Development of a Localization System Based on VLC Technique for an Indoor Environment

Abstract

In this paper, we develop an indoor localization device which embeds localization information into indoor light-emitting-diodes (LED) lighting systems. The key idea of our device is the use of the newly proposed "bit stuffing method". Through the use of stuff bits, our device is able to measure signal strengths even in transient states, which prohibits interference between lighting signals. The stuff bits also scatter the parts of the signal where the LED is turned on, thus provides quality indoor lighting. Additionally, for the indoor localization system based on RSSI and TDM to be practical, we propose methods for the control of LED lamps and compensation of received signals. The effectiveness of the proposed scheme is validated through experiments with a low-cost implementation including an indoor navigation task.

1. Introduction

Indoor lighting using white LEDs has drawn much attention due to its energy efficiency [1]. Also, due to the advantage of LEDs being able to operate in high frequencies, communication using visible light (VLC) with these LEDs has drawn much interest as well [1-4]. Over the past few decades, many indoor localization systems (ILSs) based on IR, Ultrasound, RFID, WLAN, Bluetooth, and UWB have been proposed, but these systems have limitations such as the necessity of additional infrastructure, low accuracy, noises, electromagnetic interference, multi-path effects, low security, and etc [5, 6]. Naturally, much research has been conducted recently on using these indoor LED lighting and VLC techniques as an alternative indoor localization signal for places where satellite global-positioning-system (GPS) signals are not available [6-15]. The biggest advantage of using VLC techniques for indoor localization is that no additional infrastructure is required for the installation since the system uses existing infrastructure of LED ceiling lamps, which reduces the cost of the ILS.

For VLC-based ILSs, to acquire the position of an object, positioning techniques based on angle of arrival (AoA), time of arrival (ToA), time difference of arrival (TDoA), and received signal strength indication (RSSI) are widely used [7, 8]. AOA [9], ToA, TDoA [7, 10] approaches have disadvantage such as deployment of an array of image sensors, synchronization problem of all transmitters and receivers, precise delay control, and sensitivity to the phase noise [8], [11]. These in turn bring cost and complexity problems when implementing the actual system. RSSIbased localization systems using RF carrier[12], dual-tone multi-frequency [13], time allocation [14], RSSR [11] and wavelength allocation [15] have been reported in the literature. Although RF carrier and dual-tone multifrequency techniques have very good estimation accuracy, they have limitations. RF carrier technique requires additional compensation since frequency response of Tx and Rx is unstable depending on the frequency. Dual-tone multi-frequency technique needs to consider signal-tonoise ratio (SNR) for the actual system to work, since the SNR and the positioning error are inversely proportional. Additionally, these techniques based on the frequency modulation require a high-speed analog-to-digital converter (ADC), a broadband low-noise amplifier (LNA), and more computing power, thus increasing the cost of the system and the overall complexity. In case of time allocation and RSSR techniques based on time division multiplexing (TDM), their drawback is the necessity of synchronization process [6]. In addition, the maximum number of LED lamps could be limited in TDM method to prevent LED flickering. Wavelength allocation technique requires additional devices such as an optical filter, multiple Rx [6].

Here, we focus on RSSI and TDM based localization technique for VLC-based ILS due to the fact that they do not require precise delay control, additional device, and high-performance hardware. For this type of method, identifications (IDs) are embedded to each LED lamp control signal during a given time slot, and the way they are embedded is critical for the accuracy of localization and the quality of lighting. The embedding should be done such that IDs are easily retrievable and have minimal impact on human visible light perception (i.e., having a repetitive signal of over 50 Hz to reduce eye fatigue). Also, the length of the pulse for ID needs to be kept at minimum in order to maximize the time duration of the LED being on, which prevents LED flickering. However, as the number of LED lamps increase, more bits are required to embed ID. This in turn means more bits require shorter pulse intervals, which brings cost problems when choosing devices for implementation.

Another problem when implementing the RSSI-TDM method is interference between ID pulses. The interference make it difficult to measure the RSSI value accurately because the previous lamp state (ID pulse on or off) affects the RSSI value of the receiver. Moreover, in real world scenarios, the RSSI value at the receiver is affected by the environment such as ambient light, roughness of the floor, measurement error, and so on, unlike simulation.

