Int. J. Sensor Networks, Vol. X, No. Y, XXXX
The impact of fading and shadowing on the network performance of wireless sensor networks Bao Hua Liu* Land and Joint Systems, Thales Australia E-mail: [email protected]
Brian Otis Department of Electrical Engineering, University of Washington, Seattle, USA E-mail: [email protected]
Subhash Challa National Information and Technology Australia (NICTA), Victorian Research Labs The University of Melbourne E-mail: [email protected]
Paul Axon Land and Joint Systems, Thales Australia E-mail: [email protected]
Chun Tung Chou and Sanjay K. Jha School of Computer Science and Engineering, University of New South Wales, Sydney, Australia E-mail: [email protected]
E-mail: [email protected]
Abstract: Most ad hoc and sensor network research assumes idealised radio propagation models without considering fading and shadowing effects. Experimental results have shown that many well-designed protocols will fail simply because of fading and shadowing experienced in a realistic wireless environment. While fading and shadowing for radio propagation are well understood in the wireless communication community, they are rarely studied in network level research for wireless sensor networks. This paper studies the fading and shadowing effects on the performance of different systems for wireless sensor networks. We show that fading and shadowing can have a significant influence on network performance. We study and compare network performance for three different systems: (1) a multichannel CDMA system; (2) a pure CDMA system; (3) a contention-based system. Through discrete event simulation (using ns-2), we show that the multichannel CDMA system outperforms the pure CDMA system as well as the contention-based system under fading and shadowing environments. Keywords: wireless sensor networks; fading; shadowing; CDMA; FDMA; Multiple Access Interference; MAI; Media Access Control; MAC. Reference to this paper should be made as follows: Liu, B.H., Otis, B., Challa, S., Axon, P., Chou, C.T. and Jha, S.K. (XXXX) ‘The impact of fading and shadowing on the network performance of wireless sensor networks’, Int. J. Sensor Networks, Vol. X, No. Y, pp.XXX–XXX. Biographical notes: Bao Hua Liu is the Manager of Network Enabled Warfare Laboratory (NEWLAB), Land and Joint Systems, Thales Australia at Garden Island, New South Wales, Australia. Currently, he manages and leads several R&D projects in supporting Thales Australia’s solutions for future Australian Defence Forces’ (ADF) Network Centric Warfare (NCW) concept. He received a PhD in Computer Science and Engineering from University of New South Wales. He also received an MSc in Internetworking from the University of New South Wales, and a BE in Copyright © XXXX Inderscience Enterprises Ltd.
B.H. Liu et al. Electronics and Communications Engineering from Tianjin University, PR China. This work was done when Dr. Bao Hua Liu was employed as a Research Fellow at the University of Technology Sydney, Australia. Brian P. Otis received a BS in Electrical Engineering from the University of Washington and his MS and PhD from the University of California at Berkeley. He joined the University of Washington as Assistant Professor of Electrical Engineering in August 2005. His research interests include ultra-low power analog, digital and RF circuits for ubiquitous sensing and communication. He is an Associate Editor of the IEEE Transactions on Circuits and Systems Part II. He was the recipient of the 2003 U.C. Berkeley Seven Rosen Funds award and was co-recipient of the 2002 ISSCC Outstanding Technology Directions Paper. Subhash Challa is a Senior Principal Researcher at NICTA in the University of Melbourne where he leads a number of user inspired research in data fusion and tracking with applications in Life Sciences. Before joining NICTA, he was the Professor of Computer Systems at the University of Technology, Sydney, and leads the Networked Sensor Technologies (NeST) Lab. He is also the CTO and co-founder of SenSen Networks Pty Ltd - a venture capital backed innovative start-up company that delivers intelligent video-centric monitoring and surveillance software solutions. He received his PhD from QUT, Brisbane, Australia in 1999. This work was done when Prof. Subhash Challa worked at the University of Technology Sydney, Australia. Paul Axon is Technical Architect for the design of Naval wireless communications systems at Thales Australia for telemetry and navigation and high-speed Naval Tactical IP Trunk links. His experience includes communications architecture, analysis, design; satellite links for telemetry control and interactive remote video learning, interactive digital TV, communications modelling, e-Business process and architecture, secure networks digital switching research. He graduated of from Cambridge University (UK) and London University (UK). He is a Chartered Engineer and Member of the IEAust, IEE and ACM. Chun Tung Chou is an Associate Professor in the School of Computer Science and Engineering at the University of New South Wales, Sydney, Australia. He received his BA in Engineering Science from the University of Oxford, UK and his PhD in Control Engineering from the University of Cambridge, UK. He has published extensively in wireless networking, computer networking and system identification. His current research interests are in Wireless Mesh Networks, Wireless Sensor Networks and Network Optimisation. Sanjay K. Jha received his PhD from the University of Technology, Sydney, Australia. He is a Professor and Head of the Network Group at the School of Computer Science and Engineering at the University of New South Wales. His research activities cover a wide range of topics in networking, including wireless sensor networks, ad hoc/community wireless networks, resilience/quality of service (QoS) in IP networks, and active/programmable networks. He is the Principal Author of the book Engineering Internet QoS and a co-editor of the book Wireless Sensor Networks: A Systems Perspective. He is an Associate Editor of the IEEE Transactions on Mobile Computing.
