Challenges in wireless communication for connected sensors and wearable devices used in sport biofeedback applications

Accepted Manuscript Challenges in wireless communication for connected sensors and wearable devices used in sport biofeedback applications Anton Kos, Veljko Milutinovi´c, Anton Umek

PII: DOI: Reference:

S0167-739X(17)31669-2 https://doi.org/10.1016/j.future.2018.03.032 FUTURE 4044

To appear in:

Future Generation Computer Systems

Received date : 30 July 2017 Revised date : 25 February 2018 Accepted date : 17 March 2018 Please cite this article as: A. Kos, V. Milutinovi´c, A. Umek, Challenges in wireless communication for connected sensors and wearable devices used in sport biofeedback applications, Future Generation Computer Systems (2018), https://doi.org/10.1016/j.future.2018.03.032 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Challenges in Wireless Communication for Connected Sensors and Wearable Devices Used in Sport Biofeedback Applications Anton Kos1, Veljko Milutinović2,3, and Anton Umek1 1

Faculty of Electrical Engineering, University of Ljubljana, Tržaška cesta 25, 1000 Ljubljana, Slovenia 2 School of Electrical Engineering, University of Belgrade, Bulevar kralja Aleksandra 73, 11000 Belgrade, Serbia 3 IPSI Belgrade, Dalmatinska 55, 11120 Belgrade, Serbia E-mail: [email protected] (corresponding), [email protected], [email protected].

Abstract: Sensors, wearables, wireless networks, and other Internet of Things technologies are ever more present in our daily life. We study their applicability and use in biofeedback systems and applications in sport. Biofeedback systems are important in motor learning where a person in the loop uses the feedback information to influence the execution performance. Sensors, actuators, and wireless technologies come in great varieties regarding their properties. We describe the most common groups of sensors and actuators that are used in sport and list the most widespread and easily available wireless technologies. We present the most important constraints of a biofeedback system operation and define a number of fundamental architectures of biofeedback systems. Taking into account all of the above, we present a number of different biofeedback application scenarios in sports. We match the scenarios to the most appropriate existing wireless technology that is expected to sustain scalability in the number of nodes or increased data rates for the expected application lifetime. We find out that currently none of the existing wireless technologies can satisfy the variety of demands of different biofeedback application scenarios.

Index Terms: biofeedback systems; wireless technologies; wearable sensors; embedded devices; Internet of Things; distributed applications

1 INTRODUCTION A great number of researchers are putting their efforts into research connected to the Internet of Things (IoT) – a field of research with great prospects and plans for the future. IoT is present in systems of all scales; from large systems such as smart energy and smart city, medium sized systems such as industry automation and smart home, to small sized systems such as connected car and smart wearables. IoT can offer benefits also to more traditional areas, such as agriculture, and healthcare [1]. IoT can also help with promoting and enabling the concept of smart and connected communities [2]. We are particularly interested in the role of IoT in human wellbeing, which includes, but is not limited to healthcare, rehabilitation, recreation, and sports. Sport and recreation are identified as some of the most rapidly growing areas of personal and consumer Internet of Things applications [3]. Smart devices or smart wearables such as wrist band activity trackers, heart rate monitors, motion tracking devices, and others are penetrating our daily life; they are readily available, affordable, and fast growing in numbers, and spatial density. The practical use of such embedded devices heavily depends on the development of wireless and sensor technology, corresponding applications, and a close cooperation with sports experts. This paper studies the possible use of embedded devices and (distributed) mobile applications in recreational and professional sports. We try to make a leap from toys and gadgets that mostly offer approximate, statistically based, activity and biometric measurements, to devices, tools, and applications that would offer precise and timely information for motor skill or training improvement.

Simply throwing better sensors and more powerful processing units into embedded devices will not do the trick, at least not completely. In acquisition and analysis of sports movements, anything can be a limiting factor, from sensor ranges, precision and sampling frequency, to computational power of the processing device and communication channels between the system elements. Gadgets such as wrist bands, that give statistical parameters and count event, can afford to acquire movements or physiological processes with low frequency and precision. They can easily calculate the results locally. This is not the case with devices used in sports where high sampling frequency and precision is required. Here the parameter and results calculation must be done remotely in a processing device with high computational power. Demand for remote processing includes the need for sensor signal transmission; wireless communication channels are the obvious choice in sports. While toys and gadgets are primarily intended for the recreational use, advanced devices, tools and applications would be primarily used by professionals and highly skilled and motivated amateurs. When trying to implement a real-time system for a high dynamic sport, a number of serious obstacles can be identified in sensor, processing, and communication technologies and devices. For example, sensors may have insuficient dynamic range, sampling frequency and/or accuracy, processing devices may have insuficient processing power or consume too much energy, batteries have insuficient capacity and do not support long enough autonomity of wearable devices, actions of sport activity are taking place in an area that cannot be covered by the chosen wireless technology, etc. An important part of the problem are obstacles that are most commonly found in available wireless communication technologies; some of them are described in [4]. The physical communication channel limitations drive the search for the best balance of achievable levels of bit rate and range against the available or prescribed transmission power. This paper focuses on the issues of building sensory systems in sport that can help with (faster) learning of motor skills, more efficient training, prevention of sport injuries, and other activities where technology can be included. While focusing on sport and recreation, most of the solutions can be beneficially used also in rehabilitation and healthcare. In sport, sensor technology has been predominantly used for motion tracking and analysis with wearable motion sensors attached to the athlete’s body. In recent years biofeedback applications are appearing in various sports, and sensors are being built into sport equipment as well. Biofeedback systems are important in motor learning in sports where users employ the feedback information to influence their execution performance [5]. During the practice, the natural (inherent) feedback information is provided internally through human sense organs. Augmented feedback is provided by external source, traditionally by instructors and trainers, recently also by technical equipment and devices. In the majority of research work about augmented feedback, the feedback information is given with a delay after the performed activity, which is defined as terminal feedback. The same is true for the majority of sport applications already available on smartphones; they offer post processing with a presentation of some of the vital or important parameters. The concurrent feedback, which is given in real time within the currently performed action, has also been found useful for accelerated motor learning [6]-[8]. In many sports disciplines video recording and optical tracking are classical methods for providing augmented feedback information for post analysis and terminal feedback. An alternative to the above mentioned systems are wearable embedded systems, which use one or several wearable sensors attached to the human body or integrated into sport equipment. Wearable and integrated sensors should measure the monitored quantities, but they should not interfere with the sport activity itself. Therefore, such sensors must be lightweight and small-size and they should not physically obstruct the activity. The last requirement also implicitly defines the wireless mode of communication between the sensors, sensor nodes, processing devices, and actuators. Wireless communication is not the only possibility. Sensors can be connected to a sensor node or processing device by wires or, in the case of implants, use the human body as the propagation channel [9]. The choice of the communication channel heavily depends on the type and the dynamics of the sport’s activity being monitored. For example, static sports or sports with very low dynamics may allow the use of wired sensors, while high dynamic sports with a lot of movement would not. Concerning the requirement for minimum obstruction of the user the most appropriate are systems with wireless communication. Although few manufacturers have started developing such products, they seem to be slowed down by wireless network architectural standard deficiencies. A selection of the most suitable set of sensors,

