**2. Related Works**

Until now, there are numerous studies that investigate the effects of the human body shadowing on wireless signal transmission to characterise and model the wireless channel and antenna radiation by focusing on various aspects of body-centric radio frequencybased communication. Moreover, human BSEs have been analysed and modelled for a variety of applications, including people counting [7,8], fall detection [9,10], and activity recognition [11], as well as proximity detection for coronavirus contact tracing application [12]. Although several studies have been performed on analysing the effect of human body shadowing on radio signal transmission targeting indoor localisation applications, they have mostly neglected the derivation of a compensation model and/or the integration of a compensation model to implement a real-time localisation system. Table 1 presents an overview of the existing literature that discusses human BSEs on wireless signal transmission for indoor localisation and tracking applications.

In the literature, researchers utilised several wireless technologies; this study only focuses on the systems that exploited RSSI as their measurement approach and/or fingerprinting as their localisation method. An RFID-based system is presented in [13], where the

authors demonstrated the improvement in indoor localisation accuracy by compensating the errors caused by human body shadowing. Channel models for both the LOS and NLOS cases were derived, and RFID RSSI-based Monte Carlo localisation was implemented to achieve an accuracy of 1.18 m. However, this approach has very limited applicability for real-time location tracking applications because the differentiation between the LOS and NLOS conditions were assessed manually.

**Table 1.** Comparison among existing studies focused on human body shadowing effects on wireless signal for indoor localisation and tracking applications.


1 An = analysis; Mo = modelling; Co = compensation. 2 Ex = experiment; Si = simulation.

From Institute of Electrical and Electronic Engineers (IEEE) 802.11 family of standards, the impact of the human body on RSSI-based ranging measurements for cooperative localisation is presented in [15]. The authors investigated both the body and hand grip effects on RSSI among the neighbouring nodes. This study demonstrated that there is no significant improvement in cooperative localisation, compared to the noncooperative case, if the BSEs are not mitigated correctly. In [18], a mathematical model is proposed to mitigate the user's BSE on RSSI of WiFi signals for improving indoor positioning accuracy. Handheld mobile devices were used to collect WiFi signals in a multipath-free environment, both for the LOS and NLOS cases, to analyse the BSEs. Finally, a model was derived that can intensify the strength of the signals which are coming through the NLOS states caused by the user body. Still, the authors did not discuss the methods of orientation estimation for real-time applicability of the proposed model. Moreover, this study only considers the handheld mobile devices for the experiment, and thus the model may not be compatible with body-attached wearable devices. In [5], the authors presented the first fingerprintingbased indoor localisation system that considered the subject's BSEs on signal RSSI for position estimation. A radio map was created by using both the empirical measurements of RSS and a signal propagation model. To reduce the location estimation error that is cause by the user's body, RSSI fingerprints were collected for four orientations of the subject's body, compared to the RNs, in terms of four directions, i.e., north, south, east, or west. A K-NN search algorithm was employed, and a median accuracy of 2–3 m was achieved after compensating the human BSE. However, this solution only analyses the effect of user orientation on location estimation and falls short of proposing any compensation model with its real-time applicability to mitigate that effect. To solve the issue of estimating user orientation in real time, King et al. described an approach named COMPASS, where

the authors utilised a digital compass to acquire the user's orientation during both the offline and online phases [14]. During the offline phase, radio fingerprints were collected from each RN for eight orientations in every 45◦ angle position. In the online phase, a subset of fingerprints from the radio map was preselected based on the user orientation, and a probabilistic algorithm was applied to the subset to calculate the user position. Results demonstrated that considering the body orientation improved the localisation accuracy, where the average accuracy was 1.65 m. Yet, the radio map becomes highly redundant as eight radio fingerprints corresponding to eight directions, i.e., in every 45◦, were collected for a single RP. As a result, the search space increases by eight times, which can cause an extra burden on the system performance in terms of computation cost and memory requirement for a large environmental area. It may even become infeasible for resource-constrained wearable devices for edge computing in the case of real-world applications. Moreover, it also increases the cost of the offline phase in terms of time and labour. Additionally, using COMPASS may produce high errors in orientation estimation for indoors, especially around the objects that have electromagnetic radiations. A similar approach was applied in [19], where the authors used mobile phone integrated compass and collected radio fingerprints for four orientations in the offline phase. During the online phase, they narrowed down the search space by applying a clustering method that used both the signal domain and spatial domain. An adaptive weighted K-NN algorithm was developed, which achieved an accuracy of 2.0 m for the 50th percentile; however, this study only considers four orientations of the human body in four directions, which is not enough to explore the complete variations of RSS values around a body.

From Zigbee-based indoor localisation solutions, a body-worn device is used in [16], where the authors analysed the BSEs on RSSI of 2.4 GHz ZigBee signals. Two tags were attached on the chest and back of a wearer, and data were collected at different angular positions. The arc tangent function is used for orientation estimation, and a simple cosine model is used to compensate the user's BSE. Another improved version for BSE compensation in indoor localisation is proposed by the same group in [17], where the authors presented two solutions for improving localisation accuracy. In the first solution, a subject requires multiple wearable tags that need to be mounted in different positions to calculate RSSI. Then, their means are used as the input for fingerprint matching with reference fingerprints from a radio map generated using the WHIPP tool [21]. In the second solution, an arc tangent function is used to estimate the orientation of the target, where the target's current location is calculated by averaging four previous positions. To mitigate the BSE, two compensation models are proposed: one is a basic over/underestimation model, and the other is a simulation-based three-dimensional model. Results demonstrated that the proposed models could compensate the BSE to improve localisation accuracy from 3.48 m to 2.99 m (50th percentile for chest). However, this approach requires a subject to wear multiple tags, which may limit its scope for real-world application. Moreover, as the radio map created during the offline phase did not consider the BSE, the estimated location accuracy will be low and, eventually, the orientation estimator's performance will degrade with time. Furthermore, the system assumes a subject always walks forward, and is unable to infer rotation and moving direction, which can cause a significant difference between the estimated orientation and actual orientation.

Recently, Deng et al. reported an IMU-aided system to compensate body shadowing error for BLE-based indoor positioning [20]. The effects of the human body on BLE signal RSSI were analysed. A compensation model was proposed which considers the distance and angle between an RN and the unknown target to calculate the error. The distance is calculated from the signal propagation model, and the user's heading is approximated from the IMU. Finally, an algorithm was proposed to estimate the location of an unknown target by mitigating the body shadowing error in real time. Results demonstrated that the system could achieve an average accuracy of 0.77 m for location estimation. However, the use of IMU exclusively can produce wrong heading estimation because of error accumulation issues with IMU. Moreover, the use of a signal propagation model solely to measure

distance can produce high distance error, especially indoors. Thus, the described body shadowing detection strategy can lead to erroneous output as a consequence of the errors from the heading and distance estimation.

In this study, the user's orientation is estimated by applying a unique approach using a BSE compensation model that is proposed to mitigate the user's body shadowing error in real time to improve indoor localisation accuracy.

#### **3. Analysis of User's Body Shadowing Effect on RSSI**

Several experiments were performed to investigate the effects of the user's body on radio signal RSS values. This section describes the experiments and provides observations from the experiments.
