On the development of a wireless self localizing streetlight monitoring system

C.M. De Dominicis, A. Flammini, E. Sisinni

Dept. of Information Engineering University of Brescia 25123 Brescia Italy emiliano.sisinni@ing.unibs.it
AbstractThe growing interest towards green and efficient use of electrical energy has recently pushed the industry of street light control systems. In the past, very simple on/off switching mechanism based on daylight sensing and cyclic preventive maintenance procedures were adopted. Nowadays, intelligent control systems offering remote supervision have strongly contributed to a change in perspective in maintenance engineering form a traditional Fail and Fix view to a Predict and Prevent approach. However, most of the infrastructure is already in place and replace street light with improved units has to be economical feasible. Traditional solutions exploit Power Line Communications (PLCs). In this paper we propose a wireless retrofitting of lamps, which has some advantages due to the independence from power line faults. In addition, the service provider usually ignores the exact location of the lamp poles and the integration with Geographic Information System (GIS) databases requires a preliminary on-site survey. The novelty of our proposal relies on the exploitation of nodes self location capability. Radios based on Chirp Spread Spectrum (CSS) modulation are used in order to achieve a good spatial resolution without the adoption of expensive GPS modules. In particular, experimental results are focused on the ranging capabilities of such devices, showing the feasibility of the proposed approach. Open field trials, mimicking real world applications, have shown an overall accuracy on the order of one meter. Keywords-street light control system, wireless sensor network, wireless ranging, chirp spread spectrum

L. Fasanotti, F. Floreani
CELS Dept. of Industrial Engineering University of Bergamo 24044 Dalmine (BG) Italy {luca.fasanotti;fabio.floreani}@intellimech.it additional transmission line, preserving the already available infrastructure. The supervisory software of the remote controller systems usually can graph power consumption, can switch the lamp status and can locate devices on a map by means of a Geographic Information System (GIS) database. The latter is a very useful capability, since it allows the maintainer to be informed where is located the faulty lamp. However, the streetlight position can be retrieved only by means of a preliminary on-site survey or installing a GPS unit for each lamp. Both the approaches are quite expensive, even if automatic position extraction from ortho-photos has been also suggested, as in [2]. Nevertheless, PLC has also some disadvantages. For instance, it is not possible to ensure the link if a short circuit in the line occurs, or if the line is definitively broken. Therefore, a wireless network could be a preferable solution. The interest for this kind of solutions is confirmed by the huge number of patents; as an example consider [3]. In the past, solutions based on Zigbee were proposed [4] and some proprietary commercial systems are available, as the LumenIQ from StreetLight or the SiCA platform from EPISenso [5]. The original contribution of this paper is the design of a location aware streetlight node based on a low cost wireless ranging device that can also be used for data communication. The ranging is performed by means of pseudoranges estimation (as it occurs for GPS), using the Chirp Spread Spectrum (CSS) modulation, a well known technology used in the past for radar applications. The paper is structured as follows. In the next section a brief description of the possible architecture for the realization of a remote wireless streetlight controller is proposed. In the third section, an overview of the CSS technology is sketched. In the Section IV the actual implementation is described, while in Section V some experimental results confirming the feasibility of the proposed approach are reported. Finally, some conclusions are drawn. II. ARCHITECTURE OF THE REMOTE STREETLIGHT
CONTROL SYSTEM

I.

INTRODUCTION

In many countries around the world lighting is provided along the streets by large numbers of lamp units. Clearly, street lighting is a huge electricity consumer; installations in Europe require about 60 TWh each year. Clearly, some strategies in order to increase the efficiency must be applied. It is not sufficient to turn off lights, since several researches showed that this approach leads to an increase in night-time accidents, with more people injured, and also in crime. For all these reasons an intelligent control and management systems based on dimmable luminaries, advanced control and communication systems and administrative tools must be implemented. The European project E-street estimated that only in Europe an annual saving potential of 38TWh is affordable [1]. Most of commercial available solutions exploit power line communications (PLCs) to implement a local network among streetlights. Such a solution does not require installing

