated reference systems. A system is said to be uniformly accelerated if and only if all of the clocks in the system can be synchronized to each other, and the clocks remain synchronized as long as the acceleration remains uniform. The formal definition uses a constant antisymmetric acceleration tensor $\alpha>0$7 in the evolution equation

$\alpha>0$8

with the $\alpha>0$9 decomposition

$(X,\mathbb Q_a)_{a>0}$0

where $(X,\mathbb Q_a)_{a>0}$1 is a 3-vector of linear acceleration and $(X,\mathbb Q_a)_{a>0}$2 is a 3-vector interpreted as angular velocity. The frame itself is built from a one-parameter family of instantaneously comoving inertial frames whose tetrad evolves by generalized Fermi–Walker transport,

$(X,\mathbb Q_a)_{a>0}$3

(Scarr, 2019).

The synchronization argument depends on the paper’s time-dilation law for a point fixed in the accelerated frame. The key claim is that for such a point the time dilation between the clock at spatial point $(X,\mathbb Q_a)_{a>0}$4 and the clock at the origin is constant in time, because $(X,\mathbb Q_a)_{a>0}$5, $(X,\mathbb Q_a)_{a>0}$6, and $(X,\mathbb Q_a)_{a>0}$7 are constant in the frame’s own description. Once the initial readings are synchronized by a light-signal exchange and the remote clock’s rate is adjusted by this constant factor, the clocks remain synchronized indefinitely so long as the acceleration tensor remains constant. Any two clocks can then be synchronized by composition through the origin (Scarr, 2019).

The rotating-disk case is presented as a direct application. For a disk rotating with constant angular velocity, the motion is treated as a uniformly accelerated frame with nonzero $(X,\mathbb Q_a)_{a>0}$8 and suitable $(X,\mathbb Q_a)_{a>0}$9. The paper argues that the usual claim of an unavoidable “time gap” results from applying the inertial special-relativistic time-dilation law rather than the position-dependent law appropriate to accelerated motion. Within that framework, clocks on a rotating disk can therefore be synchronized and kept synchronized, and the “uniform clock process” is the coherent family of rest clocks with fixed relative rates and fixed synchronization offsets (Scarr, 2019).

This usage differs sharply from the stochastic one. Uniformity no longer concerns weak convergence or ergodic averaging; it concerns a global simultaneity structure induced by a constant acceleration tensor. A plausible implication is that “uniform” here refers to invariance of calibration data across time rather than equality of tick intervals in any inertial sense.

4. Uniform phase partition in digital timing hardware

In digital instrumentation, a uniform clock process can be realized by a multiphase interpolation architecture for time-to-digital conversion. A high-precision TDC implemented in a single FPGA uses a PLL, discriminator circuit, coarse module, fine time module, and data combination/output block. The input reference clock is 125 MHz, and the PLL multiplies it by 8 to produce a 1 GHz internal clock. That 1 GHz clock is used both for coarse counting and as the basis for fine interpolation (Qi et al., 2015).