Future applications will increasingly depend on embedded wireless sensor networks. A sensor network consists of numerous sensor/actuator devices. These devices consist of one or more integrated sensing units, embedded microprocessors, low-power communication radios, on-board energy, with location awareness and organised in an ad hoc multihop network. Sensors are normally untethered and communicate over short distances using wireless media. Sensor network applications are generally expected to utilise low data rate (e.g. 1–100 kbps), have short data packet lengths (e.g. 50 bytes), and normally have limited on board energy, processing capability, buffer space and other resources. Contention-based protocols may not be a suitable choice for Media Access Control (MAC) layer as they suffer from both low network throughput and long packet latency. Associating with each short data packet transmission, the
RTS-CTS-DATA-ACK handshake sequence can constitute up to 40% overhead in sensor networks (Woo and Culler, 2001). Furthermore, contention-based protocols also suffer from the well-documented hidden node and exposed node problems. It is well known that energy consumption is the crucial factor in sensor network design. This may lead to sensor network protocols which prioritise energy savings over network throughput and packet latency. However, we argue that both network throughput and packet latency are critical for many sensor network applications, such as battlefield surveillance, real-time monitoring seismic waves, machine operations and bush fires. Accurate and timely delivery of the sensed data in these potentially life-threatenting situations is of great importance. While a sensor network is expected to operate with low duty cycle (e.g. 1% on average) and may remain silent for long periods, a communication ‘hot region’ can emerge quickly due to simultaneous events (Stankovic
The impact of fading and shadowing on the network performance of wireless sensor networks et al., 2003). Such unpredictable traffic patterns require highly adaptive protocols to achieve both real-time guarantee as well as energy efficiency. In order to achieve significant improvement in network performance, a novel multichannel CDMA system which can simultaneously achieve high network throughput, high system capacity, low packet latency and low communication energy consumption have been proposed (Liu et al., 2004, 2005a) for wireless sensor networks. These advantages are the result of applying frequency division (multichannel) to reduce the Multiple Access Interference (MAI) in a pure CDMA system. The original idea has been designed as a MAC layer protocol for wireless sensor networks, hence CSMAC (CDMA Sensor MAC). Our former works in Liu et al. (2004, 2005a) proposed the original idea of the CSMAC system (and a mathematical modelling of the mean MAI at a given node) but lack extensive simulation studies and analysis. In particular, like many wireless ad hoc and sensor network research, the former studies of CSMAC assume idealised radio propagation model (e.g. ‘disk-model’) and did not consider the influence of fading and shadowing effects. This paper attempts to bridge this gap by exploring the fading and shadowing influence on the proposed CSMAC system. We also study and evaluate the influence of fading and shadowing on a pure CDMA system as well as a contention-based system – SMAC (Ye et al., 2002) – and compare their performance with CSMAC. Sophisticated physical layer parameters and radio propagation models have been used to simulate the realistic wireless environment in our study. We demonstrate that fading and shadowing can have significant influence on the sensor network performance and must be considered seriously in any protocol and algorithm design (e.g. MAC, routing, topology control, etc.) for wireless sensor networks. Without considering fading and shadowing effects, well developed protocols and algorithms by using idealised radio propagation models will simply fail in a realistic wireless environment. While fading and shadowing for radio propagation have been well understood in wireless communication community for a long time, their impact on the network performance has been rarely studied in network level research for wireless sensor and ad hoc networks. The rest of this paper is organised as follows. Section 2 describes some background for radio propagation. Section 3 reviews the related work. Section 4 gives a brief review of CSMAC system and the design challenges at the physical layer. Section 5 elaborates the simulations and analysis. Section 6 concludes this paper with future directions.
Radio propagation background
The mechanisms of electromagnetic wave propagation are diverse and can generally be attributed to reflection, diffraction and scattering (Rappaport, 2002). Radio propagation models can be broadly classified into two categories: large-scale propagation (or fading) models and small-scale fading models. Large scale propagation models are used to predict the mean signal strength decays as
a function of the Transmitter-Receiver (T-R) separation distance raised to some power (i.e. a power law function). Small-scale fading models are used to characterise rapid fluctuations of the received signal strength over very short distances or very short time durations. In this section, we provide a brief review for the propagation models that have been used in our studies.
Large scale propagation models
Large-scale variations are typically broken down into distance attenuation and shadow fading. Distance attenuation is the empirically observed long-term trend in signal loss as a function of distance, typically proportional to the range raised to some power. Shadowing is the variation about this trend line caused by objects that are many wavelengths in size (Pottie and Kaiser, 2005). Two most commonly used large-scale propagation models in wireless sensor and ad hoc network research are Friis free space model and two-ray ground reflection model. Friis free space model is the simplest propagation model where the transmitter and receiver have a clear, unobstructed line-of-sight path between them. The signal strength received by a receiver antenna which is separated from a radiating transmitter antenna by a distance d, is given by the Friis free space equation: Pr (d) =
Pt Gt Gr λ2 (4π )2 d 2 L
where Pt is the transmitted power, Pr (d) is the received power, Gt is the transmitter antenna gain, Gr is the receiver antenna gain, d is the T-R separation distance in metres, L is the system loss factor (L ≥ 1) and λ is the wavelength in metres. Equation (1) reveals that the radio propagation loss in free space is directly proportional to the square of distance. A single line-of-sight path between the transmitter and receiver is seldom the only physical means for propagation. The two-ray ground reflection model considers both the direct path and a ground reflect propagation path between transmitter and receiver. The received power at distance d is predicted by: Pr (d) =
Pt Gt Gr h2t h2r d 4L
where ht and hr are the heights of transmit and receive antenna, respectively. The above equation shows a faster power loss than Friis free space Equation (1). A simple observation of Equations (1) and (2) reveal that both Friis free space and two-ray ground reflection model predict the received power as a deterministic function of distance – where the communication range is represented as an ideal circle sphere (without considering shadow fading). A more sophisticated large-scale propagation model that considers the shadow fading is the log-normal shadowing model. Theoretical and measurement-based propagation models indicate that average received signal power decreases logarithmically with distance, whether in outdoor or indoor radio channels (Rappaport, 2002). The mean path loss for an
B.H. Liu et al.
arbitrary T-R separation can be expressed as a function of distance n d PL(d) ∝ (3) d0
carrier frequency. The baseband complex impulse response is then given by Pottie and Kaiser (2005): αi (t)ej θi (t) δ τ − τi (t) (7) h(t; τ ) =
PL(d) is often stated in decibels
Subsequent rays typically experience larger attenuation than the first arrived ray. Consequently, the impulse response generally decays with delay. When a large number of rays arrive with similar delay and no dominant path, the Central Limit Theory dictates that the statistics of their sum will be a complex Gaussian distribution with zero mean. When there is a dominant path, it will be a complex Gaussian with non-zero mean. Now let X be the in-phase component of this Gaussian and Y be the quadrature component. The √ signal amplitude for this delay can be expressed as Z = X 2 + Y 2 . If the two components have zero mean and variance σ 2 , the resulting distribution is Rayleigh: z2 z , z>0 (8) f (z) = 2 exp − σ 2σ 2 If the means of X and Y are mX and mY , respectively, the resulting distribution is Ricean: z2 + A2 Az z I0 2 , z > 0 (9) f (z) = 2 exp − σ 2σ 2 σ where A = mX +mY and I (·) is the modified Bessel Function of the first kind and zero-order. The Ricean distribution is often described in terms of a parameter K = A2 /(2σ 2 ). As A → 0, K → 0, the Ricean distribution approaches the Rayleigh distribution. Figure 1 illustrates the combined effects of large-scale propagation and small-scale fading. It shows that the radio signal attenuation as a function of distance from the source may be conceived as the superposition of distance loss, shadowing and multipath effects.
where n is the path loss exponent, d0 is the reference distance corresponding to a point located in the far field of the transmit antenna and d is the T-R separation distance. PL(d) represents the average path loss for a given value of d. The value of n lies in between 2 and 6 depending on the specific propagation environment. For example, n = 2 for free space propagation, but n will have larger value when obstructions are presented. The model given by Equation (4) is called log-distance path loss model. The bars in Equation (4) denote the average value of all possible path loss values for a given value of d. The path loss versus distance expressed in Equation (4) does not consider the surrounding environmental clutter that may be different at two locations with the same T-R separation. This leads to the measured signals that are vastly different from the average value predicted by Equation (4). Empirical studies have shown that with a specific value of d, the path loss PL(d) at a particular location is random and distributed log-normally (normal in dB) about the mean distance-dependent value PL(d)[dB] = PL(d)(dB) + Xσ = PL(d0 )(dB) d +10nlog + Xσ d0
and the received signal power can be expressed as Pr (d)[dBm] = Pt [dBm] − PL(d)[dB]
where Xσ denotes a zero mean, Gaussian random variable (in decibels) with standard deviation σ (also in decibels). Xσ is site and distance dependent. The log-normal distribution describes the shadowing effect which occurs over a large number of measurement locations that have the same T-R separation, but have different levels of clutter in the propagation path. This phenomenon is referred to as log-normal shadowing.