The deliverables of the European project called E-street, Intelligent Road and Street lighting in Europe, ended in the 2008, are a good reference for the state of the art of remote streetlight control systems. The project defines a monitoring system for adaptive street lighting made up of five subsystems;

Authors acknowledge partial funding from MIUR PRIN Metodologie e tecniche di misura per la localizzazione spazio-temporale in reti di sensori wireless, N. 2008TK5B55_003,

978-1-4244-8064-7/11/$26.00 2011 IEEE

roadside equipment; power system; local control system; central supervisory control system and communication system/network [1]. A typical installation is depicted in Fig 1.
Central Supervisory Control System Internet Roadside equipment (W)WAN based Communication System Power Supply & Local Control System (PLC based)

the commercial systems implement the human-machine interface as an intuitive GIS system. An example of the interface developed in the E-street project is furnished in the Fig.3 [1].

Remote client

Figure 1. A sketch of the adaptive street light controller.

As stated in Section I, the apparently obvious solution based on the use of PLCs for the local communication system has some cons with respect to the wireless approach. The price of the device is higher and the network availability can be compromised by a line fault. For these reasons, a two-tier network is a preferable cost-effective solution (Fig.2): inexpensive short range radios, operating in a free band as the ISM, are used locally, at the street level; infrastructure-based wireless or wired wide area network interconnects the cabinet with the remote control room.
Central Supervisory Control System Internet (W)WAN based Communication System Power Supply System Wireless Local Network Roadside equipment

Figure 3. An example of GIS interface for streetlight monitoring (from Estreet project deliverable).

Remote client

Figure 2. A two-tier wireless adaptive street light controller.

In addition, it must be highlighted that typical deployment of adjacent streetlights mimics a line-of-sight (LOS) environment. In fact, streetlights are mounted above high poles, and their spacing guarantees that many lights are in range of each other. Hence, path redundancy is easily accomplished even with a reduced power link budget. Moreover, the streetlights must be powered, so this energy is also available for the wireless streetlight controller during normal operation. The controller can be also powered by a battery pack, recharged by means of a solar harvester, in order to counteract power line faults. The number of streetlights within a municipality may be very large; it is therefore crucial to have an asset management and optimization system. Such a system should help service providers to properly manage the maintenance of streetlights. Thanks to the availability of a communication system that allows collecting data coming from an instrumented streetlight, pro-active maintenance is possible. The asset database should have a web interface, in order to be accessed easily not only by the operators, but also by the public. In this way, final users could immediately notify a lamp fault. For this reason, most of

The georeferencing process is normally performed by hand; therefore it is a very time consuming step in the development of the streetlight control system. In principle, the process could be automated installing an additional GPS module for every lamp pole; however, this approach is economical unfeasible in most of applications and cannot be applied for indoor lighting systems. In many cases absolute geolocalization is not needed; often it is sufficient to locate the lamp position in a coordinate system relative to the supply cabinet, which is a natural anchor point. Such a consideration suggests providing each lamp with a local wireless ranging device and in case to equip only the cabinet with a GPS receiver. Simple and cheap Local Positioning Systems based on the received signal strength could be suitable, but are strongly affected by the environment and require the estimation of the path loss [6]. For this reason, the use of an alternative wireless ranging system is suggested and implemented in this work. A Chirp Spread Spectrum (CSS) system seems to be a good solution, since it allows measurement with less than 1 m error bound in open space, as confirmed by experimental results. In the next section, some details about this technology are furnished. III. THE CHIRP SPREAD SPECTRUM MODULATION AND ITS
RANGING CAPABILITY

CSS is a particular modulation technique derived from linear frequency modulated continuous wave (FMCW) radar systems, a.k.a. as pulse compression. Many of the features that make pulse compression suitable from a radar point of view translate directly in the development of a radio communication system. In particular, processing gain derived from the time spreading of the pulse allows for an interference robust modulation; also the effect of multipath is mitigated. A. The pulse compression method The CSS is based on the properties of a linear frequency modulated pulse. The baseband version of such a signal is described by (1), where TC is the chirp duration, is the chirp rate, d = [-1,1] defines the chirp direction (down, instantaneous frequency decreases, or up, instantaneous frequency increases). The chirp rate represents the rate of change of the instantaneous frequency; for a linear chirp is a constant. The

chirp bandwidth B is usually defined as the range of the instantaneous frequency, i.e. B = TC.
T S(t ) = exp jd t 2 , t < C 2 2

a) UpChirp generator u[k] (data bit) DownChirp generator Out

(1)

