Analytic Comparison of Wake-up Receivers for WSNs and Benefits over the Wake-on Radio Scheme

Analytic Comparison of Wake-up Receivers for WSNs and Benefits over the Wake-on Radio Scheme Vana Jelicic, Michele Magno #, Davide Brunelli, Vedran Bilas and Luca Benini # Faculty of Electrical Engineering and Computing, University of Zagreb, Croatia # DEIS, University of Bologna, Italy University of Trento, Povo, Italy {vana.jelicic vedran.bilas}@fer.hr; {michele.magno luca.benini}@unibo.it; davide.brunelli@disi.unitn.it ABSTRACT Since in most wireless sensor network (WSN) scenarios nodes must operate autonomously for months or years, power management of the radio (usually consuming the largest amount of node s energy) is crucial. In particular, reducing the power consumption during listening plays a fundamental role in the whole energy balance of a sensor node, since shutting down the receiver when no messages are expected can remarkably increase the autonomy. Idle listening is a hard challenge because incoming messages are often unpredictable and developers have to trade off low power consumption and high quality of service. This paper is focusing on benefits of introducing a wake-up receiver over simple duty-cycling (wake-on radio). We analyze and compare the existing wake-up receiver prototypes and explore their benefits using simulations of two typical scenarios: with and without addressing requirements. A particular approach outperforms other solutions in terms of lifetime extension because of its very low power consumption (1µW). We also evaluate the overhead of the addressing capability, which sometimes has a non-negligible impact on the performance. Categories and Subject Descriptors H.4 [Information Systems Applications]: Miscellaneous General Terms Performance Keywords Wireless sensor networks, duty cycle, wake-on radio, wakeup receiver. 1. INTRODUCTION Wireless sensor networks (WSNs) have been a research focus in various engineering disciplines for more than a decade. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. PM2HW2N 12, October 21 22, 2012, Paphos, Cyprus. Copyright 2012 ACM 978-1-4503-1626-2/12/10...$15.00. Strict energy constraints of battery-powered wireless sensor nodes have introduced the necessity of energy awareness in both software and hardware solutions. Energy harvesters capturing energy from environmental sources such as solar, wind or thermal can strongly extend the lifetime of batteries. On the other hand, the reduction of power consumption through power management is very important to further extend the lifetime or to reduce the form factor and cost of harvesters. Moreover, power management helps the node not to waste energy doing useless operations. Since a transceiver consumes generally the largest amount of a node s energy, reducing its activity when it s not necessary brings higher power savings. In fact, the radio usually consumes about 20 ma in transmission (e.g. for CC2420 RX mode consumes 18.8 ma and TX mode at 0 dbm consumes 17.4 ma). Most of the time the transceiver is in idle state, listening to the channel if there is an incoming message. However, unfortunately, a message can be received only if the radio is in RX state. Since the node receives messages relatively rarely, lots of effort have been done to reduce the useless idle listening of the transceiver, usually by periodically switching the radio on and off as a part of MAC protocol [1]. In addition, a hot topic has been how to enable waking the node (and the entire network) on a message reception, by implementing a hardware solution in a form of a low-power continuously active wake-up receiver that wakes up the node and the main transceiver upon message detection [2]. Main contributions of this paper are the following: - we evaluate, assess and compare the majority of wakeup implementations available in literature; - we simulate the wake-up receiver performances using two scenarios, with and without addressing, and compare it against the wake-on radio scheme (i.e. simple transmitter duty-cycling). The following section presents radio topologies in transceiver power management (cycled receivers and separate wake-up receivers). Section 3 puts this work into the perspective of the related work on wake-up receivers. In Section 4 we group the existing prototypes regarding the circuit implementation and compare the performance of each group representative. Section 5 presents the conclusion and future work ideas. 2. BGROUND This section refers to different communication scenarios and radio topologies. When two nodes are to communicate, the receiver node must be awake when the sender initiates the communication, which is referred to as a rendez-vous [2]. There are three types of rendez-vous schemes: 99