Combined effects of distance loss, shadowing and multipath fading
Distance loss Shadowing Multipath fading
d PL(d)(dB) = PL d0 (dB) + 10nlog d0
Small scale fading models
Small-scale fading is caused by movement of transmitter, receiver or other object in the environment. Two common small-scale fading models are Rayleigh and Ricean. A Rayleigh distribution is normally used to describe the statistical time-correlation nature of the received signal envelope, or the envelope of an individual multipath component. When there is a dominant stationary (non-fading) signal component present, such as line-of-sight propagation path, the small scale fading envelop distribution is Ricean. The received signal is normally modelled as the sum of multiple rays, each of which is characterised by a propagation delay τ , attenuation α and phase shift θ . The phase shift is dependent on the delay with θ = 2πfc τ , where fc is the
Literature survey have shown that most related works focus on studying a specific problem under the fading or shadowing assumption. The fading and shadowing are normally modelled with simple statistical property and studied separately. Unlike their works, our study focuses
The impact of fading and shadowing on the network performance of wireless sensor networks on the effects of fading and shadowing on the network performance. We use sophisticated radio propagation model and combined the fading and shadowing effects to simulate the realistic wireless environment (see Section 5). To the best of our knowledge, ‘combined studies of fading and shadowing on the network performance for wireless sensor networks’ has not been found in the literature. Bettstetter (2004) addressed the topological design of wireless multihop networks that are robust against node failures under shadowing environments. His study investigates the minimum node density required to ensure that all nodes inside a randomly chosen area are k-connected with high probability. Chen et al. (2005) studied the effects of cooperative communication via transmission diversity and multihopping as well as optimal power allocation schemes in fading channels. A simple quasi-static Rayleigh fading channel is assumed in their study. Liu and Haenggi (2005) presented closed-form expressions of the average link throughput for sensor networks with a slotted ALOHA MAC protocol in Rayleigh fading channels. The Rayleigh fading is modelled with simple statistical behaviour based on path loss. Mullen and Huang (2005) studied several Mobile Ad Hoc Network (MANET) behaviours such as MAC retries on the performance and route stability using a simple stochastic Rayleigh fading model of the received power. Souryal and Moayeri (2005) describes a cross-layer approach for relaying messages in multihop ad hoc networks that adapts to time-varying channel effects and exploits spatial diversity by opportunistically selecting an appropriate relay from a set of candidate relays at each hop. The main idea is to use Multicast RTS (MRTS) and multiple replied CTS (from multiple potential relaying neighbours) to estimate the instantaneous channel state information. The sender then selects the next hop node with maximum expected progress, which is calculated based on the instantaneous measured Signal to Interference and Noise Ratio (SINR) at both the sender and receiver. Besides the employment of MRTS and multiple replied CTS, this approach also enlarges the packet size of RTS and CTS to accommodate the current channel state information such as SINR. This approach in fact violates the design goal of CSMAC, where control packet exchange (e.g. RTS/CTS) are expected to be removed to achieve high system performance. In most of these works, the effects of fading are approximated using random number generation with the appropriate statistical behaviour. The drawback of this approach is that it does not describe the time-correlation of the signal envelope. In comparison, we have adopted the small-scale fading model (Rayleigh and Ricean, see Section 2) proposed by Punnoose et al. (2000). This model has the appropriate statistics as well as time correlational properties obtained from the Doppler spectrum. A similar work to ours has been presented by Han and Abu-Ghazaleh (2005). They studied the effect of shadow fading (small-scale fading has not been studied) on ad hoc networks by using log-normal shadowing model. They have shown that the effect of shadowing on the MAC (e.g. IEEE 802.11) performance can be a dominant factor. One interesting conclusion of their study is that eliminating RTS/CTS packets results in more effective operation. RTS/CTS hence has been considered harmful
in a shadow fading environment since four transmissions must succeed for a packet to be delivered (RTS, CTS, DATA, ACK). At low transmission success probability (due to shadow fading) it becomes highly improbable for four consecutive transmissions to succeed. Several subtle interactions between the MAC and routing layer have also been observed. For example, the criteria for determining the best path should not only consider the link status but also the link order. In addition, because routing protocols rely on MAC level transmission failure (when the retry limit is exceeded), route failure errors are often generated unnecessarily. Moreover, because MAC level broadcasts are unreliable, they are especially vulnerable to shadow fading.
CSMAC: a multichannel CDMA system
CSMAC has been designed to achieve high network performance in comparison with a contention-based system – SMAC (Ye et al., 2002) – as well as a pure CDMA system. CSMAC uses a combination of CDMA and FDMA (multiple channels) techniques to reduce channel interference and consequently improves system capacity and network throughput. The motivation behind CSMAC is to remove the control packet exchange (e.g. RTS/CTS/ACK) and CSMA/CA Distributed Coordination Function (DCF) exponential backoff scheme to achieve low packet latency, low energy consumption and high network throughput. CSMAC allows each node to start transmitting immediately when a packet is received from upper layer. While a pure CDMA system may also remove control packet exchange and DCF, it suffers from the well known MAI problem. Considering a Direct Sequence Spread Spectrum/Binary Phase Shift Keying (DSSS/BPSK) system, let P0 denote the average received power of the desired signal at the detector. Further assume that there are k interferers with received powers P1 , P2 , . . . , Pk , the total MAI power is calculated as the sum of all interference powers, that is, MAI Power =
The effective bit energy-to-noise ratio at the detector is then given by Pursley (1977): Eb µ = N0eff
−1 k 2 i=1 Pi 1 + 3LP0 µ0
where L is the processing gain, and µ0 = Eb /N0 equals Eb /N0eff at the detector in the absence of interferers. The probability of bit error Pe with a given µ is then given √ by Pe = 1/2erfc( µ). In a cellular CDMA system, MAI may be well controlled by a base station. But in a multihop CDMA sensor network, the MAI may be uncontrollable due to the randomised network topology and lack of a base station. Unless some sort of coordination (e.g. DCF, RTS/CTS) is used in an interference vicinity, MAI may cause significant degradation in network performance for a pure CDMA system.
B.H. Liu et al.
