Hidden Rhythms of Team Sport Movement Reveal What Wearable Trackers Can Actually Capture

Every sprint, cut and collision on a team sport field carries a hidden signature: a rhythm of oscillations that unfolds at frequencies far too fast for the eye to follow. A new study has mapped that rhythm for the first time across both laboratory-style circuits and a full professional match, and its central finding is strikingly simple. When elite athletes run, jump, change direction and absorb tackles, nearly all of the meaningful frequency content of the acceleration signal measured at the upper body sits below roughly 5 Hz, with only the most explosive movements pushing meaningful power up toward 8 Hz. The work, published in Sports Engineering by Susanne Ellens, Codey Moran and Matthew C. Varley, offers one of the clearest frequency-domain references yet for anyone designing, selecting or interpreting data from the wearable tracking devices that now saturate professional sport.

The reason frequency content matters so much comes down to a foundational rule of signal processing. The Nyquist-Shannon sampling theorem states that to faithfully reconstruct a signal, a device must sample at least twice as fast as the highest frequency present in that signal. Fail that test and the data falls victim to aliasing, a treacherous artefact in which the sampled record fabricates a spurious, slower oscillation that never existed in the athlete’s true movement. The consequences for sport science are not abstract. The authors illustrate the problem with a 1 Hz device asked to record a maximal deceleration, a 180-degree turn and a maximal acceleration. Such a device may log data points just before and just after the event while missing the turn itself, potentially misleading judgements about an athlete’s conditioning or training status. In an era where load management and injury risk decisions hinge on tracking data, an aliased signal is not merely inaccurate, it can quietly invert the story the data tells.

Previous frequency analyses of human movement, largely built on high-sampling-rate inertial measurement units because of their portability and speed, had already established reference points for gentler activities. Roughly 99 percent of the signal power during walking falls below 15 Hz, and 98 percent below 10 Hz, while everyday actions such as stand-to-squat, sit-to-stand, walking or body rotation occupy a band between 0.3 and 3.5 Hz. Team sports, however, involve far more dynamic accelerations, sprints and changes of direction that may contain higher-frequency components, including phases of foot strike that systems sampling below 40 Hz may fail to capture. Meanwhile, the tracking systems actually worn by athletes, Global Navigation Satellite Systems and Local Positioning Systems, sample anywhere from 1 to 50 Hz, and evidence shows that systems below 10 Hz demonstrate lower agreement and validity than those sampling at 10 Hz or above. What remained unknown was the true spectral content of genuine team sport movement.

To close that gap, the researchers ran a two-part investigation. In the controlled environment, thirteen competitive male and female athletes, all engaged in team sports such as futsal and soccer at a trained or developmental level, completed a circuit representative of common team sport movements. Each wore a research-grade inertial measurement unit sampling at approximately 1136 Hz, a rate exceeding any reported peak frequency component of human movement, positioned on the upper back at the level of the sixth thoracic vertebra between the shoulder blades, a location that approximates the body’s centre of mass. The circuit comprised fourteen movement trials: walking, runs of 10 and 15 metres, all-out efforts over 10 and 25 metres, changes of direction at 45, 90 and 180 degrees, a zigzag, jumps, a simulated tackle in which participants dropped their body to the ground as quickly as possible, ball handling and backwards movement. Maximal speeds reached 6.0 to 7.9 metres per second, with accelerations of 3.4 to 5.2 metres per second squared.

The game environment brought the analysis into the real world. Movement data from one full Australian Football League match were analysed for twenty-three elite male players from a single club, all injury free, all playing at least 60 minutes of effective time and together representing every positional group. Because club regulations approved only the match-day hardware, the researchers used a GNSS-embedded inertial measurement unit sampling at 100 Hz rather than the research-grade sensor. The match recording was analysed as a single continuous signal covering active playing time, with breaks and bench time excluded, providing a global characterisation of the match’s overall spectral content rather than a resolution of individual actions. In both settings, the raw three-axis accelerometer data were processed identically in essence: axis means were subtracted to remove the gravitational component, a zero-phase Butterworth high-pass filter removed low-frequency sensor drift, and the filtered axes were combined into a resultant acceleration using the Euclidean norm.

From there, the team computed the power spectral density of each signal using a one-sided fast Fourier transform, after applying a Hann window to suppress spectral leakage. Two spectral metrics were derived. The dominant movement frequency was defined as the frequency at peak power, effectively the most prominent cyclic oscillation in the signal. The upper limit frequency at half peak power captured the highest-frequency bin at which mean spectral power remained at least half of the peak, irrespective of whether that power sat adjacent to the dominant peak or in a separate secondary region of the spectrum. Together, these metrics describe both the central rhythm of a movement and the upper reaches of its frequency content, the very region where under-sampling would begin to corrupt the record.

