Why GPS Underreports Sprint Max Speed and How to Fix It

A winger's GPS report after Saturday's match reads 28.9 km/h peak speed. Three days earlier, the same player ran 32.6 km/h on the same field during a flying 20m test with a radar gun locked on his hip. That 3.7 km/h gap is close to what separates a starting winger from a bench player on most speed leaderboards. The staff's first instinct is usually to blame effort or the unpredictable nature of match play. Sometimes that is the real story. More often, especially when the same gap shows up across the whole squad against every reference test run that season, the GPS unit is quietly clipping the top of the velocity curve before the number ever reaches the report.

This is not a broken unit. It is three things stacking on top of each other: a sampling rate too coarse to catch a true instantaneous peak, a smoothing filter doing exactly what it was built to do, and satellite geometry that shifts from field to field. What follows is what is actually happening inside a sub-10Hz unit, or a nominally 10Hz unit with a poor satellite lock that day, plus a field-testable protocol for telling the three causes apart instead of guessing which one is eating your numbers.

How GPS Actually Loses Peak Speed

GPS receivers do not measure velocity directly. They measure position at fixed intervals, then derive velocity by differentiation. A 10Hz unit produces one fix every 100ms; a 5Hz unit, one every 200ms. True peak velocity typically lasts well under half a second, sometimes barely 150-250ms in team-sport athletes who touch max speed only briefly between direction changes, so the single fastest window can fall between two fixes and never get sampled. The reported peak becomes the fastest interval the unit happened to catch, mathematically guaranteed to sit at or below the athlete's true max.

Sampling rate is only the first cut. Every unit applies smoothing, usually a moving average or Kalman filter, to the raw stream before calculating velocity, because raw position data is noisy enough to produce nonsensical spikes on its own. The filter removes that noise, but it also flattens genuine short-duration peaks the same way it flattens fake ones, since the algorithm cannot tell a real spike apart from a positioning error. Then there is satellite geometry, summarized in horizontal dilution of precision, or HDOP: a unit with ten or more satellites and an HDOP under 2.0 produces meaningfully tighter fixes than the same unit run near a grandstand roof or tree cover, where satellite count can drop to six or fewer and HDOP can climb past 4 or 5, compounding whatever filtering runs afterward. A session near a covered stand can post a lower max speed than an identical sprint at midfield with nothing about the athlete's output actually different.

What the Research Actually Shows

Varley, Fairweather, and Aughey (2012), in the Journal of Sports Sciences, tested 1Hz, 5Hz, and 10Hz GPS units against a radar-based reference during maximal acceleration, deceleration, and constant-velocity efforts. Across acceleration and deceleration, where true peaks are narrowest, 1Hz units produced the largest errors by a wide margin, frequently missing the radar-timed peak by well over a kilometer per hour, while 10Hz units performed meaningfully better at every phase but still underestimated instantaneous peak velocity during the sharpest parts of a sprint. Their own trials were straight-line efforts on one surface with one manufacturer's units, so exact error magnitudes will not transfer automatically to multidirectional movement or every brand on the market.

Buchheit, Al Haddad, Simpson, and colleagues (2014), in the International Journal of Sports Physiology and Performance, took the filtering side head-on. Their analysis showed that the moving-average smoothing manufacturers apply to raw velocity data before reporting it attenuates real high-magnitude accelerations and decelerations, the same short, sharp events that define a genuine sprint peak, with attenuation scaling to the width of the smoothing window. Their conclusion was blunt: filtering meant to make data look cleaner was, in some configurations, aggressive enough to quietly discard real athletic output rather than noise. The limitation worth carrying forward is that their analysis centered on one system's proprietary algorithm, so a different brand's filter will attenuate peaks by a different amount, which is exactly why the protocol below tests your own fleet rather than someone else's published number. Malone, Lovell, Varley, and Coutts (2017), reviewing GPS use in team sports, recommend a satellite count at or above 10 with an HDOP below 2.0 as the floor for trustworthy data, the threshold used below to define a clean versus compromised testing location.

Equipment and Test Setup

You need a true reference velocity, visibility into your GPS unit's output before and after filtering, and two locations that differ in satellite visibility while everything else about the sprint stays identical.

Wind matters more than most staff account for. A tailwind above roughly 2.0 m/s inflates true ground speed enough to distort every comparison here, so check conditions before each block and discard gusting sessions rather than averaging around them.

Step-by-Step Control Protocol

1. Warm-up (10-12 minutes): Standard sprint warm-up, dynamic mobility, finishing with two progressive strides at roughly 90% effort.
2. Establish the reference peak: Three maximal 30m sprints in the open-sky location, reference device running continuously. Record the single highest instantaneous velocity as the athlete's true reference peak. Thirty meters, not twenty, matters here: many team-sport athletes have not touched top speed by 20m, understating the reference itself before any GPS comparison starts.
3. Block A, clean satellite conditions: Same sprint, same location, GPS worn simultaneously with the reference device, three trials. Pull both the filtered and raw velocity trace where the vendor allows it. Record reference peak, GPS filtered peak, GPS raw peak, satellite count, and HDOP for every trial.
4. Block B, compromised satellite conditions: Identical protocol and device settings, moved to the compromised-sky location, same three trials and variables.
5. Compute the three gaps: Reference minus GPS raw isolates sampling-interval error. GPS raw minus GPS filtered isolates the filtering contribution. Block B's filtered peak minus Block A's filtered peak isolates the satellite-geometry contribution, since device and filter settings never change between blocks.
6. Discard criteria: Drop any trial with a false start, a tailwind beyond 2.0 m/s, or an HDOP above 5 recorded in the location meant to be clean, a sign the site needs relocating before rerunning.

Total time runs roughly 35-40 minutes per athlete across both blocks, most of it rest between maximal efforts, at least 3 minutes between each 30m trial.

Reading the Gap: What Each Pattern Means

Once both blocks are run, the pattern across the three gaps points to a specific fix.

A gap that moves session to session with no matching HDOP change is usually genuine day-to-day variation in the athlete, not the device. That is the useful negative result.

Mistakes That Blur the Diagnosis

Making This Part of Standard Practice

Treat 10Hz as a floor, not a target, for anything where absolute max speed matters rather than relative session-to-session tracking, and do not put a sub-10Hz unit on an athlete you plan to compare against a radar-timed test. Log HDOP and satellite count on every session file alongside distance and speed, the same way you would log surface or weather, so a low reading has context attached the moment someone asks about it later.

Re-run this control protocol against your GPS fleet once or twice a season rather than once at purchase, since firmware updates change filter behavior without much notice. And when a single session's max speed looks low, check the satellite log for that session before treating it as a training-quality problem. A 3-5% change sits well inside the range this article's cited research reports for sampling and filtering error alone; it takes a documented, HDOP-controlled gap larger than that, repeated across trials, before a low number is worth a conversation with the athlete rather than a look at the device.