In this paper, we propose a new method for embedding IDs into LED lighting using the newly proposed “ID stuffing” to avoid lighting flickering and to prevent inter-pulse interference. In addition, we present a column-row lamps driving method, which minimizes the off-state of the lamp and interference between adjacent lamps. To do so, the proposed method activates only one column or row lamps at a given time slot when sending the VLC_Packet. Additionally, for RSSI/TDM-based ILS to be more practical, we embed the compensation functionality for the ambient light and the roughness of the floor. Compared to other VLC-based ILSs, our system has the following advantages; Simplicity of the implementation, low-cost, auto-synchronization of the receiver, scalability for a number of LED lamps, and good accuracy in real world scenarios.

To demonstrate the effectiveness of the proposed system, we implement an indoor localization system using LED lamps and a mobile robot equipped with an optical receiver. The LED lighting system is configured as a 4 × 4 grid with light sources placed 40 cm apart from each other on the ceiling. The receiver mounted on the robot is located on the floor (170 cm from the ceiling). With this system configuration, we have the indoor localization result of position error less than 3 cm. In addition, we show that the mobile robot can move autonomously using the proposed method.

 

2. System Description

As shown in Fig. 1, our indoor localization system consists of 16 LED lamps configured as 4 × 4 grids placed 40 cm apart from each other on the ceiling, and a low-cost receiver located on the floor (170 cm from the ceiling). Each lamp is fabricated with 18 white LEDs to ensure an illuminance level of 60 lux. The viewing angle of LED (NBL-R3W) is 30 degree and the typical luminous intensity is 5.0 cd. The 16 lamps are controlled by a single 8-bit microcontroller ATmega128 (Atmel Co.), and the lamp’s IDs are encoded as rows and columns to reduce the number of bits in representation. In short, the 50 Hz repetitive lamp driving signal is composed of the illuminating interval (which is always 1), the 12-bit signal interval (START_BIT:010, ID_BIT:CcccRrrr, END_BIT:0). The signal interval is referred to as the VLC_Packet. In case of the lamp located at column 1 row 1, the ID_BIT and VLC_Packet will be [10001000] and [010100010000] respectively, whereas in case of column 3 row 2, [00100100] and [010001001000].

Fig. 1.Indoor localization system with LED lamps and optical receiver mounted on the robot.

The receiver uses a low-cost photo-diode (SFH-213) to measure the strength of the light source in the visible light spectrum. The half angle of photo-diode is ±10 degree and the radiant sensitive area is 1 mm2. As the START_BIT is detected in the receiver, the receiver decodes the following ID_BIT and measures the signal strength. In this manner, the receiver is able to recognize a start point of VLC_Packet. Thus the proposed system does not require synchronization between the transmitters and the receiver. When decoding, if the pulse width is large, it is easy to obtain the signal strength, but the overall signal interval becomes long, thus resulting in less illuminating interval and poor lighting. This is because, in the VLC_Packet, at least nine of the pulses are ‘0’, and therefore provides poor lighting in the signal interval. Conversely, if the pulse width is short, then the illuminating interval becomes long, providing good lighting, but measuring the strength of the light coming from each lamp becomes hard. Our core contribution is the signal encoding method which overcomes this trade-off. We use short pulse intervals but insert small pieces of illuminating intervals between each pulse in the ID_BIT, which we call “ID stuffing” (borrowing the concept of bit stuffing from the Control Area Network Protocol [16]). This insertion of stuff bits acts as if we scatter our ID signal interval onto the entire driving signal, which has an important effect. This scattering separates ID signal bits and keeps them from interfering with each other, allowing shorter pulse intervals.

 

3. Measuring RSSI

To show that a simple straightforward implementation without stuff bits is not feasible, we experimented with two pulse widths, 700 us and 100 us. We will also show how the stuff bits work even though we choose short pulse width.

3.1 RSSI without ID stuffing

In Fig. 2, with pulse interval of 700 us, the strength of each row and column LED lamps can be easily measure by observing their strengths at their steady state. RSSI values of each column lamps are 0.35, 3.80, 1.40, and 0.20 volts, respectively. However, using 700 us gives poor illumination, since the VLC_Packet is too long (9 zeros take out 6,300 us of the 20,000 us, i.e., only 68.5% being the illuminating interval). In case of using pulse interval of 100 us, the signals do not reach their steady states and are in transient states even at the end of each pulse intervals. Thus, we used these final transient values to obtain the RSSI, the values of each column lamps are 0.35, 3.25, 1.55, and 0.35 volts, respectively. However, the quality of the measured RSSI is poor in this case.

Fig. 2.Measured signal strength of lights to driving pulse width of 700 us (left) and 100 us (right).

