5G-Based Body Sensor Network for Real-Time Feedback in Running

Sebastian Mayr

and Harald Rieser

Human Motion Analytics, Salzburg Research Forschungsgesellschaft mbH, Salzburg, Austria

Keywords:

5G, Body Sensor Network, Running, Real-Time Feedback.

Abstract:

Immediate feedback is vital in sports and motor learning. Traditionally, coaches provide this feedback. How-

ever, recent developments in wearable sensing and feedback devices, state-of-the-art processing algorithms and

sensor data integration allow for new applications and insights. While this allows for more direct feedback, it

also introduces new demands on feedback systems in terms of network transmission speed, energy consump-

tion and needed calculation power. The 5G networking standard is a promising technology to address these

issues. It promises reduced energy usage, high availability, latency in the low millisecond range and high data

rates up to 10 Gbps. Despite these characteristics, there has not been much research on the use of 5G in sports

biofeedback systems. Therefore, this work aims to assess the usability of 5G for a real-time feedback system

in the sports context. We did this by developing a distributed feedback system for the assessment of running

gait symmetry. The system utilizes internal smartphone IMU as sensors and employs a 5G infrastructure for

data transmission to an offsite server for the calculation of running metrics and generation of the feedback

signal. Pilot tests, using the feedback system, showed mean roundtrip times of 140.14ms (sd = 14.47ms) for

outdoor use and 128.92ms (sd = 25.83ms) for indoor use. These results indicate that promised low latencies

by the 5G standard are currently not reachable outside of controlled testing environments. Nonetheless, as

these times are below average reaction times in recreational athletes (150ms) it is still promising for real-time

feedback applications in running.

1 INTRODUCTION

Immediate feedback is an essential part of motor

learning. It is especially beneﬁcial to get feedback

directly after a particular movement is performed

(Sigrist et al., 2013). For an athlete, this feedback task

was traditionally performed by a coach. However, in

recent years wearable sensing devices, called wear-

ables, have been adopted by consumers and sports sci-

entists alike. The huge beneﬁt of these devices is their

mobility. They allow for data acquisition during ex-

ercise and often provide simple descriptive statistics

and summaries of these activities in a wide range of

applications.

One particular example in the ﬁeld of athlete

health monitoring is described by James et al. In their

study a combination of temperature-, heart rate- and

movement sensors were used for the detection of hy-

perthermia in running under extreme conditions dur-

ing the Tokyo Olympics in 2020. Here the collected

data was transmitted in real-time to stakeholders like

https://orcid.org/0000-0002-7540-155X

https://orcid.org/0000-0003-1407-4601

coaches and medical staff for monitoring of the ath-

letes health status. (James et al., 2024).

The use of wearable sensors for concurrent feed-

back to the athlete during movement is less frequently

considered. So-called real-time feedback systems try

to perform such an observation and correction task on

a quantitative level. Generally, they consist of lay-

ers for sensing, data processing, transmission of data,

and delivery of feedback. Depending on the use case

these tasks are performed by networks of indepen-

dent units distributed on the human body that com-

municate via wireless channels (e.g. Bluetooth or

WiFi). These networks are called Body Area Net-

works or Body Sensor Networks (BSN). Commonly,

BSN are constructed in a star topology (Lai et al.,

2013), which means they incorporate multiple types

of sensors for data acquisition that transfer data to a

central hub where computations are performed, and

the feedback is generated. A separate actuator unit

delivers this feedback in a suitable way to the user

who can optimize the assessed action by adapting

to the external feedback. Although BSN have been

around for a long time there are still challenges. Lai

et. al list improvement of wearability and reduction of

180

Mayr, S. and Rieser, H.

5G-Based Body Sensor Network for Real-Time Feedback in Running.

DOI: 10.5220/0012942500003828

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 12th International Conference on Sport Sciences Research and Technology Support (icSPORTS 2024), pages 180-184

ISBN: 978-989-758-719-1; ISSN: 2184-3201

energy consumption as the most pressing challenges

(Lai et al., 2013). Engineers and researchers are try-

ing to ﬁnd the right balance between optimal sensor

sample rates, optimized data transmission between

nodes, efﬁcient data processing on-site, and downsiz-

ing of the devices and therefore reduction of battery

capacity.