a) pure synchronous: The nodes clocks are presynchronized so the wake-up time of each node is known in advance. This scheme requires recurrent time synchronization that consumes considerable energy. Moreover, the sensors wake up even if there is no packet to transmit or receive, causing idle listening or overhearing. b) pseudo-asynchronous (or cycled receiver): Source nodes wake up and emit a preamble signal that indicates the intention of data transmission. The preamble time has to be set long enough to coincide with the wake-up schedule of the destination node (i.e. longer than its sleep time). In this scheme time synchronization is not required, but sensors follow a duty cycle and consume considerable energy with preamble signaling. c) pure asynchronous: Sensor nodes reside in deep sleep and can be woken up by their neighbors on demand with very low-power wake-up receivers. Whenever a node intends to send a packet, first it wakes up the destination node and then sends the packet. Therefore, wake-up receivers are a solution to the redundant energy consumption caused by rendez-vous. Although there are WSN MAC protocols that employ synchronization (e.g. SMAC and TMAC), we will study the asynchronous duty-cycled MAC protocols that remove the synchronization energy overhead and are easier to implement as they do not require clock synchronization (e.g. B- MAC, X-MAC). IEEE 802.15.4/ZigBee supports both synchronous (beaconed) and asynchronous (non-beaconed) mode [3]. Low-power radio schemes without synchronization can be roughly divided into two categories: the cycled receiver scheme and the separate wake-up receiver scheme. 2.1 Cycled receiver The cycled receiver scheme employs duty cycle control on the main radio to decrease the power consumption, while suffering a penalty on the latency performance. The cycled receiver schemes can be in general classified into two groups, depending upon who (transmitter or receiver) initiates the rendez-vous. These schemes are called Transmitter/Receiver Initiated CyclEd Receivers, respectively, or TICER and RICER and are shown in Fig. 1(a) and 1(b) [4]. In TICER scheme, as soon as a node has a data packet to transmit, it wakes up and monitors the channel. If it does not hear any ongoing transmissions on the channel, it starts transmitting request-to-send (RTS) signals to the destination node, and monitors the channel for responses. The destination node, upon waking up according to its regular wake-up schedule, immediately acquires the RTS s, upon which it responds with a clear-to-send (CTS) signal. After reception of the CTS signal, the source node transmits the data packet. The frequency of transmitting RTS s should be as low as possible to save transmit power. However, the period cannot be longer than active time of destination node, otherwise the destination node might miss the rendez-vous. Similar to the TICER scheme, in RICER a sensor node with no data packet to transmit wakes up with period T. It then transmits a short wake-up beacon to announce that it is awake, and monitors the channel for a response. If there is no response, the node goes back to sleep. A source node with data to transmit stays awake awaiting a wake-up beacon. Upon reception, it starts transmitting the data packet. In TICER, the source node has to retransmit until it receives the CTS message. It can cause a larger power consumption for the source node compared to the RICER source Src Src Src Src TX RX WURx ON WURx RTS T (a) TICER scheme T Wake-up beacon Low RSSI Main radio T T_wait (b) RICER scheme Data T (c) WOR scheme Wake-up signal (d) WURx scheme CTS Data Data Data Figure 1: TICER, RICER, WOR and WURx illustration of communication between two nodes node that transmits only after receiving the wake-up beacon (and waits in idle state until it gets it). In RICER, the destination node has to send each time a wake-up beacon, consuming more than the TICER destination node. TICER and RICER have comparable overall performance (taking into account power consumptions of both the source and the destination node) and it depends on period T. RICER performs better than TICER under strong fading conditions (being a three way handshake protocol as opposite to the four way handshake in TICER) [4]. An implementation of a simple transceiver duty-cycling (TICER) is presented in [5]. The wake-on sensor network uses the Wake-On Radio (WOR) capability that enables the radio (TI CC1101) to periodically wake up from sleep mode and listen for incoming packets without MCU interaction. In [6], WOR is used to explore advantages of duty-cycling radio s activity when a network of PIR nodes detects an event and notifies the camera, instead of having a PIR sensor on the camera board. Currently commercially available transceivers with WOR possibility are CC1000, CC1101, CC1100E, CC2500, CC430 from TI, TRX2 from Quasar UK, and ATmega128RF from Atmel. The WOR functionality may also be used in combination with CC1100/CC2500 RSSI function. This function will perform an initial RSSI level measurement when entering RX mode, and if it does not exceed a programmable threshold, the RX will terminate immediately and return to SLEEP (remaining in WOR mode). This function can reduce the time in RX and lower the power consumption if no signal is present. This scheme is shown in Fig. 1(c), where the source node immediately 100