wireless technologies, and services for each application is a demanding task. It requires deep knowledge of properties, capabilities, demands, shortcomings, and benefits of each of the elements of such systems. We study some of the most relevant issues of wireless communication for sensor and actuators in biofeedback systems and give some guidelines about the choice of existing and widely adopted technologies. The main contributions of this paper are in the systematic approach to the current challenges in biofeedback systems in sport. To the best of our knowledge, this is the first paper that collectively studies the use of sensors, wireless communication technologies, processing devices, and actuators in connection with their possible use in various architectures and implementations of biofeedback systems in sport. We can sum up the most important contributions as follows: (a) The choice of the most appropriate technologies is of a paramount importance for a successful sport biofeedback system implementation; we list and describe the most represented sensors, actuators, and wireless technologies, together with their most important properties. (b) Sport is a very diverse activity taking place in various environments and under various conditions. We identify the most important constraints that are present in sport biofeedback system design and offer definitions and description of a number of fundamental system architectures. (c) We present the classification and comparison guidelines for the studied sport biofeedback system architectures. (d) The variety of possible sport biofeedback system implementations is very high. We present some of the typical biofeedback application scenarios in sport and match application demands with sensors, actuators, and communication possibilities. The remainder of the paper is organized as follows. Section 2 describes the operation, constraints, architectures, and other properties of the biofeedback systems, Section 3 is dedicated to the properties of sensors used in sport and wireless technologies are listed and described in Sections 4. Section 5 studies various application scenarios for wireless sensor based biofeedback systems. Section 6 presents current biofeedback systems and future trends and Section 7 concludes the paper.

2 BIOFEEDBACK SYSTEM Biofeedback systems include sensors, processing devices, actuators, and users who (try to) react to the given feedback information; as shown in Figure 1. In general, biofeedback systems operate as follows; user’s activity is driving sensor signals, which are (wirelessly) sent to the processing device for analysis. The results are (wirelessly) sent to the actuator, which provides the feedback information to the user through one of the modalities (auditory, visual, and haptic). The user closes the feedback loop by reacting to the feedback information. Sensor

Se ns or s

ign als

Processing device

User

Actuator

ls na g i s ck ba d e Fe

Figure 1. Biofeedback system elements and feedback loop operation.

2.1 System elements Sensors detect actions, state, and physiological parameters of users as well as various parameters of their physical environment. They produce signals that are sent to the processing device for analysis. Sensors in biofeedback systems are mostly attached to the user’s body or equipment, but they can also be in the vicinity of the user. Accelerometers and gyroscopes in the wrist band are an example of the former, high speed cameras and radars in the stadium are an example of the later. Each biofeedback system may include one or more sensors of the same or different type. The processing device receives sensor signals, analyses them, and generates feedback signals that are sent to actuators. The processing device can be a separate device (i.e., laptop or tablet), a component of a multifunction body-attached device (i.e., smartphone or smart watch), or a virtual device (i.e., cloud virtual appliance). Actuators are devices that actuate or incite action of a user with their activity. Actuators employ human senses to communicate feedback information to the user. The most commonly used senses are hearing, sight, and touch; commonly referred to as auditory, visual, and haptic modalities. For example, headphones are an actuator that uses auditory modality for feedback. Similarly to sensors, each biofeedback system may include one or more actuators of the same or different type and they can be placed in the vicinity of the user or attached to the user’s body. A large display showing user’s results, an actuator using the visual modality, is usually placed near the user. A vibrotactile actuator, using the haptic modality, must be attached to the appropriate body part of the user. Users and communication channels can also be considered as biofeedback systems elements. User’s (re)actions are necessary to close the feedback loop and communication channels are needed for the transmission of sensor and feedback signals between system elements. Although wired technologies can be used in practice, e.g., to send a feedback signal from the body-attached processing device to the nearby vibrotactile actuator, wireless communication technologies are most commonly used and are one of the focuses of this paper.

2.2 Feedback loop delay One of the important parameters of the biofeedback system is the feedback loop delay. The total feedback loop delay includes sensor delay, sensor signal transmission delay (also reffered to as communication delay), processing delay, feedback signal transmission delay, actuator delay, and user reaction time delay, as can be comnprehended from Figure 1. The feedback loop can be divided into a human (bio) and technical part. From the human perspective the technical part provides augmented feedback and the loop operation mode, depending on the timing, is concurrent or terminal. From the perspective of the technical part, the human provides biofeedback and the system operates in real time or post processing mode.

2.3 Constraints The biofeedback system is subjected to a number of constraints that define and bound its operation. We believe that the most important constraints in sport biofeedback systems are related to space, time, and computational power. Other constraints, such as energy and accuracy, are less aggravating because efficient mechanisms for their mitigation exist. 2.3.1 Space constraint The space constraint defines the possible locations of biofeedback system elements. We define three basic types of biofeedback systems: (a) personal space system, where all system elements are attached to the user, (b) confined space system, where the elements are distributed within a defined and limited space, and (c) open space system, where elements are not restricted in space. In most cases sensors and actuators are attached to the user. The most diverse is the location of the processing device that can be attached to the user, close to the user or at any other location connected to the system. 2.3.2 Time constraint A biofeedback system can work only if the feedback loop is closed. That means that the user

