Assessing accuracy and reliability

Sep 4, 2016 | ACADEMY

Assessing the accuracy and reliability of the GPEXE to determine the time course of speed and acceleration

In recent years the use of tracking systems has become widespread in team sports monitoring. The technology in this field improved rapidly, the costs have become affordable and the researchers have investigated various aspects of this topic, highlighting the advantages and the limits of these systems. One of the most common of them is the Global Navigation Satellite System receiver (commonly referred to simply as a GPS) which allow to receive informations from satellites and calculate position and speed.

The research in this field was initially focused on the total distance covered (assuming that this was related to the overall energy expenditure of the player); a little later, the distance covered in different speed categories was addressed (assuming that the higher was the speed, the greater would be the intensity). To this end, 5 Hz and 10 Hz units provided a sufficient accuracy in order to calculate the parameters mentioned above.
More recently, as this approach fails to account for the demands of changing in speed (i.e. a greater energy consumption for accelerations and a greater muscular load for decelerations), the validity and reliability of the acceleration data obtained from GPS has become crucial. 10 Hz units offer a better accuracy to quantify accelerations/decelerations but it is not entirely clear whether the results can be considered satisfactory under all conditions (e.g. maximal speed, sudden changes of direction, etc…).

The purpose of this pilot investigation was therefore to compare a high sampling frequency GPS system (GPEXE PRO 18.18 Hz, Exelio, Udine, Italy) and a radar unit (ATS II 46.875 Hz, Stalker Radar, Plano, Texas, USA) during maximal straight line sprints with the aim of examining the differences between these two devices in estimating peak speed and peak acceleration.


8 trained males (soccer and rugby players) were involved in this study. Each subject completed a different number of bouts on three distances (e.g. 20, 30 and 40 meters) for a total of 65 sprints (all subjects included) (12 trials were discarded due to technical troubles with both tools):

  • 20 meters (N = 9; discarded = 3);
  • 30 meters (N = 31; discarded = 3);
  • 40 meters (N = 25; discarded = 6).

The following steps were required to compare the two devices at the same sampling frequency:

The raw speed signal was filtered with a specific designed algorithm (property of Exelio, Udine, Italy) operating at the original sampling frequency of each device (i.e. 18.18 Hz and 46.875 Hz for the GPEXE PRO and the Stalker ATS II, respectively), see figure 1;

the-raw-speed-signal-was-filtered-with-a-specific-designed-algorithm-gpexe_0figure 1: time course of the raw speed (thin continuous line) and of the filtered speed (thick dashed line); the left panel (red tracks) provides data from the GPEXE and the right panel (green tracks) from the radar.

the radar’s filtered speed was down-sampled at the same sampling frequency of the GPS (the reverse process would have required to “construct” several additional samples!), see figure 2;

the-radars-filtered-speed-was-down-sampled-at-the-same-sampling-frequency-of-the-gps-gpexefigure 2: time course of the filtered speed at the original sampling frequency of the radar (46.875 Hz, dark green dashed line) and after the down-sampling at the GPS frequency (18.18 Hz, bright green dashed line). Blue arrows highlight the speed offset at the beginning and at the end of the bout.

the time course of the filtered speed from both systems was synchronized by using a Root Mean Square Deviation algorithm (RMSD); the radar shows a speed offset at the beginning and at the end of the sprint (≈ 0.5 m⋅s-1, see blue arrows in figure 2) even if the subject started from and finished at a still position; the RMSD synchronization was needed because it was not possible to determine the exact start of the test and, as a consequence, only the data between dashed blue lines were considered, see figure 3;

root-mean-square-deviation-algorithm-gpexe-vs-radar-synchronization-method figure 3: comparison between GPEXE (red dashed line) and radar (green dashed line) time course of the filtered speed after the RMSD synchronization.

the acceleration was calculated from the filtered speed of each device, see figure 4;

comparison-between-gpexe-red-line-and-radar-green-line-time-course-of-acceleration-gpexe figure 4: comparison between GPEXE (red line) and radar (green line) time course of acceleration.

The parameters that follow were calculated as described below and are reported in figure 5:

  • the difference between radar peak speed (↑vrad) and GPEXE peak speed (↑vgps):

↑vdiff = ↑vrad – ↑vgps  [1]

      • the difference between the time corresponding to radar peak speed (time ↑vrad) and the time corresponding to GPEXE peak speed (time ↑vgps):

time ↑vdiff = time ↑vrad – time ↑vgps  [2]

      • the difference between radar peak acceleration (↑arad) and GPEXE peak acceleration (↑agps):

↑adiff = ↑arad – ↑agps [3]

      • the difference between the time corresponding to radar peak acceleration (time ↑arad) and the time corresponding to GPEXE peak acceleration (time ↑agps):

time ↑adiff = time ↑arad – time ↑agps [4]

      • RMSD of speed (RMSDv).
figure 5: parameters used to compare GPEXE and radar (see text for details); on the left panel the speed and on the right panel the acceleration.

The average values (± SD) of all variables described above are reported in table 1:

table 1: data comparison between GPEXE and radar (see text for details).


Aim of the study: to assess the accuracy and reliability of the GPEXE to determine the time course of speed and acceleration during 20 to 40 meters straight line sprint.
Reference method: a radar system that, however, showed a marked speed offset at the beginning and at the end of each trial (see figure 2).

Assessing accuracy reliability gpexe pearsons-correlation-left-panel-and-bland-altman-plot-right-panel-of-all-speed-data-gpexefigure 6: Pearson’s correlation (left panel) and Bland-Altman plot (right panel) of all speed data (N = 8165).
Assessing accuracy reliability gpexe pearsons-correlation-left-panel-and-bland-altman-plot-right-panel-of-all-acceleration-data-gpexe_0
figure 7: Pearson’s correlation (left panel) and Bland-Altman plot (right panel) of all acceleration data (N = 8165).


neglecting this perplexing speed offset, the data obtained by the two instruments are essentially superimposable (see table 1 and figure 6 and 7).

Practical remarks:

the data show that a gps with a high sampling frequency is perfectly adequate to monitor speed and acceleration during maximal linear sprints. This supports the view that the variables derived therefrom, such as the energy expenditure and metabolic power [3], as well as the individual mechanical characteristics (horizontal force, ratio of force) [4], are basically correct within the limits of the models described and discussed elsewhere [1, 2].


  1. di Prampero PE, Fusi S, Sepulcri L, Morin JB, Belli A, Antonutto G (2005). Sprint running: a new energetic approach. J Exp Biol 208: 2809-2816.
  2. Nagahara R, Botter A, Rejc E, Koido M, Shimizu T, Samozino P, Morin JB (2016). Concurrent validity of GPS for deriving mechanical properties of sprint acceleration. Int J Sports Physiol Perform [ahead of print] doi: 10.1123/ijspp.2015-0566
  3. Osgnach C, Poser S, Bernardini R, Rinaldo R, di Prampero PE (2010). Energy cost and metabolic power in elite soccer: a new match analysis approach. Med Sci Sports Exerc 42(1): 170-178.
  4. Samozino P, Rabita G, Dorel S, Slawinski J, Peyrot N, Saez de Villarreal E, Morin JB (2016). A simple method for measuring power, force, velocity properties, and mechanical effectiveness in sprint running. Scand J Med Sci Sports 26(6): 648-658.

Author: Cristian Osgnach

Related Contents
Related Contents