sends data (useful for short data packets), without sending the RTS first. 2.2 Separate wake-up receiver By adding a separate wake-up receiver (WURx) to monitor the communication channel continuously, the main radio is kept in the sleep mode most of the time. When a node wants to communicate, it sends a wake-up signal, usually containing the address of the destination node to awake only the desired neighbor (Fig. 1(d)). For this mode of operation to be effective, the power consumption of the wake-up device must be quite low. This paper explores the benefits of the separate wake-up receiver scheme over the cycled receiver (in particular the WOR). In continuation we present the related work confronting these two communication schemes, as well as the main features comparison of different wake-up receiver prototypes found in literature. 3. RELATED WORK The contribution of this paper is the evaluation of different wake-up receiver prototypes in terms of lifetime prolongation compared to the WOR scheme. Publications with similar subjects present only a qualitative study [2], or refer to the wake-up receiver as a general concept [4], using one of the prototypes as an example, without comparisons with others or justifying the choice [7]. Furthermore, in [8], the power budget of a wake-up receiver is explored in order for the WURx MAC to outperform the X-MAC and Static TDMA protocols, with Nordic nrf24l01 as the main radio. In [9], a duty-cycled WURx (very poorly characterized prototype) is presented and compared with always on WURx, showing that they outperform X-MAC in low-traffic scenarios. We, on the other hand, study all wake-up receivers published to date, compare their characteristics more profoundly (qualitative and quantitative) and explore their benefits over the WOR in terms of energy savings. Gu and Stankovic [10] (2004) were first to present the design goals for a WURx: low power consumption, high sensitivity, resistance to interference and fast wake-up and also proposed the idea of passive zero-powered wake-up receivers that harvest energy from the received EM signal. Lin et al. [4] give a more realistic consumption estimation if the WURx power consumption is greater than 50 µw, the overall performance of the purely asynchronous protocol will be worse than that of pseudo-synchronous schemes. In [7], an analytical model has been proposed to compare the energy consumption of the separate WURx scheme and the cycled receiver in a typical application scenario. The cycled receiver scheme introduces latency in data communication (due to duty cycling and necessity of retransmitting in case the message doesn t reach the destination node while awake). In applications that don t require low latency, the main radio can apply very low duty cycle and thus reduce the power consumption to even lower value than the one of WURx. Assigning the same energy consumption to both schemes, they analyze and simulate a situation using Nordic nrf24l01 as the main radio and a 50 µw WURx (proposed by Pletcher [11]) with probabilities of missing a wake-up and false alarm both being 0.1. Results show that the separate WURx scheme outperforms the cycled receiver for the systems with maximal allowed wake-up latency up to 700 ms for low packet arrival rate (10 2 packets per second). Our work is focused on energy savings of the WURxs compared Figure 2: WURx scheme [17] 1 1 0 1 to the WOR scheme (connected to the duty cycle of the main radio in WOR). On the other hand, latency is bound to the WOR wake-up period and strongly depends on the application requirements. We will tackle this issue in the further stages of our work. Besides the advantages of the separate WURx scheme virtually eliminates idle listening on the main radio (presuming that only the desired node wakes up), reduces latency (as receivers are woken up when they are needed) and reduces collisions (as transmissions are not scheduled into discrete communication periods) there are several crucial challenges in design and implementation of WURxs (limited reception range, false wake-ups caused by interference from other sources etc.). Thus, trade-offs expected in WSNs with WURxs are wake-up range vs. energy consumption, wakeup range vs. delay and in-band vs. out-of-band WURx [2]. 3.1 Wake-up receiver prototypes The first WURx prototype was done by van der Doorn et al. [12] in 2007, but had a large power consumption (819 µw). Ansari et al. [13] design an external low-cost hardware wakeup circuit consuming 876 na and attach it to the microcontroller of a sensor node. In recent years a significant progress has been recorded. Table 1 lists the published prototypes (since 2008), comparing the most important parameters. The communication needs of the wake-up scheme are radically different from the usual ones. Instead of data rate and spectral efficiency, the primary goal is the power efficiency at the node and the network level. First two solutions ([14, 15]) are only simulation based. Others are mostly fabricated in CMOS technology or built out of the off-the-shelf components. In addition, only the solution from van Langevelde [16] is a wake-up transceiver (we ll refer to that design as WUR), and the others are only receivers (WURx). 3.1.1 Address decoding (AD) Fig. 2 depicts a general WURx scheme: filtering the signal arrived at the antenna circuitry, envelope detection, sample and hold, digitalization and thresholding, and in some cases the digital baseband circuitry to detect the address from the wake-up signal. Many papers present only the analog frontend of the wake-up device giving little or no information on the rest of the WURx (most important how the decoding of the node s address is performed and how much it consumes). In [17], Zhang et al. present a 3.72 µw ultra-low power digital baseband for WURxs, that detects the address of the node from the wake-up message and wakes the node up only if the message is dedicated to it. It introduces only 20 µs delay. In [15], [18] [20], WURxs have a dedicated HW included to decode the address. Gamm et al. [21] embed 101