The controlled-environment results painted a movement-specific picture. Dominant frequencies ranged from 1.2 Hz, the lowest reportable bin above the 1 Hz high-pass filter cutoff, for the zigzag and all-out jump movements, up to 6.2 Hz for backwards movement, the highest dominant frequency of any task, likely reflecting the quicker cyclic stepping pattern of moving backwards. Walking, intriguingly, showed a higher dominant frequency of 3.8 Hz than the 10 and 15 metre runs at 2.8 Hz each, demonstrating that dominant frequency does not automatically scale with step cadence across movement types. Peak power told a complementary story: walking had the lowest value at 1.2 units, whereas high-effort linear running produced dominant frequencies between 2.8 and 3.4 Hz but the highest peak power, above 23 units, indicating slower cyclic rates accompanied by much larger accelerations. Changes of direction clustered between 3.4 and 3.8 Hz, and for most movements the upper half-power limit fell between 2.8 and 5 Hz, except for backwards movement, the 180-degree change of direction and the 10 and 25 metre all-out efforts, which carried stronger power into frequencies of 7.8 to 8.2 Hz.

The match data converged remarkably on the laboratory findings. The game environment produced a dominant movement frequency of 2.8 Hz with a peak power of 11 units, directly comparable to the combined spectrum of all controlled movements, which also peaked at 2.8 Hz with a peak power of 8.6 units. The authors suggest this shared dominant frequency may reflect broadly similar cyclic movement patterns between contexts, potentially tied to stride-related oscillations, and note that in Australian Football roughly 88 percent of a player’s movement is linear walking and jogging, which likely shaped the match spectrum and its slightly higher peak power. But they issue a clear caution: relying solely on game spectra to determine sampling requirements could overlook higher movement frequencies, because high-intensity movements, impacts and multidirectional actions form only about 4 percent of a match and are underrepresented in aggregate data. The frequencies of the specific movements of interest must also be weighed.

So what should practitioners actually buy? Applying the Nyquist criterion to the observed upper range of roughly 8 Hz yields a theoretical lower bound of approximately 16 to 18 Hz for the upper-trunk acceleration signals examined here. The researchers are explicit that this is not a practical sampling-rate recommendation, since anti-aliasing filter roll-off, measurement noise, temporal precision and oversampling margins all push the genuinely required rate higher. It also reflects whole-body movement captured at the trunk; higher frequencies are likely present in distal body segments, where sampling rates of 100 Hz or more are commonly used for technique-specific analysis such as step frequency, swim strokes and joint kinematics. Lower-frequency 10 Hz GNSS and LPS systems, typically placed near the trunk and designed for broader metrics like distance covered, align well with the lower requirement identified here. The choice, the authors argue, hinges on the objective: detailed technique analysis or whole-body movement monitoring, since higher sampling frequencies provide more data but not always more meaningful data.

The study also carries a message for the wearable industry itself. Because frequency requirements proved context-dependent, the authors argue that a single universal minimum sampling-frequency standard would be inappropriate. Instead, they call for manufacturers to transparently disclose true sampling frequency and whether onboard filtering, resampling or interpolation has been applied to raw data, proposing this as a candidate addition to the Specifications checklist of the Sports Technology Quality Framework. Such disclosure would let practitioners judge device adequacy against context-specific evidence rather than a blunt universal threshold. The findings come with honest limitations: the two environments used different sensors, a cross-sensor comparison involving only three participants showed identical peak frequencies and closely overlapping normalised spectra but cannot be generalised to match variability, sensor fixation differed between settings, neither device was validated against motion capture or force plates, and the game analysis covered a single match from one team. The controlled participants also ran slower than elite athletes, whose speeds near 8.5 metres per second may shift frequency characteristics. As preliminary evidence, though, the study delivers a conservative frequency-domain reference: for whole-body team sport movement monitored at the upper trunk, the action lives below about 5 Hz, and only the most violent accelerations reach toward 8 Hz.

Subject of Research: Sports Science

Article Title: Frequency characteristics of team sport movements: an inertial measurement unit accelerometer analysis

Article References: Ellens, S., Moran, C., & Varley, M. C. (2026). Frequency characteristics of team sport movements: an inertial measurement unit accelerometer analysis. Sports Engineering, 29(2), Article 40. https://doi.org/10.1007/s12283-026-00575-2

Image Credits: AI Generated

DOI: 10.1007/s12283-026-00575-2

Keywords: wearable technology, athlete tracking, inertial measurement units, sampling frequency, Nyquist-Shannon theorem, Australian Football, sports engineering, accelerometry, Frequency, characteristics, team, sport