Conference Agenda

As spectrum use becomes more crowded, a basic monitoring problem becomes harder to solve: how to tell when behavior in a band has changed in a meaningful way, and how to describe that change when the identity of the active user is not known. This problem matters both for research and for policy. In many real monitoring settings, passive measurements can be collected continuously over long periods of time, but those measurements do not come with reliable labels identifying which transmitter, service, or system is active at a given moment. This is especially true outside clean, well-separated licensed channels. In broader or more heavily shared spectral environments, analysts often rely on simple occupancy summaries, threshold rules, or manual inspection of plots. These approaches may show that activity increased or decreased, but they usually do not show what kind of change occurred, whether it was brief or sustained, or whether it reflected a meaningful shift in how the spectrum was being used. This paper asks whether long-term passive spectrum measurements can be used to detect meaningful changes in spectrum use and summarize those changes in a form that analysts can interpret, even when transmitter identity is not known.

This paper studies an interpretable method for detecting and summarizing change in long-term passive spectrum monitoring data without requiring transmitter labels. The approach is designed for multiple spectral regions with different usage conditions. The data cover several bands between 44 MHz and 900 MHz, including 44–48 MHz, 100–200 MHz, 400–460 MHz, 460–463 MHz, 463–516 MHz, 700–800 MHz, and 800–900 MHz. Earlier analysis of the 460–463 MHz portion of the dataset focused on a set of 19 narrow, non-overlapping licensed channels and showed that repeated patterns of channel behavior could be identified over time. The present study extends that line of work to a broader and more realistic monitoring setting, where spectral structure is less clean and channel attribution is not always available in advance.

The central idea is to represent local spectrum behavior through short time windows and to describe each window using physically meaningful features rather than raw measurements alone. These features capture median and peak power, duty cycle, burst structure, impulsiveness, sweep-to-sweep fluctuation, and spectral spread. Windows with similar behavior are grouped into a small number of common pattern types. Once these patterns are learned, longer intervals such as days can be summarized by the distribution of pattern types they contain. Change is then defined not as a shift in a single power statistic, but as a shift in the distribution of recurring behaviors relative to a historical baseline. This makes it possible to detect periods whose internal structure differs from normal conditions even when no transmitter identity is known.

A main contribution of the framework is that it does not stop at flagging unusual intervals. For each detected event, it also provides a compact visual summary of how the event differs from the baseline and from nearby time periods. In practice, this allows an analyst to see whether the detected change reflects a sudden burst of activity, a sustained increase in use, the disappearance of previously common behavior, or a broader redistribution across patterns of activity. This distinction matters in policy-relevant monitoring settings. In shared or uncertain spectral environments, the key question is often not simply whether a band became more occupied, but whether its usage pattern changed in a way that deserves closer review.

The paper does not claim to identify exactly who transmitted or why a change occurred. Instead, it shows that long-term passive measurements can provide a useful screening tool. By scanning large volumes of unlabeled data, the method can highlight statistically unusual changes and present them in a form that analysts can quickly interpret. This may be especially useful in shared-spectrum environments, where identifying specific transmitters is often difficult but recognizing unusual changes in activity still matters for monitoring and oversight.