One approach to reduce energy consumption is to

outsource the calculations. In general, local com-

putation is more efﬁcient than transmission of data

and off-site calculation. However, with the increas-

ing computing requirements of new data processing,

and, in some cases, machine learning algorithms, the

limited processing power of local sensor nodes might

be no longer sufﬁcient. In such a case, sending raw

data to a high-performance off-site server, for cal-

culating results and sending them back for feedback

delivery, seems to be an interesting concept to con-

sider. This approach allows for scaling computation

resources based on current demand.

Sending high amounts of data between real-time

feedback systems and an off-site calculation server

needs a stable connection with high throughput, low-

latency and, for data collection in the ﬁeld, this con-

nection also needs to be wireless. The 5G mobile

telecommunication network standard offers new pos-

sibilities for BSN implementations. Its eight speciﬁ-

cation requirements promise a plethora of improve-

ments over current mobile communication network

technology: (i) 100% coverage, (ii) Up to 10 Gbps

data rate, (iii) one-millisecond latency, (iv) 1000×

bandwidth per unit area, (v) Up to 100x number of

connected devices per unit area, (vi) 99.999% avail-

ability, (vii) 90% reduction in network energy usage

and (viii) Up to 10-year battery life for IoT devices

(Attar et al., 2022).

Low latency, reduced energy usage and the pro-

posed increase in battery life, as well as the increased

transmission range in comparison to Wi-Fi, Bluetooth

or BLE, are interesting propositions for the develop-

ment of new 5G-based BSN systems.

Therefore, in this work we want to assess the suit-

ability of 5G for body area networks in a sports con-

text. The remainder of this paper is structured in 4

sections. In section 2 we describe the sensors, 5G in-

frastructure, how the BSN was implemented and the

target metric of the feedback system. In section 3 the

results are presented and discussed. Section 4 pro-

vides ﬁnal conclusions and a future outlook.

2 METHODS

2.1 Infrastructure

The system consists of three main parts (Figure 1).

First, the local BSN, that is made up of two smart-

phones that act as sensors, actuators, and transmission

devices. Second, an off-site server that is connected

to the BSN via a 5G network. Here, all calculations of

running metrics and feedback are performed. Third,

a data-logging service and a dashboard to display the

collected data, which is accessible via public internet.

Figure 1: Overview of system components.

Data is transmitted over a commercial-grade pri-

vate 5G-Standalone (5G-SA) network that operates in

Time Division Duplex (TDD) mode within the n78

frequency band (3500 MHz) with 80 MHz bandwidth.

This TDD setup utilizes a subcarrier spacing (SCS) of

15 kHz and follows Time Division-Long Term Evolu-

tion (TD-LTE) frame conﬁguration 2, incorporating

subframe conﬁguration 6 with a downlink-uplink ra-

tio of 3:1, in compliance with regulatory requirements

(Telekom-Control-Kommission (TKK), 2018; Tanen-

baum and Wetherall, 2010). The 5G-SA network con-

sists of MIMO-capable outdoor and indoor cells in a

remote radio head setup.

The off-site server is connected to the 5G core

with a low latency, high-speed cable connection to

minimize transmission time in this part of the net-

work. All time-critical back-end components are lo-

cated on this server:

• A messaging subsystem/server (based on MQTT).

For this server, we opted for a low-latency conﬁg-

uration. Therefore low quality of service (QoS)

levels were chosen.

• The algorithms for data analysis (see Section 2.3)

For non-time-critical sections in the back-end, a log-

ging service and a dashboard providing information

in near-real-time on an additional server were imple-

mented. The logging service was connected to the

messaging service and collected all sent data. The

dashboard provides a web-based user interface for vi-

sual display of the collected data. Physically, these

services were provided on a server accessible by a

standard internet connection.