receives, understands, and reacts to the given feedback information. The feedback information can be given at different times: (a) terminal feedback is given after the activity has been performed; (b) concurrent feedback is given during the activity. An important parameter is the feedback loop delay that consists of communication delays for the transmission of sensor and feedback signals, processing delay, and user reaction delay, as shown in Figure 1. Acceptable feedback loop delay depends on the basic type of feedback. For the terminal feedback the timing is not that important; it can vary from a few seconds (coach advice given immediately after the activity has been completed) to a few hours or even days (video analysis of a recorded activity). For the concurrent feedback, the delay in the technical part of the feedback loop (communication and processing delay) should be shorter than the human reaction time. 2.3.3 Computation constraint Computation constraint is closely related and dependent on the space and time constraints as well as on the properties of sensors and actuators. Processing in the biofeedback loop can be done in: (a) post processing mode, where sensor signals acquired during the performed activity are processed after the activity finishes, and (b) in real-time mode, where sensor signals are processed concurrently with the execution of the activity. While the post processing mode does not represent a computational problem to the most of the processing devices and communication technologies, real time operation mode can many times be a very difficult problem. For example, in the post processing mode the absolute communication delay is not of paramount importance providing that the processing device receives and analyses all the data at a specified time after the completion of the activity. This time delay can vary from a few seconds, to minutes, hours, days or even longer time periods, depending on the intended ways of use. In the real time mode the processing device has to finish the processing within the time frame of one sensor sampling period, which can be as low as 1 ms or even less. While real time processing is generally not a problem for devices with relatively high computation power such as laptops, desktops, and cloud based computing, it can prove too demanding for mobile and/or embedded devices such as microcontrollers and smartphones. 2.3.4 Other constraints Apart from the abovementioned constraints biofeedback systems also have a number of other constraints. Among them the most important are energy and accuracy constraints, which are characteristic in all (wearable) sensor systems. Energy constraint limits the autonomy of wireless sensors or wireless sensor devices. This is especially important in (a) wireless sensor networks, where sensors can be in an inaccessible or very remote location or (b) in medical use, where sensors can be implanted into the human body. In biofeedback systems in sport this constraint has only very limited influence as sensor or sensor device battery is easily accessible, therefore it can also be easily recharged or replaced when needed. Accuracy constraint limits the usability and validity of measured quantities acquired by sensors and results derived from them. While sensor inaccuracies may be critical for some applications, that is rarely true for biofeedback appliactions in sport. The main reasons are: (a) in most sport activities frequent and high quality sensor calibration can be performed periodicaly or on as-needed basis [10], (b) accurate sport activity measurements in biofeedback systems are mostly performed in short time frames where sensor inaccuracy does not exceed the required accuracy threshold. 2.3.5 Communication delay One important parameter, that has not yet been addressed, is the communication delay within the biofeedback loop. This parameter is connected to all of the constraints studied above. Communication delay is heavily dependent on the communication technology used. For the real-time biofeedback systems the communication delay must be only a fraction of the user reaction delay. For the postprocessing biofeedback systems the communication delay is generally not the limiting factor. The communication delay Td [s] is closely connected to the properties of the communication technology used. It is proportional to the transmitted data length L [b] and inversly proportional to the bit rate of the communication tecnology R [b/s] through the expression Td = L/R. Lower communication delay can therefore be achieved by choosing a technology with a higher bit rate or by reducing the amount of data being sent. The detailed study of the communication delay within the biofeedback loop

can be found in [11] and more information about wireless communication technologies are presented in Section 4.

2.4 Architectures Numerous biofeedback system architectures can be devised based on different combinations of space, time and computation constraints. While these constraints define the base architecture of the system, its details are defined by its intended functionality. In the terms of structure biofeedback systems can be divided into compact and distributed system types. In compact biofeedback systems all of its elements (sensors, actuators, processing devices, and communication channels) are very close to each other; integrated into one device or attached to the same person. In distributed biofeedback systems its elements are at the arbitrary positions, providing that they can perform their functions as intended. For example, sensors for motion acquisition can be attached to the user’s body (accelerometer), or they can be located away from the user (video camera). Based on the intended functionality we can define a number of different system types. In this paper we define the user, instructor, and cloud type of the biofeedback system. In the user type all the elements and functionalities of system are under user’s control. In the instructor type the instructor monitors and controls the biofeedback system operation; for example, analyses the performed action in post-processing, gives the terminal feedback information to the user, and check statistical parameters and history of the user. Cloud type process and/or store the results in the cloud. Various types of data, from raw sensor signals to complex data analysis results, are available to anyone with appropriate access rights. Cloud data is usually accessed through a web application. Cloud system can also support a participatory concept of data acquisition, processing and sharing. In the terms of physical extent the biofeedback systems can be divided into personal, confined, and open space system types. In the personal type all system elements are in or on the user’s body. Confined space systems have elements in the vicinity of the user; for example, in the same room or in the same playing field. Open space systems have no limitation of distances between the system elements; for example, alpine skiing, marathon run, cycling, and other open space sports. In the latter two types, the element that is away from the user is practically always the processing device. 2.4.1 User architecture We show three examples that represent biofeedback systems based on different combinations of structure, functionality and physical extension types described in the above paragraphs. Figure 2 shows an architecture of a biofeedback system that belongs to the user functionality type. Its physical extend corresponds to the personal space type system, its structure resembles a compact type system. The monitoring device is used to control the operation of the biofeedback system and to view the results. The user architecture can give either concurrent or terminal feedback. Because such systems are most commonly implemented on the embedded devices or on the smartphone, the computational power can be a problem if the concurrent feedback is required. An example of a user biofeedback system is a relatively undemanding application for gait control, where the entire system is implemented within a smartphone. Its sensors feed the application algorithm that concurrently outputs audio feedback signal correlated to one gait parameter (step symmetry, for example).

Figure 2. An architecture of a biofeedback system with the user functionality type, its physical extend corresponds to the personal space type, and by structure it resembles a compact type of the biofeedback system.

2.4.2 Instructor architecture Biofeedback system shown in Figure 3 belongs to the instructor functionality type, its physical extend corresponds to the confined space type, and by structure it resembles a distributed type of the biofeedback system. Its main difference to the user architecture is the distributed structure. In most cases the sensors and actuators are on the user, while processing and monitoring devices are with the instructor at the remote location. This architecture requires wireless communication channels. If concurrent mode of operation is required, then the communication channels must have low latency and the processing device must be capable of performing all the necessary processing in real time. An example of an instructor biofeedback system is a running monitoring application, where the instructor monitors the athlete’s performance on a stadium in real time. Processing and monitoring device can be, for example, a laptop, a tablet, or even a desktop. The feasibility of a concurrent feedback in the major part depends on the properties of wireless communication channels.

Figure 3. An architecture of a biofeedback system that belongs to the instructor functionality type, its physical extend corresponds to the confined space type, and by structure it resembles a distributed type of the biofeedback system.

2.4.3 Cloud architecture Figure 4 represents an architecture of a biofeedback system that belongs to the cloud functionality type, its physical extend corresponds to the open space type, and by structure it resembles a distributed type of the biofeedback system. The core of the system is in the cloud that performs the functions of storage and data analysis, if terminal feedback is sufficient, the processing can also be done in the cloud. Figure 4 indicates that the entire biofeedback loop can be at the user’s location. In this case the loop operates in the user mode and the loop signals (sensor and feedback) are sent to the cloud during the activity or after the activity for possible complex data and statistical analysis. Results on a different level of detail are available to anybody with sufficient access rights – user, instructor or even general public.

