Abstract
The present work investigates two properties of level crossings of a stationary Gaussian process $X(t)$ with autocorrelation function $R_X(\tau)$. We show firstly that if $R_X(\tau)$ admits finite second and fourth derivatives at the origin, the length of up-excursions above a large negative level $-\gamma$ is asymptotically exponential as $-\gamma \to -\infty$. Secondly, assuming that $R_X(\tau)$ admits a finite second derivative at the origin and some defined properties, we derive the mean number of crossings as well as the length of successive excursions above two subsequent large levels. The asymptotic results are shown to be effective even for moderate values of crossing level. An application of the developed results is proposed to derive the probability of successive excursions above adjacent levels during a time window.
Introduction
Understanding the behavior of random signals as they cross specific thresholds is a fundamental problem in stochastic process theory with wide-ranging applications in engineering. While level-crossing theory has been studied since the 1940s, two specific scenarios remained insufficiently characterized: the behavior of signals above very low "floors" and the statistical properties of "successive excursions"—where a signal hits a second, higher peak after already crossing an initial threshold.
In fields like wireless communications, engineers must determine not just if a signal will drop or spike, but the duration of such events. For instance, a radio link failure typically occurs only when a signal remains below a threshold for a minimum duration. This paper fills these gaps by providing new mathematical tools to predict these durations and crossing probabilities.
Methodology and Validation
The study employs a combination of classical stochastic calculus and asymptotic analysis. To address low-level excursions, the author exploits the symmetry of Gaussian processes (centered at zero) to map the behavior of excursions above deep negative levels to the known behavior of intervals between high-level crossings. By leveraging the fact that high-level up-crossing intervals asymptotically follow a Poisson process, the paper proves that low-level excursion durations must follow an exponential distribution due to the memoryless property.
For the successive levels problem, the analysis utilizes the property that high-level excursions tend to resemble downwards-opening parabolas. By treating the signal's peak as a Rayleigh random variable, the author calculates the probability and duration of a signal intersecting a second, higher threshold ($\gamma_2$) given it has already crossed a lower one ($\gamma_1$).
The theoretical results were validated using numerical simulations of two types of stochastic processes: Gauss-Markov processes (characterized by exponential autocorrelation) and Gaussian processes with squared exponential autocorrelation, which produce smoother sample paths.
Key Insights
A central finding of this work is the dichotomy between predictability and randomness in signal trajectories. At high peaks, a signal's path is highly predictable and parabolic. Conversely, when a signal hovers above a very low floor, its duration is governed by an exponential distribution, making it essentially "memoryless."
The author also introduces the concept of a "critical length" ($\tau_1^*$). If a signal remains above the first threshold ($\gamma_1$) for longer than this critical time, it is mathematically guaranteed (with probability 1) to have crossed the higher threshold ($\gamma_2$) during that window.
Results and Applications
The paper establishes several key formal results:
- Theorem 5: Proves that the length of an up-excursion above a large negative level $-\gamma$ is asymptotically exponential with a rate equal to the up-crossing rate.
- Theorem 6 & Proposition 8: Derive the exact mean number of crossings and the distribution of the duration ($T_2$) spent above a second level $\gamma_2$ given an excursion above $\gamma_1$.
A practical application is demonstrated in the context of wireless networking. The derived formulas allow for calculating the probability of successive excursions within a specific time window $T$. This is directly applicable to optimizing "handover" triggers in LTE or 5G networks, where devices must decide to switch base stations based on signal strength and duration to maintain connectivity.
Limitations and Future Work
While the results are robust even at moderate signal levels, they rely on the assumption of a stationary Gaussian process. Physical phenomena like thermal noise are well-modeled this way, but other radio channels (e.g., Nakagami or Rayleigh fading) may require further investigation. Additionally, the proofs require the signal's autocorrelation function to have finite second and fourth derivatives at the origin, which may exclude certain "rough" stochastic processes.
Summary updated