Energy-Aware Spectrum Sensing in Cognitive Wireless Sensor Networks: a Cross Layer Approach Luca Stabellini and Jens Zander Wireless@KTH, The Royal Institute of Technology, Electrum 418, SE-164 40 Kista, Sweden Email: {lucast,jenz}@kth.se Abstract—Low-power transmissions of sensor nodes are eas- ily corrupted by interference generated by co-located wireless terminals that leading to packet losses might increase energy consumption and result in unreliable communications. Dynamic spectrum access mechanisms can mitigate these problems allow- ing cognitive sensor devices to sense the spectrum and access the wireless medium in an opportunistic way. With this respect, energy efficient algorithms for spectrum sensing have to be designed in order to meet the power constraints of wireless sensor networks. In this paper we consider an energy constrained system comprising two sensor nodes that avoid interference by exploiting spectrum holes in the time domain. We design the algorithm used for spectrum sensing so as to minimize the average energy required for the successful delivery of a packet. While carrying our this task we adopt a cross layer approach that accounts for the average channel occupancy and the power of interfering transmissions at the physical layer as well as for the size of packets used by sensors at the transport layer. Our results show that accounting for the short length of packets commonly used in sensor networks can significantly improve energy efficiency leading to gains of up to 50% if compared to other spectrum sensing algorithms envisaged in the literature. I. I NTRODUCTION Communications of low-power sensor nodes are easily corrupted by interference induced by transmissions of other co-located wireless devices operating in the same frequency band. Such interference might lead to packet losses and require retransmissions thus eventually increasing energy consumption and data latency. These problems result in unreliable commu- nications and are nowadays further enhanced by the increasing proliferation of wireless terminals in the few available and always more crowded unlicensed bands. In fact, recent surveys conducted in the context of industrial automation [1] have shown that the potential unreliability of wireless communica- tions is perceived as one of the major barriers to the adoption of wireless technologies for sensing and control applications. In order to avoid interference and mitigate the induced performance degradation cognitive radio solutions have been recently envisaged in the context of wireless sensor networks: the basic idea tailored by these schemes is to allow sensor devices to monitor the available frequency bands and op- portunistically select for their transmissions unused pieces of spectrum (also called white spaces or spectrum holes [2]). Two different techniques can be implemented for this purpose. A first approach aims at exploiting spectrum holes in the fre- quency domain through frequency agility (see for instance [3], [4]): once interference is detected on a certain channel, nodes switch their radio to a different one and re-establish links with their neighbors. A different solution instead is to take advantage of white spaces in the time domain ( [5]) accessing the medium during intra-packet idle periods, when interfering devices are not transmitting. Both approaches require sensor nodes to identify suitable spectrum holes through spectrum sensing. This results in extensive usage of the radio unit which represents the major source of energy consumption for low-power wireless motes [6] and thus introduces significant energy overhead. In order to meet power constraints of sensor networks and enable the adoption of interference avoidance schemes based on dynamic spectrum access energy efficient algorithms for spectrum sensing have thus to be designed. In this context a relevant question is how much energy should be spent on channel sensing. High energy budgets will result in accurate sensing outcomes but on the other hand, investing too much on this task might not be worth if interfering signals are sporadic or are perceived with very high power and are thus easy to detect. Furthermore, in order to achieve the highest energy efficiency, the sensing energy budget should be dimensioned accounting for the size of trans- mitted packets. The loss of long packets caused by sensing errors results in costly retransmissions and has to be avoided by achieving sensing outcomes as reliable as possible; when transmitting shorter packets instead, lower energy budgets might originate sensing inaccuracies but result in lower overall energy consumption. It should be noted that this problem is of extreme interest for wireless sensor networks: sensor packets can be as small as a few tens of bytes 1 therefore selecting the right amount of energy that has to be devoted to spectrum sensing might significantly improve energy efficiency. We remark that several works have been recently dealing with the problem of spectrum sensing (the reader is referred to [8] for a survey on this topic). Two issues have mainly been investigated: those are the accurate detection of primary signals (for instance by means of dedicated signal processing techniques [9] or cooperation among different sensing units [10], [11]: these works mainly consider licensed settings) and the fast identification of spectrum opportunities [12]. We here investigate a different dimension and aim at designing the sensing scheme to minimize the energy consumption. To this purpose we consider an energy constrained system comprising two sensor nodes, one transmitter and one receiver: these nodes share their communication channel with a set of interfering devices and opportunistically access the medium exploiting white spaces in the time domain. For this system we compute the optimal spectrum sensing strategy, i.e. the optimal energy budget and sensing decision threshold that 1 The IEEE 802.15.4 radio standards, that is currently used by a large fraction of the deployed sensor networks [7] specifies a maximum packet size of 128 bytes and suggests an average frame length of 22 bytes. This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2010 proceedings. 978-1-4244-6398-5/10/$26.00 ©2010 IEEE