The interpolation mechanism uses four clocks of the same frequency with phase shifts $$\left(\{b\,X_{b^{-\alpha}t},\, t\ge 0\}, \mathbb Q_a\right)\stackrel{d}{=} \left(\{X_t,\, t\ge 0\}, \mathbb Q_{ba}\right),$$0, $$\left(\{b\,X_{b^{-\alpha}t},\, t\ge 0\}, \mathbb Q_a\right)\stackrel{d}{=} \left(\{X_t,\, t\ge 0\}, \mathbb Q_{ba}\right),$$1, $$\left(\{b\,X_{b^{-\alpha}t},\, t\ge 0\}, \mathbb Q_a\right)\stackrel{d}{=} \left(\{X_t,\, t\ge 0\}, \mathbb Q_{ba}\right),$$2, and $$\left(\{b\,X_{b^{-\alpha}t},\, t\ge 0\}, \mathbb Q_a\right)\stackrel{d}{=} \left(\{X_t,\, t\ge 0\}, \mathbb Q_{ba}\right),$$3, each with duty ratio $$\left(\{b\,X_{b^{-\alpha}t},\, t\ge 0\}, \mathbb Q_a\right)\stackrel{d}{=} \left(\{X_t,\, t\ge 0\}, \mathbb Q_{ba}\right),$$4. Because the adjacent phase difference is $$\left(\{b\,X_{b^{-\alpha}t},\, t\ge 0\}, \mathbb Q_a\right)\stackrel{d}{=} \left(\{X_t,\, t\ge 0\}, \mathbb Q_{ba}\right),$$5, one full period is divided into eight equal parts: $$\left(\{b\,X_{b^{-\alpha}t},\, t\ge 0\}, \mathbb Q_a\right)\stackrel{d}{=} \left(\{X_t,\, t\ge 0\}, \mathbb Q_{ba}\right),$$6 With the 1 GHz clock, $$\left(\{b\,X_{b^{-\alpha}t},\, t\ge 0\}, \mathbb Q_a\right)\stackrel{d}{=} \left(\{X_t,\, t\ge 0\}, \mathbb Q_{ba}\right),$$7, so the theoretical resolution becomes $$\left(\{b\,X_{b^{-\alpha}t},\, t\ge 0\}, \mathbb Q_a\right)\stackrel{d}{=} \left(\{X_t,\, t\ge 0\}, \mathbb Q_{ba}\right),$$8. The paper states that “By utilizing four multiphase clocks to make up the interpolation clock, one clock period can be divided into eight uniform parts” (Qi et al., 2015).

The measurement model combines coarse and fine timing according to

$$\left(\{b\,X_{b^{-\alpha}t},\, t\ge 0\}, \mathbb Q_a\right)\stackrel{d}{=} \left(\{X_t,\, t\ge 0\}, \mathbb Q_{ba}\right),$$9

where $0$0 is the coarse counter value and $0$1 are the fine offsets of the start and stop signals relative to clock edges. The fine module uses four D flip-flops and one encoder. At the arrival of the measured signal, the flip-flops sample the four phase-shifted clocks, generating a 4-bit pattern that is converted into a 3-bit code identifying one of eight time regions within the 1 ns period. The coarse module consists of a 9-bit Gray-code counter, a D flip-flop, and a Gray-code-to-binary converter, with Gray coding used to reduce transition ambiguity (Qi et al., 2015).

Uniformity in this setting is a design property produced by three conditions working together: PLL-based clock generation, equal $0$2 phase spacing, and $0$3 duty cycle. The duty cycle is decisive because it makes the rising and falling edges evenly spaced, so the bins are uniform. The discriminator circuit, composed of a pulse stretching circuit and an identification circuit, allows the TDC to operate even when the order of the two input signals is not known in advance.

The reported experiments support the theoretical bin width. The measured standard deviations were 113 ps and 107 ps for intervals in the neighborhood of the 125 ps bin width, and the paper states that these results are consistent with the theoretical value. For linearity, the interval was varied from 10 ns to 300 ns in steps of 10 ns; the linear fit yielded correlation coefficient $0$4 and nonlinear error $0$5, with fitted relation essentially $0$6 (Qi et al., 2015). Here, “uniform clock process” is therefore a hardware timing lattice: a deterministic subdivision of a master period into equal interpolation sectors.

5. Uniform and non-uniform activation clocks in gossip algorithms

In distributed consensus, the clock is the random activation mechanism for asynchronous pairwise updates. Each node $0$7 has a Poisson clock with rate $0$8. When the clock of node $0$9 ticks, it chooses a neighbor $T^{(X)}(t)=\int_0^t \frac{ds}{X_s^\alpha},$0 with probability $T^{(X)}(t)=\int_0^t \frac{ds}{X_s^\alpha},$1, and the two states are averaged via

$T^{(X)}(t)=\int_0^t \frac{ds}{X_s^\alpha},$2

Because Poisson processes are memoryless, the waiting time to the next tick has exponential distribution $T^{(X)}(t)=\int_0^t \frac{ds}{X_s^\alpha},$3, which makes the model natural for asynchronous gossip. When the node clocks are merged, the merged process is Poisson with rate $T^{(X)}(t)=\int_0^t \frac{ds}{X_s^\alpha},$4, and the probability that the $T^{(X)}(t)=\int_0^t \frac{ds}{X_s^\alpha},$5-th merged tick came from node $T^{(X)}(t)=\int_0^t \frac{ds}{X_s^\alpha},$6 is

$T^{(X)}(t)=\int_0^t \frac{ds}{X_s^\alpha},$7

(Jafarizadeh, 2015).