5G-Based Body Sensor Network for Real-Time Feedback in Running

181

2.2 Sensors

Dedicated consumer grade 5G IMU sensors are not

available on the market yet. As modern smartphone-

integrated IMU deliver reliable and valid measure-

ment results compared with a gold standard (Grouios

et al., 2022), we used mobile phones as sensing, actu-

ation and transmission devices in the BSN. Most cur-

rent smartphones are capable of receiving and sending

5G-SA on a hardware level. However, manufactur-

ers often fail to deliver the required software updates.

For data collection, we used a Moto g5G and Moto

edge30 smartphone (Motorola Mobility LLC, USA)

to collect acceleration and angular velocity in three

axes as well as the GPS signal. The devices were

placed on the lower back at the 5. lumbar vertebrae

and the lateral side of the upper Tibia – directly un-

der the knee. Positioning of the sensors is depicted in

Figure 2.

Figure 2: Sensor position on the body: Sensors are placed

at the 5. lumbar vertebrae and lateral side of the upper Tibia

– directly under the knee (left).

Schematic overview of calculation of running gait metrics

and feedback (right).

2.3 Movement Metric – Gait Symmetry

In their review of feedback systems in running van

Hooren et. al pointed out that modiﬁcation of running

technique via real-time feedback can reduce injury

risk and increase performance. Suitable for feedback

are parameters related to injuries or performance, that

can be measured accurately and are easily modiﬁable

(Van Hooren et al., 2020). Hence, for testing this

prototype feedback system, we used the running gait

symmetry metric employed by Lee et. al, where a sin-

gle sacral-mounted IMU is used to assess the symme-

try of the COM vertical acceleration between the left

and right step within each stride (Lee et al., 2010).

This metric is easily measurable, and according to

Radzak et. al, increasing asymmetry is linked to fa-

tigue (Radzak et al., 2017).

Gait symmetry is calculated as the ratio of COM

vertical acceleration at the time of foot strike of the

left and right step during a single stride.

Symmetry =

COM right

− a

COM le f t

COM right

(1)

A positive symmetry score indicates a greater ac-

celeration during the right step, whereas a negative

score denotes a dominance of the left step. Values

near 0 occur when the acceleration on the left and the

right are similar. To foster comprehensibility for the

users, the symmetry score was min-max normalized

between a = −1 and b = 1. The normalization was

performed according to equation 2.

Symmetry

normalized

2 · (x − min(x))

max(x) − min(x)

− 1 (2)

For foot-strike detection we used two methods.

First, we used the mediolateral hip acceleration to dis-

tinguish between left and right steps. According to the

inverted pendulum model by Zijlstra and Hof, COM

is accelerated to the right during a left step and to

the left during a right step due to the tilting motion

of the hips (Zijlstra and Hof, 2003). In addition to

this method we apply a second IMU on the right shin

for the detection of a right reference step by a sim-

ple peak ﬁnding algorithm based on the comparison

of neighboring values (Virtanen et al., 2020), which

we consider as the starting point of a stride.

Data is continuously streamed from the sensors to

the off-site server. In regular intervals, the buffered

data is evaluated for step events. A subsequent calcu-

lation of the step symmetry is performed on the last 10

steps. Figure 2 and Algorithm 1 show this schemati-

cally.

Data: IMU Data

Result: Feedback Signal

while Data is sent to server do

Collect data from cache

Calculate anterior-posterior- and

mediolateral acceleration

Calculate total acceleration of leg sensor

data

Calculate steps from acceleration data of

shin-mounted IMU

Calculate steps from acceleration data of

sacrum-mounted IMU

Calculation of step symmetry (equation

Scale step symmetry between -1 and 1

(equation 2)

Publish feedback to device

end

Algorithm 1: Schematic calculation of the running gait

symmetry.

2.4 Android Application

The implemented android application (see Figure 3)

had four main purposes:

icSPORTS 2024 - 12th International Conference on Sport Sciences Research and Technology Support

182

1. Collect IMU data (accelerometer and gyroscope)

and GPS (not used by the algorithms) from the

smartphone

2. Synchronize data collection between devices

(start/stop of data collection, local time of the de-

vices)

3. Send data to the server to further analyze the IMU

data

4. Log Round Trip Times for upload and feedback

For data collection, the standard Android Sensor APIs

were used. The sample interval for the accelerometer

and gyroscope was conﬁgured with 10ms (100Hz),

for GPS the interval was 1s. Collection was started

only on the main device, the secondary device was

started automatically by a control interface that was

provided via the 5G network. Additionally, time

synchronization (based on Network Time Protocol -

NTP) via this control interface was realized to mini-

mize time drift between devices involved. Collected

data were sent to the server by both devices, where

it was analyzed using the aforementioned algorithms.