Figure 4. An architecture of a biofeedback system that belongs to the cloud functionality type, its physical extend

corresponds to the open space type, and by structure it resembles a distributed type of the biofeedback system.

An example of a cloud biofeedback system is an application that controls gait, but also send the gait signal to the cloud. In the cloud the data are processed, analysed and results are available to: (a) users to track their activity, (b) to instructors to give comments on performance and skill, and (c) to the public to see, for example, high level statistical data such as number of steps made each day. 2.4.4 Classification and comparison of system architectures Figure 5 shows the classification of biofeedback systems architectures according to different aspects presented in section 2.4 and different constraints defined in section 2.3. For example, when functionality is taken as the key aspect, the instructor architecture of the biofeedback system can be devised: (a) as confined or open space system according to the space constraint and physical extent aspect, (b) as a distributed structure system, (c) it offers terminal feedback in confined and open space, it offers concurrent feedback only in confined space, (d) the available processing power ranges from moderate to high. Similar examples can be presented for other key aspect selections.

User Instructor

Functionality Cloud

Personal

Confined

Open

Structure

Compact

Distributed

Uncommon

Terminal feedback

Concurrent feedback Processing power

Physical extent Space constraint

Hard to assure

Time constraint Processing constraint

Figure 5. Classification of biofeedback systems according to the constraints defined in section 2.3 and architectures presented in section 2.4.

As already explained in sections 2.4.1 to 2.4.3, system architectures are defined based on intended functionalities of the biofeedback system: user, instructor, and cloud. While the detailed comparison of presented architectures based on all of the above presented aspects is out of the scope of this paper, a direct comparison between individual architectures can be done based on Figure 5. To assist the reader in this task, we present some of the most important advantages and disadvantages originating from the constraints described in sections 2.3.1 to 2.3.4. The user architecture has the most advantages when implemented as a personal space system with a compact structure. All sensors, processing device and actuators are on the user’s body, many times as a part of the same (embedded) system, i.e. smartphone. In general, sensor data synchronization and feedback loop delay are not a problem. Alternatively, to overcome the shortcomings of wireless transmission between the system elements, a wired communication medium can be used for data transfer. The most notable disadvantage of such implementations is the limited processing power that can be an obstacle for realization of concurrent feedback. The strongest advantage of the instructor architecture is its flexibility and ability of solving the problems imposed by the given constraints. Since the processing device is away from the user it has no important limitations about the processing power needed for the implementation of concurrent feedback. It also allows the instructor to actively and timely participate in the user’s activity. The main disadvantage of such implementations is the required use of wireless communication channels that

may cause increased feedback loop delay or data loss. Especially in open space systems the transmission range or/and bit rate of the wireless technology can be a limitation factor. The most notable advantage of the cloud architecture are its immense storage and processing capabilities that can be used for detailed analysis and data mining of the collected user activity data. The main disadvantage of the cloud system at this time is the hard-to-control communication delay that makes concurrent feedback practically impossible.

2.5 Processing and communication The need for processing and communication resources of the biofeedback system depends on a number of factors; from the number of sensors, their sampling frequency and bit rate to the available battery power, distance between system element, available communication technologies, protocol stacks, and other. Processing power and processing delay are especially important in real-time biofeedback systems where the processing device is receiving streamed sensor data at every sensor sample period. We assume that the communication technologies used have high enough bit rate to support the transmission of sensor data to the processing device. The processing of received sensor data must be done within one sampling period. In a pipelined processing algorithm the processing delay can be several sampling periods, but a new processing result must be availably at every sampling period. A binary classification of processing devices is possible into (a) devices capable of real-time processing and (b) devices capable of post processing. In connection to the processing requirements, the bit rate of communication channels is of limited importance, providing that it is high enough to assure the transmission of all data created by sensors. More about related issues can be found in [11]. An important factor in processing and communication demands of real-time biofeedback systems is the number of sensors in the system. The processing power and bit rate of communication technology of multi-sensor systems must be proportional to the number of sensors. Apart from the increased processing and communication demands, other problems can become an issue. For example, sensor data stream synchronization or sensor node density. The former is important for a quality sensor data processing, the latter is important for the correct selection of communication technology, which must allow concurrent operation of a large enough number of nodes. Some more details about these issues can be found in Section 5.3. Some of the many possible combinations of biofeedback system architectures presented in section 2.4 can be explained by Figure 6. For the connection with wireless technologies presented in section 4, networks classification by area is included.

Figure 6. Sensors, communication and processing resources in the biofeedback loop.

A personal space biofeedback system can be realized within the Body Area Network (BAN), where body attached sensors and actuators use wired or wireless communication channel, and equipment sensors use wireless communication channel. The functionality roughly corresponds to the user architecture of the biofeedback system and the BAN processing unit collects data for post processing or performs real-time processing for concurrent feedback.

A confined space biofeedback system with the instructor functionality can be easily realized within the Local Area Network (LAN) or within the Personal Area Network (PAN). The sensors and actuators are wirelessly connected to a laptop or a notepad over the wireless LAN technology (WLAN). Similarly to the previous example the LAN processing unit performs post processing or real-time processing. An open space personal biofeedback system can be realized as a cloud application. Body attached sensors, sport equipment sensors and actuators communicate with the cloud over the Wide Area Network (WAN), or they are wirelessly connected to a gateway that does the WAN communication with the cloud. Real-time operation is made difficult due to larger delays in communication with the cloud. Post processing options are practically unlimited.

3 SENSORS AND ACTUATORS IN SPORT Sensors and actuators used in sport are very heterogeneous and can be grouped according to several criteria; measured quantity, delay demands, bit rates, etc. Table 1 lists the most used sensors and actuators with their corresponding bit rates or bit rate ranges and the delay constraints [9]-[13]. One distinctive group of sensors measure low dynamic human physiological parameters such as temperature, heart rate, breathing, blood pressure, blood sugar, oxygenation, and others. Their common characteristic is low bit rate. Most of the above mentioned sensors measure the parameter with the frequency of up to 1 Hz and communicate it to the system even in longer time periods (5 s, 10 s, 60, etc.). The produced bit rate of physiological sensors is estimated to be below 100 bit/s. Table 1. Bit rates and delay constraints for sensors and actuators used in sport. Sensor Temperature sensor Heart rate sensor Oximeter CO2 sensor Blood sugar sensor Blood pressure sensor ECG Accelerometer Gyroscope Magnetometer Altimeter Strain gage

Bit rate < 100 bit/s < 100 bit/s < 100 bit/s < 100 bit/s < 100 bit/s < 100 bit/s 20-100 kbit/s 1-200 kbit/s 1-200 kbit/s 1-200 kbit/s < 1 kbit/s 1-50 kbit/s