CSMAC system architecture
A CDMA system uses spread spectrum modulation technique, in which the baseband signals are spread using different Pseudo Noise (PN) codes to enable multiple access. The performance of CDMA system is primarily limited by MAI. MAI occurs because, unlike FDMA and TDMA channels, CDMA PN codes are not completely orthogonal. Completely orthogonal codes are normally used in synchronous systems. However, in asynchronous systems, perfect orthogonal codes are suboptimal and exhibit high cross-correlation. Since MAI is caused by the non-perfect orthogonality of CDMA codes, the rationale of CSMAC design is to orthogonalise the reception in the vicinity of a sensor node. CSMAC uses frequency division to reduce MAI in a CDMA sensor network (Liu et al., 2005a), where each node is assigned a unique receiving frequency which is different from its one hop and two hop neighbours. Because most sensor network applications utilise low data rate, it is possible to use a relatively narrow band CDMA system operating over multiple frequency channels. For example, assuming a data rate of 20 kbps with 50 chip/bit spreading, the resulting chip rate of the signal will be 1 Mcps. Assuming that the bandwidth of this spread signal is 1 MHz, if the system were to use the 2.4 GHz ISM band (2.4–2.4835 GHz), we can have more than 80 FDMA channels. Moreover, our studies reveal that a much smaller number of channels are required in a real sensor network deployment. For example, with a uniformly randomly deployed sensor network, only 13 channels are required when the node degree is 6 and 18 channels are required when the node degree is 9 (Liu et al., 2005b). Figure 2 illustrates the conceptual steady state architecture of CSMAC with a regular triangular sensor network topology, where each sensor has exactly six neighbours and the receiving frequency of each node is shown next to the node. With a random deployment, the number of neighbours and network topology can be determined by a topology control protocol (e.g. K-Neigh (Blough et al., 2003)). There are two code assignment schemes that can be used in CSMAC: 1 Node-based, where each neighbour of a given node is assigned a unique PN code that is different from other neighbours. 2 Link-based, where each directed link is assigned a unique PN code that is different from its adjacent links (two directed links are adjacent if they have a common end node). As an example, Figure 2 shows the link based PN code assignment, where the code assigned to each directed link is shown along the link. When a node wants to transmit a packet to a neighbour, it synthesises its transmitter to the correspondent receiving frequency of the neighbour and uses the predetermined PN code with the neighbour to spread baseband signal. For example, when A wants to transmit to D, its transmitter synthesises frequency f 3 and uses PN1 to spread the baseband signal. Note this transmission does not cause interference to other neighbours such as C, B, G, F, E because their receivers are tuned to different receiving frequencies. Also note that B, F and D can transmit to A simultaneously because A’s multiuser detection
receiver can distinguish all its neighbours’ transmissions concurrently. Note that A does not need to monitor the whole set of PN codes but only the set of codes that are employed by its immediate neighbours. A’s transmission to D will not destroy A’s reception from B, F and D even if they are happening in parallel because A’s transmission is operating on f 3 but A’s reception is operating on f 5. Furthermore, the transmission signal from G to B will not contribute to the noise floor at A because it is operating on f 4. To this end, we notice that MAI only occurs at a given node (e.g. A) when multiple neighbours (e.g. B, F, D) transmit to this node (e.g. A) simultaneously. When these simultaneous transmissions occur, they are actually desired signals since they are all addressed to the given node (e.g. A). With proper power control, the resulting MAI at the given node (e.g. A) can be controlled at the lowest level. Those uncontrollable MAI presented in a pure CDMA system, which is caused by the transmission between an interference node (e.g. G) and its neighbour (e.g. B but not A), is reduced significantly due to the employment of frequency division. Note it is possible that some other nodes are transmitting with the same frequency in the network (as frequencies are reused spatially). But assuming that the steady state architecture shown in Figure 2 can be achieved,1 those transmissions are normally far enough and the resulting interference is negligible. Because a node need not consider the interference caused by its transmission on the unintended receivers, it is much easier for the node to control its transmission power to assure that the transmitted signal arrives at the intended receiver with a certain power level, for example, the lowest receiving threshold. Figure 2
System architecture of CSMAC f3
f1 G f7
f2 E f4
N O f7
An interesting observation of the communication paradigm described is that the simultaneous transmissions (from different neighbours) to a node are no longer an undesirable but preferred mode. Simultaneous receptions at a given node means that the ‘ON’ time of the receiver may be reduced significantly, which implies lower energy consumption. In contrast, multiple transmissions have to be received in sequence for a contention-based system which implies longer receiver ‘ON’ time and higher energy consumption (also
The impact of fading and shadowing on the network performance of wireless sensor networks longer latency). For example, assume that each neighbour of node A in Figure 2 wants to send a message to A. If all neighbours can start their transmissions simultaneously, the ‘ON’ time of A’s receiver is only one message time Tm (assume same message size and data rate). If all these messages have to be sent in sequence by using a contentionbased system, the ‘ON’ time of A’s receiver will be 6Tm (without considering RTS/CTS/ACK, DCF etc.). Even if we assume that the multiuser detection circuits consumes more power than a single user detection circuits, the receiver frontend synthesiser and filters will consume less energy due to the shorter ‘ON’ time. This example demonstrates that sacrifice latency may not always achieve energy savings in wireless sensor networks. By employing frequency division, CSMAC can achieve a significant reduction in MAI and consequently less channel contention. Reduction in MAI makes it possible to forego the control packet (e.g. RTS/CTS/ACK) exchange which leads to lower protocol overhead, higher network throughput, lower packet latency and lower energy consumption. It is worth noting the trade-offs of the proposed CSMAC architecture. Besides the physical layer challenges described in next Section 4.2, the multichannel design of CSMAC means higher bandwidth is required. Our argument is that, in an energy-constrained wireless sensor network, bandwidth is not as scarce a resource as energy. For the purpose of this work, sharing of spectrum by too many users could cause excessive interference, but it is also a waste of spectrum if the bandwidth is not used by any user. In addition, CSMAC may employ shorter PN sequence (without sacrifice the network performance significantly) since it already employs frequency division. Shorter PN code chip size will result in smaller spread signal bandwidth. If we assume that the total available bandwidth is fixed (e.g. 2.4 GHz ISM band has 83.5 MHz), smaller spread signal bandwidth means more frequency channels are available. On the other hand, if we assume that the spread signal bandwidth is fixed, shorter PN code chip size means that we can use higher data rate.