The results were sent back to all devices using a result

channel, and results (with timestamps) were logged

and presented to the users.

Figure 3: Android Application for data collection.

2.5 Testing

For a demonstration of the functionality of the sys-

tem, it was tested by a single subject in an indoor and

outdoor environment. For the outdoor test, the sub-

ject ran a round course with a length of 130 meters 5

times. The indoor test was performed 10 times in a

straight corridor of 20 meters.

3 RESULTS AND DISCUSSION

The average time taken to deliver a feedback sig-

nal to the user varied between the settings. Dur-

ing the outdoor run, the mean time was 140.14ms

(sd = 14.47ms). In contrast, with the indoor antenna,

the mean time was 128.92ms (sd = 25.83ms).

In real-time feedback, there is a limit for the min-

imal achievable feedback delivery time, which Umek

and Kos call the biofeedback delay (Umek and Kos,

2016). It consists of the feedback loop delay (FLD)

and the reaction time delay (RTD). The former is

composed of the communication delays from the sen-

sors to the processing unit and from there to an actua-

tor as well as the processing time delay itself. By us-

ing a 5G network for data transmission the communi-

cation delays can be reduced and by using an off-site

server for time-critical data processing the processing

delay can be minimized.

Average marathon runners run with a cadence of

180 steps per minute (Tenforde et al., 2019), which

roughly amounts to making a step every 330ms.

As the mean reaction time of an average athlete is

150ms (Umek and Kos, 2016) a FLD smaller than

330ms–150ms = 180ms is needed to enable the run-

ner to react properly to the feedback signal within the

next stride. The biofeedback delay measurements in

the in- and outdoor setting fall below this theoreti-

cal threshold by around 40ms to 50ms (see Figure 4).

These results indicate the suitability of 5G data trans-

mission in distributed BSN applications in running.

Figure 4: Minimal time to initiation of movement change

is the sum of feedback loop delay (FLD) and reaction time

delay (RTD).

4 CONCLUSION

Although 5G does not meet expectations in terms of

low-latency, measurements with the developed real-

time feedback system show its usability for running

applications. Since the described BSN-based biofeed-

back system is merely a demonstrator, there are sev-

eral areas that could be improved. One limitation is

seen in the used hardware. Naturally, smartphones

have a higher weight and exhibit different inertia than

smaller, dedicated IMU measurement systems. We

5G-Based Body Sensor Network for Real-Time Feedback in Running

183

tried to counteract this behavior by ﬁxing the sensor

as securely as possible to the subject without hinder-

ing the runner. Since there are no consumer-ready 5G

IMU sensors available the choice of smartphones as a

replacement was only pragmatic.

One of the beneﬁts of a distributed BSN is the

increased processing power on the off-site process-

ing device. Nonetheless, the gait-symmetry algorithm

can hardly be described as very resource-intensive. It

would certainly be possible to do the same calcula-

tions on the smartphones themselves. This was done

intentionally, as the point of this work was mainly to

assess the overall performance of a distributed 5G-

based BSN biofeedback system.

In the future it would be interesting to leverage

the increased (off-site) processing power and include

more resource intense machine learning algorithms

and new physiological sensor data like skin temper-

ature, power or heart-rate.

Another critical point is the fact that the system

was tested with only one BSN in the 5G network. This

was done to eliminate potential interactions of devices

and effects on the performance of the 5G network. In

the future, a comparison of the behavior of the net-

work and the time delays in scenarios with several si-

multaneously active devices would be interesting.

ACKNOWLEDGEMENTS

This project is partially funded by the Austrian state

of Salzburg under the program “WISS 2025” contract

number 20102-F2001049-FPR.