To figure out the relation between the receiver location and RSSI of the lamp, we measured RSSI of each column and row lamps while the receiver location varies from 0 to 120 cm along the perpendicular direction of the column and row lamps.

We only present the characteristic curves of the column lamps (Fig. 3: left) because the row lamps showed almost same characteristics. From these curves, it is possible to obtain the characteristic curves between the receiver location and the RSSI as shown in Fig. 3 (We only denoted the 2nd column lamps case at right hand side, and we named it RSSI curve of the 2nd column lamps.).

Fig. 3.Relation between receiver location and RSSI of the column lamps.

In Fig. 3(left), the measured results of RSSI with pulse interval of 100 us, we can see a peak at 40 cm where the 2nd column lamps (thick line) are placed. To get the distance from the RSSI, we have to use the right hand side figure, re-plotting the data according to distances. However, as we can see in the case of 2nd column lamps (Fig. 3: right), it is not feasible to get the distance from the RSSI because the displacements of the receiver from the peak point are not symmetric, that is, the RSSI changes in right hand side of the peak and the changes of the other side are different (left hand side signal is strong). This phenomenon results from the interference of the previous ID pulse.

3.2 RSSI with ID stuffing

To prevent the interference of the previous ID pulse, we employed bit stuffing method between ID pulses.

In Fig. 4, stuff bits of duration 500 us (red dashed line) are inserted just before the row id pulses of duration 100 us, which results in a longer VLC_Packet without losing the illumination interval (illumination interval is scattered inside the signal interval). Unlike the results of the straightforward approach in Fig. 2 and Fig. 3, using stuff bits gives accurate measurements as shown in Fig. 5.

Fig. 4.LED lamps driving signal with 9 stuff bits (Column-1 and Row-1 turn on).

Fig. 5.Measured signal strength of the lamps with stuff bits.

The measured RSSI values for each row lamps, under peak values of the arrow marked interval, are 0.95, 3.75, 1.85, and 0.70 volts, respectively. The measurement results for the receiver position changes are similar to Fig. 3 but with symmetric results. Thus we can have unique RSSI for a given distance as shown in right hand side figure. This allows accurate estimation of the distance from the measured signal strength.

However, the characteristic curve shown in Fig. 6 varies depending on the location of the receiver or ambient lighting. Specifically, the signal shown in Fig. 5 is shifted up without changing shape in daytime, whereas it shifted down in nighttime. This effect could be minimized with simple addition since we know the RSSI value of the END_BIT interval where all LEDs are turned off, when the characteristic curve was established. We also have to compensate RSSI curve, because the illuminance of the floor is not even everywhere. Simple experiment reveals this problem as shown in Fig. 7.

Fig. 6.Relation between receiver location and RSSI of 2nd column lamps after normalization.

Fig. 7.RSSI curve of 2nd column lamps for different locations.

In Fig. 7, the receiver location varies from 0 to 120 cm along perpendicular direction of 1st row lamps (red line) and the center of two row lamps (blue line). It is obvious that we have to use different RSSI curve to get more exact location. However we cannot decide which RSSI curve is fit for the current measure because the receiver location is unknown. Thus we employed normalization technique to get the unified RSSI curve that is the average curve of shown in Fig. 7 (right), even though averaging errors exist.

The normalization of the signal is performed with the maximum value and the minimum value of the signal measured. The maximum value is obtained by averaging the value of illuminating or stuff pulse intervals, and the minimum value is set with the negative peak value in STOP_BIT interval. The detailed normalization process is as follows:

STEP 1: Compensate ambient lighting

established

STEP 2: Normalize the compensated signal

 

4. Indoor Localization

After the RSSI measurement, we convert the measurement to the distance between the receiver and column or row lamps. For this conversion, we need a function representing the RSSI curves shown in Fig. 7. Here we state how to obtain the function, and in turn, the location of the receiver from the measured signals.

4.1 RSSI function

According to the light diffusion equation [17], this relationship can be approximated as follows:

where l is the distance between the light sources s and the receiver and s is the signal strength measure. Here, we have two exponential terms since we have multiple lamps contributing to the measurements (one row or column at each time). The number of exponential term should be equal to the number of lamps contributing to the measurement, but we found experimentally that two was sufficient. In our experimental setup, a1=1112, b1 = 0.0341, a2 = 175.6, b2 = −4.1E−05.