Delay Not critical Seconds Seconds Seconds Not critical Not critical <1s < 50 ms < 50 ms < 50 ms Not critical < 50 ms

Actuator Tactile actuator Headphones/Loudspeaker (Voice) Headphones/Loudspeaker (Audio) Display (Video)

Bit rate < 100 bit/s 50-100 kbit/s <1 Mbit/s < 10 Mbit/s

Delay < 50 ms < 50 ms < 50 ms < 50 ms

With more dynamic physiological processes, like electrocardiogram (ECG), the entire signal at higher sampling frequencies is required. For example, ECG signal sampled at 1000 Hz with 24 bit precision, would produce a 24 kbit/s data flow. Even with more demanding physiological processes it can be estimated that the single sensor (device) does not produce a data flow of more than a few hundreds of kbit/s. Sport sensors are most commonly associated with measuring the kinematic parameters of the performed activity. For those purpose accelerometers, gyroscopes, and magnetometers are commonly combined in one miniature inertial measurement unit (IMU). Different sensors in different devices operate at different sampling frequencies (from 10 Hz to 2000 Hz) and with different precision (from 12 to 24 bits). For example, a 9 DOF sensor device, operating in the above mentioned ranges, would produce data flows from 1.08 kbit/s to 432 kbit/s. When more than one IMU devices are used at the

same time, the bit rate increases proportionally to the number of active IMU devices. In addition to a well-known IMU devices, sport equipment can include integrated resistive and semiconductor strain gage sensors for indirect measurement of force, torque, pressure, and bend. Similarly to IMU, the bit rate depends on sampling frequency and precision. For example, a strain gage sensor with sampling frequency of 2000 Hz and 16 bit precision generates 32 kbit/s. Collective bit rate increases linearly with the number of active sensors. The simplest actuators, such as buzzers (tactile) or beepers (audio), essentially use only one bit of feedback information, which tells the actuator to be active or idle. The bit rate of such simple actuators can be well below 100 bit/s. The situation is very different when the feedback signals are “natural” human sense signals (voice, audio, and video). Those signals are quite demanding and their bit rates can be anything between a few kbit/s (speech) to a few Mbit/s (video). Different modalities can be combined in the feedback information; for example, haptic and auditory (or video), where more than one haptic actuator can be used, but only one audio (or video) signal in the feedback. By combining one or more physiological, one or more kinematic sensor, one or more sensor integrated into equipment, together with actuators, the required bit rates and delay constraints can be very high. The selection of the most appropriate wireless technology is of paramount importance for achieving a high enough quality of service of the biofeedback system operation.

4 WIRELESS TECHNOLOGIES Wireless technologies are, like sensors and actuators, also very heterogeneous. Here we do not intend to give a detailed comparative analysis of all wireless technologies, but limit the discussion on the standardized and employed technologies that can be used for biofeedback applications in sport. Table 2 lists the most widespread wireless technologies for BAN, PAN, LAN and MAN (Metropolitan Area Network). We list only the parameters relevant for further discussion, details can be found in [1], [4], and [13]-[21]. The complete list of all wireless technologies and their variants that are available, under development or under testing, is much more extensive. We choose to list primarily the technologies that can be acquired on the market without major obstacles or the technologies that are expected to have commercial products available soon. Table 2. Standardized wireless technologies with potential use in sport Technology

Frequency

Range

Bit rate

TX Power

Bluetooth 2.1 + EDR

2.4 GHz

10 - 100 m

1 - 3 Mbit/s

2.5 - 100 mW

Bluetooth 4.0 + LE

2.4 GHz

10 m

1 Mbit/s

2.5 mW

868 MHz and 2.4 GHz

10 - 100 m

20 - 250 kbit/s

1 - 100 mW

IEEE 802.11n

2.4 and 5 GHz

70 m

600 Mbit/s

100 mW

IEEE 802.11ac

5 GHz

35 m

6.93 Gbit/s

160 mW

IEEE 802.11ad

60 GHz

10 m

6.76 Gbit/s

10 mW

IEEE 802.11ah

900 MHz

1 km

40 Mbit/s

100 mW

IEEE 802.11af

54 - 790 MHz

>1 km

1.8 - 26.7 Mbit/s

100 mW

LoRaWAN

868 - 928 MHz

up to 100 km

250 - 5470 bit/s

1.5 - 100 mW

ZigBee

The selection of the most appropriate wireless technology depends heavily on the given architecture of the biofeedback system from section 2.4 and the place of sensor signal processing from section 2.5. See section 5 for a more detailed sensor and wireless technology selection discussion. The

heterogeneity of wireless technologies and the variety of biofeedback system architectures suggests the use of multi-radio concepts [22]. Two main parameters are taken into consideration when selecting the most appropriate wireless technology for a biofeedback system in sport: range (coverage) and bit rate. Both parameters are highly dependent on the selected biofeedback system architecture and place of sensor signal processing. Some examples of considerations when choosing the most appropriate wireless technology:  For sending the sensor data from an open space system architecture a LoRaWAN or a IEEE 802.11ah technologies are needed.  ZigBee and LoRaWAN are not suitable for the transmission of high dynamic IMU signals.  Bluetooth is suitable for personal biofeedback applications, but not for confined and open space biofeedback applications. For biofeedback applications in sport the power and battery life are not of the primary concern. Contrary to sensor networks, batteries of sensor devices in sport are easily accessible for changing or recharging. Since the operation time of a biofeedback system in sport is limited by human endurance, battery life time can be much shorter than in some other sensor based applications, where batteries must last for days, months or even years. For this reason the Table 2 lists not only the well-known sensor network technologies such as Bluetooth, ZigBee, and LoRaWAN, but also several LAN technologies with various bit rates and ranges. LAN technologies fill (a) the range gap between 100 m and 1 km, and (b) the bit rate gap above 2 Mbit/s. LAN technologies are medium power, medium bit rate, and medium range. For example, a quick calculation shows that a small and lightweight battery with the 500 mAh capacity could support the operation of a LAN transceiver with a peak power of 100 mW for 5 to 10 hours; more than enough for most professional sport use. Not mentioned until now are the mobile wireless technologies such as GPRS, EDGE, 3G, and 4G. They are not widely employed in current sensor devices in sport. They are usable for terminal feedback systems, but due to their latency, they are not suitable for concurrent feedback with real-time biofeedback systems. This may change with the promises of 5G standards [23]. It is envisioned that 5G mobile networks will have much lower latencies, much more capacity, with much higher spectral efficiency, connecting the entire world by achieving communications between anybody, anything, anywhere, anytime, and with whatever device. But until this comes to be, we are bound to technologies which are available today.

