When evaluating the speed of a shortwave (HF) link, looking at the mean latency alone isn’t enough; it doesn’t tell the whole story. To get a complete picture, we need to consider latency variance as well.
Unlike fiber-based links, which offer consistent and predictable latency thanks to dedicated pathways, low interference, and minimal signal processing delays, shortwave links are subject to the following key sources of latency variance:
Multipath Phenomenon
Also known as “ionospheric fading” or “skywave multipath,” this occurs when signals carrying the same message take different paths before reaching the receiver. As a result, the end-to-end travel time varies. A “delay spread” in the professional engineering jargon.
(For example, a 10,000 km HF signal path can experience significant variation depending on conditions.)
The images below show two examples of the multipath phenomenon.
Daytime Changes
The ionospheric layers that propagate shortwave signals change in height between day and night—rising higher at night. As a result, signals travel farther up in the sky during nighttime, leading to slightly higher latency compared to daytime conditions.
How to account for daytime and nighttime effects on long cross-Atlantic or cross-Pacific paths—where daylight hours vary significantly—is a topic of its own. But one thing is clear: these daily latency shifts contribute to overall variance.
Naturally (pun not intended), as daylight hours change throughout the year, there’s also a seasonal cycle on top of the daily cycle. Read more about that—and a third 11-years cycle! – here.
Signal Detection Time
Another major contributor to latency variance is the time it takes to detect a signal. In latency-optimized systems, this time fluctuates based on HF conditions and any local “man-made {electrical} noise”—poor HF conditions or interference result in slightly longer detection times.
The result?
A rather wide and right-skewed latency dress, where longer delays are far more likely than shorter ones. This is because latency can always increase due to adverse conditions, but there’s a hard lower limit dictated by physics.
For example, on RAFT links, latency standard deviation can range from 0.4ms to 1ms, depending on the link (the chart below is our Chicago to Tokyo link).
So, when requesting latency figures, the full answer should include not just the mean but also the min, max, standard deviation (SD), and a histogram—all essential for accurate backtesting simulations.
