Visible light positioning (VLP) is an emerging and promising technology in the indoor positioning system (IPS) landscape that provides centimeter-level accuracy at a low cost. However, it needs to be more thoroughly investigated, particularly in industrial environments where the presence of metal imposes a significant multipath efect and degrades the VLP accuracy. The absence of literature studies that evaluate VLP in the presence of multipath prompts us to conduct our own set of experiments. The main goal of this paper is to provide insight regarding the spatial extent to which the positioning performance is noticeably afected. Moreover, the efects of shadowing as well as reflections are considered, with the term Non-Line-of-Sight (NLoS) referring to both efects. The experimental results demonstrate that the increase in received light intensity due to reflections depends on the relative location of the light-emitting diode (LED), the photodiode (PD), and the reflecting surface. For the investigated configuration, it is observed that the localization error can be impacted up to 1.5m from a metal closet, with a pronounced specular component. Finally, it is experimentally confirmed that NLoS efects are rather deterministic, which allows them to be modeled.
Visible Light Positioning (VLP), which uses light-emitting diode (LED) light signals, is an
emerging technology for the next generation of low-cost indoor positioning systems (IPSs).
It has the potential to address some of the limitations of RF-based localization technologies,
such as high cost in either use or installation, electromagnetic (EM) interference, privacy
violations, and the problem of an increasingly crowded RF spectrum [
Research on the evaluation of VLP in the presence of multipath are scarcely found. Xu et al.
[
In this paper, we will investigate experimentally the influence of multipath reflections caused by the presence of a closet and storage racks on VLP performance. In what follows, we provide a description of the measurement configuration and introduce the experimental scenarios in Section 2. In Section 3, the propagation model and localization approach for our case study will be presented. Section 4, discusses the performance of VLP in NLoS environments based on the obtained results. Finally, conclusions and future works are discussed in Section 5.
To evaluate VLP system’s positioning accuracy, a 4m x 4m x 5m lab was used, whose outskirts are covered with black cloths to minimise external light and uncontrolled reflections. Fig.1 shows the lab and its features. Its inside houses a height-adjustable platform of cable trays, from which various configurations of LED transmitters can be suspended. Below, the wooden lfoor is strategically perforated to form a uniform square grid with a 0.5 m spacing. The holes’ purpose is to accommodate the measuring platform.
Various LED transmitter configurations are at one’s disposal. In this work, with reproducibility
in mind, most often a combination of commercial components is used. The COB LEDs used in
the experiments are of the Bridgelux BXRE-50C3001-D-24 type, having a Lambertian radiation
(a)
(b)
(c)
pattern. Experiments are also performed with linear LEDs not having non-Lambertian radiation
patterns, namely ETAP’s E4010/LED1N060D LEDs. Additionally, both of the employed LED
types are intensity modulated to transmit 50% duty cycle square waves with frequencies , =
2−1 0, ( = 1, 2, .., ). 0 is set to 500 Hz to exceed the flicker threshold [
To ensure scientific reproducibility, the default receiver resorts to commercially available devices. It couples a Thorlabs amplified photodetector module with tunable gain, namely the PDA36A2 or PDA100A2 with an active area of = 13 mm2 and = 75.4 mm2, respectively, to a National Instruments USB-6212 for data acquisition (DAQ). The optical receiver under test is iftted on top of a 2D slider system consisting of motor-Driven Velmex’ BiSlides ® (See Fig.2 (b)). This measurement system covers a 1m by 1m zone with a predefined (here mostly 2.5 cm) granularity and a submillimetre accuracy. By sequentially displacing it, for which the alignment is ensured by the holes in the floor, the complete 4m x 4m receiver plane can be traversed. A MATLAB® (or Python) backend is responsible for addressing the sliders, the sensor read-out, and the data manipulation.
To investigate the NLoS influence on the VLP performance, we used both a greige-green lacquered metallic closet and storage racks with carton boxes as typical obstacles in our experimental measurements, depicted in Fig.2 (c) and Fig.3 (a-d), respectively. To induce more homogeneous reflections, the closet is turned with its backside to the positioning area (See Fig.2 (c)). The arrangement of each experiment is shown in Table.1.