The detection of the incoming signal is performed by means of the matched filter, whose impulse response corresponds to the up-chirp signal for the down-chirp signal (and vice versa), i.e. H(t)=S*(t) . It is a well known result that the matched filter output is given by (2):
sin t(TC t ) 2 ,t <T . G(t) = S(t) H(t) = C t 2

b) I DQPSK mapper
Q

u[k] ( data bit )

Symbol mapper

Chirp Generator Out

(2)
Figure 4. Block diagram of CSS modulation schema: a) BOK, b) DM .

The filter output is clearly a sinc like function with the maximum value at t = 0 and a main lobe width of 2/B; the use of a (implicit) rectangular window ensures a sidelobe rejection of about 13 dB, that can be improved applying a different windowing function. The pass-band version s(t,i,Ti) of the chirp signal is centered around the frequency i and the time instant Ti, respectively, as shown in (3):
2 s(t, i , Ti ) = exp j i (t Ti ) + j d(t Ti ) W (t Ti ) 2 t W(t) = rect T C

(3)

B. The ranging procedure In addition, the response (4) is very useful in order to implement a ranging system, overcoming limits of traditional UWB systems. In particular, a tradeoff between the fine grain time localization of the incoming signals and the chirp bandwidth B exists, which enables the realization of devices working in the ISM band at 2.4GHz. Indeed, a ranging system based on the time-of-flight (TOF), or pseudorange evaluation, measures the propagation delay, i.e. the time elapsed between the sending and the reception of the RF signal when line-ofsight propagation occurs. The TOF is evidently equal to the originator/recipient separation distance divided by the propagation velocity. As a rule of thumb, this means that a spatial resolution of 0.3 m implies a time resolution of 1 ns. The propagation delay measurement needs strict clock synchronization between the sender and the receiver, which has a high cost and prevents the implementation in very simple device. It is possible to simplify the system measuring the round trip time (RTT), which requires the use of only the local clock, i.e. without a common time reference. The RTT technique measures the propagation time due to a first request packet (propagating from one node to a second one) followed by a reply packet (from the second node back to the original one). Generally, reply packets are automatically generated by means of a highly predictable hardware, only adding an offset that can be easily removed. Since both the paths are assumed symmetrical, the TOF is half the RTT. Nevertheless, the RTT technique does not remove the effect of the local clock drift, which obliges to use high stability oscillators. As shown in [7], this requirement can be mitigated with the Symmetric Double Sided Two-Way Ranging (SDS-TWR) proposed in [8]. It consists in repeating the request/reply packets exchange twice, inverting the role of the two devices in the second exchange, as shown in Fig.5. TREPLY is the time that elapses between the request packet reception and the reply packet sending (hardware imposed). It is possible to state that the TOF, assumed equal in all the packet exchange, is obtainable as:
TOF = 1 (RTTAB TREPLY,AB + RTTBA TREPLY,BA ) 4

In this case, the matched filter output still has a sinc-like shape but reaches the maximum value at t = Ti , as shown in (4):
sin (t Ti )(TC t Ti ) 2 exp[ j (t T )] g(t, i , Ti ) = i i (t Ti ) 2 t Ti < TC

(4)

When used for transmitting data, the CSS technique is classified into two broad categories: binary orthogonal keying (BOK) and the direct modulation (DM).

The classification criterion is the role of the baseband chirp in the modulation process In the former, the chirp direction also codes the data bit streaming; for instance, bits 1 and 0 can be represented by chirp signals with positive and negative instantaneous frequency change rates, respectively. On the other hand, in DM only the spreading effect of pulse compression is exploited and data bits are coded using different properties of the signal, e.g. modifying the chirp phase. A graphical representation of both the modulation schema is depicted in Fig.4.