5 APPLICATION SCENARIOS Given the heterogeneity of sensors and actuators and heterogeneity of wireless technologies, countless scenarios of their use for biofeedback systems in sport are possible. From the network topology point of view two basic variants are common: (a) Star - sensors and actuators each communicate directly with the processing device. (b) Relay - sensors and actuators communicate with a gateway (relay) which relays the data to the processing device. A gateway is any device that can communicate with one or more sensors and actuators through one of the available wireless technologies. It should also be capable of relaying the synchronized and multiplexed signals to or from the processing device. Communication with the processing device can be done using the same wireless technology as with the sensors and actuators, but in most cases the technology used for relaying is different. Two examples of gateways and their communication in downlink and uplink are shown in Figure 7; a gateway on the user’s body and a mobile device located on or near the user. Based on Figure 7, there are a number of possible communication scenarios that can be used by biofeedback applications. One example of a relayed communication type is a biofeedback application with several body-attached sensors that are connected to the gateway by wires. The gateway, which is

also attached to the user’s body, synchronizes all sensor signals and relays them over the 3G/4G/5G mobile link to the cloud. When the processing in the cloud is completed, results can be monitored by the instructor or/and the feedback can be communicated back to the user; depending on the biofeedback system architecture (see section 2.4). An example of a star communication type is a biofeedback application with several IMU devices attached to the user or to the sport equipment. IMU devices are communicating with the smartphone over Bluetooth. The smartphone is performing the functions of a processing device, monitoring device, and possibly even the function of an actuator by giving the feedback to the user over the screen, loudspeakers or vibration. Many more scenarios are possible by combining the elements and communication paths from Figure 7. Below we present some characteristic application scenarios. They are explained through Figure 7, which shows various connectivity options. We include various sensors, from wearables to sensors integrated into medical and sport equipment. We match the scenarios to the most appropriate existing wireless technology that is expected to sustain scalability in number of nodes or increased data rates for the expected application lifetime. We present three application scenarios that are based on various sensor groups presented in section 3 and assessed through available wireless technologies listed in Table 2.

Mobile network /5G /4G G 3

Cloud storage/processing/app

5.4 802.1

BTLE 802.1

BAN

Internet

1ad

WLAN

802.11

PAN

Standalone sensor device

LAN

On-body sensor

MAN/WAN

Equipment sensor Gateway BAN wirless link

Figure 7. Some of the communication possibilities for biofeedback systems used in healthcare and sports.

5.1 Low dynamic application scenario A biofeedback application using signals from low dynamic physiological sensors is not demanding in terms of communication. Bit rates, even with a larger number of sensors, do not exceed 1 kbit/s. Such applications can be implemented in any system architecture discussed in section 2.4 For example, a personal architecture is implemented with Bluetooth LE, confined space architecture with IEEE 802.11af, and open space architecture with LoRaWAN. Also, star and relay topologies are equally viable. An example of usage in sport is monitoring the physiological parameters of an athlete during running on a treadmill in fitness, a track, or in nature.

5.2 High dynamic application scenario A biofeedback application using signals from high dynamic IMU is more demanding. Its bit rates can easily reach several hundred kbit/s - when more IMU devices are used simultaneously even several Mbit/s. In the star topology almost all technologies listed in Table 2 satisfy the demands when used in the personal or confined space architecture (see section 2.4). For the open space architecture a relay over the gateway or mobile device with mobile network connectivity is possible, providing a high enough bit rate is available (see Figure 7). An example of usage in sport is giving auditory feedback to a gymnast performing practice on a vaulting horse.

5.3 High dynamic multiple sensor application scenario A biofeedback application using signals from a combination of high dynamic IMU, sport equipment sensors, and actuators with video feedback can be very demanding. Bit rates can exceed 10 Mbit/s when a number of IMU devices are on the user’s body, a number of sensors integrated into the sport equipment, and 3D video is used for the feedback. Even in the star topology only LAN technologies listed in Table 2 can satisfy the demands of the personal architectures. Confined space architectures are limited to different ranges, depending on the LAN technology used. Open space architectures are not viable. In the system with more than one gateway a synchronization can be a major problem. An example of usage in sport is giving real time video feedback to a skier with a personal architecture of the system (all devices in the backpack). Even more demanding scenarios are possible when using biofeedback applications in group sports. One such example is a high performance real-time biofeedback application for a football match. The scenario includes 22 active players, 3 judges, 10 IMU devices per person, 1000 Hz sampling rate, 9 DoF data, each with 16 bits per measured quantity. The nett bit rate of all sensors achieves 36 Mbit/s in the direction to the processing unit. Implementing a real-time biofeedback application in this scenario is an extremely demanding task for all elements of the system; from sensors and processing unit, to communication channels.

5.4 Properties and requirements of different sports From the above biofeedback application scenarios it is evident that the type of sport or a particular sport discipline is one of the major deciding factors for the correct choice of the most appropriate communication technology. The number of sport disciplines is far too large to be able to discuss the technical challenges in terms of biofeedback systems for each of them. Similarly to the classification of biofeedback system in sport, presented in Figure 5, sport disciplines can be classified from the technical aspect based on the following criteria: (a) place (fixed, bounded, unbounded), (b) number of athletes (individual, group, mass), (c) movement dynamics (low, medium, high), (d) movement type (defined single aperiodic movement, cyclic-periodic movement, not defined free movements), (e) used equipment (none, simple, complex), (f) environment (indoor, outdoor, water, etc.).

6 CURRENT POPULAR BIOFEEDBACK SYSTEMS Measuring and quantification of human body activities and processes had grown very popular in recent years. A number of very different systems have with varied commercial success have been developed for that purpose. We list and describe a few systems that in our opinion form a representative set of the current popular, commercially available, biofeedback systems. Entry level systems and devices manly use only accelerometers to measure the energy expenditure of athletes. It has been confirmed that basic time-motion parameters in many sports can be analyzed sufficiently well only by using 3D accelerometers. Most wearable devices include MEMS inertial sensors (accelerometers, gyroscopes and magnetometers) integrated into one IMU. Those inertial sensors are small and low-cost devices. For that reason, many inexpensive wristband fitness trackers are currently present in the market and popular among general population [24]-[25]. Many wearable devices are commercially available and used in amateur and professional sport activities. Most popular wristband fitness trackers such as FitBit, Nike Fuelband, and Microsoft Band are supported by several different (IoT) applications. A number of studies confirm that such devices correctly measure