As the area is confined by black cloths, we assume the receiving photodiode (PD) will receive
light either along the Line-of-Sight (LoS) path or along a first-order reflection of the closet
or rack. Fig.4 shows how the optical light channel is usually modeled in the case of NLoS
contributions [
()
[
(, ) ℎ, = (′, ′) ·
2,1 · 2,2
cos( ′) · · ( , )
where (, ) is the radiation pattern of the LED, which is axially symmetric in the case of
a Lambertian emitter and for order , reduces to +1 cos(), with the angle of irradiance
(see Fig.4) [
()
(1)
(2)
(3)
(4)
radiation pattern, and it is typically governed by Phong’s model consisting of both difuse and
specular parts [12]. It should be noted that the NLoS components are the same LoS variables
with the symbol prime ′. The PD generates photocurrent contributions , in response to
incident light and the measured , values are linked to , using the nominal responsivity
̂︁ of the photodiode as follows [
,
· ̂︁
(5)
where is defined as a gain mismatch factor. It accounts for the wavelength mismatch
between transmitter and receiver [
For localization and measuring ,, the LED i’s photocurrent magnitude-based RSS (received
signal strength) values, we installed the PDA36A2 or PDA100A2 on top of the 2D slider system.
Moreover, 15 , values are averaged per measurement location to minimize the noise impact.
For demultiplexing the contributions of the diferent LEDs, we employed the Fast Fourier
Transform (FFT)-based demodulation as presented in [
Fig.5 depicts the 4 LEDs’ , planar distributions, obtained with the PDA36A2 in the presence of the closet. From Fig.5 (a)-(d), it is clear that due to light’s particle/ray (dual) nature NLOS efects are much more deterministic, and hence predictable for VLP than for sub-GHz radio frequency signals. Shadowing locally efectuates a large , reduction. In Fig.6 (a), it is visualised by the cluster of points with an , that is smaller than what it would be in LoS conditions. With light the , magnitude originating from difraction is limited. It efectuates that in the shadow areas , nears the noise floor. Moreover, multipath’s , enlarging contribution (i) depends on the relative positioning of the LED and the closet through the irradiance angle during reflection , (ii) is spatially confined to around a 1.5m vicinity of the closet, and (iii) exhibits a pronounced specular component. Furthermore, diferentiating LED 2 and 3’s − , curves of Fig.6 (a)/(b) shows both that the , of the LEDs closer to the obstacle are relatively larger afected, and that the multipath contribution can be of significant magnitude. From a trilateration point-of-view, the obstacle-induced , surplus entails an underestimating of the inter-LED-PD distance. This underestimation, and − ,’s noninvertiblility in general, will impede highly-accurate localisation around the obstacle.
Fig.7 quantifies the spatial extent of the impact of closet-induced multipath on the 2D positioning performance. The NLoS influence mainly manifests locally around the obstacle, and specifically around the specular reflection vector of each of the LEDs. Expectedly, positioning outliers are found in the regions of , surplus. These do not exceed the 1m error bound. Moreover, all in all, MBF copes well when a single LoS link obstruction occurs. The errors in the left top and bottom corners are mainly the result of the , (the radiant flux of LED ) calibration of LEDs 2 and 4, which did not correct their NLoS contribution. Typically, this , calibration would be performed beforehand and in controlled circumstances. Finally, Fig.8 demonstrates that the obstacle-induced NLoS is also dependent on the LED fixture’s nature. It shows the ,2 and ,4 distribution in the presence of ETAP’s E4010/LED1N060D. Similarly, as for the COB LED case, a strong RSS reduction due to shadowing is observed, as well as an increased RSS due to closet-induced specular reflections. Notice that, due to practical constraints, we were unable to measure right up to the closet, leaving some space. This is especially visible in Fig.8, and we should not interpret it as shadows.