(5)

Supposing that lock clocks of originator and recipient suffer from drift eA and eB it is possible to estimate the TOF as:
TOF = 1 [(RTTAB TREPLY,AB )(1 + e A )(RTTBA TREPLY,BA )(1 + e B )] (6) 4

Assuming that TREPLY,AB= TREPLY,BA+ and TOF<< then the <TOF> can be approximated as in (7):
1 < TOF >= TOF + (e A e B ) 4

high directivity antennas has suggested the idea of DisMAC, where a sort of bucket brigade between adjacent nodes is established [11]. As stated in [12], a localized relative coordinate system can be obtained mixing ranging information with angle of arrival and/or digital compass estimations, enabling some position centric applications as flooding, discovery or source-sink communication. These capabilities can be easily added to the wireless node, but the development of the routing protocol is outside the scope of this work. For this reason a simplified routing strategy has been implemented. The ranging measurement normalized by the received signal strength is used to identify neighbours at the MAC Layer in the preliminary discovery phase. When this phase ends, all the nodes send their neighbour table to the coordinator, which reconstructs the actual nodes arrangement and assigns Network Layer addresses. In the operating phase, each node forwards packets coming from one- or two-hops nodes to increase reliability. At the top of the stack, the application layer defines two different messages according to the data forwarding direction; MSGCDATA, from the coordinator to the streetlights and MSGLDATA, from the streetlights to the coordinator. Both the data packets are acknowledged using a MSGCACK and MSGLACK packet; respectively. A pictorial description of the data flow is shown in Fig.6. For sake of simplicity, all the packets have the same fixed length; the description of data fields within the packet is outside the scope of this work
MSGCACK MSGCACK MSGLDATA MSGLDATA

(7)

This result implies that a spatial resolution on the order of 1 m, that requires a 3 ns time resolution, can be obtained with a low-quality crystal (e.g. eA,eB 100ppm) and TREPLY on the order of several microseconds.
Originator A Recipient B

TOF RTTAB TREPLY,AB

TREPLY,BA

RTTBA

Figure 5. Illustration of the SDS- TWR mechanism.

IV.

THE IMPLEMENTATION OF NETWORK NODES

Coordinator

The streetlight scenario considered in this work poses some ties on the network topology of the wireless network. The streetlights are roughly deployed along a straight line, thus also the network has a linear topology. The communication is bidirectional and occurs from the cabinet to the light pole (control mode) and viceversa (diagnostic mode). The node in the cabinet is also the network coordinator and acts as the gateway toward the WAN tier-two network. A simple proprietary protocol stack has been developed to experimentally test the performance of the system. The physical layer is clearly based on the CSS radio. The Medium Access Control (MAC) is based on contention and implements Carrier Sensing Multiple Access with Collision Avoidance (CSMA/CA). It must be remembered that the overall throughput required by the application is quite low and does not pose sever constraints. In addition, in this peculiar application nodes are normally mains powered and consumption due to retransmissions is not critical. At this level, each node is univocal identified by its MAC address, which is also its serial number. The role of the network protocol is to translate MAC addresses into network addresses and to forward the messages from the source to the destination hop by hop. In the considered scenario, it is clear that a privileged dimension exists and it is natural to take advantage of it; benefits of a position centric routing approach for street lighting systems are already underlined in the literature [9]. Basically, the idea is to use the distance knowledge in order to satisfy a minimum transmission energy cost function; performance bound of this approach are shown in [10], where the MERR protocol is described. In addition, the availability of

Node 1
MSGCDATA MSGLACK

Node N

Figure 6. Data packets exchange across the network.

The actual network nodes have been designed around the NanoLOC (NA5TR1) device from Nanotron. The core of the system is a monolithic fully digital implementation of a CSS transceiver. This transceiver adopts the BOK modulation and allows for a gross data rate in the range from 2 Mbps to 125 kbps. The transceiver is controlled by an ATmega 128L microcontroller from Atmel. The nominal transfer rate of 1 Mbps has been chosen; the baseband chirp has a 1 s duration and nominally occupies 80 MHz, i.e. the whole ISM band @ 2.4 GHz. The coarse time resolution in the detection of the RTT is determined by the sample rate of the ADC in the receiving path; it is about 4 ns, i.e. 1.2 m. Sub meter spatial resolution is obtained thanks to dithering & averaging technique. Random and evenly distributed time shifts are added to the chirp position (e.g. by the noise or intentionally); then several consecutive chirp pulses are averaged at the receiver side. In this way the overall resolution is improved; such an operation is performed in hardware and it is transparent for the final user. V. EXPERIMENTAL RESULTS An extensive measurement campaign has been carried out in order to estimate ranging performance of the proposed