Table 1: Wake-up receiver prototypes Authors Year f [GHz] Rate [kbps] S [dbm] d [m] P [µw] AD l [ms] Implementation Le-Huy [15] 2008 2.4 50-50 NA 20 Y NA simulation Yu [14] 2008 2.4 100-75 NA 53 N NA simulation Langevelde [16] 2009 0.868 45-89 NA 2.4 N 1.36 130 nm Pletcher [11] 2009 2 100-72 NA 52 N NA 90 nm Durante [23] 2009 2.4 100-53 NA 12.5 Y, FPGA NA 120 nm Gamm [21] 2010 0.868 NA -52 40 2.78 Y 13 120 nm Drago [32] 2010 2.4 250/500-87/-82 NA 415 N NA 65 nm Fraunhofer [20] 2010 0.868 1-60 30 33 Y 32 180 nm Huang [27] 2010 2.4/0.915 100-64/-75 NA 51 N NA 90 nm Huang [31] 2011 0.915 10-86 NA 123 N NA 90 nm Marinkovic [24] 2011 0.433 5.5-51 10 0.270 N, (MCU) 9 off-the-shelf Shih [18] 2011 0.9165 0.370-122 1000 1153 Y NA off-the-shelf Hambeck [19] 2011 0.868 100-71 304 2.4 Y 40 110 130 nm into their design a low-power, low-frequency WURx with an integrated correlator which compares the received signal to a byte pattern saved in a configuration register [22]. Durante et al. [23] use an FPGA to decode the address. Other solutions only detect the wake-up signal based on its energy and wake up the MCU. If the wake-up signal transmits the address to the MCU (e.g. via SPI [24]), the MCU can decide whether or not to wake-up the main transceiver. Otherwise the address, if necessary, has to be transmitted in a message via main radio. 3.1.2 Sensitivity (S) and range (d) Sensitivity is a very important parameter of a WURx. The Friis transmission equation describes the power received by one antenna under idealized conditions given another antenna some distance away transmitting a known amount of power: P r P t = G tg r ( ) 2 λ 1 4π d, (1) n where P t is the transmitted power, G t and G r the antenna gain on the transmitter and receiver side respectively, d the transmission distance, λ the wavelength of the frequency used, and n the path loss exponent. For a medium-density WSN, a wake-up range around 4 5 m would be quite acceptable. With that in mind, the received power at 4 m, assuming operation in the 2.4 GHz ISM band, a radiated power P t of 0 dbm and pseudo-omnidirectional antennas at both ends, is -49 and -55 dbm, for a path loss exponent of 2 and 3 respectively. At 5 m, it would fall to -51 and -58 dbm. Higher antenna gains, both at the receiver and the transmitter, could significantly increase the received power, improving the effective sensitivity, or the range, of the WURx [15]. The reported ranges from Table 1 vary from 10 m [24] to 1 km [18]. 3.1.3 Resistance to interference Another important characteristic of a WURx is resistance to interference. The main idea with the WURx is to avoid waking up the node by mistake. There are two possible sources of wake-up errors: 1) nodes waking up because of a wake-up signal intended for another node, and 2) nodes decoding their address code from the noise or interference. Another problem is missing a wake-up signal. An example of tested values is presented in [25] (99% detection probability and a false wake-up rate of 10 3 /s). Shih et al. [18] report low packet error rate of 0.0159 at SNR 4 db. 3.1.4 Latency (l) The time necessary for the WURx to receive and decode the wake-up signal has been very scarcely addressed in the papers. As seen from Table 1, there are data for only few prototypes (from cca 1 ms to 110 ms). 3.1.5 Power consumption (P) Power consumption of a WURx depends mostly on the sensitivity and data rate. The lowest power consumption has the Marinkovic et al. WURx (only 270 nw) [24]. The largest power consumption report Shih et al. (1.1 mw) because of the high sensitivity [18]. 3.2 Applications with WURx There are no examples in the literature of real-world applications with WURx. In [26], an implementation of an ultralow power event-driven radio is proposed to minimize the power consumption of a building automation system. Eventdriven receiver (consisting of the WURx from [27] and the low power transmitter from [28]) is compared against other commercial low power radios (nrf24l01 and TI CC2420) and the possibility of implementing an autonomous radio (with power scavengers) is investigated. Marinkovic et al. [29] propose WURx implementation to synchronize the TDMA communication protocol in a single-hop star WBAN. Comparing to the very low power TDMA protocol, for longer measurement intervals, using the WURx ensures approximately 14 times lower communication power consumption. 4. WURX COMPARISON AND BENEFITS We will extract common features of existing WURx prototypes (Table 1) and compare their performance, focusing on the power consumption. 4.1 Wake-up schemes Based on the wake-up circuitry design, there are following two possibilities of the node wake-up: 4.1.1 WURx receiving the wake-up signal and main transceiver communicating the data In a common simple implementation, the source node sends the wake-up signal and the destination node detects it 102