the number of steps and energy consumption, but are not reliable for more complex movement patterns or activities [26]-[28]. Several sport smartwatches are also equipped with similar sensors as wristbands. In addition to the physical body activity parameters, some wristbands and smart-watches also measure various physiological parameters, such as body temperature, heart rate, and blood oxygen level. Beside the popular activity trackers for general population or for amateur users, other specialized devices are used in elite sports. They are able to accurately quantify specific athlete’s activity and even reduce the number and severity of injuries. Catapult system has a leading role in elite sport performance analytics with their monitoring system for elite sport [29]. Catapult wearable devices are designed to evaluate the sport-specific athlete’s activity (player load) in more than thirty different sports, also in most popular team sports, such as football, rugby, and hockey. Players’ load can be calculated from the acceleration data; however the Catapult wearable devices are also equipped with several other sensors: gyroscopes, magnetometers, and GPS. A wristband with integrated accelerometer, heart rate sensor, and body temperature designed by Whoop is intended for a permanent athlete’s activity monitoring [30]. Acquired data is used to predict the level of athlete’s personal readiness, exertion and sleep need. Smartphone application notifies the athlete to follow the advised daily rhythm for strain, recovery and sleep. Early concussion detection can help reduce the brain damage in contact sports such as football. The Linx wearable device is designed to measure the impact forces on the head [31]. It transfers the event alert in real-time to responsible team medical professionals, who can diagnose and treat the head injuries. In the near future, some new sensor technologies are needed to minimize the sport injuries [24]. For example the non-invasive measurement of blood sugar and lactic acid levels could help a lot to prevent injury during the sport activity. Sensors for non-invasive anterior cruciate ligament (ACL) strain measurements can also be helpful in reducing the frequency of knee injury [24]. It is relatively straightforward to classify the abovementioned systems according to the rules presented in Section 2.4.4 and Figure 5. Wristbands and smartwatches fall into the user functionality system type with compact structure and predominantly terminal feedback. Professional systems, such as Catapult, Whoop, and Linx, fall into the instructor functionality system type with distributed structure and terminal feedback. All above described systems have elements of the cloud functionality because the data can be stored and processed in the cloud. It can be noticed that real-time biofeedback systems, which are the focus of this paper, are quite rare at present. Two examples on the prototype level for golf and swimming are presented in [32] and [33], respectively. It is expected that in the future biofeedback systems will incorporate real-time feedback in much larger part. It is also expected that the next generation of real-time biofeedback systems will use a larger number of sensors and will therefore also need more powerful processing units and advanced communication technologies.

7 CONCLUSION The main question in wireless biofeedback application design is which wireless technology to choose. There is no straightforward answer to the above question. Sensors, actuators, and also biofeedback systems and applications are very heterogeneous; they can produce bit rates from a few bit/s for measuring a patient’s heart rate to a few Mbit/s for a high resolution display in sport. Similar heterogeneity is expressed in the variety of wireless technologies; from technologies that cover body area (BAN) to technologies that cover metropolitan and wide area (MAN and WAN); from technologies with bitrates of a few kbit/s to technologies with bitrates of a few Gbit/s; from frequencies of 54 MHz to 60 GHz, etc. Regarding the above said, it is no surprise that none of the currently available, widespread, and commercially accessible wireless technologies can satisfy all possible demands of different biofeedback application scenarios. We discuss the possible solutions in terms of a compromise between the range, bit rate, and biofeedback system architecture. We are aware that there is only a minor chance that one wireless technology can satisfy all the demands of various biofeedback applications. If we were to choose one from Table 2, then IEEE 802.11ah would be our candidate: the range of 1 km, bit rates of up to 40 Mbit/s, and promises of high node density. It could cover most of the different architectures of biofeedback systems; the problem is that the devices using this standard are not available yet. One of the reflections of the current state of technology used for biofeedback in sport can be found

in [34]. The author identifies a clear deficiency of available wireless sensors, multi-sensor wireless IMU devices, and the appropriate user friendly and easy to use biofeedback applications. We believe that the abovementioned deficiencies have roots in inadequacy of available wide-spread wireless technologies. In our future research and work we intend to focus also on defining a set of quality of service (QoS) parameters that would be tailored to biofeedback applications, with the emphasis on real-time architectures providing concurrent feedback. We believe that based on these QoS parameters choosing or maybe even developing the adequate wireless technology for biofeedback systems would be much easier.

ACKNOWLEDGMENT This work was supported in part by the Slovenian Research Agency within the research program Algorithms and Optimization Methods in Telecommunications (research core funding No. P2-0246).

AUTHOR CONTRIBUTIONS All authors conceived and designed the study, all authors contributed to writing the paper.

CONFLICTS OF INTEREST The authors declare no conflict of interest. The founding sponsors had no role in the design of the study, in the writing of the manuscript, and in the decision to publish the results.

REFERENCES [1]

Internet of Things (IoT), Keysight Technologies, 5992-1217EN, www.keysight.com, May 4, 2016, (accessed 24.7.2017).

[2]

Sun, Y., Song, H., Jara, A. J., & Bie, R. (2016). Internet of things and big data analytics for smart and connected communities. IEEE Access, 4, 766-773.

[3]

Lightman K. Silicon gets sporty. 10.1109/MSPEC.2016.7420400.

[4]

Pyattaev, A., Johnsson, K., Andreev, S., & Koucheryavy, Y. (2015). Communication challenges in high-density deployments of wearable wireless devices. IEEE Wireless Communications, 22(1), 12-18.

[5]

Giblin, G., Tor, E., & Parrington, L. (2016). The impact of technology on elite sports performance. Sensoria: A Journal of Mind, Brain & Culture, 12(2).

[6]

Giggins, O. M., Persson, U. M., & Caulfield, B. (2013). Biofeedback in rehabilitation. Journal of neuroengineering and rehabilitation, 10(1), 60.

[7]

Sigrist, R., Rauter, G., Riener, R., & Wolf, P. (2013). Augmented visual, auditory, haptic, and multimodal feedback in motor learning: a review. Psychonomic bulletin & review, 20(1), 21-53.

[8]

Baca, A., & Kornfeind, P. (2006). Rapid feedback systems for elite sports training. IEEE Pervasive Computing, 5(4), 70-76.

[9]

Cavallari, R., Martelli, F., Rosini, R., Buratti, C., & Verdone, R. (2014). A survey on wireless body area networks: technologies and design challenges. IEEE Communications Surveys & Tutorials, 16(3), 1635-1657.

IEEE

Spectrum.