Relative Frequency [%]

wireless node. Typical application scenario resembles free space propagation; for this reason experiments have been performed outdoor in static condition. The network setup has been keep as simple as possible; only the coordinator and one node are present. Ranging data are retrieved from the coordinator by a host system using a serial link, as shown in Fig.7. Each measurement point requires about 6 ms to be evaluated; since node position is fixed, performance can be improved using a simple moving average filter.
Directional Antennas RS232 NanoLOC C Host PC Coordinator Fluke 411D NanoLOC C Node

Bias (offset + error): 0.91 m

12 10 8 6 4 2 0 9

9.5

Estimated distance [m]

10

10.5

11

Standard deviation: 0.18m

Figure 9. Distribution of the experimental measurements with omnidirectional antennas in outdoor free-space environment; the reference position is 9 m.

Omnidirectional Antennas

Figure 7. Block diagram of the test bed.

Estimated distance [m]

Statistical analysis in the following refers to 3000 consecutive single point distance estimation. The reference distance is evaluated by means of a laser Rangefinder 411D from Fluke, which has a measurement range of 30 m and an overall accuracy on the order of 0.25 mm/m. A first trial was carried out using omnidirectional antennas (TP-LINK TLANT2405CL; nominal gain 5dBi; horizontal beam width: 360; vertical beamwidth: 32). Node distance varies in the range [1,20] m in 1 m step, as shown in Fig.8.
Estimated distance [m]
25 20 15 10 5 0 0 5 10 15 20

Subsequently, a new trial in the same LOS condition but using directional antennas has been performed (TPLINK TLANT2414A; nominal gain 14dBi; horizontal beam width: 30; vertical beamwidth: 30). Node distance varies in the range [1,10] m, as shown in Fig.10. The linearity error is 0.04 m. As in the previous case, the linear fitting shows a non zero intercept due to the analog front end and to the cable (in this configuration about 3.70 m). The error bars show the 99%confidence interval; in this condition the measurements follow a Gaussian distribution with a standard deviation of about 0.18 m, as shown in Fig.11 for 6.9 m - reference position.
15

10

0 0 2 4 6 8 10

Reference distance [m]

Reference distance [m]

Figure 8. ranging performance with free space propagation and omnidirectional antennas; error bars show 99%-confidence interval.

Figure 10. ranging performance with free-space prooagation and directional antennas; error bars show 99%-confidence interval.
Bias (offset + error): 3.88 m

14 Relative Frequency [%] 12 10 8 6 4 2 0 9 9.5 10 10.5 11 Estimated distance [m] 11.5 12

The linearity error is 0.17 m. The linear fitting shows a non zero intercept; this offset term (about 0.70 m) is due to the analog front end and to the cable; an in-factory calibration procedure is needed in order to compensate it. The error bars show the 99%-confidence interval; in this condition measurements follow a Gaussian distribution with a standard deviation of about 0.18 m. As an example, consider Fig.9 that refers to measurements carried out with a reference distance of 9 m. As explained before, the measurement bias includes two contributions; the measurement inaccuracy and an offset due to the delay introduced by the analog front end.

Reference: 6.935m

Standard deviation: 0.18m

Figure 11. Distribution of the experimental measurements with directional antennas in outdoor free-space environment; the reference position is 6.935 m.