Table 2: Prototype representatives of different WURx circuit implementations WURx circuit implementation Prototype representative analog front-end + MCU Marinkovic WURx with address decoding by MCU [24] analog front-end + FPGA Durante s WURx with address decoding by FPGA [23] analog front-end + digital baseband SoC Huang WURx [31] with address decoding by ULP DBB [17] from Zhang et al. [17] van Langevelde WUR [16] with address decoding by ULP DBB [17] analog front-end + other dedicated HW Hambeck WURx, with address encoding included in dedicated HW [25] with the WURx that activates the main transceiver for data transmission. The main transceiver (e.g. IEEE 802.15.4/Zig- Bee transceiver) generates the wake-up signal usually with the OOK modulation. The simplest wake-up circuits wake up the node upon detecting the signal, i.e. all the nodes within the transmission range wake up. More sophisticated solutions decode the address from the wake-up signal and wake the node up if the addresses match. There are several ways to engage that kind of WURx into a WSN, depending on the implementation of the wake-up receiver circuit i.e. how does it detect and decode a wake-up signal (first column of Table 2). The representatives of each group are in the second column of Table 2. The first design wakes up the microcontroller each time the analog front-end detects a wake-up signal. The microcontroller decodes the address and goes back to sleep if it isn t the dedicated target. In [24], they propose a solution with transmitting the packet via SPI to the microcontroller in order to dechipher the address. In addition, since that WURx works on 433 MHz, they use a 433 MHz radio as main, in order to avoid adding another transmitter for the wake-up signal. Other solutions embed address decoding into the WURx circuitry trading the higher complexity (thus also higher quiescent power consumption) for lower wake-up frequency of the rest of the node (in ideal case only when the address corresponds to the node). 4.1.2 WUR both receiving and transmitting the wakeup signal In order for the node to be able to wake up other nodes without waking the main radio, the wake-up circuitry has to be able also to transmit. In case that data transmitted between nodes is very brief (a couple of bits) and can be embedded within the wake-up signal, the node can contain only the wake-up radio, without the main transmitter. The implementation of van Langevelde [16] achieves -89 dbm receiver sensitivity and -6 dbm transmitter output power while consuming 1.6 ma and 1.8 ma, respectively, from a 1.2 to 1.5 V supply. A similar idea is presented in [26], where the transmitter with a power amplifier consumes about 333 µw, having an output power of -10 dbm and efficiency of 30% [28]. The proposed 2.4 GHz direct modulation transmitter radiates 1 mw with 3.88 mw power consumption, and it supports OOK and ASK modulation up to 10 Mbps. In February 2012 Imec and Holst Centre announced a 2.3/2.4 GHz transmitter for wireless sensor applications compliant with 4 wireless standards (IEEE802.15.6/4/4g and Bluetooth Low Energy). The transmitter has been fabricated in a 90 nm CMOS process, and consumes only 5.4 mw from a 1.2 V supply (2.7 nj/bit) at 0 dbm output. This is 3 to 5 times more power-efficient than the current state-of-the-art Bluetooth- LE solutions. The multi-standard transceiver is highly reconfigurable and supports the required modulations and data WURx Microcontroller Main radio Sensor unit Figure 3: A typical wireless sensor node with a WURx attached rates from 50 k 2 Mbps [30]. Thus, one can expect further development of ultra-low-power transceivers. In this work we will simulate the perferomance of van Langevelde implementation [16], taking into account only its receiver part. 4.2 Simulation We will test a general scenario of attaching a WURx to a WSN node (Fig. 3), with approximate power consumption while active to be 100 mw (20 ma main radio, 10 ma sensor unit, 3 ma microcontroller, with 3 V supply). These values are realistic for a node consisting of a CC2420 radio, a MSP430 microcontroller and a LSM303DLM 3-axis accelerometer and 3-axis magnetometer. The WURx can be any of the five implementations from Table 2. Power consumption of a node comprising a WURx can be expressed as: P node = P wup + P run, (2) where P wup denotes power consumption of detecting a wakeup signal (static WURx power consumption and dynamic WURx power consumption including address decoding), and P run denotes power consumption of the running node after waking up (usually consisting of microcontroller, sensors and main transceiver activity). It is important to notice that the wake-up power depends only on the number of events N ev (number of wake-up signals that arrive), while power of the running node depends also on the number of nodes N nodes in the network: P wup = f (N ev), ( P run = f N ev, 1 N nodes ). More precisely, all nodes in the network (in the communication range) detect the wake-up signal, but it is dedicated only to one of them. The more nodes there are in the network, the lower is the possibility of activating a certain node. The running node s power consumption (P run) is the same for each node, regardless of the WURx type (the node is activated when it is addressed). The wake-up power consumption (P wup) differs depending on the implementation. The solutions with dedicated HW (DBB SoC and others) (3) 103