Physical layer and idle energy consumption
CSMAC works on a fundamentally different physical layer platform compared to most protocols that have been proposed for sensor networks. The following hardware capabilities are assumed: 1 the transceiver operates in full-duplex mode2 2 a multiuser detection receiver is required but must only monitor a limited number (e.g. number of neighbours) of PN codes. In this section, we will argue that such an architecture is feasible and can be implemented in an energy-efficient manner if new transceiver technologies and techniques are utilised. One of the great challenges of this architecture will be implementing the transceiver with a reasonable amount of power consumption and complexity. Substantial work has been undertaken on reducing the power consumption
of relatively simple, low datarate narrowband transceivers (Otis et al., 2005), but the proposed CSMAC architecture demands additional functionality. Full-duplex transceiver operation in a sensor network environment will necessitate significant modifications to existing transceiver architectures. The typical duplexer design assumes an asymmetric FDM link. This system allows a fixed duplexer filter, and is typically implemented as a ceramic, Surface Acoustic Wave (SAW) or Bulk Acoustic Wave (BAW) filter bank. However, in the proposed system, all links are symmetric and all nodes are identical. Thus, the duplexers must be reconfigurable to accommodate an arbitrary transmit/receive frequency assignment. The main purpose of a duplexer is to provide sufficient isolation from the transmitter to the receiver. This is necessary to prevent the large transmitted carrier from leaking into the receiver and de-sensitising the front-end, corrupting the desired low-level signal. It is also needed to block the leaked transmitted spectrum from contributing directly to the weak received signal. This would be extremely difficult using standard transceiver techniques, but recent work on Micro-Electronic Mechanical System (RF MEMS) high quality factor (Q) resonators could allow such a system. For example, advances in BAW devices have allowed miniature implementations of RF duplexers (Ruby et al., 2001) for traditional mobile phone transceivers. It is possible to use individual high Q resonators to shunt energy in the transmitted RF band and attenuate its effects on the sensitive receiver. Performing high Q passive RF filtering using BAW resonators was demonstrated in Otis et al. (2004) in a low power two channel transceiver. The proposed physical layer would need a reconfigurable duplexer to protect the receiver front-end from the transmitted frequency. An adaptive duplexer design has been presented in O’Sullivan et al. (2005) for use with SAW duplexers. However, this system tunes a null within the receive band of an asymmetric link, and does not address the reconfigurability requirements of a symmetric link. A solution to this problem requires ongoing research in the fields of MEMS and transceiver system design. A reconfigurable duplexer for WSN applications constructed of passive MEMS filters would add little power dissipation to the transceiver. Additionally, since CSMAC relies on frequency-division as well as code-division, the transmitter and receiver will need separate, agile frequency synthesisers that can hop independently and simultaneously throughout the network frequency band. A dual frequency synthesiser architecture is common in full-duplex transceivers. Alternately, a dual up-conversion and down-conversion architecture could be used. This would allow a shared high frequency synthesiser and a variable transmitter Intermediate Frequency (IF) to achieve the desired frequency duplex spacing. Besides these hardware issues, idle energy consumption is also critical in sensor networks. To achieve significant energy savings, a node should power down its radio when it does not participate in data forwarding or no event occurs. A number of sleep-wakeup schemes have been proposed (Guo et al., 2001; Schurgers, 2002; Ye et al., 2002). Although CSMAC design targets high network throughput and low packet latency, we would like to emphasise that
B.H. Liu et al.
this design can also accommodate one of the sleep-wakeup schemes. The combination of the proposed architecture design with one of these schemes can achieve both high network throughput and low energy consumption.
Simulations and analysis
Simulations have been performed using the discrete event simulator ns-2. Our studies have two objectives: 1 Demonstrate the fading and shadowing effects on network performance 2 Compare the performance of different systems – CSMAC, pure CDMA and SMAC (Ye et al., 2002) – under fading and shadowing environment.
It is important to have good models for all aspects of the communication systems. In our simulation, FDMA is implemented at the wireless physical layer in ns-2. If a packet is received within the allocated frequency band for a given node, the packet will be passed to the upper layers (MAC) for additional processing with energy consumption; otherwise the packet is discarded without further energy consumption. CDMA is implemented as a PN code attribute (not a real PN code) in packet header. When a packet is received, the PN code is checked against the PN codes which are monitored by the receiver. If a match is found, the packet is passed to the next step for further processing. If no match is found, the packet is discarded. This procedure is used to simulate the demodulation process. All simulations are conducted based on a randomly generated network topology where 500 nodes are uniformly randomly deployed in a square area with 225 m × 225 m (with node density around 0.01 node/m2 ) according to a Poisson point process. A simple topology control protocol K-Neigh (Blough et al., 2003) has been adopted to limit the number of neighbours for a given node. The resulted network topology with node degree of six has been illustrated in Figure 3. This topology has been used in all the simulation studies.
The simulation parameters are shown in Table 1. The physical layer parameters such as data rate, Rx threshold, Rx power, Maximum Tx power are derived from a super regenerative transceiver especially designed for wireless sensor networks (Otis et al., 2005). Note that the architecture of this transceiver is not directly compatible with CSMAC and we only use these parameters as references in our studies. We will focus more on the comparative result rather than the absolute result. For example, we will be more interested in the relative energy consumed by CSMAC in comparison with SMAC and pure CDMA, rather than the absolute energy consumption. The carrier sense threshold is set according to the following equation: RI ≈ 2RR , where RI is the carrier sense range and RR is the communication range. We assume that the transceiver power can be adjusted with a 1 dBm step quantisation, simulating hardware power control. The data packet size, control packet size and contention window size are all taken from the SMAC ns-2 implementation. The SMAC retry limit represents the maximum number of RTS (and DATA) retries if the message is lost. One of the important parameters in CDMA simulation is the MAI threshold, which is defined as the maximum ratio between the total interference signal power and the desired signal power. The MAI threshold has been calculated based on a simple CDMA system using BPSK modulation and a convolution code with rate 1/2. We assume that the processing gain is 50 and the required Eb /N0eff is 5 dB. The MAI threshold is calculated according to Equation (11) k Pi MAI Threshold = i=1 = 23.72 = 13.75 dB (12) P0 This value represents that when the ratio between the interference signal power and the desired signal power is larger than 13.75 dB, the Forward Error Correction (FEC) will not function properly and the required Bit Error Rate (BER) cannot be achieved. The consequence is packet damage due to MAI. Unlike the MAI Threshold in a CDMA system, we use Capture Threshold in a contention-based spread spectrum environment. Capture Threshold is defined as the ratio between the intended signal power versus the interference power. P0 Capture Threshold = k i=1
Network topology used in simulation
With spread spectrum modulation, it is possible for the strongest signal to successfully capture the intended receiver, even when there are many other transmission signals. Normally, the closest transmitter is able to capture the receiver because of its small propagation path loss. This is called the near-far effect. In our simulation, we assume SMAC also employs spread spectrum modulation similar to IEEE 802.11. The capture threshold is set to 10 dB which is the default value of the CMU wireless extensions to ns-2. Note that the MAI Threshold is used in a CDMA system where multiple PN codes are employed for multiple access. Thus, multiuser detection technique is employed at the receiver side to distinguish multiple simultaneous transmissions to a same node. Capture Threshold is used in a spread spectrum system where a single PN code is employed
The impact of fading and shadowing on the network performance of wireless sensor networks Table 1
Parameters used in simulations
Data Rate Processing Gain Rx Threshold Carrier Sense Threshold Rx Power Max Tx Power Tx Power Step Data Packet Size SMAC Control Packet Size SMAC Contention Window Size SMAC Retry Limit Capture Threshold MAI Threshold Path Loss Exponent (n) Propagation Model Shadowing Deviation (σ ) Shadowing Reference Distance (d0 ) Fading Model Ricean K factor Max Velocity Antenna Gain System Loss ISM Frequency Band
5 kbps 50 −100 dBm (with BER 1 × 10−3 ) −112 dBm 400 µW 1.75 mW 1 dBm 50 Bytes 10 Bytes 63 5 10 dB 13.75 dB 4 Log-normal shadowing 4.0 dB 1m Ricean 6 (7.78 dB) 2.5 m/s 1 l 2.4–2.4835 GHz
by all nodes. Spread spectrum is used to achieve better channel performance in jamming and multipath environment, but not to provide multiple access. In our simulation, the small scale fading envelope is used to modulate the results of a large scale propagation model. The log-normal shadowing is used as the large scale model to reflect the shadowing effect. The Ricean model proposed in Punnoose et al. (2000) is used as the small scale model to reflect the fading effect. A dataset containing the components of a time-sequenced fading envelope is precomputed in the Ricean model. The Max Velocity represents the maximum velocity of the movement of transceiver, receiver or other object (e.g. human body) in the environment. This parameter is used to calculate the Doppler frequency which is then used to calculate the time index in the dataset for the correspondent fading envelope. Other radio propagation parameters such as path loss exponent, shadowing deviation, reference distance, Ricean K factor, have been explained in Section 2.