Uniform clock distribution means all nodes have identical Poisson rates, so $T^{(X)}(t)=\int_0^t \frac{ds}{X_s^\alpha},$8. Non-uniform clock distribution means the $T^{(X)}(t)=\int_0^t \frac{ds}{X_s^\alpha},$9 are adjusted so that the activation probabilities $A^{(X)}$0 are not uniform. This enlarges the feasible set of the optimization problem. The paper formulates the Fastest Classical Gossip Algorithm (FCGA) as a semidefinite program minimizing the second-largest eigenvalue of the mean gossip operator $A^{(X)}$1. It states that the optimal results obtained for uniform clock distribution are suboptimal compared to those of the non-uniform one, and that for non-uniform distribution the optimal answer is not unique (Jafarizadeh, 2015).

One constructive route to an optimal non-uniform solution uses the optimal continuous-time consensus weights together with detailed balance: $A^{(X)}$2 so that

$A^{(X)}$3

and $A^{(X)}$4. The paper states that this can make the gossip convergence rate match the optimal continuous-time consensus rate. It also notes that for some topologies, such as the wheel with $A^{(X)}$5, the uniform-clock optimum is strictly suboptimal compared to the non-uniform-clock optimum (Jafarizadeh, 2015).

The same clock-distribution idea extends to quantum gossip. By expanding the density matrix in generalized Gell-Mann matrices, the quantum update is transformed into a classical gossip-like state update on the coefficient vector, and the Fastest Quantum Gossip (FQG) problem becomes a convex optimization problem addressable by semidefinite programming. In this literature, then, a “uniform clock process” is not a deterministic metronome but the special case $A^{(X)}$6 of a Poisson-driven activation law; the main result is that relinquishing this uniformity can strictly improve convergence.

6. Minimal clocks, regular counters, and operationally well-behaved signals

Minimal-clock models treat uniformity as a property of information-bearing state progression rather than as exact synchronization or equal activation probability. In a framework of minimal abstract pure clocks, the clock is a discrete-time finite-state Markov chain optimized to encode time itself and nothing else, with performance measured by the mutual information $A^{(X)}$7. The clocks have discrete state spaces, a global synchronized tick, no environmental input, minimal memory capacity, and fixed transition kernels; the initial condition is $A^{(X)}$8 (Robu et al., 2019).

The basic 1-bit examples are the oscillator clock and the drop clock. The oscillator,

$A^{(X)}$9

is interpreted as a local clock: it resolves phase, such as odd versus even step, but not position in a long time horizon. The drop clock,

$0^\circ,45^\circ,90^\circ,135^\circ$00

produces exponential decay $0^\circ,45^\circ,90^\circ,135^\circ$01 for remaining in $0^\circ,45^\circ,90^\circ,135^\circ$02 and is interpreted as a global or lifetime-scale clock, distinguishing early from late but not phase. The paper’s central dichotomy is therefore oscillator for local time and drop clock for global elapsed time (Robu et al., 2019).

Within that framework, the nearest candidates to a uniform clock process are the deterministic oscillator at $0^\circ,45^\circ,90^\circ,135^\circ$03, deterministic cascades, and especially the composite clock at maximal allowed information flow $0^\circ,45^\circ,90^\circ,135^\circ$04, where the two-bit construction becomes a four-state counter. The optimized cascades, however, often show a condensation effect rather than uniform temporal spacing, and the bag of independent clocks becomes a heterogeneous mixture of one oscillator and multiple drop clocks. The paper’s bottom line is that no single minimal clock is uniformly best across all time scales; optimization is regime-dependent (Robu et al., 2019).