N MCU LMP4.5 Main radio sleep WURx idle 1 uw WAKE-UP SIGNAL MCU LPM3 Main radio sleep WURx RX 11 uw 9 ms This node addressed? ACTIVATE NODE Y Wake-up signal PIR node Camera node Figure 4: Waking up with Marinkovic et al. WURx [24]; node address within wake-up signal have a constant power consumption (we can disregard the very short increments during signal reception). The solutions that activate an FPGA or a microcontroller spend a significant time and power (t decod and P decod respectively) to decode the wake-up signal, consuming energy which depends on the N ev: E wup,dyn = N ev P decod t decod. (4) WURx design implementing an FPGA [23] provides only a worst case power consumption value, thus we can calculate only the average power consumption for the solution implementing a microcontroller. In [24], a solution implementing the WURx with the MSP430 microcontroller is presented. It is a solution with 5.5 kbps and cca 10 m range. This solution requires a microcontroller to decode the address from the wake-up signal. MSP430 is one of the lowest power microcontrollers and it can decode the address without entering the active state. That is the reason we consider a general node comprising the MSP430 (Fig. 3). The WURx detects the wake-up signal, generates the interrupt which then wakes up the MSP430 from power down mode (LPM4.5) to Low Power Mode 3 (LPM3), to read the demodulated wake-up packet as a digital stream on the SPI. Only if the packet is intended for it, it will go into active mode, switching on the main receiver. Fig. 4 shows the power state diagram of detecting the wake-up signal with MSP430. We analyze two types of application scenarios in order to evaluate the benefits of radio trigger in terms of energy consumption and lifetime prolongation of sensor nodes, as well as to evaluate different WURx prototypes. 4.2.1 Addressing not required In the first scenario the main node sends a message to wake up the nodes in the communication range, thus there is no need for addressing in the wake-up signal. In case of Marinkovic et al. WURx [24] there is no need for using the SPI to transfer the address information to the micocontroller, the wake-up interrupt is sufficient. A specific application in a WBAN is presented in [29], where the WURx is used to synchronize all the nodes for the TDMA communication that follows. Another example could be from the surveillance application, where a continuously active Pyroelectric InfraRed (PIR) node detects an event and sends the wake-up signal to WURxs of all the neighbor camera nodes within the communication range (Fig. 5). We consider a general case and assume that a camera node receives 200 wake-up signals in an hour, upon which the WURx activates the node (sensors acquire the data and send them to the main node). A node is active for 500 ms and after that goes back to sleep state until the next wake-up signal. Fig. 6 shows the average wake-up power consumption of the node with different WURx solutions. The node with Figure 5: A WSN for surveillance application without addressing requirements Marinkovic et al. WURx has the lowest power consumption (only 1 µw). Marinkovic+uC is the solution presented in [24], with SPI. Marinkovic, no addr is a solution without SPI and addressing capability. In addition, if we consider a solution without the WURx where a sensor node has a main transceiver consuming 60 mw when active, in WOR with 10% duty cycle and receives 200 requests per hour from the coordinator (waking up and consuming 100 mw for 500 ms) its lifetime is 420 hours with 2 AA batteries providing each 1.5 V and 1000 mah. Fig. 7 shows the lifetime prolongation of the solutions with WURxs compared to the solution where the node doesn t comprise the WURx but duty cycles the main transceiver. We see that engaging a WURx prolongs the node s lifetime for more than 150%. The best performance has the Marinkovic et al. WURx due to its lowest power consumption (at Fig. 7 designated as Marinkovic, no addr). Nevertheless, as seen from the graph, the performances of all the WURxs in terms of lifetime prolongation are similar, due to the fact that the biggest influence on the node s lifetime has the time that the node spends in active state (after being woken up by the WURx). 4.2.2 Addressing required The second typical application requires addressing of the node within the wake-up signal, i.e. the sender node doesn t want to wake up all the nodes within the communication range, but only a specific one. An example from the surveillance application is presented in [6], where the main node gathers the information from the low power densely deployed PIR sensor nodes detecting people presence and wakes up the specific camera node to acquire the image. There are number of other applications with similar approach. We consider here a general case. If a network consists of N nodes P_avg [uw] 60 50 40 30 20 10 0 1.0 1.0 12.5 Marinkovic Marinkovic, no Durante Huang Langevelde Hambeck +uc addr (+ FPGA) + DBB + DBB (+ dedicated HW) Figure 6: Average wake-up power consumption of the node for different WURx solutions with 200 wake-ups per hour. 54.7 6.1 2.4 104