10-6 10-7 10-83 3.5 4 4.5 5 ] m [
Additionally, we employed the Rician K-factor () to visualize the NLoS contribution in our experiment. is defined as the ratio between the LoS power ( ), determined via an identical measurement without a closet, and the power in all other scattered paths ( ), determined as the power in the closet scenario minus the power in the without-closet scenario, = 1010︀(
)︀ () [13]. Notice that as this approach for determining the LoS and NLoS contributions is only valid when the received power in the no-closet scenario is smaller than or equal to the received power in the closet scenario, we here exclude the areas that are shadowed by the closet. The grey areas in Fig.9, therefore, indicate the shadowed areas (as well as the areas right next to the closet where measurements were practically impossible). Figs.9 (a) and (b) show for LED2 (top right LED) and LED4 (bottom right LED), respectively. From both figures, generally drops when moving away from the LED. In Figs.9 (a) and (b), zones C indicates that further away from the LEDs, the NLoS contribution falls within the noise lfoor, while the LoS contribution decreases with distance. The highlighted areas of A and B, on the other hand, illustrate that there has been a sharp drop in the because of the larger contributions of NLOS due to the closet. 10-6 3 1 0 -1 -2-2 -1 0 1
2 (b) 10-6
To provide insight into the expected NLoS extent encountered during industrial deployments,
Fig.3’s imaged storage rack features in two configurations. Fig.10 shows the , found when
the storage rack is located at the outside of the positioning zone, while Fig.11 does so for a
scenario in which the storage rack is located in between the LED arrangement. Fig.10 and
Fig.11 display the more complex multipath and obstruction nature of the storage rack. It is a
consequence of the heterogeneity of the storage rack (and its contents). The narrow,
highlyreflective metal structures introduce very directive , contributions, as can be remarked from
Fig.10 (c) and (e) for instance. Luckily, the associated influence on the positioning is spatially
confined as well. The accuracy diminishing efect of the first storage rack configuration is treated
by Almadani et al. in [
6.5
(e)
the points are located more towards the negative y-coordinates than the storage rack is, the
former’s light is blocked by, while the latter’s propagates between, the metallic structures. A
ifnal remark constitutes that to guarantee a suficient amount of LoS links in the presence of
closely spaced rows of tall storage racks, the exact location and shape of obstacles are needed
for VLP with a fine granularity, which for RF is not the case for example [
A 3 [ 2 1 2 0 10-6 5 4 1 1 0 2 1 0 -1 -2-2 -1 0
1 10-6 10-74 (d) 10-5 10-6 10-7 4 4.5 5 5.5
This paper conducts an experimental investigation into the accuracy of VLP in environments with induced obstacles, such as metallic closet and storage racks. Two VLP’s accuracy degradation factors, including multipath and shadowing, are considered, which are both collectively referred to as NLoS impacts. The experimental results demonstrate that the obstacle-induced NLoS influences (i) are much more deterministic and therefore predictable for VLP than for the sub-GHz radio frequency (RF) signals, (ii) are vastly dependent on the local deployment of the LED transmitters and the location of the obstacles through the irradiance angle during reflection, (iii) are spatially confined to around a 1.5m vicinity of the obstacles because of attenuation due to high path loss, and finally (iv) are dependent on the nature of the LED fixture. Future work could include experimental modelling of the NLoS contributions and incorporating these into MBF-based localization approaches to mitigate NLoS-induced positioning errors. Angular Characteristics on RSS-Based VLP Accuracy, in: IEEE Access, vol. 8, 2020, pp. 83116-83130. doi: 10.1109/ACCESS.2020.2991298. [12] C.R. Lomba, R.T. Valadas, A.M. de Oliveira Duarte, Experimental characterisation and modelling of the reflection of infrared signals on indoor surfaces, in: IEE Proceedings Optoelectronics, vol. 145, no.3, 1998, pp. 191 – 197. DOI: 10.1049/ip-opt:19982020. [13] C. Tepedelenlioglu, A. Abdi, G. B. Giannakis, The Ricean K factor: estimation and performance analysis, in: IEEE Transactions on Wireless Communications, vol. 2, no. 4, July 2003, pp. 799-810. doi: 10.1109/TWC.2003.814338.