Simulation results and analysis
This section compares the one hop performance of three different systems; CSMAC, pure CDMA and SMAC (Ye et al., 2002); with and without fading and shadowing effects. To simulate non-fading and non-shadowing environment, we set the shadowing deviation (σ ) to 0 and Ricean K factor to a large value. Our discussions will mainly focus on the network performance with fading and shadowing environment and take the non-fading and non-shadowing scenario as reference. Note that SMAC is CSMA/CA-based, where each node needs to contend for the media by using a DCF with contention windows and
back-off its transmission. On the contrary, CSMAC and CDMA can transmit a packet immediately without using DCF. In our simulation, we did not enable the synchronisation flag in SMAC ns-2 implementation so there is no synchronisation overhead for SMAC. We also assume infinite buffer space so there is no packet drop due to buffer overflow. The idle energy consumption is not included in this study since it only adds a constant mean value to the results and it may hide the effect of the communication energy consumption. The performance is measured based on different packet generation rates, which scales from the lowest 0.01 packet/s to the highest 10 packet/s. Three parameters are measured: ‘packet delivery ratio, one hop latency and communication energy consumption’. These performance parameters are measured based on 10 runs. In each run, each node transmits 100 packets and each packet is transmitted to a randomly selected neighbour. Each packet’s departure time is randomised to approximate a Poisson process. A total of 50,000 packets (500 nodes × 100 packets/node) are generated in each run. The simulation start time is also randomised according to the packet transmission interval. This is to ensure that each node can start sending at similar time but not exactly the same time which may have significant influence on the performance of SMAC. In our simulation, we assume that each node can adjust its Tx power for different neighbours. If the Tx power for a given neighbour is set too high, the transmission will cause larger interference to other neighbours. Moreover, higher power results in higher energy consumption. Conversely, if the Tx power is set too low, the transmission may be lost due to the weak signal strength at the receiver. In our simulation, the
B.H. Liu et al.
power level is calculated and set accordingly at the transmitter by using the log-normal shadowing model dependent on the distance between the transmitter and receiver. A value for the inverse Q-function corresponds to 95% of receiving-rate (see ns-2 manual for details) has been used in the Tx power calculation. In other words, the Tx power is set with 95% of confidence that the signal can be correctly received by the receiver when the shadowing is considered (note that small-scale fading is not considered). In practice, the power level may be set to a reasonable value through negotiation between the transmitter and receiver.
Packet delivery ratio
Figure 4 illustrates the average packet delivery ratio comparison with and without fading and shadowing. The delivery ratio is measured as the ratio between the number of packets that have been successfully received and the total number of packets that have been transmitted. Note the degraded performance when fading and shadowing are enabled in comparison with the non-fading and non-shadowing scenario. Both CSMAC and CDMA suffered from nearly 10% degradation in the delivery ratio. Moreover, the Tx power is set higher in the fading and shadowing environment compared to the non-fading and non-shadowing environment. But we see that higher Tx power does not guarantee the reliability of packet transmissions. Figure 4
Average packet delivery ratio comparison Average packet delivery ratio comparison
Average packet delivery ratio
0.6 CSMAC with fading and shadowing CDMA with fading and shadowing SMAC with fading and shadowing CSMAC without fading and shadowing CDMA without fading and shadowing SMAC without fading and shadowing
Packet generation rate (packet/s)
Another interesting observation is that SMAC performs better than CSMAC and CDMA when the traffic is very low (e.g. less than 0.05 packet/s). This behaviour benefits from SMAC’s acknowledgement and retransmission scheme. SMAC has five-time retries for RTS (or DATA) transmission if the sender cannot receive a CTS (or ACK) from the receiver. SMAC also uses DCF, RTS/CTS to reduce the collision probability. In contrast, CSMAC and CDMA do not have these schemes. The consequence is that a packet loss (due to fading, shadowing and MAI) cannot be recovered for CSMAC and CDMA. This prompts us for several approaches to address the reliability issue for CSMAC and CDMA. The first approach is to employ an acknowledgement scheme. Either positive or negative acknowledgement can be used. But
acknowledgement will increase both the energy consumption and packet latency, which violate our original design goal. The second approach is to increase the Tx power. But each node may have a limited maximum power and it is also difficult to determine the correct power level to combat fading and shadowing since the environment is always changing. Higher power not only increases the energy consumption, but also causes more interference to neighbours. The third approach is to reduce the data rate adaptively when the channel quality is low (and increase the data rate when the channel quality is high). This requires that each node has the capability to estimate the instantaneous channel quality and adjust its data rate dynamically, which may increase the hardware complexity. The fourth approach is to leave the reliability issue to upper layer. Since sensor networks are normally deployed with redundancy, the upper layer application such as a data fusion algorithms may not require 100% delivery ratio. To achieve certain level of confidence on the distributed detection and estimation, such an algorithm may tolerate a certain level of packet loss. This approach requires the codesign of data fusion algorithms and wireless sensor networking. The design an antifading, antishadowing protocol and comparison of different approaches remain a subject of future work. Although SMAC achieves better performance than CSMAC and CDMA when the traffic is very low, its delivery ratio drops below CSMAC and CDMA when the traffic reaches 0.05 packet/s. With our parameter settings, 0.05 packet/s correspondents to a 0.85% duty cycle on average. Note that we have randomised the departure time of each packet in our simulation. This randomisation makes our simulation somewhat different from the reality in sensor networks – where simultaneous events may cause a communication ‘hot region’ but the average duty cycle is still less than 1%. In this scenario, the traffic may be much higher than 0.05 packet/s when the simultaneous events occur. Figure 4 reveals that the SMAC delivery ratio saturates at around 74% when the traffic increases. While this seems to be a reasonable value, we will see later that the one hop latency of SMAC in fact reaches an unacceptable high level. The continuous degradation of a CDMA system demonstrates the influence of MAI as well as fading and shadowing. In a higher traffic scenario, many concurrent transmissions can easily exceed the MAI threshold. The consequence is that more and more packets are damaged due to MAI. Note that CDMA fails over SMAC when the traffic is over 1 packet/s. This degradation trend saturates at 39% when the system reaches the full transmission capacity (e.g. 5.91 packet/s). In contrast, we observe that MAI has almost no influence on CSMAC due to its multichannel design. CSMAC can always maintain its delivery ratio at around 90% even when the system reaches full transmission capacity. Packet loss is mainly caused by fading and shadowing effects rather than MAI.