Node lifetime prolongation [%] 180 160 With addressing Without addressing 140 120 100 80 60 40 20 0 Marinkovic Marinkovic, no Durante Huang Langevelde Hambeck +uc addr (+ FPGA) + DBB + DBB (+ dedicated HW) Figure 7: Lifetime prolongation of the node with different WURxs (scenario with addressing requirement and scenario without addressing requirement) compared to the node with 10% duty cycled main transceiver (WOR). P_avg [uw] 100 10 1 1,0300 1,0280 1,0260 1,0240 1,0220 1,0200 1,0180 1 10 100 1000 10000 100000 1 10 100 1000 Number of events [ev/h] number of nodes comprising the Marinkovic et al. WURx [24], and the there are N ev,i events/h for a certain node i, all other nodes (N nodes 1) in the communication range are also woken up by the wake-up signal and go back to sleep. We can assume that the events are uniformly distributed among nodes: N ev,i = Nev N nodes. (5) Then a node k (k N nodes, k i), wakes up in vain its microcontroller to check the destination address in the wakeup call for the following number of times: Nev N wup,v = N ev. (6) N nodes Other solutions from Table 2 wake up the microcontroller only when the wake-up signal is intended for it (N ev,i times). In a case study where the main node sends asynchronous demands (N ev = 1000) to 5 sensor nodes with addressing included in the wake-up signal (N ev,i = 200), including a WURx reduces idle listening of the main transceiver. If we include the Marinkovic et al. WURx without addressing capabilities (without the SPI), the destination address has to be communicated and decoded just after the node wakeup using the main transceiver and microcontroller in active state. We assume that activity to last for 100 ms while consuming 70 mw. Lifetime prolongation of the node with a WURx instead of duty-cycling is depicted in Fig. 7. The best performance has the Marinkovic et al. WURx solution, but it is necessary to include the SPI and addressing capabilities. Otherwise, the lifetime prolongation is only 51% instead of 158%. We compare all those addressing comprising solutions to the Marinkovic et al. WURx without addressing. In that case the WURx wakes up the node each time it detects a wake-up signal, regardless if it is intended for it or not (N ev times). In applications without addressing requirements that solution has the best performance regarding energy consumption. In applications with addressing requirements Marinkovic et al. WURx with addressing capabilities has lower power consumption than other WURxs (Fig. 7). Fig. 8 shows the average wake-up power consumption of the node P wup for different WURx solutions (Table 2) and different number of events per hour. The P wup of the Marinkovic et al. WURx with addressing is increasing with the num- Figure 8: Average wake-up power consumption of the node for different WURx solutions and different number of events per hour ber of events due to dynamic energy consumption from (4). Other solutions have a constant average P wup. We see that even for 100 000 wake-up signals per hour Marinkovic et al. WURx has lower power consumption than other WURxs with addressing capabilities. 5. CONCLUSION In this paper we explore benefits of WURxs over simple duty cycling (WOR). Simulation results for a typical reallife scenario with 200 events per hour in situations with and without addressing requirements show the benefits of the WURXs in terms of node lifetime prolongation compared to the solution with 10% duty-cycled wake-on radio scheme. Moreover, the analysis and comparison of the WURx prototypes show advantages of Marinkovic et al. [24] WURx, in both cases. Due to its very low average power consumption (1µW), it prolongs the node s lifetime for 158%. In application requiring addressing of the destination node, address has to be included within the wake-up signal and decoded in MCU. Otherwise the lifetime prolongation of the node would be 3 times lower. Increasing number of wake-ups, Marinkovic et al. WURx with addressing capabilities increases the wake-up power consumption due to MCU decoding activity. All other solutions have a constant wakeup power consumption. But even with 100 000 wake-ups per hour, its average power consumption is still lower than other solutions with address detection in dedicated HW instead of in MCU, making it most appropriate solution for most real-life applications. In future work we will address the specific case studies and analytically explore the benefits of Marinkovic et al. WURx compared to the wake-on radio, in terms of both power consumption and latency. We will also explore the case studies engaging a WUR both receiving and transmitting a wake-up signal. 6. NOWLEDGMENTS This research has received funding from GENESI Project (EU 7th Framework Programme, grant agreement n. 257916). 105