4.2 Measuring receiver location

To obtain the receiver location using the RSSI function, since we use the projected displacement along the ground plane instead of the distance between the lamps and the receiver, we have to use the following trigonometry relation.

where 170 is the vertical distance between the lamps and the ground plane, l is the distance between the lamps and the receiver, and r is the projected displacement from the column(or row) lamps and the receiver.

Only three projected displacements are chosen for calculating the distance d because the slope of the RSSI curve shown in Fig. 7 is steep where the signal is weak. In other words, the weakest signal has low signal-to-noise ratio, and adding it only harms the overall estimation. Consequently, we can obtain the position of the receiver with this three selected displacements using the following equation.

where si indicates the selected displacement. For instance, if r1 is the projected displacement of the weakest signal among them, we set r2, r3, and r4 to s1, s2, and s3, respectively. The ‘n’, the number of starting column lightings chosen, serves to add the offset to the strongest lightings from the origin. As we can see in Fig. 8, it is obvious that we have to add two times of the lighting layout span, 40 cm, to the average of the selected displacements, the 2nd term of the right hand side (5). If r4 is discarded when the receiver locates near 2nd LED lamps, only one lighting layout span should be added.

Fig. 8.Selection method of the displacement d.

As a special case, the receiver located near the edge of the lighting area, any side of the Fig. 8, only two distances are available if the others are exceeded their limits too much because their signal strengths are weak. For these cases, the position of the receiver is obtained by the following equation.

The position of the receiver from row lamps is obtained by the same approach, which gives us 2D location of the receiver on the ground plane.

4.3 Indoor location example

To demonstrate the usefulness of the proposed method, we implemented the 4 × 4 LED lighting and an optical receiver on the micro-mouse as shown in Fig. 9. In order to check the position of the receiver, the orthogonal projection of the position of LEDs onto the ground plane was marked on the floor.

Fig. 9.Indoor localization system with LED lighting

We conducted two kinds of experiments. Firstly, we checked the performance of the localization capability. We manually placed the receiver on a 10 cm grid and evaluated the localization error at each grid point. We then performed navigation control of the mobile robot with the position data measured by the receiver

As we can see in Fig. 10, the error, Euclidean norm of the error of each axis, remains under 3 cm. Although the manual localization of the receiver may contain some error since we located the receiver by hand, we can say that the performance of the proposed system is reasonably accurate for the indoor localization comparing the results in [11], the maximum and mean values of location errors were 3.89 cm and 1.68 cm, respectively in simulation, whereas, in experiments, they had the error of 10.29 cm and 3.24 cm.

Fig. 10.Distribution of the localization error

For the navigation, we developed a program (Fig. 11, we call it host after this) that can communicate with the mobile robot (micro-mouse built with stepping motors) via network. The grid of the screen is set by 5 cm, and the 16 circles denote the projection of LEDs. The five red circles shown in the figure are the target points which the robot has to move through.

Fig. 11.Example of the indoor navigation

The receiver was mounted on the mobile robot to measure the location as shown in Fig. 9(right). The orientation of the robot was estimated in the host computer based on the changes of the robot location, since the proposed localization system can only measure the position of the robot. The initial orientation of the robot was calculated by moving the robot 5 cm backwards, which is a bit longer than the precision of the receiver, and then the robot was moved forward to the origin P1. Then, the host initiated the rotation command to correct the heading of the robot and the move command to step forward. In this manner, robot passed through the target points marked on the screen. The trajectories of the robot are marked with dark circles around straight lines connecting 5 target points.

The position error (Fig. 11: right, small dots stand for error at the target points P1-P5) did not exceed 9 cm, which is different from the measured error of the first experiment. The major reason cause of difference is the changes of the receiving angle of the light, which comes from the vibration of the robot on the uneven floor. However the robot completed its mission with the error less than 3 cm.

Currently, we are conducting a follow-up study on recognizing the direction and compensating the inclination of the receiver to make the proposed system more reliable.

 

5. Conclusion

We have developed an indoor localization device that embeds localization information into an indoor LED lighting system. The key idea of our scheme was to use “bit stuffing method.” Using stuff bits, our method was able to measure signal strengths independent of the previous signal. This allowed our implementation to use a simple function to measure the location of the receiver. The stuff bits also scattered the parts in the signal where the LED was turned on, enabling our scheme to provides quality indoor lighting as well, which was underconsidered in previous works. The proposed bit stuffing method would be useful for RSSI/TDM-based indoor localization applications. The effectiveness of the proposed scheme was validated through experiments with an actual low-cost implementation and navigation example, showing promising results.