One hop latency
Figure 5 illustrates the average one hop packet latency comparison. Each value in the figure is calculated based on all packets that have been successfully received. The latency
The impact of fading and shadowing on the network performance of wireless sensor networks
Average one hop latency comparison Average one hop latency comparison
Average one hop latency (s)
CSMAC with fading and shadowing CDMA with fading and shadowing SMAC with fading and shadowing CSMAC without fading and shadowing CDMA without fading and shadowing SMAC without fading and shadowing
10 10 Packet generation rate (packet/s)
It is expected that CSMAC and CDMA outperforms SMAC significantly since CSMAC and CDMA do not use carrier sense, RTS/CTS, DCF, etc. For CSMAC and CDMA, a node can start transmitting immediately when a packet is received from the upper layer. Figure 5 shows that the latency of CSMAC and CDMA is almost constant and is not influenced by the effects of fading and shadowing and the increases of traffic load. The latency of CSMAC and CDMA increases sharply when the packet generation rate reaches the full transmission capacity where the queueing delay becomes dominant. On the contrary, the latency of SMAC increases gradually with the increases of traffic load. We also note that the effects of fading and shadowing have significant influence on SMAC. It can be seen that at 0.05 packet/s, the one hop latency of SMAC with fading and shadowing is 15 times higher than the latency without fading and shadowing! The reason behind this behaviour can be explained as follows. Since SMAC uses RTS-CTS-DATA-ACK handshake for each data packet transmission, any control or data packet loss during this handshake due to fading and/or shadowing will initiate an extra retransmission (until the retry limit is exceeded). For example, if the CTS or ACK is lost, the sender must retransmit the RTS or DATA packet. This will not only increase the latency for this data packet, but also the queueing delays for other packets in the buffer. Moreover, all neighbours of the sender and receiver must wait until this handshake finishes before they can initiate their own transmissions. Figure 5 shows that the SMAC latency is well above 10 sec at 0.05 packet/s and even above 100 sec at 0.06 packet/s! This one-hop latency is unacceptable for many practical applications.
Figure 6 illustrates the average communication energy consumption with and without fading and shadowing. We
note that the energy consumption has almost no relation with traffic load for CSMAC and CDMA since the number of packets generated in each run is the same (50,000). Traffic load has a small impact on SMAC and the energy consumption increases with the increase of packet generation rate. This trend saturates when the packet generation rate reaches 0.05 packet/s (with fading and shadowing). Figure 6 also shows that CSMAC consumes the least energy and SMAC consumes the most, while CDMA consumes between CSMAC and SMAC. Several reasons cause SMAC to consume more energy. Firstly, a control packet (RTS/CTS/ACK) exchange consumes large amount of energy. With our parameter settings in Table 1, the control packet contributes 37.5% overhead. Secondly, overhearing consumes energy. In fact, SMAC has an overhearing avoidance scheme to turn-off a node’s transceiver when it overhears an RTS or CTS that is not destined to itself. However, the problem is that the interference range of a transmission is much larger than the normal transmission range, for example, RI ≈ 2RR . A node may detect an RTS or CTS but cannot correctly receive it. The result is that overhearing avoidance scheme does not function for those nodes located between RR and RI . Thirdly, packet collision wastes energy. Although RTS/CTS are used, packet collision cannot be fully avoided (including collisions of RTS/CTS). Collided packets need retransmission and consume extra energy. Finally, packet loss due to fading and shadowing induces significant retransmissions and hence, more energy is consumed. CDMA consumes more energy than CSMAC because of the overhearing. Since all nodes are operating within the same frequency channel, the overhearing in CDMA is more severe than in CSMAC. Figure 6
Average communication energy consumption comparison Average energy consumption per node comparison
CSMAC with fading and shadowing CDMA with fading and shadowing SMAC with fading and shadowing CSMAC without fading and shadowing CDMA without fading and shadowing SMAC without fading and shadowing
0.3 Average energy consumption per node (J)
is measured between the packet generation time and the time when this packet is received by the intended receiver. This means that the packet queueing time in the buffer is also counted.
0 −2 10
Packet generation rate (packet/s)
Figure 6 shows that CSMAC consumes less than 10% communication energy than SMAC. In our simulation, the Tx and Rx power are set to the same value for both CSMAC and SMAC. But a multiuser detection receiver may consume slightly more power than a single user detection receiver. What we wish to show is the comparative result. Unless a multiuser detection receiver consumes 10 times more power than a single user detection receiver, it is possible to use multichannel to achieve energy savings.
B.H. Liu et al.
An interesting observation is the influence of fading and shadowing on different systems. Figure 6 reveals that fading and shadowing cause a 57% increase in energy consumption for CSMAC, 103% for CDMA and 142% for SMAC. The main reason for this trend is that the transmission power is calculated based on the log-normal shadowing model. When shadowing is assumed, the Tx power is set to a higher level to combat the shadowing (and fading) effect. The increased energy consumption for CSMAC is mainly due to the increased Tx power. Besides the increased Tx power, overhearing is another major contribution for CDMA. Since the Tx power is increased, the interference range is also increased. The result is that more neighbours overhear the signal and more energy is consumed. Thus, CDMA shows a higher level of influence by fading and shadowing than CSMAC. For SMAC, besides all above discussed reasons, fading and shadowing cause significant retransmissions and hence SMAC shows the highest level of influence by fading and shadowing.