A complementary operational perspective comes from clock tomography. There, a finite physical clock is modeled as a clockwork $0^\circ,45^\circ,90^\circ,135^\circ$05 coupled to a classical tick register $0^\circ,45^\circ,90^\circ,135^\circ$06, with tick statistics $0^\circ,45^\circ,90^\circ,135^\circ$07 derived from the state

$0^\circ,45^\circ,90^\circ,135^\circ$08

The key structural result is that if the clockwork lives on a finite state space and the clock is guaranteed to tick eventually, then its waiting-time distribution must decay at least exponentially up to a finite polynomial factor: $0^\circ,45^\circ,90^\circ,135^\circ$09 for some $0^\circ,45^\circ,90^\circ,135^\circ$10. Equivalently, the moment generating function exists in a neighborhood of the origin, so the moments determine the distribution uniquely in that region (Nurgalieva et al., 2024).

Operational reconstruction is carried out by comparing the ordering of ticks of the target clock against those of a Poisson reference process. The reference has count distribution

$0^\circ,45^\circ,90^\circ,135^\circ$11

and, for a reset target clock with waiting-time density $0^\circ,45^\circ,90^\circ,135^\circ$12, the relative count probabilities are

$0^\circ,45^\circ,90^\circ,135^\circ$13

The relative moments $0^\circ,45^\circ,90^\circ,135^\circ$14 and the target moments $0^\circ,45^\circ,90^\circ,135^\circ$15 are connected by Stirling transforms, so the waiting-time distribution can be reconstructed from tick-order statistics. The paper describes the Poisson process as the simplest possible reference because it is memoryless, of minimal precision among i.i.d. clocks with $0^\circ,45^\circ,90^\circ,135^\circ$16, physically ubiquitous, and analytically tractable (Nurgalieva et al., 2024).

Taken together, these works distinguish two advanced senses of uniformity. One is counter-like progression through states, as in deterministic oscillators or the composite four-state counter. The other is statistical regularity sufficient for moment-based reconstruction, as in finite physical clocks with decay envelopes. In neither case is “uniform” synonymous with exact periodicity in every circumstance.

7. Conceptual synthesis and recurring distinctions

Several recurring distinctions organize the literature.

First, there is a distinction between uniformity of observation scale and uniformity of intrinsic dynamics. The pssMp result studies the clock under logarithmic observation $0^\circ,45^\circ,90^\circ,135^\circ$17, and uniformity refers to the functional limit over compact time intervals in that transformed scale (Caballero et al., 2020). By contrast, the FPGA TDC realizes uniformity directly in the underlying timing lattice by dividing one physical period into eight equal bins (Qi et al., 2015).

Second, there is a distinction between uniformity of rates and uniformity of calibration. In gossip algorithms, uniform clocks mean equal activation probabilities $0^\circ,45^\circ,90^\circ,135^\circ$18, and the main message is that such uniformity may be suboptimal (Jafarizadeh, 2015). In accelerated frames, the essential property is not equal ticking in an inertial sense but stable relative calibration through a constant position-dependent dilation law (Scarr, 2019).

Third, there is a distinction between uniform clocks as canonical objects and clock families as modeling devices. The diffusion penalization literature explicitly treats clocks as parameterized random times $0^\circ,45^\circ,90^\circ,135^\circ$19, and the limit law depends on the chosen clock (Profeta et al., 2016). The stochastic-clock invariance principle, by contrast, identifies a specific uniform rescaled process with a Brownian limit (Caballero et al., 2020).

Finally, there is a distinction between regularity as periodicity and regularity as reconstructibility. Minimal-clock models show that the most informative small clocks are often not uniform across scales: oscillators are local, drop clocks are global, cascades condense, and composite clocks only become clean counters in a maximal-coupling regime (Robu et al., 2019). Clock tomography shows that finite physical clocks need not be perfectly regular to be operationally tractable; it is enough that their waiting-time laws have a decay envelope and finite relative-moment structure against a Poisson reference (Nurgalieva et al., 2024).

The strongest general conclusion supported by these literatures is therefore not that there exists a unique uniform clock process, but that clock uniformity is domain-specific. In stochastic scaling limits it appears as Brownian universality on the $0^\circ,45^\circ,90^\circ,135^\circ$20 scale; in relativistic frame theory as persistent synchronization under constant $0^\circ,45^\circ,90^\circ,135^\circ$21; in hardware as equal phase subdivision; in gossip as a specific Poisson-rate constraint; and in minimal or physical clock theory as either regular counter structure or mathematically controlled waiting-time statistics.