It is a well known result that multipath affects RTT measurements; the CSS modulation mitigates its effects but cannot avoid it. For this reason a directional antenna, which focuses most of the radiated energy into a particular direction possibly free of obstacles, allows for better performance. However, the installation of directional antennas could be a mechanically cumbersome task, vanishing advantages for the considered application. In order to improve the reliability, an additional test based on spatial or antenna diversity has been performed. The environment has been changed in order to highlight the effect of multipath. The test has been conducted outdoor in a construction site. In this case, two omnidirectional antennas few centimetres apart (i.e. one wavelength referred to centre channel frequency) are used. Such an arrangement mitigates the multipath since each antenna experiences a different interference, i.e. if one antenna is subjected to destructive interference due to multipath, the other is likely to receive a good signal. Rather than simultaneously analyze the RF signals coming from both the antennas, two different but co-located nodes have been sequentially interrogated for the RTT estimation, as shown in Fig.12. Results are depicted in Fig.13; the error bars show 99%-confidence interval.
RS232 NanoLOC C Host PC Coordinator Fluke 411D Node A NanoLOC NanoLOC C C Node B

possible to lower the linearity error from 0.58 m (when a single antenna is used) down to 0.23 m. VI. CONCLUSIONS In this work a wireless network for streetlight remote control is discussed. In particular, the novelty of the proposal is in the location awareness of nodes, which can self localize themselves. Simple prototypes have been built using commercial hardware. The capability of the ranging measurements, the basis for localization, is deeply characterized, showing an overall accuracy on the order of one meter. In near future, location aware routing algorithms will developed that will improve the efficiency of the network. ACKNOWLEDGMENT Authors would like to acknowledge Intellimech consortium for help in the realization of the experimental setup. REFERENCES
E-street project deliverables available online: http://www.estreetlight.com/ [2] M Abdalla: 3GR for Road Safety Integration of GIS, GPS, GSM and Remote Sensing for Road Safety, in Proc. Of the 8th International conference, Application of Advanced Technologies in Transportation Engineering, ASCE: Pp 255-260, 2004 [3] Jr. Budike, E.S. Lothar (Powerweb Technologies), Wireless internet lighting control system, US patent 7,167,777, Jan 23, 2007. [4] J.D. Lee, K.Y. Nam, S.H. Jeong, S.B. Choi, H.S. Ryoo, D.K. Kim, Development of Zigbee based Street Light Control System, in Proc. Of IEEE Power Systems Conference and Exposition 2006, Pp.22362240, Oct. 29 2006-Nov. 1 2006. [5] Companies web site: www.episensor.com; www.streetlightiq.com, [6] K. Tateishi, T. Ikegami, Estimation Method of Attenuation Constant During Localization in RSSI, in Proc. Of ISCIT 2008, Pp.482-487, 2123 Oct. 2008. [7] IEEE Standard 802.15.4a-2007 (Amendment to IEEE Std 802.15.42006). [8] R. Hach, Symmetric Double Sided Two-Way Ranging, IEEE P802.15 Working Group for Wireless Personal Area Networks (WPAN), Doc. IEEE P.802.15-05-0334-00-004a, 2005 [9] R. Palucci Pantoni, D. Brando, Analysis and Proposal of a Geocast Routing Algorithm Intended For Street Lighting System Based On Wireless Sensor Networks, International Journal of Computer and Network Security, Vol. 2, No. 4, April 2010, Pp.:65-70 [10] M. Zimmerling, W. Dargie, J.M. Reason, Energy-Efficient Routing in Linear Wireless Sensor Networks, in Proc. Of IEEE MASS 2007, Pp.13, 8-11 Oct. 2007. [11] Karveli, T. Voulgaris, K. Ghavami, M. Aghvami, A.H., DiS-MAC: A MAC protocol for sensor networks used for roadside and highway monitoring, in Proc. Of ICUMT09, 12-14 Oct. 2009, Pp. 1 6, St. Petersburg [12] Niculescu, D., Nath, B., Localized positioning in ad hoc networks, Proceedings of the First IEEE International Workshop on Sensor Network Protocols and Applications, 11 May 2003, Pp. 42 - 50 [1]

Figure 12. Block diagram of the setup for spatial diversity test.

Estimated distance [m]

10 8 6 4 2 0 0 2 4 6 8 10

Reference distance [m]

Figure 13. ranging performance with spatial diversity when omnidirectional antennas are used in outdoor construction site environment; error bars show 99%-confidence interval.

In this case, for each measurement point performed by the two devices, 2000 consecutive estimation have been carried out and the actual value has been chosen in order to minimize the standard deviation of the population. In this way it has been