7. REFERENCES [1] R. Jurdak, P. Baldi, and C. Videira Lopes. Adaptive Low Power Listening for Wireless Sensor Networks. In IEEE Trans. on Mobile Comp., 6(8):988 1004, 2007. [2] I. Demirkol, C. Ersoy, and E. Onur. Wake-up Receivers for Wireless Sensor Networks: Benefits and Challenges. In IEEE Wireless Comm., pp. 88 96, 2009. [3] C. Fischione, S. Coleri Ergen, P. Park, K. H. Johansson, and A. Sangiovanni-Vincentelli. Medium Access Control Analytical Modeling and Optimization in Unslotted IEEE 802.15.4 Wireless Sensor Networks. In 6th IEEE Comm. Society Conf. on Sensor, Mesh and Ad Hoc Comm. and Networks (SECON), pp. 1 9, 22 26 June 2009. [4] E.-Y. Lin, J. Rabaey, and A. Wolisz. Power-efficient rendez-vous schemes for dense wireless sensor networks. In IEEE ICC, Paris, 20 24 June 2004. [5] G. Lu, D. De Mingsen, and X. W.-Z. Song. TelosW: Enabling Ultra Low-Power Wake-On Sensor Networks. In Proc. 7th Internat. Conf. on Networked Sensing Systems (INSS), 15 18 June 2010. [6] V. Jelicic, M. Magno, D. Brunelli, V. Bilas, and L. Benini. An energy efficient multimodal wireless video sensor network with ez430-rf2500 modules. In Proc. 5 th Internat. Conf. on Pervasive Comp. and Appl.(ICPCA), Maribor, Slovenia, pp. 161 166, Dec. 1 3, 2010. [7] Y. Zhang, L. Huang, G. Dolmans, and H. de Groot. An Analytical Model for Energy Efficiency Analysis of Different Wakeup Radio Schemes. In 20th Personal, Indoor and Mobile Radio Comm. Symp. (PIMRC), Tokyo, Japan, pp. 1148 1152, 16 Sept. 2009. [8] M. Lont, D. Milosevic, P. G. M. Baltus, A. H. M. van Roermund, and G. Dolmans. Analytical Models for the Wake-up Receiver Power Budget for Wireless Sensor Networks. In GLOBECOMM, 2009. [9] N. S. Mazloum and O. Edfors. DCW-MAC: An energy efficient medium access scheme using duty-cycled low-power wake-up receivers. 2011. [10] L. Gu and J. A. Stankovic. Radio-Triggered Wake-Up Capability for Sensor Networks. In Proc. 10th IEEE Real-Time and Embedded Techn. and Appl. Symp. (RTAS), pp. 27 36, 25 28 May 2004. [11] N. M. Pletcher, S. Gambini, and J. Rabaey. A 52 µw Wake-Up Receiver With 72 dbm Sensitivity Using an Uncertain-IF Architecture. In IEEE Journ. of solid-state circuits, 44(1):269 280, Jan. 2009. [12] B. van der Doorn, W. Kavelaars, and K. Langendoen. A prototype low-cost wakeup radio for the 868 MHz band. In Int. J. Sensor Networks, 5(1), 2009. [13] J. Ansari, D. Pankin, and P. Mahonen. Radio-Triggered Wake-ups with Addressing Capabilities for Extremely Low Power Sensor Network Applications. In RTWAC-PIMRC, 2008. [14] X. Yu, J.-S. Lee, C. Shu, and S.-G. Lee. A 53µW Super-Regenerative Receiver for 2.4GHz Wake-up Application. 2008. [15] P. Le-Huy and S. Roy. Low-Power 2.4 GHz Wake-Up Radio for Wireless Sensor Networks. In Proc. IEEE Internat. Conf. on Wireless & Mobile Comp., Networking & Comm., pp. 13 18, 2008. [16] R. van Langevelde, M. van Elzakker, D. van Goor, H. Termeer, J. Moss, and A. J. Davie. An Ultra-Low-Power 868/915 MHz RF Transceiver for Wireless Sensor Network Applications. In IEEE Radio Frequency Integrated Circuits Symp., 2009. [17] Y. Zhang, S. Chen, N. F. Kiyani, G. Dolmans, J. Huisken, B. Busze, P. Harpe, N. van der Meijs, and H. de Groot. A 3.72µW Ultra-low Power Digital Baseband for Wake-up Radios. 2011. [18] W.-C. Shih, R. Jurdak, B.-H. Lee, and D. Abbott. High sensitivity wake-up radio using spreading codes: design, evaluation, and applications. In EURASIP Journ. on Wireless Comm. and Networking, 2011. [19] C. Hambeck, S. Mahlknecht, and T. Herndl. A 2.4µW Wake-up Receiver for Wireless Sensor Nodes with -71dBm Sensitivity. 2011. [20] Fraunhofer Institute. Ultra low-current wakeup receiver, 2010. [21] G. U. Gamm, M. Sippel, M. Kostic, and L. M. Reindl. Low Power Wake-up Receiver for Wireless Sensor Nodes. In ISSNIP, pp. 121 126, 2010. [22] Austriamicrosystems. AS3931 3D Low Power Wakeup Receiver. Datasheet, 2010. [23] M. Spinola Durante and S. Mahlknecht. An ultra low power Wakeup Receiver for Wireless Sensor Nodes. In 3rd Internat. Conf. on Sensor Tech. and Appl., 2009. [24] S. Marinkovic and E. Popovici. Nano-Power Wireless Wake-Up Receiver With Serial Peripheral Interface. In IEEE Jour. on selected areas in comm., 29(8):1641 1647, Sept. 2011. [25] C. Hambeck, S. Mahlknecht, and T. Herndl. A 2.4µW Wake-up Receiver for Wireless Sensor Nodes with -71 dbm Sensitivity. 2011. [26] Y. Zhang, A. Breeschoten, X. Huang, N. Kiyani, A. Ba, P. Harpe, K. Imamura, R. de Francisco, V. Pop, G. Dolmans, and H. de Groot. Improving Energy-Efficiency in Building Automation with Event-Driven Radio. 2011. [27] X. Huang, P. Harpe, G. Dolmans, and H. de Groot. A 915MHz Ultra-Low Power Wake-Up Receiver with Scalable Performance and Power Consumption. 2011. [28] X. Huang, P. Harpe, X. Wang, G. Dolmans, and H. de Groot. A 0dBm 10Mbps 2.4GHz Ultra-low Power ASK/OOK Transmitter with Digital Pulse-Shaping. In RFIC, CA, USA, May 2010. [29] S. Marinkovic and E. Popovici. Power Efficient Networking Using a Novel Wake-up Radio. In 5th Internat. Conf. on Pervasive Comp. Tech. for Healthcare (PervasiveHealth) and Workshops, 2011. [30] IMEC news, 2012, http://www2.imec.be/be_en/press/imec-news/ imecisscclowpowertransmitter.html [31] X. Huang, S. Rampu, X. Wang, G. Dolmans, and H. de Groot. A 2.4GHz/915MHz 51µW Wake-Up Receiver with Offset and Noise Suppression. In ISSCC, 2010. [32] S. Drago, D. M. W. Leenaerts, F. Sebastiano, L. J. Breems, K. A. A. Makinwa, and B. Nauta. A 2.4GHz 830pJ/bit Duty-Cycled Wake-Up Receiver with -82dBm Sensitivity for Crystal-Less Wireless Sensor Nodes. 2010. 106