Concluding remarks and future works
In this paper, we studied the fading and shadowing effects on the network performance for wireless sensor networks. Our simulation results have shown that fading and shadowing can have significant impact on the network performance. We compared three different systems: a multichannel CDMA system (CSMAC), a pure CDMA system and a contentionbased system (SMAC). Our studies have shown that CSMAC outperforms the pure CDMA system as well as SMAC. One problem of the CSMAC is the reliability issue in a fading and shadowing environment. This demonstrates from one aspect that fading and shadowing must be considered seriously in any system or protocol design for wireless sensor networks. In addition to the comparison of different reliability approaches that have been discussed in Section 5.2, there are a number of challenges stimulated by the proposed CSMAC system which require ongoing research. For example, comparison of different topology control protocols on the performance of CSMAC is an interesting problem. There are enormous topology control protocols proposed in literature for ad hoc and sensor networks and current CSMAC only adopts a simple topology control protocol K-Neigh (Blough et al., 2003). Furthermore, we believe that K-Neigh may not be a suitable choice for topology control. A simple observation of Figure 3 has shown that this topology may not provide efficient support for geographic routing protocols. For example, nodes A and B (see Figure 3) are in fact very close to each other. But they are not neighbours because of the employment of K-Neigh. So they cannot communicate with each other at MAC layer. Assume that A has a packet needs to be send to B, this packet has to be routed around with multiple hops in order to reach node B. This is very inefficient and significant network resources (energy, bandwidth, etc.) have been wasted. One can easily image that how the network performance can be improved if a link exists between node A and node B. We believe that the joint development of MAC (CSMAC), routing and topology control in order to achieve optimised network performance would be a challenging topic and requires extensive research effort.
Our next step is to prototype the proposed CSMAC system and further evaluate the network performance based on real hardware platform. Successful development and evaluation of the CSMAC system based on hardware platform will further trigger potential commercialisation effort.
Acknowledgement This research has been supported by the ARC Linkage Grant LP0561200 and ADI Limited, Australia (now Thales Australia).
References Bettstetter, C. (2004) ‘Failure-resilient ad hoc and sensor networks in a shadow fading environment’, Proceedings of IEEE/IFIP International Conference on Dependable Systems and Networks (DSN), DIWANS, Florence, Italy, June. Blough, D.M., Leoncini, M., Resta, G. and Santi, P. (2003) ‘The K-Neigh protocol for symmetric topology control in ad hoc networks’, Proceedings of IEEE MobiHoc, Annapolis, Maryland, USA, pp.141–152. Chen, S.H., Mitra, U. and Krishnamachari, B. (2005) ‘Cooperative communication and routing over fading channels in wireless sensor networks’, WirelessCom, 2005. Guo, C., Zhong, L.C. and Rabaey, J.M. (2001) ‘Low power distributed MAC for ad hoc sensor radio networks’, Proceedings of IEEE GlobeCom 2001, San Antonio, 25–29 November 2001, Vol. 5, pp.2944–2948. Han, S.Y. and Abu-Ghazaleh, N.B. (2005) ‘On the effect of fading on ad-hoc networks’, Available at: http://wwwlib.umi.com/cr/binghamton/fullcit?.p.1422359. Liu, B.H., Bulusu, N., Pham, H. and Jha, S. (2004) ‘CSMAC: a novel CDMA based MAC protocol for wireless sensor networks’, IEEE GLOBECOM Wireless Ad hoc and Sensor Networks, 29 November – 3 December 2004, Dallas Texas, USA, pp.33–38. Liu, B.H., Chou, C.T., Lipman, J. and Jha, S. (2005a) ‘Using frequency division to reduce MAI in CDMA wireless sensor networks’, IEEE Wireless Communications and Networking Conference (WCNC), 13–17 March, New Orleans, LA, pp.657–663. Liu, B.H., Chou, C.T. and Jha, S. (2005b) ‘A multi-channel DS-CDMA media access control protocol for wireless sensor networks’, Technical Report, UNSW-CSE-TR-0503. Liu, X.W. and Haenggi, M. (2005) ‘Throughput analysis of fading sensor networks with regular and random topologies’, EURASIP Journal on Wireless Communications and Networking, Vol. 4, pp.554–564. Mullen, J. and Huang, H. (2005) ‘Impact of multipath fading in wireless ad hoc networks’, The 2nd ACM International Workshop on Performance Evaluation of Wireless Ad Hoc, Sensor, and Ubiquitous Networks (PE-WASUN 2005), 10–13 October, Montreal, Canada. O’Sullivan, T., York, R., Noren, B. and Asbeck, P. (2005) ‘Active duplexer implemented using single-path and multipath feedforward techniques with BST phase shifters’, IEEE Transactions on Microwave Theory and Techniques, Vol. 53, No. 1. Otis, B.P., Chee, Y.H., Lu, R., Pletcher, N.M. and Rabaey, J.M. (2004) ‘An ultra-low power MEMS-based two-channel transceiver for wireless sensor networks’, Symposium on VLSI Circuits, Honolulu, HI, June.
The impact of fading and shadowing on the network performance of wireless sensor networks Otis, B.P., Chee, Y.H. and Rabaey, J.M. (2005) ‘A 400µW-RX, 1.6mW-TX Super-Regenerative Transceiver for Wireless Sensor Networks’, Proceedings of 2005 IEEE International Solid-State Circuits Conference (ISSCC 2005), 9 February, 2005. Pottie, G. and Kaiser, W. (2005) Principles of Embedded Networked Systems Design, Cambridge University Press. Punnoose, R.J., Nikitin, P.V. and Stancil, D.D. (2000) ‘Efficient simulation of ricean fading within a packet simulator’, Proceedings of IEEE VTS-Fall Vehicular Technology Conference, VTC 2000, 52nd, Vol. 2, pp.764–767. Pursley, M.B. (1977) ‘Performance evaluation for phase-coded spread-spectrum multiple-access communications – Part I: system analysis’, IEEE Transactions on Communication, Vol. COM-25, August, pp.795–799. Rappaport, T.S. (2002) Wireless Communications, Principles and Practice, 2nd Edition, Prentice Hall. Ruby, R., Bradley, P., Larson, J., Oshmyansky, Y. and Figueredo, D. (2001) ‘Ultra-miniature High-Q filters and duplexers using FBAR technology’, Proceedings of IEEE International Solid-State Circuits Conference (ISSCC), February, pp.120–121. Schurgers, C. (2002) ‘Optimizing sensor networks in the energy-latency-density design space’, IEEE Transactions on Mobile Computing, Vol. 1, No. 1, Janaury–March, pp.70–80.
Souryal, M.R. and Moayeri, N. (2005) ‘Channel-adaptive relaying in mobile ad hoc networks with fading’, Proceedings IEEE SECON 2005. Stankovic, J.A., Abdelzaher, T.F., Lu C.Y., Sha L. and Hou, J.C. (2003) ‘Real-time communication and coordination in embedded sensor networks’, Proceedings of the IEEE, Vol. 91, No. 7, July, pp.1002–1022. Woo, A. and Culler, D. (2001) ‘A transmission control scheme for media access in sensor networks,’ Proceedings of ACM MobiCom, July 16–21, Rome, Italy, pp.221–35. Ye, W., Heidemann, J. and Estrin, D. (2002) ‘An energy-efficient MAC protocol for wireless sensor networks’, IEEE Proceedings of Infocom, June, pp.1567–1576.
This require a channel allocation protocol during the network formation phase. Please refer our technical report (Liu et al., 2005b) for details. 2 Since CSMAC does not use control message exchange (e.g. RTS/CTS), full duplex mode is required to allow Tx and Rx happen simultaneously in two channels.