2016

Mar,

53(3),

48-53,

DOI:

[10] Kos, A., Tomažič, S., & Umek, A. (2016). Suitability of smartphone inertial sensors for real-time biofeedback applications. Sensors, 16(3), 301. [11] Umek, A., & Kos, A. (2016). The Role of High Performance Computing and Communication for

Real-Time Biofeedback in Sport. Mathematical Problems in Engineering, 2016. [12] Chen, M., Gonzalez, S., Vasilakos, A., Cao, H., & Leung, V. C. (2011). Body area networks: A survey. Mobile networks and applications, 16(2), 171-193. [13] Cao, H., Leung, V., Chow, C., & Chan, H. (2009). Enabling technologies for wireless body area networks: A survey and outlook. IEEE Communications Magazine, 47(12), 84-93. [14] Lin, J. R., Talty, T., & Tonguz, O. K. (2015). On the potential of bluetooth low energy technology for vehicular applications. IEEE Communications Magazine, 53(1), 267-275. [15] Adame, T., Bel, A., Bellalta, B., Barcelo, J., & Oliver, M. (2014). IEEE 802.11 ah: the WiFi approach for M2M communications. IEEE Wireless Communications, 21(6), 144-152. [16] Movassaghi, S., Abolhasan, M., Lipman, J., Smith, D., & Jamalipour, A. (2014). Wireless body area networks: A survey. IEEE Communications Surveys & Tutorials, 16(3), 1658-1686. [17] Ayub, K., & Zagurskis, V. (2015). Technology Implications of UWB on Wireless Sensor Network-A detailed Survey. International Journal of Communication Networks and Information Security, 7(3), 147. [18] Kwak, K. S., Ullah, S., & Ullah, N. (2010, November). An overview of IEEE 802.15. 6 standard. In 2010 3rd International Symposium on Applied Sciences in Biomedical and Communication Technologies (ISABEL 2010) (pp. 1-6). IEEE. [19] Zhang, Y., Sun, L., Song, H., & Cao, X. (2014). Ubiquitous WSN for healthcare: Recent advances and future prospects. IEEE Internet of Things Journal, 1(4), 311-318. [20] Deslise, J. J. (2015). What’s the difference between IEEE 802.11 af and 802.11 ah?. Microwaves RF, 54, 69-72. [21] Baños-Gonzalez, V., Afaqui, M. S., Lopez-Aguilera, E., & Garcia-Villegas, E. (2016). IEEE 802.11 ah: A Technology to Face the IoT Challenge. Sensors, 16(11), 1960. [22] Hämäläinen, M., Paso, T., Mucchi, L., Girod-Genet, M., Farserotu, J., Tanaka, H., ... & Ismail, L. N. (2015, March). ETSI TC SmartBAN: Overview of the wireless body area network standard. In 2015 9th International Symposium on Medical Information and Communication Technology (ISMICT) (pp. 1-5). IEEE. [23] Wang, C. X., Haider, F., Gao, X., You, X. H., Yang, Y., Yuan, D., ... & Hepsaydir, E. (2014). Cellular architecture and key technologies for 5G wireless communication networks. IEEE Communications Magazine, 52(2), 122-130. [24] Seshadri, D. R., Drummond, C., Craker, J., Rowbottom, J. R., & Voos, J. E. (2017). Wearable devices for sports: New integrated technologies allow coaches, physicians, and trainers to better understand the physical demands of athletes in real time. IEEE pulse, 8(1), 38-43. [25] Li, R. T., Kling, S. R., Salata, M. J., Cupp, S. A., Sheehan, J., & Voos, J. E. (2016). Wearable performance devices in sports medicine. Sports health, 8(1), 74-78. [26] Diaz, K. M., Krupka, D. J., Chang, M. J., Peacock, J., Ma, Y., Goldsmith, J., ... & Davidson, K. W. (2015). Fitbit®: An accurate and reliable device for wireless physical activity tracking. International journal of cardiology, 185, 138-140. [27] Takacs, J., Pollock, C. L., Guenther, J. R., Bahar, M., Napier, C., & Hunt, M. A. (2014). Validation of the Fitbit One activity monitor device during treadmill walking. Journal of Science and Medicine in Sport, 17(5), 496-500. [28] Tucker, W. J., Bhammar, D. M., Sawyer, B. J., Buman, M. P., & Gaesser, G. A. (2015). Validity and reliability of Nike+ Fuelband for estimating physical activity energy expenditure. BMC rts Science, Medicine and Rehabilitation, 7(1), 14. [29] The Monitoring System of Choice for Elite Sport, https://www.catapultsports.com/products, accessed on 25.2.2018. [30] WHOOP Elite, https://whoop.com/elite/, accessed on 25.2.2018. [31] Advanced Technology to Make Sports Safer, https://linxias.com/features/, accessed on

25.2.2018. [32] Umek, A., Tomažič, S., & Kos, A. (2015). Wearable training system with real-time biofeedback and gesture user interface. Personal and Ubiquitous Computing, 19(7), 989-998. [33] Li, R., Cai, Z., Lee, W., & Lai, D. T. (2016, August). A wearable biofeedback control system based body area network for freestyle swimming. In Engineering in Medicine and Biology Society (EMBC), 2016 IEEE 38th Annual International Conference of the (pp. 1866-1869). IEEE. [34] Howard, Roisin. Wireless Sensor Devices in Sports Performance. IEEE Potentials, 2016, 35.4: 40-42.

Anton Kos received his Ph.D. in electrical engineering from University of Ljubljana, Slovenia, in 2006. He is an assistant professor at the Faculty of Electrical Engineering, University of Ljubljana. He is a member of the Laboratory of Information Technologies at the Department of Communication and Information Technologies. His teaching and research work includes communication networks and protocols, quality of service, dataflow computing and applications, usage of inertial sensors in biofeedback systems and applications, signal processing, and information systems. He is the (co)author of twenty-seven papers appeared in the international engineering journals and of more than fifty papers presented at international conferences.

Veljko Milutinović is a professor at the School of Electrical Engineering, University of Belgrade, Serbia. During the 80's, for about a decade, he was with the Purdue University in the U.S.A, where he co-authored the architecture and design of the world's first DARPA GaAs microprocessor. During the 90's, after returning to Serbia, he took part in teaching and research at a number of major EU schools. He also delivered lectures at Stanford and MIT, and has about 20 books published by leading publishers in the USA. He is a Fellow of the IEEE and a Member of Academia Europaea.

Anton Umek received his Ph.D. in electrical engineering from University of Ljubljana, Slovenia, in 1999. He is an asisstant professor at the Faculty of Electrical Engineering, University of Ljubljana. His teaching and research work during the last ten years includes signal processing, digital communications, communication security and access networks. He is the (co)author of more than thirty conference and journal papers. His current research interest includes biofeedback systems and wireless sensor networks.

Highlights 







Importance of IoT and embedded devices for biofeedback systems in sport. Definition of biofeedback system constraints, versions, processing, and communication. Sensors, actuators, and wireless technologies used in sport are very heterogeneous. None of the wireless technologies satisfies all demands of biofeedback applications.