REFERENCES

Attar, H., Issa, H., Ababneh, J., Abbasi, M., Solyman, A.

A. A., Khosravi, M., and Said Agieb, R. (2022). 5G

System Overview for Ongoing Smart Applications:

Structure, Requirements, and Speciﬁcations. Compu-

tational Intelligence and Neuroscience, 2022:1–11.

Grouios, G., Ziagkas, E., Loukovitis, A., Chatzinikolaou,

K., and Koidou, E. (2022). Accelerometers in Our

Pocket: Does Smartphone Accelerometer Technology

Provide Accurate Data? Sensors (Basel, Switzerland),

23(1):192.

James, C., Lam, G. W., Guppy, F., Mu

niz-Pardos, B., An-

geloudis, K., Keramitsoglou, I., Knopp, M., Ruiz, D.,

Racinais, S., and Pitsiladis, Y. (2024). The integra-

tion of multi-sensor wearables in elite sport. Gatorade

Sports Science Exchange, 37:1–10.

Lai, X., Liu, Q., Wei, X., Wang, W., Zhou, G., and Han, G.

(2013). A Survey of Body Sensor Networks. Sensors

(Basel, Switzerland), 13(5):5406–5447.

Lee, J. B., Sutter, K. J., Askew, C. D., and Burkett, B. J.

(2010). Identifying symmetry in running gait using a

single inertial sensor. Journal of Science and Medicine

in Sport, 13(5):559–563.

Radzak, K. N., Putnam, A. M., Tamura, K., Hetzler, R. K.,

and Stickley, C. D. (2017). Asymmetry between lower

limbs during rested and fatigued state running gait in

healthy individuals. Gait & Posture, 51:268–274.

Sigrist, R., Rauter, G., Riener, R., and Wolf, P. (2013). Aug-

mented visual, auditory, haptic, and multimodal feed-

back in motor learning: A review. Psychonomic Bul-

letin & Review, 20(1):21–53.

Tanenbaum, A. S. and Wetherall, D. J. (2010). Computer

Networks. Pearson, Boston Amsterdam, 5 edition.

Telekom-Control-Kommission (TKK) (2018). Ausschrei-

bungsunterlage im Verfahren betreffend Frequenz-

zuteilungen im Frequenzbereich 3410 bis 3800 MHz.

Tenforde, A. S., Borgstrom, H. E., Outerleys, J., and Davis,

I. S. (2019). Is Cadence Related to Leg Length and

Load Rate? Journal of Orthopaedic & Sports Physi-

cal Therapy, 49(4):280–283.

Umek, A. and Kos, A. (2016). The Role of High Perfor-

mance Computing and Communication for Real-Time

Biofeedback in Sport. Mathematical Problems in En-

gineering, 2016:1–11.

Van Hooren, B., Goudsmit, J., Restrepo, J., and Vos, S.

(2020). Real-time feedback by wearables in run-

ning: Current approaches, challenges and sugges-

tions for improvements. Journal of Sports Sciences,

38(2):214–230.

Virtanen, P., Gommers, R., Oliphant, T. E., Haberland, M.,

Reddy, T., Cournapeau, D., Burovski, E., Peterson, P.,

Weckesser, W., Bright, J., van der Walt, S. J., Brett,

M., Wilson, J., Millman, K. J., Mayorov, N., Nel-

son, A. R. J., Jones, E., Kern, R., Larson, E., Carey,

C. J., Polat,

I., Feng, Y., Moore, E. W., VanderPlas,

J., Laxalde, D., Perktold, J., Cimrman, R., Henriksen,

I., Quintero, E. A., Harris, C. R., Archibald, A. M.,

Ribeiro, A. H., Pedregosa, F., van Mulbregt, P., and

SciPy 1.0 Contributors (2020). SciPy 1.0: Fundamen-

tal algorithms for scientiﬁc computing in python. Na-

ture Methods, 17:261–272.

Zijlstra, W. and Hof, A. L. (2003). Assessment of spatio-

temporal gait parameters from trunk accelerations

during human walking. Gait and Posture, 18(2):1–10.

icSPORTS 2024 - 12th International Conference on Sport Sciences Research and Technology Support

184